ops: add MUSA

Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com>

.clang-format

@@ -22,14 +22,7 @@ AllowShortIfStatementsOnASingleLine: Never
 AllowShortLambdasOnASingleLine: Inline
 AllowShortLoopsOnASingleLine: false
 AlwaysBreakBeforeMultilineStrings: true
-# Treat CUDA keywords/attributes as "attribute macros" and avoid breaking lines inside them
-AttributeMacros:
-  - __host__
-  - __device__
-  - __global__
-  - __forceinline__
-  - __launch_bounds__
-BinPackArguments: true
+BinPackArguments: false
 BinPackParameters: false # OnePerLine
 BitFieldColonSpacing: Both
 BreakBeforeBraces: Custom # Attach

.clang-tidy

@@ -17,7 +17,6 @@ Checks: >
     clang-analyzer-*,
     -clang-analyzer-security.insecureAPI.DeprecatedOrUnsafeBufferHandling,
     performance-*,
-    -performance-enum-size,
     portability-*,
     -portability-simd-intrinsics,
     misc-*,

.devops/cann.Dockerfile

@@ -1,129 +0,0 @@
-# ==============================================================================
-# ARGUMENTS
-# ==============================================================================
-
-# Define the CANN base image for easier version updates later
-ARG CHIP_TYPE=910b
-ARG CANN_BASE_IMAGE=quay.io/ascend/cann:8.3.rc2-${CHIP_TYPE}-openeuler24.03-py3.11
-
-# ==============================================================================
-# BUILD STAGE
-# Compile all binary files and libraries
-# ==============================================================================
-FROM ${CANN_BASE_IMAGE} AS build
-
-# -- Install build dependencies --
-RUN yum install -y gcc g++ cmake make git libcurl-devel python3 python3-pip && \
-    yum clean all && \
-    rm -rf /var/cache/yum
-
-# -- Set the working directory --
-WORKDIR /app
-
-# -- Copy project files --
-COPY . .
-
-# -- Set CANN environment variables (required for compilation) --
-# Using ENV instead of `source` allows environment variables to persist across the entire image layer
-ENV ASCEND_TOOLKIT_HOME=/usr/local/Ascend/ascend-toolkit/latest
-ENV LD_LIBRARY_PATH=${ASCEND_TOOLKIT_HOME}/lib64:${LD_LIBRARY_PATH}
-ENV PATH=${ASCEND_TOOLKIT_HOME}/bin:${PATH}
-ENV ASCEND_OPP_PATH=${ASCEND_TOOLKIT_HOME}/opp
-ENV LD_LIBRARY_PATH=${ASCEND_TOOLKIT_HOME}/runtime/lib64/stub:$LD_LIBRARY_PATH
-# ... You can add other environment variables from the original file as needed ...
-# For brevity, only core variables are listed here. You can paste the original ENV list here.
-
-# -- Build llama.cpp --
-# Use the passed CHIP_TYPE argument and add general build options
-ARG CHIP_TYPE
-RUN source /usr/local/Ascend/ascend-toolkit/set_env.sh --force \
-    && \
-    cmake -B build \
-        -DGGML_CANN=ON \
-        -DCMAKE_BUILD_TYPE=Release \
-        -DSOC_TYPE=ascend${CHIP_TYPE} \
-        . && \
-    cmake --build build --config Release -j$(nproc)
-
-# -- Organize build artifacts for copying in later stages --
-# Create a lib directory to store all .so files
-RUN mkdir -p /app/lib && \
-    find build -name "*.so*" -exec cp -P {} /app/lib \;
-
-# Create a full directory to store all executables and Python scripts
-RUN mkdir -p /app/full && \
-    cp build/bin/* /app/full/ && \
-    cp *.py /app/full/ && \
-    cp -r gguf-py /app/full/ && \
-    cp -r requirements /app/full/ && \
-    cp requirements.txt /app/full/
-    # If you have a tools.sh script, make sure it is copied here
-    # cp .devops/tools.sh /app/full/tools.sh
-
-# ==============================================================================
-# BASE STAGE
-# Create a minimal base image with CANN runtime and common libraries
-# ==============================================================================
-FROM ${CANN_BASE_IMAGE} AS base
-
-# -- Install runtime dependencies --
-RUN yum install -y libgomp curl && \
-    yum clean all && \
-    rm -rf /var/cache/yum
-
-# -- Set CANN environment variables (required for runtime) --
-ENV ASCEND_TOOLKIT_HOME=/usr/local/Ascend/ascend-toolkit/latest
-ENV LD_LIBRARY_PATH=/app:${ASCEND_TOOLKIT_HOME}/lib64:${LD_LIBRARY_PATH}
-ENV PATH=${ASCEND_TOOLKIT_HOME}/bin:${PATH}
-ENV ASCEND_OPP_PATH=${ASCEND_TOOLKIT_HOME}/opp
-# ... You can add other environment variables from the original file as needed ...
-
-WORKDIR /app
-
-# Copy compiled .so files from the build stage
-COPY --from=build /app/lib/ /app
-
-# ==============================================================================
-# FINAL STAGES (TARGETS)
-# ==============================================================================
-
-### Target: full
-# Complete image with all tools, Python bindings, and dependencies
-# ==============================================================================
-FROM base AS full
-
-COPY --from=build /app/full /app
-
-# Install Python dependencies
-RUN yum install -y git python3 python3-pip && \
-    pip3 install --no-cache-dir --upgrade pip setuptools wheel && \
-    pip3 install --no-cache-dir -r requirements.txt && \
-    yum clean all && \
-    rm -rf /var/cache/yum
-
-# You need to provide a tools.sh script as the entrypoint
-ENTRYPOINT ["/app/tools.sh"]
-# If there is no tools.sh, you can set the default to start the server
-# ENTRYPOINT ["/app/llama-server"]
-
-### Target: light
-# Lightweight image containing only llama-cli and llama-completion
-# ==============================================================================
-FROM base AS light
-
-COPY --from=build /app/full/llama-cli /app/full/llama-completion /app
-
-ENTRYPOINT [ "/app/llama-cli" ]
-
-### Target: server
-# Dedicated server image containing only llama-server
-# ==============================================================================
-FROM base AS server
-
-ENV LLAMA_ARG_HOST=0.0.0.0
-
-COPY --from=build /app/full/llama-server /app
-
-HEALTHCHECK --interval=5m CMD [ "curl", "-f", "http://localhost:8080/health" ]
-
-ENTRYPOINT [ "/app/llama-server" ]

.devops/cloud-v-pipeline

@@ -0,0 +1,22 @@
+node('x86_runner1'){            // Running on x86 runner containing latest vector qemu, latest vector gcc and all the necessary libraries
+    stage('Cleanup'){
+        cleanWs()               // Cleaning previous CI build in workspace
+    }
+    stage('checkout repo'){
+        retry(5){               // Retry if the cloning fails due to some reason
+            checkout scm        // Clone the repo on Runner
+        }
+    }
+    stage('Compiling llama.cpp'){
+        sh'''#!/bin/bash
+            make RISCV=1 RISCV_CROSS_COMPILE=1 # Compiling llama for RISC-V
+        '''
+    }
+    stage('Running llama.cpp'){
+        sh'''#!/bin/bash
+            module load gnu-bin2/0.1            # loading latest versions of vector qemu and vector gcc
+            qemu-riscv64 -L /softwares/gnu-bin2/sysroot  -cpu rv64,v=true,vlen=256,elen=64,vext_spec=v1.0 ./llama-cli -m /home/alitariq/codellama-7b.Q4_K_M.gguf -p "Anything" -n 9 > llama_log.txt            # Running llama.cpp on vector qemu-riscv64
+            cat llama_log.txt                   # Printing results
+        '''
+    }
+}

.devops/cpu.Dockerfile

@@ -4,6 +4,8 @@ FROM ubuntu:$UBUNTU_VERSION AS build
 
 ARG TARGETARCH
 
+ARG GGML_CPU_ARM_ARCH=armv8-a
+
 RUN apt-get update && \
     apt-get install -y build-essential git cmake libcurl4-openssl-dev
 
@@ -11,8 +13,10 @@ WORKDIR /app
 
 COPY . .
 
-RUN if [ "$TARGETARCH" = "amd64" ] || [ "$TARGETARCH" = "arm64" ]; then \
+RUN if [ "$TARGETARCH" = "amd64" ]; then \
         cmake -S . -B build -DCMAKE_BUILD_TYPE=Release -DGGML_NATIVE=OFF -DLLAMA_BUILD_TESTS=OFF -DGGML_BACKEND_DL=ON -DGGML_CPU_ALL_VARIANTS=ON; \
+    elif [ "$TARGETARCH" = "arm64" ]; then \
+        cmake -S . -B build -DCMAKE_BUILD_TYPE=Release -DGGML_NATIVE=OFF -DLLAMA_BUILD_TESTS=OFF -DGGML_CPU_ARM_ARCH=${GGML_CPU_ARM_ARCH}; \
     else \
         echo "Unsupported architecture"; \
         exit 1; \
@@ -20,7 +24,7 @@ RUN if [ "$TARGETARCH" = "amd64" ] || [ "$TARGETARCH" = "arm64" ]; then \
     cmake --build build -j $(nproc)
 
 RUN mkdir -p /app/lib && \
-    find build -name "*.so*" -exec cp -P {} /app/lib \;
+    find build -name "*.so" -exec cp {} /app/lib \;
 
 RUN mkdir -p /app/full \
     && cp build/bin/* /app/full \
@@ -68,7 +72,7 @@ ENTRYPOINT ["/app/tools.sh"]
 ### Light, CLI only
 FROM base AS light
 
-COPY --from=build /app/full/llama-cli /app/full/llama-completion /app
+COPY --from=build /app/full/llama-cli /app
 
 WORKDIR /app
 

.devops/cuda.Dockerfile

@@ -25,7 +25,7 @@ RUN if [ "${CUDA_DOCKER_ARCH}" != "default" ]; then \
     cmake --build build --config Release -j$(nproc)
 
 RUN mkdir -p /app/lib && \
-    find build -name "*.so*" -exec cp -P {} /app/lib \;
+    find build -name "*.so" -exec cp {} /app/lib \;
 
 RUN mkdir -p /app/full \
     && cp build/bin/* /app/full \
@@ -61,7 +61,7 @@ RUN apt-get update \
     python3 \
     python3-pip \
     && pip install --upgrade pip setuptools wheel \
-    && pip install --break-system-packages -r requirements.txt \
+    && pip install -r requirements.txt \
     && apt autoremove -y \
     && apt clean -y \
     && rm -rf /tmp/* /var/tmp/* \
@@ -74,7 +74,7 @@ ENTRYPOINT ["/app/tools.sh"]
 ### Light, CLI only
 FROM base AS light
 
-COPY --from=build /app/full/llama-cli /app/full/llama-completion /app
+COPY --from=build /app/full/llama-cli /app
 
 WORKDIR /app
 

.devops/intel.Dockerfile

@@ -1,8 +1,8 @@
-ARG ONEAPI_VERSION=2025.2.2-0-devel-ubuntu24.04
+ARG ONEAPI_VERSION=2025.1.1-0-devel-ubuntu24.04
 
 ## Build Image
 
-FROM intel/deep-learning-essentials:$ONEAPI_VERSION AS build
+FROM intel/oneapi-basekit:$ONEAPI_VERSION AS build
 
 ARG GGML_SYCL_F16=OFF
 RUN apt-get update && \
@@ -21,7 +21,7 @@ RUN if [ "${GGML_SYCL_F16}" = "ON" ]; then \
     cmake --build build --config Release -j$(nproc)
 
 RUN mkdir -p /app/lib && \
-    find build -name "*.so*" -exec cp -P {} /app/lib \;
+    find build -name "*.so" -exec cp {} /app/lib \;
 
 RUN mkdir -p /app/full \
     && cp build/bin/* /app/full \
@@ -31,7 +31,7 @@ RUN mkdir -p /app/full \
     && cp requirements.txt /app/full \
     && cp .devops/tools.sh /app/full/tools.sh
 
-FROM intel/deep-learning-essentials:$ONEAPI_VERSION AS base
+FROM intel/oneapi-basekit:$ONEAPI_VERSION AS base
 
 RUN apt-get update \
     && apt-get install -y libgomp1 curl\
@@ -73,7 +73,7 @@ ENTRYPOINT ["/app/tools.sh"]
 FROM base AS light
 
 COPY --from=build /app/lib/ /app
-COPY --from=build /app/full/llama-cli /app/full/llama-completion /app
+COPY --from=build /app/full/llama-cli /app
 
 WORKDIR /app
 

.devops/llama-cli-cann.Dockerfile

@@ -23,12 +23,11 @@ ENV LD_LIBRARY_PATH=${ASCEND_TOOLKIT_HOME}/runtime/lib64/stub:$LD_LIBRARY_PATH
 RUN echo "Building with static libs" && \
     source /usr/local/Ascend/ascend-toolkit/set_env.sh --force && \
     cmake -B build -DGGML_NATIVE=OFF -DGGML_CANN=ON -DBUILD_SHARED_LIBS=OFF -DLLAMA_BUILD_TESTS=OFF  && \
-    cmake --build build --config Release --target llama-cli && \
-    cmake --build build --config Release --target llama-completion
+    cmake --build build --config Release --target llama-cli
 
 # TODO: use image with NNRT
 FROM ascendai/cann:$ASCEND_VERSION AS runtime
-COPY --from=build /app/build/bin/llama-cli /app/build/bin/llama-completion /
+COPY --from=build /app/build/bin/llama-cli /llama-cli
 
 ENV LC_ALL=C.utf8
 

.devops/llama-cpp-cuda.srpm.spec

@@ -37,7 +37,6 @@ make -j GGML_CUDA=1
 %install
 mkdir -p %{buildroot}%{_bindir}/
 cp -p llama-cli %{buildroot}%{_bindir}/llama-cuda-cli
-cp -p llama-completion %{buildroot}%{_bindir}/llama-cuda-completion
 cp -p llama-server %{buildroot}%{_bindir}/llama-cuda-server
 cp -p llama-simple %{buildroot}%{_bindir}/llama-cuda-simple
 
@@ -69,7 +68,6 @@ rm -rf %{_builddir}/*
 
 %files
 %{_bindir}/llama-cuda-cli
-%{_bindir}/llama-cuda-completion
 %{_bindir}/llama-cuda-server
 %{_bindir}/llama-cuda-simple
 /usr/lib/systemd/system/llamacuda.service

.devops/llama-cpp.srpm.spec

@@ -39,7 +39,6 @@ make -j
 %install
 mkdir -p %{buildroot}%{_bindir}/
 cp -p llama-cli %{buildroot}%{_bindir}/llama-cli
-cp -p llama-completion %{buildroot}%{_bindir}/llama-completion
 cp -p llama-server %{buildroot}%{_bindir}/llama-server
 cp -p llama-simple %{buildroot}%{_bindir}/llama-simple
 
@@ -71,7 +70,6 @@ rm -rf %{_builddir}/*
 
 %files
 %{_bindir}/llama-cli
-%{_bindir}/llama-completion
 %{_bindir}/llama-server
 %{_bindir}/llama-simple
 /usr/lib/systemd/system/llama.service

.devops/musa.Dockerfile

@@ -1,6 +1,6 @@
 ARG UBUNTU_VERSION=22.04
 # This needs to generally match the container host's environment.
-ARG MUSA_VERSION=rc4.3.0
+ARG MUSA_VERSION=rc4.2.0
 # Target the MUSA build image
 ARG BASE_MUSA_DEV_CONTAINER=mthreads/musa:${MUSA_VERSION}-devel-ubuntu${UBUNTU_VERSION}-amd64
 
@@ -32,7 +32,7 @@ RUN if [ "${MUSA_DOCKER_ARCH}" != "default" ]; then \
     cmake --build build --config Release -j$(nproc)
 
 RUN mkdir -p /app/lib && \
-    find build -name "*.so*" -exec cp -P {} /app/lib \;
+    find build -name "*.so" -exec cp {} /app/lib \;
 
 RUN mkdir -p /app/full \
     && cp build/bin/* /app/full \
@@ -81,7 +81,7 @@ ENTRYPOINT ["/app/tools.sh"]
 ### Light, CLI only
 FROM base AS light
 
-COPY --from=build /app/full/llama-cli /app/full/llama-completion /app
+COPY --from=build /app/full/llama-cli /app
 
 WORKDIR /app
 

.devops/nix/package.nix

@@ -34,7 +34,6 @@
   rocmGpuTargets ? builtins.concatStringsSep ";" rocmPackages.clr.gpuTargets,
   enableCurl ? true,
   useVulkan ? false,
-  useRpc ? false,
   llamaVersion ? "0.0.0", # Arbitrary version, substituted by the flake
 
   # It's necessary to consistently use backendStdenv when building with CUDA support,
@@ -129,6 +128,10 @@ effectiveStdenv.mkDerivation (finalAttrs: {
   };
 
   postPatch = ''
+    substituteInPlace ./ggml/src/ggml-metal/ggml-metal.m \
+      --replace '[bundle pathForResource:@"ggml-metal" ofType:@"metal"];' "@\"$out/bin/ggml-metal.metal\";"
+    substituteInPlace ./ggml/src/ggml-metal/ggml-metal.m \
+      --replace '[bundle pathForResource:@"default" ofType:@"metallib"];' "@\"$out/bin/default.metallib\";"
   '';
 
   # With PR#6015 https://github.com/ggml-org/llama.cpp/pull/6015,
@@ -176,7 +179,6 @@ effectiveStdenv.mkDerivation (finalAttrs: {
       (cmakeBool "GGML_METAL" useMetalKit)
       (cmakeBool "GGML_VULKAN" useVulkan)
       (cmakeBool "GGML_STATIC" enableStatic)
-      (cmakeBool "GGML_RPC" useRpc)
     ]
     ++ optionals useCuda [
       (

.devops/rocm.Dockerfile

@@ -1,10 +1,10 @@
 ARG UBUNTU_VERSION=24.04
 
 # This needs to generally match the container host's environment.
-ARG ROCM_VERSION=7.0
-ARG AMDGPU_VERSION=7.0
+ARG ROCM_VERSION=6.4
+ARG AMDGPU_VERSION=6.4
 
-# Target the ROCm build image
+# Target the CUDA build image
 ARG BASE_ROCM_DEV_CONTAINER=rocm/dev-ubuntu-${UBUNTU_VERSION}:${ROCM_VERSION}-complete
 
 ### Build image
@@ -13,14 +13,18 @@ FROM ${BASE_ROCM_DEV_CONTAINER} AS build
 # Unless otherwise specified, we make a fat build.
 # List from https://github.com/ggml-org/llama.cpp/pull/1087#issuecomment-1682807878
 # This is mostly tied to rocBLAS supported archs.
-# gfx803, gfx900, gfx906, gfx1032, gfx1101, gfx1102,not officialy supported
-# check https://rocm.docs.amd.com/projects/install-on-linux/en/docs-6.4.1/reference/system-requirements.html
+# gfx803, gfx900, gfx1032, gfx1101, gfx1102,not officialy supported
+# gfx906 is deprecated
+#check https://rocm.docs.amd.com/projects/install-on-linux/en/docs-6.2.4/reference/system-requirements.html
 
-ARG ROCM_DOCKER_ARCH='gfx803;gfx900;gfx906;gfx908;gfx90a;gfx942;gfx1010;gfx1030;gfx1032;gfx1100;gfx1101;gfx1102;gfx1200;gfx1201;gfx1151'
-#ARG ROCM_DOCKER_ARCH='gfx1151'
+ARG ROCM_DOCKER_ARCH='gfx803,gfx900,gfx906,gfx908,gfx90a,gfx942,gfx1010,gfx1030,gfx1032,gfx1100,gfx1101,gfx1102'
+#ARG ROCM_DOCKER_ARCH=gfx1100
 
-# Set ROCm architectures
+# Set nvcc architectured
 ENV AMDGPU_TARGETS=${ROCM_DOCKER_ARCH}
+# Enable ROCm
+# ENV CC=/opt/rocm/llvm/bin/clang
+# ENV CXX=/opt/rocm/llvm/bin/clang++
 
 RUN apt-get update \
     && apt-get install -y \
@@ -36,16 +40,11 @@ WORKDIR /app
 COPY . .
 
 RUN HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" \
-    cmake -S . -B build \
-        -DGGML_HIP=ON \
-        -DGGML_HIP_ROCWMMA_FATTN=ON \
-        -DAMDGPU_TARGETS="$ROCM_DOCKER_ARCH" \
-        -DGGML_BACKEND_DL=ON -DGGML_CPU_ALL_VARIANTS=ON \
-        -DCMAKE_BUILD_TYPE=Release -DLLAMA_BUILD_TESTS=OFF \
+    cmake -S . -B build -DGGML_HIP=ON -DAMDGPU_TARGETS=$ROCM_DOCKER_ARCH -DGGML_BACKEND_DL=ON -DGGML_CPU_ALL_VARIANTS=ON -DCMAKE_BUILD_TYPE=Release -DLLAMA_BUILD_TESTS=OFF \
     && cmake --build build --config Release -j$(nproc)
 
 RUN mkdir -p /app/lib \
-    && find build -name "*.so*" -exec cp -P {} /app/lib \;
+    && find build -name "*.so" -exec cp {} /app/lib \;
 
 RUN mkdir -p /app/full \
     && cp build/bin/* /app/full \
@@ -94,7 +93,7 @@ ENTRYPOINT ["/app/tools.sh"]
 ### Light, CLI only
 FROM base AS light
 
-COPY --from=build /app/full/llama-cli /app/full/llama-completion /app
+COPY --from=build /app/full/llama-cli /app
 
 WORKDIR /app
 

.devops/s390x.Dockerfile

@@ -1,126 +0,0 @@
-ARG GCC_VERSION=15.2.0
-ARG UBUNTU_VERSION=24.04
-
-### Build Llama.cpp stage
-FROM gcc:${GCC_VERSION} AS build
-
-RUN --mount=type=cache,target=/var/cache/apt,sharing=locked \
-    --mount=type=cache,target=/var/lib/apt/lists,sharing=locked \
-    apt update -y && \
-    apt upgrade -y && \
-    apt install -y --no-install-recommends \
-        git cmake ccache ninja-build \
-        # WARNING: Do not use libopenblas-openmp-dev. libopenblas-dev is faster.
-        libopenblas-dev libcurl4-openssl-dev && \
-    rm -rf /var/lib/apt/lists/*
-
-WORKDIR /app
-COPY . .
-
-RUN --mount=type=cache,target=/root/.ccache \
-    --mount=type=cache,target=/app/build \
-    cmake -S . -B build -G Ninja \
-        -DCMAKE_BUILD_TYPE=Release \
-        -DCMAKE_C_COMPILER_LAUNCHER=ccache \
-        -DCMAKE_CXX_COMPILER_LAUNCHER=ccache \
-        -DLLAMA_BUILD_TESTS=OFF \
-        -DGGML_NATIVE=OFF \
-        -DGGML_BACKEND_DL=ON \
-        -DGGML_CPU_ALL_VARIANTS=ON \
-        -DGGML_BLAS=ON \
-        -DGGML_BLAS_VENDOR=OpenBLAS && \
-    cmake --build build --config Release -j $(nproc) && \
-    cmake --install build --prefix /opt/llama.cpp
-
-COPY *.py             /opt/llama.cpp/bin
-COPY .devops/tools.sh /opt/llama.cpp/bin
-
-COPY gguf-py          /opt/llama.cpp/gguf-py
-COPY requirements.txt /opt/llama.cpp/gguf-py
-COPY requirements     /opt/llama.cpp/gguf-py/requirements
-
-
-### Collect all llama.cpp binaries, libraries and distro libraries
-FROM scratch AS collector
-
-# Copy llama.cpp binaries and libraries
-COPY --from=build /opt/llama.cpp/bin     /llama.cpp/bin
-COPY --from=build /opt/llama.cpp/lib     /llama.cpp/lib
-COPY --from=build /opt/llama.cpp/gguf-py /llama.cpp/gguf-py
-
-
-### Base image
-FROM ubuntu:${UBUNTU_VERSION} AS base
-
-RUN --mount=type=cache,target=/var/cache/apt,sharing=locked \
-    --mount=type=cache,target=/var/lib/apt/lists,sharing=locked \
-    apt update -y && \
-    apt install -y --no-install-recommends \
-        # WARNING: Do not use libopenblas-openmp-dev. libopenblas-dev is faster.
-        # See: https://github.com/ggml-org/llama.cpp/pull/15915#issuecomment-3317166506
-        curl libgomp1 libopenblas-dev && \
-    apt autoremove -y && \
-    apt clean -y && \
-    rm -rf /tmp/* /var/tmp/* && \
-    find /var/cache/apt/archives /var/lib/apt/lists -not -name lock -type f -delete && \
-    find /var/cache -type f -delete
-
-# Copy llama.cpp libraries
-COPY --from=collector /llama.cpp/lib /usr/lib/s390x-linux-gnu
-
-
-### Full
-FROM base AS full
-
-ENV PATH="/root/.cargo/bin:${PATH}"
-WORKDIR /app
-
-RUN --mount=type=cache,target=/var/cache/apt,sharing=locked \
-    --mount=type=cache,target=/var/lib/apt/lists,sharing=locked \
-    apt update -y && \
-    apt install -y \
-        git cmake libjpeg-dev \
-        python3 python3-pip python3-dev && \
-    apt autoremove -y && \
-    apt clean -y && \
-    rm -rf /tmp/* /var/tmp/* && \
-    find /var/cache/apt/archives /var/lib/apt/lists -not -name lock -type f -delete && \
-    find /var/cache -type f -delete
-
-RUN curl https://sh.rustup.rs -sSf | bash -s -- -y
-
-COPY --from=collector /llama.cpp/bin /app
-COPY --from=collector /llama.cpp/gguf-py /app/gguf-py
-
-RUN pip install --no-cache-dir --break-system-packages \
-        -r /app/gguf-py/requirements.txt
-
-ENTRYPOINT [ "/app/tools.sh" ]
-
-
-### CLI Only
-FROM base AS light
-
-WORKDIR /llama.cpp/bin
-
-# Copy llama.cpp binaries and libraries
-COPY --from=collector /llama.cpp/bin/*.so /llama.cpp/bin
-COPY --from=collector /llama.cpp/bin/llama-cli /llama.cpp/bin/llama-completion /llama.cpp/bin
-
-ENTRYPOINT [ "/llama.cpp/bin/llama-cli" ]
-
-
-### Server
-FROM base AS server
-
-ENV LLAMA_ARG_HOST=0.0.0.0
-
-WORKDIR /llama.cpp/bin
-
-# Copy llama.cpp binaries and libraries
-COPY --from=collector /llama.cpp/bin/*.so /llama.cpp/bin
-COPY --from=collector /llama.cpp/bin/llama-server /llama.cpp/bin
-
-EXPOSE 8080
-
-ENTRYPOINT [ "/llama.cpp/bin/llama-server" ]

.devops/tools.sh

@@ -13,8 +13,6 @@ elif [[ "$arg1" == '--quantize' || "$arg1" == '-q' ]]; then
     exec ./llama-quantize "$@"
 elif [[ "$arg1" == '--run' || "$arg1" == '-r' ]]; then
     exec ./llama-cli "$@"
-elif [[ "$arg1" == '--run-legacy' || "$arg1" == '-l' ]]; then
-    exec ./llama-completion "$@"
 elif [[ "$arg1" == '--bench' || "$arg1" == '-b' ]]; then
     exec ./llama-bench "$@"
 elif [[ "$arg1" == '--perplexity' || "$arg1" == '-p' ]]; then
@@ -34,10 +32,8 @@ elif [[ "$arg1" == '--server' || "$arg1" == '-s' ]]; then
 else
     echo "Unknown command: $arg1"
     echo "Available commands: "
-    echo "  --run (-r): Run a model (chat) previously converted into ggml"
-    echo "              ex: -m /models/7B/ggml-model-q4_0.bin"
-    echo "  --run-legacy (-l): Run a model (legacy completion) previously converted into ggml"
-    echo "              ex: -m /models/7B/ggml-model-q4_0.bin -no-cnv -p \"Building a website can be done in 10 simple steps:\" -n 512"
+    echo "  --run (-r): Run a model previously converted into ggml"
+    echo "              ex: -m /models/7B/ggml-model-q4_0.bin -p \"Building a website can be done in 10 simple steps:\" -n 512"
     echo "  --bench (-b): Benchmark the performance of the inference for various parameters."
     echo "              ex: -m model.gguf"
     echo "  --perplexity (-p): Measure the perplexity of a model over a given text."

.devops/vulkan.Dockerfile

@@ -1,24 +1,26 @@
-ARG UBUNTU_VERSION=26.04
+ARG UBUNTU_VERSION=24.04
 
 FROM ubuntu:$UBUNTU_VERSION AS build
 
 # Install build tools
-RUN apt update && apt install -y git build-essential cmake wget xz-utils
+RUN apt update && apt install -y git build-essential cmake wget
 
-# Install cURL and Vulkan SDK dependencies
-RUN apt install -y libcurl4-openssl-dev curl \
-    libxcb-xinput0 libxcb-xinerama0 libxcb-cursor-dev libvulkan-dev glslc
+# Install Vulkan SDK and cURL
+RUN wget -qO - https://packages.lunarg.com/lunarg-signing-key-pub.asc | apt-key add - && \
+    wget -qO /etc/apt/sources.list.d/lunarg-vulkan-noble.list https://packages.lunarg.com/vulkan/lunarg-vulkan-noble.list && \
+    apt update -y && \
+    apt-get install -y vulkan-sdk libcurl4-openssl-dev curl
 
 # Build it
 WORKDIR /app
 
 COPY . .
 
-RUN cmake -B build -DGGML_NATIVE=OFF -DGGML_VULKAN=ON -DLLAMA_BUILD_TESTS=OFF -DGGML_BACKEND_DL=ON -DGGML_CPU_ALL_VARIANTS=ON && \
+RUN cmake -B build -DGGML_NATIVE=OFF -DGGML_VULKAN=1  -DLLAMA_BUILD_TESTS=OFF -DGGML_BACKEND_DL=ON -DGGML_CPU_ALL_VARIANTS=ON && \
     cmake --build build --config Release -j$(nproc)
 
 RUN mkdir -p /app/lib && \
-    find build -name "*.so*" -exec cp -P {} /app/lib \;
+    find build -name "*.so" -exec cp {} /app/lib \;
 
 RUN mkdir -p /app/full \
     && cp build/bin/* /app/full \
@@ -32,7 +34,7 @@ RUN mkdir -p /app/full \
 FROM ubuntu:$UBUNTU_VERSION AS base
 
 RUN apt-get update \
-    && apt-get install -y libgomp1 curl libvulkan1 mesa-vulkan-drivers \
+    && apt-get install -y libgomp1 curl libvulkan-dev \
     && apt autoremove -y \
     && apt clean -y \
     && rm -rf /tmp/* /var/tmp/* \
@@ -50,7 +52,6 @@ WORKDIR /app
 
 RUN apt-get update \
     && apt-get install -y \
-    build-essential \
     git \
     python3 \
     python3-pip \
@@ -68,7 +69,7 @@ ENTRYPOINT ["/app/tools.sh"]
 ### Light, CLI only
 FROM base AS light
 
-COPY --from=build /app/full/llama-cli /app/full/llama-completion /app
+COPY --from=build /app/full/llama-cli /app
 
 WORKDIR /app
 

.editorconfig

@@ -52,19 +52,3 @@ insert_final_newline = unset
 [vendor/miniaudio/miniaudio.h]
 trim_trailing_whitespace = unset
 insert_final_newline = unset
-
-[tools/server/webui/**]
-indent_style = unset
-indent_size = unset
-end_of_line = unset
-charset = unset
-trim_trailing_whitespace = unset
-insert_final_newline = unset
-
-[benches/**]
-indent_style = unset
-indent_size = unset
-end_of_line = unset
-charset = unset
-trim_trailing_whitespace = unset
-insert_final_newline = unset

.github/ISSUE_TEMPLATE/010-bug-compilation.yml

@@ -40,7 +40,7 @@ body:
     attributes:
         label: GGML backends
         description: Which GGML backends do you know to be affected?
-        options: [AMX, BLAS, CPU, CUDA, HIP, Metal, Musa, RPC, SYCL, Vulkan, OpenCL, zDNN]
+        options: [AMX, BLAS, CPU, CUDA, HIP, Metal, Musa, RPC, SYCL, Vulkan, OpenCL]
         multiple: true
     validations:
       required: true

.github/ISSUE_TEMPLATE/011-bug-results.yml

@@ -11,7 +11,7 @@ body:
         (i.e. the generated text) are incorrect or llama.cpp crashes during model evaluation.
         If you encountered the issue while using an external UI (e.g. ollama),
         please reproduce your issue using one of the examples/binaries in this repository.
-        The `llama-completion` binary can be used for simple and reproducible model inference.
+        The `llama-cli` binary can be used for simple and reproducible model inference.
   - type: textarea
     id: version
     attributes:
@@ -42,7 +42,7 @@ body:
     attributes:
         label: GGML backends
         description: Which GGML backends do you know to be affected?
-        options: [AMX, BLAS, CPU, CUDA, HIP, Metal, Musa, RPC, SYCL, Vulkan, OpenCL, zDNN]
+        options: [AMX, BLAS, CPU, CUDA, HIP, Metal, Musa, RPC, SYCL, Vulkan, OpenCL]
         multiple: true
     validations:
       required: true
@@ -74,12 +74,9 @@ body:
         Please give us a summary of the problem and tell us how to reproduce it.
         If you can narrow down the bug to specific hardware, compile flags, or command line arguments,
         that information would be very much appreciated by us.
-
-        If possible, please try to reproduce the issue using `llama-completion` with `-fit off`.
-        If you can only reproduce the issue with `-fit on`, please provide logs both with and without `--verbose`.
       placeholder: >
-        e.g. when I run llama-completion with `-fa on` I get garbled outputs for very long prompts.
-        With short prompts or `-fa off` it works correctly.
+        e.g. when I run llama-cli with -ngl 99 I get garbled outputs.
+        When I use -ngl 0 it works correctly.
         Here are the exact commands that I used: ...
     validations:
       required: true

.github/ISSUE_TEMPLATE/019-bug-misc.yml

@@ -86,7 +86,6 @@ body:
       description: >
           If applicable, please copy and paste any relevant log output, including any generated text.
           This will be automatically formatted into code, so no need for backticks.
-          If you are encountering problems specifically with the `llama_params_fit` module, always upload `--verbose` logs as well.
       render: shell
     validations:
       required: false

.github/actions/install-exe/action.yml

@@ -1,36 +0,0 @@
-name: "Install exe"
-description: "Download and install exe"
-inputs:
-  url:
-    description: "URL of the exe installer"
-    required: true
-  args:
-    description: "Installer arguments"
-    required: true
-  timeout:
-    description: "Timeout (in ms)"
-    required: false
-    default: "600000"
-
-runs:
-  using: "composite"
-  steps:
-    - name: Install EXE
-      shell: pwsh
-      run: |
-        $ErrorActionPreference = "Stop"
-        write-host "Downloading Installer EXE"
-        Invoke-WebRequest -Uri "${{ inputs.url }}" -OutFile "${env:RUNNER_TEMP}\temp-install.exe"
-        write-host "Installing"
-        $proc = Start-Process "${env:RUNNER_TEMP}\temp-install.exe" -ArgumentList '${{ inputs.args }}' -NoNewWindow -PassThru
-        $completed = $proc.WaitForExit(${{ inputs.timeout }})
-        if (-not $completed) {
-            Write-Error "Installer timed out. Killing the process"
-            $proc.Kill()
-            exit 1
-        }
-        if ($proc.ExitCode -ne 0) {
-            Write-Error "Installer failed with exit code $($proc.ExitCode)"
-            exit 1
-        }
-        write-host "Completed installation"

.github/actions/linux-setup-spacemit/action.yml

@@ -1,20 +0,0 @@
-name: "Linux - Setup SpacemiT Toolchain"
-description: "Setup SpacemiT Toolchain for Linux"
-inputs:
-  path:
-    description: "Installation path"
-    required: true
-  version:
-    description: "SpacemiT toolchain version"
-    required: true
-
-runs:
-  using: "composite"
-  steps:
-    - name: Setup SpacemiT Toolchain
-      id: setup
-      uses: ./.github/actions/unarchive-tar
-      with:
-        url: https://archive.spacemit.com/toolchain/spacemit-toolchain-linux-glibc-x86_64-v${{ inputs.version }}.tar.xz
-        path: ${{ inputs.path }}
-        strip: 1

.github/actions/linux-setup-vulkan/action.yml

@@ -1,20 +0,0 @@
-name: "Linux - Setup Vulkan SDK"
-description: "Setup Vulkan SDK for Linux"
-inputs:
-  path:
-    description: "Installation path"
-    required: true
-  version:
-    description: "Vulkan SDK version"
-    required: true
-
-runs:
-  using: "composite"
-  steps:
-    - name: Setup Vulkan SDK
-      id: setup
-      uses: ./.github/actions/unarchive-tar
-      with:
-        url: https://sdk.lunarg.com/sdk/download/${{ inputs.version }}/linux/vulkan_sdk.tar.xz
-        path: ${{ inputs.path }}
-        strip: 1

.github/actions/unarchive-tar/action.yml

@@ -1,27 +0,0 @@
-name: "Unarchive tar"
-description: "Download and unarchive tar into directory"
-inputs:
-  url:
-    description: "URL of the tar archive"
-    required: true
-  path:
-    description: "Directory to unarchive into"
-    required: true
-  type:
-    description: "Compression type (tar option)"
-    required: false
-    default: "J"
-  strip:
-    description: "Strip components"
-    required: false
-    default: "0"
-
-runs:
-  using: "composite"
-  steps:
-    - name: Unarchive into directory
-      shell: bash
-      run: |
-        mkdir -p ${{ inputs.path }}
-        cd ${{ inputs.path }}
-        curl --no-progress-meter ${{ inputs.url }} | tar -${{ inputs.type }}x --strip-components=${{ inputs.strip }}

.github/actions/windows-setup-cuda/action.yml

@@ -65,34 +65,3 @@ runs:
           echo "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.4\libnvvp" | Out-File -FilePath $env:GITHUB_PATH -Encoding utf8 -Append
           echo "CUDA_PATH=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.4" | Out-File -FilePath $env:GITHUB_ENV -Append -Encoding utf8
           echo "CUDA_PATH_V12_4=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.4" | Out-File -FilePath $env:GITHUB_ENV -Append -Encoding utf8
-
-    - name: Install Cuda Toolkit 13.1
-      if: ${{ inputs.cuda_version == '13.1' }}
-      shell: pwsh
-      run: |
-          mkdir -p "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.1"
-          choco install unzip -y
-          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/cuda_crt/windows-x86_64/cuda_crt-windows-x86_64-13.1.80-archive.zip"
-          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/cuda_cudart/windows-x86_64/cuda_cudart-windows-x86_64-13.1.80-archive.zip"
-          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/cuda_nvcc/windows-x86_64/cuda_nvcc-windows-x86_64-13.1.80-archive.zip"
-          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/cuda_nvrtc/windows-x86_64/cuda_nvrtc-windows-x86_64-13.1.80-archive.zip"
-          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/libcublas/windows-x86_64/libcublas-windows-x86_64-13.2.0.9-archive.zip"
-          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/libnvvm/windows-x86_64/libnvvm-windows-x86_64-13.1.80-archive.zip"
-          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/cuda_nvtx/windows-x86_64/cuda_nvtx-windows-x86_64-13.1.68-archive.zip"
-          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/cuda_profiler_api/windows-x86_64/cuda_profiler_api-windows-x86_64-13.1.80-archive.zip"
-          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/visual_studio_integration/windows-x86_64/visual_studio_integration-windows-x86_64-13.1.68-archive.zip"
-          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/cuda_cccl/windows-x86_64/cuda_cccl-windows-x86_64-13.1.78-archive.zip"
-          unzip '*.zip' -d "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.1"
-          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.1\cuda_crt-windows-x86_64-13.1.80-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.1" /E /I /H /Y
-          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.1\cuda_cudart-windows-x86_64-13.1.80-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.1" /E /I /H /Y
-          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.1\cuda_nvcc-windows-x86_64-13.1.80-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.1" /E /I /H /Y
-          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.1\cuda_nvrtc-windows-x86_64-13.1.80-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.1" /E /I /H /Y
-          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.1\libcublas-windows-x86_64-13.2.0.9-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.1" /E /I /H /Y
-          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.1\libnvvm-windows-x86_64-13.1.80-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.1" /E /I /H /Y
-          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.1\cuda_nvtx-windows-x86_64-13.1.68-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.1" /E /I /H /Y
-          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.1\cuda_profiler_api-windows-x86_64-13.1.80-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.1" /E /I /H /Y
-          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.1\visual_studio_integration-windows-x86_64-13.1.68-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.1" /E /I /H /Y
-          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.1\cuda_cccl-windows-x86_64-13.1.78-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.1" /E /I /H /Y
-          echo "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.1\bin" | Out-File -FilePath $env:GITHUB_PATH -Encoding utf8 -Append
-          echo "CUDA_PATH=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.1" | Out-File -FilePath $env:GITHUB_ENV -Append -Encoding utf8
-          echo "CUDA_PATH_V13_1=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.1" | Out-File -FilePath $env:GITHUB_ENV -Append -Encoding utf8

.github/actions/windows-setup-rocm/action.yml

@@ -1,15 +0,0 @@
-name: "Windows - Setup ROCm"
-description: "Setup ROCm for Windows"
-inputs:
-  version:
-    description: "ROCm version"
-    required: true
-
-runs:
-  using: "composite"
-  steps:
-    - name: Setup ROCm
-      uses: ./.github/actions/install-exe
-      with:
-        url: https://download.amd.com/developer/eula/rocm-hub/AMD-Software-PRO-Edition-${{ inputs.version }}-WinSvr2022-For-HIP.exe
-        args: -install

.github/labeler.yml

@@ -22,11 +22,6 @@ Vulkan:
         - any-glob-to-any-file:
             - ggml/include/ggml-vulkan.h
             - ggml/src/ggml-vulkan/**
-IBM zDNN:
-    - changed-files:
-        - any-glob-to-any-file:
-            - ggml/include/ggml-zdnn.h
-            - ggml/src/ggml-zdnn/**
 documentation:
     - changed-files:
         - any-glob-to-any-file:
@@ -76,10 +71,6 @@ ggml:
     - changed-files:
         - any-glob-to-any-file:
             - ggml/**
-model:
-    - changed-files:
-        - any-glob-to-any-file:
-            - src/models/**
 nix:
     - changed-files:
         - any-glob-to-any-file:

.github/workflows/build-cache.yml

@@ -1,89 +0,0 @@
-name: Build Actions Cache
-
-on:
-  workflow_dispatch: # allows manual triggering
-  schedule:
-    - cron: '0 * * * *'
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
-  cancel-in-progress: true
-
-jobs:
-  ubuntu-24-vulkan-cache:
-    runs-on: ubuntu-24.04
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v4
-
-      - name: Get latest Vulkan SDK version
-        id: vulkan_sdk_version
-        run: |
-          echo "VULKAN_SDK_VERSION=$(curl https://vulkan.lunarg.com/sdk/latest/linux.txt)" >> "$GITHUB_ENV"
-
-      - name: Setup Cache
-        uses: actions/cache@v4
-        id: cache-sdk
-        with:
-          path: ./vulkan_sdk
-          key: vulkan-sdk-${{ env.VULKAN_SDK_VERSION }}-${{ runner.os }}
-
-      - name: Setup Vulkan SDK
-        if: steps.cache-sdk.outputs.cache-hit != 'true'
-        uses: ./.github/actions/linux-setup-vulkan
-        with:
-          path: ./vulkan_sdk
-          version: ${{ env.VULKAN_SDK_VERSION }}
-
-  ubuntu-24-spacemit-cache:
-    runs-on: ubuntu-24.04
-
-    env:
-      # Make sure this is in sync with build-linux-cross.yml
-      SPACEMIT_IME_TOOLCHAIN_VERSION: "1.1.2"
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v4
-
-      - name: Setup Cache
-        uses: actions/cache@v4
-        id: cache-toolchain
-        with:
-          path: ./spacemit_toolchain
-          key: spacemit-ime-toolchain-v${{ env.SPACEMIT_IME_TOOLCHAIN_VERSION }}-${{ runner.os }}
-
-      - name: Setup SpacemiT Toolchain
-        if: steps.cache-toolchain.outputs.cache-hit != 'true'
-        uses: ./.github/actions/linux-setup-spacemit
-        with:
-          path: ./spacemit_toolchain
-          version: ${{ env.SPACEMIT_IME_TOOLCHAIN_VERSION }}
-
-  windows-2022-rocm-cache:
-    runs-on: windows-2022
-
-    env:
-      # Make sure this is in sync with build.yml
-      HIPSDK_INSTALLER_VERSION: "25.Q3"
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v4
-
-      - name: Setup Cache
-        uses: actions/cache@v4
-        id: cache-rocm
-        with:
-          path: C:\Program Files\AMD\ROCm
-          key: rocm-${{ env.HIPSDK_INSTALLER_VERSION }}-${{ runner.os }}
-
-      - name: Setup ROCm
-        if: steps.cache-rocm.outputs.cache-hit != 'true'
-        uses: ./.github/actions/windows-setup-rocm
-        with:
-          version: ${{ env.HIPSDK_INSTALLER_VERSION }}

.github/workflows/build-linux-cross.yml

@@ -4,49 +4,49 @@ on:
   workflow_call:
 
 jobs:
-  # ubuntu-24-riscv64-cpu-cross:
-  #   runs-on: ubuntu-24.04
+  ubuntu-24-riscv64-cpu-cross:
+    runs-on: ubuntu-24.04
 
-  #   steps:
-  #     - uses: actions/checkout@v4
-  #     - name: Setup Riscv
-  #       run: |
-  #         sudo dpkg --add-architecture riscv64
+    steps:
+      - uses: actions/checkout@v4
+      - name: Setup Riscv
+        run: |
+          sudo dpkg --add-architecture riscv64
 
-  #         # Add arch-specific repositories for non-amd64 architectures
-  #         cat << EOF | sudo tee /etc/apt/sources.list.d/riscv64-ports.list
-  #         deb [arch=riscv64] http://ports.ubuntu.com/ubuntu-ports/ noble main universe
-  #         deb [arch=riscv64] http://ports.ubuntu.com/ubuntu-ports/ noble-updates main universe
-  #         deb [arch=riscv64] http://ports.ubuntu.com/ubuntu-ports/ noble-security main universe
-  #         deb [arch=riscv64] http://ports.ubuntu.com/ubuntu-ports/ noble-backports main universe
-  #         EOF
+          # Add arch-specific repositories for non-amd64 architectures
+          cat << EOF | sudo tee /etc/apt/sources.list.d/riscv64-ports.list
+          deb [arch=riscv64] http://ports.ubuntu.com/ubuntu-ports/ noble main universe
+          deb [arch=riscv64] http://ports.ubuntu.com/ubuntu-ports/ noble-updates main universe
+          deb [arch=riscv64] http://ports.ubuntu.com/ubuntu-ports/ noble-security main universe
+          deb [arch=riscv64] http://ports.ubuntu.com/ubuntu-ports/ noble-backports main universe
+          EOF
 
-  #         sudo apt-get update || true    ;# Prevent failure due to missing URLs.
+          sudo apt-get update || true    ;# Prevent failure due to missing URLs.
 
-  #         sudo apt-get install -y --no-install-recommends \
-  #                 build-essential \
-  #                 gcc-14-riscv64-linux-gnu \
-  #                 g++-14-riscv64-linux-gnu
+          sudo apt-get install -y --no-install-recommends \
+                  build-essential \
+                  gcc-14-riscv64-linux-gnu \
+                  g++-14-riscv64-linux-gnu
 
-  #     - name: Build
-  #       run: |
-  #         cmake -B build -DLLAMA_CURL=OFF \
-  #                        -DCMAKE_BUILD_TYPE=Release \
-  #                        -DGGML_OPENMP=OFF \
-  #                        -DLLAMA_BUILD_EXAMPLES=ON \
-  #                        -DLLAMA_BUILD_TOOLS=ON \
-  #                        -DLLAMA_BUILD_TESTS=OFF \
-  #                        -DCMAKE_SYSTEM_NAME=Linux \
-  #                        -DCMAKE_SYSTEM_PROCESSOR=riscv64 \
-  #                        -DCMAKE_C_COMPILER=riscv64-linux-gnu-gcc-14 \
-  #                        -DCMAKE_CXX_COMPILER=riscv64-linux-gnu-g++-14 \
-  #                        -DCMAKE_POSITION_INDEPENDENT_CODE=ON \
-  #                        -DCMAKE_FIND_ROOT_PATH=/usr/lib/riscv64-linux-gnu \
-  #                        -DCMAKE_FIND_ROOT_PATH_MODE_PROGRAM=NEVER \
-  #                        -DCMAKE_FIND_ROOT_PATH_MODE_LIBRARY=ONLY \
-  #                        -DCMAKE_FIND_ROOT_PATH_MODE_INCLUDE=BOTH
+      - name: Build
+        run: |
+          cmake -B build -DLLAMA_CURL=OFF \
+                         -DCMAKE_BUILD_TYPE=Release \
+                         -DGGML_OPENMP=OFF \
+                         -DLLAMA_BUILD_EXAMPLES=ON \
+                         -DLLAMA_BUILD_TOOLS=ON \
+                         -DLLAMA_BUILD_TESTS=OFF \
+                         -DCMAKE_SYSTEM_NAME=Linux \
+                         -DCMAKE_SYSTEM_PROCESSOR=riscv64 \
+                         -DCMAKE_C_COMPILER=riscv64-linux-gnu-gcc-14 \
+                         -DCMAKE_CXX_COMPILER=riscv64-linux-gnu-g++-14 \
+                         -DCMAKE_POSITION_INDEPENDENT_CODE=ON \
+                         -DCMAKE_FIND_ROOT_PATH=/usr/lib/riscv64-linux-gnu \
+                         -DCMAKE_FIND_ROOT_PATH_MODE_PROGRAM=NEVER \
+                         -DCMAKE_FIND_ROOT_PATH_MODE_LIBRARY=ONLY \
+                         -DCMAKE_FIND_ROOT_PATH_MODE_INCLUDE=BOTH
 
-  #         cmake --build build --config Release -j $(nproc)
+          cmake --build build --config Release -j $(nproc)
 
   # ubuntu-24-riscv64-vulkan-cross:
   #   runs-on: ubuntu-24.04
@@ -141,6 +141,97 @@ jobs:
 
   #         cmake --build build --config Release -j $(nproc)
 
+  ubuntu-24-ppc64el-cpu-cross:
+    runs-on: ubuntu-24.04
+
+    steps:
+      - uses: actions/checkout@v4
+      - name: Setup PowerPC64le
+        run: |
+          sudo dpkg --add-architecture ppc64el
+
+          # Add arch-specific repositories for non-amd64 architectures
+          cat << EOF | sudo tee /etc/apt/sources.list.d/ppc64el-ports.list
+          deb [arch=ppc64el] http://ports.ubuntu.com/ubuntu-ports/ noble main universe
+          deb [arch=ppc64el] http://ports.ubuntu.com/ubuntu-ports/ noble-updates main universe
+          deb [arch=ppc64el] http://ports.ubuntu.com/ubuntu-ports/ noble-security main universe
+          deb [arch=ppc64el] http://ports.ubuntu.com/ubuntu-ports/ noble-backports main universe
+          EOF
+
+          sudo apt-get update || true    ;# Prevent failure due to missing URLs.
+
+          sudo apt-get install -y --no-install-recommends \
+                  build-essential \
+                  gcc-14-powerpc64le-linux-gnu \
+                  g++-14-powerpc64le-linux-gnu
+
+      - name: Build
+        run: |
+          cmake -B build -DLLAMA_CURL=OFF \
+                         -DCMAKE_BUILD_TYPE=Release \
+                         -DGGML_OPENMP=OFF \
+                         -DLLAMA_BUILD_EXAMPLES=ON \
+                         -DLLAMA_BUILD_TOOLS=ON \
+                         -DLLAMA_BUILD_TESTS=OFF \
+                         -DCMAKE_SYSTEM_NAME=Linux \
+                         -DCMAKE_SYSTEM_PROCESSOR=ppc64 \
+                         -DCMAKE_C_COMPILER=powerpc64le-linux-gnu-gcc-14 \
+                         -DCMAKE_CXX_COMPILER=powerpc64le-linux-gnu-g++-14 \
+                         -DCMAKE_POSITION_INDEPENDENT_CODE=ON \
+                         -DCMAKE_FIND_ROOT_PATH=/usr/lib/powerpc64le-linux-gnu \
+                         -DCMAKE_FIND_ROOT_PATH_MODE_PROGRAM=NEVER \
+                         -DCMAKE_FIND_ROOT_PATH_MODE_LIBRARY=ONLY \
+                         -DCMAKE_FIND_ROOT_PATH_MODE_INCLUDE=BOTH
+
+          cmake --build build --config Release -j $(nproc)
+
+  # ubuntu-24-ppc64el-vulkan-cross:
+  #   runs-on: ubuntu-24.04
+
+  #   steps:
+  #     - uses: actions/checkout@v4
+  #     - name: Setup PowerPC64le
+  #       run: |
+  #         sudo dpkg --add-architecture ppc64el
+
+  #         # Add arch-specific repositories for non-amd64 architectures
+  #         cat << EOF | sudo tee /etc/apt/sources.list.d/ppc64el-ports.list
+  #         deb [arch=ppc64el] http://ports.ubuntu.com/ubuntu-ports/ noble main universe
+  #         deb [arch=ppc64el] http://ports.ubuntu.com/ubuntu-ports/ noble-updates main universe
+  #         deb [arch=ppc64el] http://ports.ubuntu.com/ubuntu-ports/ noble-security main universe
+  #         deb [arch=ppc64el] http://ports.ubuntu.com/ubuntu-ports/ noble-backports main universe
+  #         EOF
+
+  #         sudo apt-get update || true    ;# Prevent failure due to missing URLs.
+
+  #         sudo apt-get install -y --no-install-recommends \
+  #                 build-essential \
+  #                 glslc \
+  #                 gcc-14-powerpc64le-linux-gnu \
+  #                 g++-14-powerpc64le-linux-gnu \
+  #                 libvulkan-dev:ppc64el
+
+  #     - name: Build
+  #       run: |
+  #         cmake -B build -DLLAMA_CURL=OFF \
+  #                        -DCMAKE_BUILD_TYPE=Release \
+  #                        -DGGML_VULKAN=ON \
+  #                        -DGGML_OPENMP=OFF \
+  #                        -DLLAMA_BUILD_EXAMPLES=ON \
+  #                        -DLLAMA_BUILD_TOOLS=ON \
+  #                        -DLLAMA_BUILD_TESTS=OFF \
+  #                        -DCMAKE_SYSTEM_NAME=Linux \
+  #                        -DCMAKE_SYSTEM_PROCESSOR=ppc64 \
+  #                        -DCMAKE_C_COMPILER=powerpc64le-linux-gnu-gcc-14 \
+  #                        -DCMAKE_CXX_COMPILER=powerpc64le-linux-gnu-g++-14 \
+  #                        -DCMAKE_POSITION_INDEPENDENT_CODE=ON \
+  #                        -DCMAKE_FIND_ROOT_PATH=/usr/lib/powerpc64le-linux-gnu \
+  #                        -DCMAKE_FIND_ROOT_PATH_MODE_PROGRAM=NEVER \
+  #                        -DCMAKE_FIND_ROOT_PATH_MODE_LIBRARY=ONLY \
+  #                        -DCMAKE_FIND_ROOT_PATH_MODE_INCLUDE=BOTH
+
+  #         cmake --build build --config Release -j $(nproc)
+
   debian-13-loongarch64-cpu-cross:
     runs-on: ubuntu-24.04
     container: debian@sha256:653dfb9f86c3782e8369d5f7d29bb8faba1f4bff9025db46e807fa4c22903671
@@ -253,46 +344,3 @@ jobs:
                          -DCMAKE_FIND_ROOT_PATH_MODE_INCLUDE=BOTH
 
           cmake --build build --config Release -j $(nproc)
-
-  ubuntu-24-riscv64-cpu-spacemit-ime-cross:
-    runs-on: ubuntu-24.04
-
-    env:
-      # Make sure this is in sync with build-cache.yml
-      SPACEMIT_IME_TOOLCHAIN_VERSION: "1.1.2"
-
-    steps:
-      - uses: actions/checkout@v4
-
-      - name: Use SpacemiT Toolchain Cache
-        uses: actions/cache@v4
-        id: cache-toolchain
-        with:
-          path: ./spacemit_toolchain
-          key: spacemit-ime-toolchain-v${{ env.SPACEMIT_IME_TOOLCHAIN_VERSION }}-${{ runner.os }}
-
-      - name: Setup SpacemiT Toolchain
-        if: steps.cache-toolchain.outputs.cache-hit != 'true'
-        uses: ./.github/actions/linux-setup-spacemit
-        with:
-          path: ./spacemit_toolchain
-          version: ${{ env.SPACEMIT_IME_TOOLCHAIN_VERSION }}
-
-      - name: Build
-        run: |
-          export RISCV_ROOT_PATH=${PWD}/spacemit_toolchain
-          cmake -B build -DLLAMA_CURL=OFF \
-                         -DCMAKE_BUILD_TYPE=Release \
-                         -DGGML_OPENMP=OFF \
-                         -DLLAMA_BUILD_EXAMPLES=ON \
-                         -DLLAMA_BUILD_TOOLS=ON \
-                         -DLLAMA_BUILD_TESTS=OFF \
-                         -DGGML_CPU_RISCV64_SPACEMIT=ON \
-                         -DGGML_RVV=ON \
-                         -DGGML_RV_ZFH=ON \
-                         -DGGML_RV_ZICBOP=ON \
-                         -DGGML_RV_ZIHINTPAUSE=ON \
-                         -DRISCV64_SPACEMIT_IME_SPEC=RISCV64_SPACEMIT_IME1 \
-                         -DCMAKE_TOOLCHAIN_FILE=${PWD}/cmake/riscv64-spacemit-linux-gnu-gcc.cmake
-
-          cmake --build build --config Release -j $(nproc)

.github/workflows/check-vendor.yml

@@ -1,52 +0,0 @@
-name: Check vendor
-
-on:
-  workflow_dispatch: # allows manual triggering
-  push:
-    branches:
-      - master
-    paths: [
-      'vendor/**',
-      'scripts/sync_vendor.py'
-    ]
-
-  pull_request:
-    types: [opened, synchronize, reopened]
-    paths: [
-      'vendor/**',
-      'scripts/sync_vendor.py'
-    ]
-
-jobs:
-  check-vendor:
-    runs-on: ubuntu-latest
-
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
-
-      - name: Setup Python
-        uses: actions/setup-python@v4
-        with:
-          python-version: '3.x'
-
-      - name: Run vendor sync
-        run: |
-          set -euo pipefail
-          python3 scripts/sync_vendor.py
-
-      - name: Check for changes
-        run: |
-          set -euo pipefail
-          # detect modified or untracked files
-          changed=$(git status --porcelain --untracked-files=all || true)
-          if [ -n "$changed" ]; then
-            echo "Vendor sync modified files:"
-            echo "$changed" | awk '{ print $2 }' | sed '/^$/d'
-            echo "Failing because vendor files mismatch. Please update scripts/sync_vendor.py"
-            exit 1
-          else
-            echo "Vendor files are up-to-date."
-          fi

.github/workflows/close-issue.yml

@@ -17,7 +17,7 @@ jobs:
     steps:
       - uses: actions/stale@v5
         with:
-          exempt-issue-labels: "refactoring,help wanted,good first issue,research 🔬,bug,roadmap"
+          exempt-issue-labels: "refactoring,help wanted,good first issue,research,bug,roadmap"
           days-before-issue-stale: 30
           days-before-issue-close: 14
           stale-issue-label: "stale"

.github/workflows/copilot-setup-steps.yml

@@ -1,57 +0,0 @@
-name: "Copilot Setup Steps"
-
-# Automatically run the setup steps when they are changed to allow for easy validation, and
-# allow manual testing through the repository's "Actions" tab
-on:
-  workflow_dispatch:
-  push:
-    paths:
-      - .github/workflows/copilot-setup-steps.yml
-  pull_request:
-    paths:
-      - .github/workflows/copilot-setup-steps.yml
-
-jobs:
-  # The job MUST be called `copilot-setup-steps` or it will not be picked up by Copilot.
-  copilot-setup-steps:
-    runs-on: ubuntu-latest
-
-    # Set the permissions to the lowest permissions possible needed for your steps.
-    # Copilot will be given its own token for its operations.
-    permissions:
-      # If you want to clone the repository as part of your setup steps, for example to install dependencies, you'll need the `contents: read` permission. If you don't clone the repository in your setup steps, Copilot will do this for you automatically after the steps complete.
-      contents: read
-
-    # You can define any steps you want, and they will run before the agent starts.
-    # If you do not check out your code, Copilot will do this for you.
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@v4
-
-      - name: ccache
-        uses: ggml-org/ccache-action@v1.2.16
-        with:
-          key: copilot-setup-steps
-          evict-old-files: 1d
-
-      - name: Dependencies
-        id: depends
-        run: |
-          sudo apt-get update
-          sudo apt-get install build-essential libcurl4-openssl-dev
-          # Install git-clang-format script for formatting only changed code
-          wget -O /tmp/git-clang-format https://raw.githubusercontent.com/llvm/llvm-project/release/18.x/clang/tools/clang-format/git-clang-format
-          sudo cp /tmp/git-clang-format /usr/local/bin/git-clang-format
-          sudo chmod +x /usr/local/bin/git-clang-format
-
-      - name: Set up Python
-        uses: actions/setup-python@v5
-        with:
-          python-version: '3.11'
-
-      - name: Install Python dependencies
-        run: |
-          python3 -m venv .venv
-          .venv/bin/activate
-          pip install -r requirements/requirements-all.txt -r tools/server/tests/requirements.txt
-          pip install flake8 pyright pre-commit

.github/workflows/docker.yml

@@ -28,7 +28,7 @@ jobs:
   push_to_registry:
     name: Push Docker image to Docker Hub
 
-    runs-on: ${{ matrix.config.runs_on }}
+    runs-on: ubuntu-22.04
     env:
       COMMIT_SHA: ${{ github.sha }}
     strategy:
@@ -39,12 +39,11 @@ jobs:
           # Note: the arm64 images are failing, which prevents the amd64 images from being built
           # https://github.com/ggml-org/llama.cpp/issues/11888
           #- { tag: "cpu", dockerfile: ".devops/cpu.Dockerfile", platforms: "linux/amd64,linux/arm64", full: true, light: true, server: true, free_disk_space: false }
-          - { tag: "cpu",    dockerfile: ".devops/cpu.Dockerfile",    platforms: "linux/amd64", full: true, light: true, server: true, free_disk_space: false, runs_on: "ubuntu-22.04" }
-          - { tag: "cuda",   dockerfile: ".devops/cuda.Dockerfile",   platforms: "linux/amd64", full: true, light: true, server: true, free_disk_space: true,  runs_on: "ubuntu-22.04" }
-          - { tag: "musa",   dockerfile: ".devops/musa.Dockerfile",   platforms: "linux/amd64", full: true, light: true, server: true, free_disk_space: true,  runs_on: "ubuntu-22.04" }
-          - { tag: "intel",  dockerfile: ".devops/intel.Dockerfile",  platforms: "linux/amd64", full: true, light: true, server: true, free_disk_space: true,  runs_on: "ubuntu-22.04" }
-          - { tag: "vulkan", dockerfile: ".devops/vulkan.Dockerfile", platforms: "linux/amd64", full: true, light: true, server: true, free_disk_space: false, runs_on: "ubuntu-22.04" }
-          - { tag: "s390x",  dockerfile: ".devops/s390x.Dockerfile",  platforms: "linux/s390x", full: true, light: true, server: true, free_disk_space: false, runs_on: "ubuntu-22.04-s390x" }
+          - { tag: "cpu", dockerfile: ".devops/cpu.Dockerfile", platforms: "linux/amd64", full: true, light: true, server: true, free_disk_space: false }
+          - { tag: "cuda", dockerfile: ".devops/cuda.Dockerfile", platforms: "linux/amd64", full: true, light: true, server: true, free_disk_space: false }
+          - { tag: "musa", dockerfile: ".devops/musa.Dockerfile", platforms: "linux/amd64", full: true, light: true, server: true, free_disk_space: true }
+          - { tag: "intel", dockerfile: ".devops/intel.Dockerfile", platforms: "linux/amd64", full: true, light: true, server: true, free_disk_space: true }
+          - { tag: "vulkan", dockerfile: ".devops/vulkan.Dockerfile", platforms: "linux/amd64", full: true, light: true, server: true, free_disk_space: false }
           # Note: the rocm images are failing due to a compiler error and are disabled until this is fixed to allow the workflow to complete
           #- {tag: "rocm", dockerfile: ".devops/rocm.Dockerfile", platforms: "linux/amd64,linux/arm64", full: true, light: true, server: true, free_disk_space: true }
     steps:
@@ -54,7 +53,6 @@ jobs:
           fetch-depth: 0 # preserve git history, so we can determine the build number
 
       - name: Set up QEMU
-        if: ${{ matrix.config.tag != 's390x' }}
         uses: docker/setup-qemu-action@v3
         with:
           image: tonistiigi/binfmt:qemu-v7.0.0-28
@@ -69,19 +67,22 @@ jobs:
           username: ${{ github.repository_owner }}
           password: ${{ secrets.GITHUB_TOKEN }}
 
-      - name: Determine source tag name
-        id: srctag
-        uses: ./.github/actions/get-tag-name
-        env:
-          BRANCH_NAME: ${{ github.head_ref || github.ref_name }}
-
-      - name: Determine image tag name
+      - name: Determine tag name
         id: tag
         shell: bash
         run: |
+          BUILD_NUMBER="$(git rev-list --count HEAD)"
+          SHORT_HASH="$(git rev-parse --short=7 HEAD)"
           REPO_OWNER="${GITHUB_REPOSITORY_OWNER@L}"  # to lower case
           REPO_NAME="${{ github.event.repository.name }}"
 
+          # determine tag name postfix (build number, commit hash)
+          if [[ "${{ env.GITHUB_BRANCH_NAME }}" == "master" ]]; then
+            TAG_POSTFIX="-b${BUILD_NUMBER}"
+          else
+            SAFE_NAME=$(echo "${{ env.GITHUB_BRANCH_NAME }}" | tr '/' '-')
+            TAG_POSTFIX="-${SAFE_NAME}-${SHORT_HASH}"
+          fi
           # list all tags possible
           if [[ "${{ matrix.config.tag }}" == "cpu" ]]; then
               TYPE=""
@@ -89,19 +90,17 @@ jobs:
               TYPE="-${{ matrix.config.tag }}"
           fi
           PREFIX="ghcr.io/${REPO_OWNER}/${REPO_NAME}:"
-          CACHETAGS="${PREFIX}buildcache${TYPE}"
-          FULLTAGS="${PREFIX}full${TYPE},${PREFIX}full${TYPE}-${{ steps.srctag.outputs.name }}"
-          LIGHTTAGS="${PREFIX}light${TYPE},${PREFIX}light${TYPE}-${{ steps.srctag.outputs.name }}"
-          SERVERTAGS="${PREFIX}server${TYPE},${PREFIX}server${TYPE}-${{ steps.srctag.outputs.name }}"
-          echo "cache_output_tags=$CACHETAGS" >> $GITHUB_OUTPUT
+          FULLTAGS="${PREFIX}full${TYPE},${PREFIX}full${TYPE}${TAG_POSTFIX}"
+          LIGHTTAGS="${PREFIX}light${TYPE},${PREFIX}light${TYPE}${TAG_POSTFIX}"
+          SERVERTAGS="${PREFIX}server${TYPE},${PREFIX}server${TYPE}${TAG_POSTFIX}"
           echo "full_output_tags=$FULLTAGS" >> $GITHUB_OUTPUT
           echo "light_output_tags=$LIGHTTAGS" >> $GITHUB_OUTPUT
           echo "server_output_tags=$SERVERTAGS" >> $GITHUB_OUTPUT
-          echo "cache_output_tags=$CACHETAGS"  # print out for debugging
           echo "full_output_tags=$FULLTAGS"  # print out for debugging
           echo "light_output_tags=$LIGHTTAGS"  # print out for debugging
           echo "server_output_tags=$SERVERTAGS"  # print out for debugging
         env:
+          GITHUB_BRANCH_NAME: ${{ github.head_ref || github.ref_name }}
           GITHUB_REPOSITORY_OWNER: '${{ github.repository_owner }}'
 
       - name: Free Disk Space (Ubuntu)
@@ -134,14 +133,11 @@ jobs:
           target: full
           provenance: false
           # using github experimental cache
-          #cache-from: type=gha
-          #cache-to: type=gha,mode=max
+          cache-from: type=gha
+          cache-to: type=gha,mode=max
           # return to this if the experimental github cache is having issues
           #cache-to: type=local,dest=/tmp/.buildx-cache
           #cache-from: type=local,src=/tmp/.buildx-cache
-          # using registry cache (no storage limit)
-          cache-from: type=registry,ref=${{ steps.tag.outputs.cache_output_tags }}
-          cache-to: type=registry,ref=${{ steps.tag.outputs.cache_output_tags }},mode=max
 
       - name: Build and push Light Docker image (tagged + versioned)
         if: ${{ (github.event_name == 'push' || github.event_name == 'schedule' || github.event_name == 'workflow_dispatch') && matrix.config.light == true }}
@@ -156,14 +152,11 @@ jobs:
           target: light
           provenance: false
           # using github experimental cache
-          #cache-from: type=gha
-          #cache-to: type=gha,mode=max
+          cache-from: type=gha
+          cache-to: type=gha,mode=max
           # return to this if the experimental github cache is having issues
           #cache-to: type=local,dest=/tmp/.buildx-cache
           #cache-from: type=local,src=/tmp/.buildx-cache
-          # using registry cache (no storage limit)
-          cache-from: type=registry,ref=${{ steps.tag.outputs.cache_output_tags }}
-          cache-to: type=registry,ref=${{ steps.tag.outputs.cache_output_tags }},mode=max
 
       - name: Build and push Server Docker image (tagged + versioned)
         if: ${{ (github.event_name == 'push' || github.event_name == 'schedule' || github.event_name == 'workflow_dispatch') && matrix.config.server == true }}
@@ -178,37 +171,8 @@ jobs:
           target: server
           provenance: false
           # using github experimental cache
-          #cache-from: type=gha
-          #cache-to: type=gha,mode=max
+          cache-from: type=gha
+          cache-to: type=gha,mode=max
           # return to this if the experimental github cache is having issues
           #cache-to: type=local,dest=/tmp/.buildx-cache
           #cache-from: type=local,src=/tmp/.buildx-cache
-          # using registry cache (no storage limit)
-          cache-from: type=registry,ref=${{ steps.tag.outputs.cache_output_tags }}
-          cache-to: type=registry,ref=${{ steps.tag.outputs.cache_output_tags }},mode=max
-
-  create_tag:
-    name: Create and push git tag
-    runs-on: ubuntu-22.04
-    permissions:
-      contents: write
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
-
-      - name: Determine source tag name
-        id: srctag
-        uses: ./.github/actions/get-tag-name
-        env:
-          BRANCH_NAME: ${{ github.head_ref || github.ref_name }}
-
-      - name: Create and push git tag
-        env:
-          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-        run: |
-          git tag ${{ steps.srctag.outputs.name }} || exit 0
-          git push origin ${{ steps.srctag.outputs.name }} || exit 0

.github/workflows/pre-tokenizer-hashes.yml

@@ -1,45 +0,0 @@
-name: Check Pre-Tokenizer Hashes
-
-on:
-    push:
-        paths:
-            - 'convert_hf_to_gguf.py'
-            - 'convert_hf_to_gguf_update.py'
-    pull_request:
-        paths:
-            - 'convert_hf_to_gguf.py'
-            - 'convert_hf_to_gguf_update.py'
-
-jobs:
-    pre-tokenizer-hashes:
-        runs-on: ubuntu-latest
-
-        steps:
-        - name: Checkout repository
-          uses: actions/checkout@v4
-
-        - name: Set up Python
-          uses: actions/setup-python@v5
-          with:
-              python-version: '3.11'
-
-        - name: Install Python dependencies
-          run: |
-              python3 -m venv .venv
-              .venv/bin/pip install -r requirements/requirements-convert_hf_to_gguf_update.txt
-
-        - name: Update pre-tokenizer hashes
-          run: |
-              cp convert_hf_to_gguf.py /tmp
-              .venv/bin/python convert_hf_to_gguf_update.py --check-missing
-
-        - name: Check if committed pre-tokenizer hashes matches generated version
-          run: |
-              if ! diff -q convert_hf_to_gguf.py /tmp/convert_hf_to_gguf.py; then
-                  echo "Model pre-tokenizer hashes (in convert_hf_to_gguf.py) do not match generated hashes (from convert_hf_to_gguf_update.py)."
-                  echo "To fix: run ./convert_hf_to_gguf_update.py and commit the updated convert_hf_to_gguf.py along with your changes"
-                  echo "Differences found:"
-                  diff convert_hf_to_gguf.py /tmp/convert_hf_to_gguf.py || true
-                  exit 1
-              fi
-              echo "Model pre-tokenizer hashes are up to date."

.github/workflows/release.yml

@@ -32,7 +32,7 @@ jobs:
           fetch-depth: 0
 
       - name: ccache
-        uses: ggml-org/ccache-action@v1.2.16
+        uses: hendrikmuhs/ccache-action@v1.2.16
         with:
           key: macOS-latest-cmake-arm64
           evict-old-files: 1d
@@ -66,16 +66,16 @@ jobs:
         id: pack_artifacts
         run: |
           cp LICENSE ./build/bin/
-          tar -czvf llama-${{ steps.tag.outputs.name }}-bin-macos-arm64.tar.gz -s ",./,llama-${{ steps.tag.outputs.name }}/," -C ./build/bin .
+          zip -r llama-${{ steps.tag.outputs.name }}-bin-macos-arm64.zip ./build/bin/*
 
       - name: Upload artifacts
         uses: actions/upload-artifact@v4
         with:
-          path: llama-${{ steps.tag.outputs.name }}-bin-macos-arm64.tar.gz
-          name: llama-bin-macos-arm64.tar.gz
+          path: llama-${{ steps.tag.outputs.name }}-bin-macos-arm64.zip
+          name: llama-bin-macos-arm64.zip
 
   macOS-x64:
-    runs-on: macos-15-intel
+    runs-on: macos-13
 
     steps:
       - name: Clone
@@ -85,7 +85,7 @@ jobs:
           fetch-depth: 0
 
       - name: ccache
-        uses: ggml-org/ccache-action@v1.2.16
+        uses: hendrikmuhs/ccache-action@v1.2.16
         with:
           key: macOS-latest-cmake-x64
           evict-old-files: 1d
@@ -108,8 +108,7 @@ jobs:
             -DCMAKE_BUILD_WITH_INSTALL_RPATH=ON \
             -DLLAMA_FATAL_WARNINGS=ON \
             -DGGML_METAL=OFF \
-            -DGGML_RPC=ON \
-            -DCMAKE_OSX_DEPLOYMENT_TARGET=13.3
+            -DGGML_RPC=ON
           cmake --build build --config Release -j $(sysctl -n hw.logicalcpu)
 
       - name: Determine tag name
@@ -120,13 +119,13 @@ jobs:
         id: pack_artifacts
         run: |
           cp LICENSE ./build/bin/
-          tar -czvf llama-${{ steps.tag.outputs.name }}-bin-macos-x64.tar.gz -s ",./,llama-${{ steps.tag.outputs.name }}/," -C ./build/bin .
+          zip -r llama-${{ steps.tag.outputs.name }}-bin-macos-x64.zip ./build/bin/*
 
       - name: Upload artifacts
         uses: actions/upload-artifact@v4
         with:
-          path: llama-${{ steps.tag.outputs.name }}-bin-macos-x64.tar.gz
-          name: llama-bin-macos-x64.tar.gz
+          path: llama-${{ steps.tag.outputs.name }}-bin-macos-x64.zip
+          name: llama-bin-macos-x64.zip
 
   ubuntu-22-cpu:
     strategy:
@@ -134,8 +133,6 @@ jobs:
         include:
           - build: 'x64'
             os: ubuntu-22.04
-          - build: 's390x'
-            os: ubuntu-24.04-s390x
           # GGML_BACKEND_DL and GGML_CPU_ALL_VARIANTS are not currently supported on arm
           # - build: 'arm64'
           #   os: ubuntu-22.04-arm
@@ -150,9 +147,9 @@ jobs:
           fetch-depth: 0
 
       - name: ccache
-        uses: ggml-org/ccache-action@v1.2.16
+        uses: hendrikmuhs/ccache-action@v1.2.16
         with:
-          key: ubuntu-cpu-cmake-${{ matrix.build }}
+          key: ubuntu-cpu-cmake
           evict-old-files: 1d
 
       - name: Dependencies
@@ -182,13 +179,13 @@ jobs:
         id: pack_artifacts
         run: |
           cp LICENSE ./build/bin/
-          tar -czvf llama-${{ steps.tag.outputs.name }}-bin-ubuntu-${{ matrix.build }}.tar.gz --transform "s,./,llama-${{ steps.tag.outputs.name }}/," -C ./build/bin .
+          zip -r llama-${{ steps.tag.outputs.name }}-bin-ubuntu-${{ matrix.build }}.zip ./build/bin/*
 
       - name: Upload artifacts
         uses: actions/upload-artifact@v4
         with:
-          path: llama-${{ steps.tag.outputs.name }}-bin-ubuntu-${{ matrix.build }}.tar.gz
-          name: llama-bin-ubuntu-${{ matrix.build }}.tar.gz
+          path: llama-${{ steps.tag.outputs.name }}-bin-ubuntu-${{ matrix.build }}.zip
+          name: llama-bin-ubuntu-${{ matrix.build }}.zip
 
   ubuntu-22-vulkan:
     runs-on: ubuntu-22.04
@@ -201,7 +198,7 @@ jobs:
           fetch-depth: 0
 
       - name: ccache
-        uses: ggml-org/ccache-action@v1.2.16
+        uses: hendrikmuhs/ccache-action@v1.2.16
         with:
           key: ubuntu-22-cmake-vulkan
           evict-old-files: 1d
@@ -235,13 +232,13 @@ jobs:
         id: pack_artifacts
         run: |
           cp LICENSE ./build/bin/
-          tar -czvf llama-${{ steps.tag.outputs.name }}-bin-ubuntu-vulkan-x64.tar.gz --transform "s,./,llama-${{ steps.tag.outputs.name }}/," -C ./build/bin .
+          zip -r llama-${{ steps.tag.outputs.name }}-bin-ubuntu-vulkan-x64.zip ./build/bin/*
 
       - name: Upload artifacts
         uses: actions/upload-artifact@v4
         with:
-          path: llama-${{ steps.tag.outputs.name }}-bin-ubuntu-vulkan-x64.tar.gz
-          name: llama-bin-ubuntu-vulkan-x64.tar.gz
+          path: llama-${{ steps.tag.outputs.name }}-bin-ubuntu-vulkan-x64.zip
+          name: llama-bin-ubuntu-vulkan-x64.zip
 
   windows-cpu:
     runs-on: windows-2025
@@ -259,7 +256,7 @@ jobs:
           fetch-depth: 0
 
       - name: ccache
-        uses: ggml-org/ccache-action@v1.2.16
+        uses: hendrikmuhs/ccache-action@v1.2.16
         with:
           key: windows-latest-cmake-cpu-${{ matrix.arch }}
           variant: ccache
@@ -298,7 +295,7 @@ jobs:
         run: |
           Copy-Item $env:CURL_PATH\bin\libcurl-${{ matrix.arch }}.dll .\build\bin\Release\
           Copy-Item "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Redist\MSVC\14.44.35112\debug_nonredist\${{ matrix.arch }}\Microsoft.VC143.OpenMP.LLVM\libomp140.${{ matrix.arch == 'x64' && 'x86_64' || 'aarch64' }}.dll" .\build\bin\Release\
-          7z a -snl llama-bin-win-cpu-${{ matrix.arch }}.zip .\build\bin\Release\*
+          7z a llama-bin-win-cpu-${{ matrix.arch }}.zip .\build\bin\Release\*
 
       - name: Upload artifacts
         uses: actions/upload-artifact@v4
@@ -331,7 +328,7 @@ jobs:
         uses: actions/checkout@v4
 
       - name: ccache
-        uses: ggml-org/ccache-action@v1.2.16
+        uses: hendrikmuhs/ccache-action@v1.2.16
         with:
           key: windows-latest-cmake-${{ matrix.backend }}-${{ matrix.arch }}
           variant: ccache
@@ -380,7 +377,7 @@ jobs:
       - name: Pack artifacts
         id: pack_artifacts
         run: |
-          7z a -snl llama-bin-win-${{ matrix.backend }}-${{ matrix.arch }}.zip .\build\bin\Release\${{ matrix.target }}.dll
+          7z a llama-bin-win-${{ matrix.backend }}-${{ matrix.arch }}.zip .\build\bin\Release\${{ matrix.target }}.dll
 
       - name: Upload artifacts
         uses: actions/upload-artifact@v4
@@ -393,7 +390,7 @@ jobs:
 
     strategy:
       matrix:
-        cuda: ['12.4', '13.1']
+        cuda: ['12.4']
 
     steps:
       - name: Clone
@@ -401,7 +398,7 @@ jobs:
         uses: actions/checkout@v4
 
       - name: Install ccache
-        uses: ggml-org/ccache-action@v1.2.16
+        uses: hendrikmuhs/ccache-action@v1.2.16
         with:
           key: windows-cuda-${{ matrix.cuda }}
           variant: ccache
@@ -434,7 +431,7 @@ jobs:
       - name: Pack artifacts
         id: pack_artifacts
         run: |
-          7z a -snl llama-bin-win-cuda-${{ matrix.cuda }}-x64.zip .\build\bin\Release\ggml-cuda.dll
+          7z a llama-bin-win-cuda-${{ matrix.cuda }}-x64.zip .\build\bin\Release\ggml-cuda.dll
 
       - name: Upload artifacts
         uses: actions/upload-artifact@v4
@@ -448,7 +445,6 @@ jobs:
           $dst='.\build\bin\cudart\'
           robocopy "${{env.CUDA_PATH}}\bin" $dst cudart64_*.dll cublas64_*.dll cublasLt64_*.dll
           robocopy "${{env.CUDA_PATH}}\lib" $dst cudart64_*.dll cublas64_*.dll cublasLt64_*.dll
-          robocopy "${{env.CUDA_PATH}}\bin\x64" $dst cudart64_*.dll cublas64_*.dll cublasLt64_*.dll
           7z a cudart-llama-bin-win-cuda-${{ matrix.cuda }}-x64.zip $dst\*
 
       - name: Upload Cuda runtime
@@ -465,7 +461,7 @@ jobs:
         shell: bash
 
     env:
-      WINDOWS_BASEKIT_URL: https://registrationcenter-download.intel.com/akdlm/IRC_NAS/24751ead-ddc5-4479-b9e6-f9fe2ff8b9f2/intel-deep-learning-essentials-2025.2.1.25_offline.exe
+      WINDOWS_BASEKIT_URL: https://registrationcenter-download.intel.com/akdlm/IRC_NAS/7cd9bba0-7aab-4e30-b3ae-2221006a4a05/intel-oneapi-base-toolkit-2025.1.1.34_offline.exe
       WINDOWS_DPCPP_MKL: intel.oneapi.win.cpp-dpcpp-common:intel.oneapi.win.mkl.devel:intel.oneapi.win.dnnl:intel.oneapi.win.tbb.devel
       ONEAPI_ROOT: "C:/Program Files (x86)/Intel/oneAPI"
 
@@ -475,7 +471,7 @@ jobs:
         uses: actions/checkout@v4
 
       - name: ccache
-        uses: ggml-org/ccache-action@v1.2.16
+        uses: hendrikmuhs/ccache-action@v1.2.16
         with:
           key: windows-latest-cmake-sycl
           variant: ccache
@@ -508,7 +504,6 @@ jobs:
           cp "${{ env.ONEAPI_ROOT }}/mkl/latest/bin/mkl_tbb_thread.2.dll" ./build/bin
 
           cp "${{ env.ONEAPI_ROOT }}/compiler/latest/bin/ur_adapter_level_zero.dll" ./build/bin
-          cp "${{ env.ONEAPI_ROOT }}/compiler/latest/bin/ur_adapter_level_zero_v2.dll" ./build/bin
           cp "${{ env.ONEAPI_ROOT }}/compiler/latest/bin/ur_adapter_opencl.dll" ./build/bin
           cp "${{ env.ONEAPI_ROOT }}/compiler/latest/bin/ur_loader.dll" ./build/bin
           cp "${{ env.ONEAPI_ROOT }}/compiler/latest/bin/ur_win_proxy_loader.dll" ./build/bin
@@ -517,19 +512,12 @@ jobs:
           cp "${{ env.ONEAPI_ROOT }}/compiler/latest/bin/svml_dispmd.dll" ./build/bin
           cp "${{ env.ONEAPI_ROOT }}/compiler/latest/bin/libmmd.dll" ./build/bin
           cp "${{ env.ONEAPI_ROOT }}/compiler/latest/bin/libiomp5md.dll" ./build/bin
-          cp "${{ env.ONEAPI_ROOT }}/compiler/latest/bin/sycl-ls.exe" ./build/bin
-          cp "${{ env.ONEAPI_ROOT }}/compiler/latest/bin/libsycl-fallback-bfloat16.spv" ./build/bin
-          cp "${{ env.ONEAPI_ROOT }}/compiler/latest/bin/libsycl-native-bfloat16.spv" ./build/bin
 
           cp "${{ env.ONEAPI_ROOT }}/dnnl/latest/bin/dnnl.dll" ./build/bin
           cp "${{ env.ONEAPI_ROOT }}/tbb/latest/bin/tbb12.dll" ./build/bin
 
-          cp "${{ env.ONEAPI_ROOT }}/tcm/latest/bin/tcm.dll" ./build/bin
-          cp "${{ env.ONEAPI_ROOT }}/tcm/latest/bin/libhwloc-15.dll" ./build/bin
-          cp "${{ env.ONEAPI_ROOT }}/umf/latest/bin/umf.dll" ./build/bin
-
           echo "cp oneAPI running time dll files to ./build/bin done"
-          7z a -snl llama-bin-win-sycl-x64.zip ./build/bin/*
+          7z a llama-bin-win-sycl-x64.zip ./build/bin/*
 
       - name: Upload the release package
         uses: actions/upload-artifact@v4
@@ -540,71 +528,42 @@ jobs:
   windows-hip:
     runs-on: windows-2022
 
-    env:
-      HIPSDK_INSTALLER_VERSION: "25.Q3"
-
     strategy:
       matrix:
         include:
           - name: "radeon"
-            gpu_targets: "gfx1151;gfx1200;gfx1201;gfx1100;gfx1101;gfx1102;gfx1030;gfx1031;gfx1032"
+            gpu_targets: "gfx1100;gfx1101;gfx1102;gfx1030;gfx1031;gfx1032"
 
     steps:
       - name: Clone
         id: checkout
         uses: actions/checkout@v4
 
-      - name: Grab rocWMMA package
-        id: grab_rocwmma
+      - name: Clone rocWMMA repository
+        id: clone_rocwmma
         run: |
-          curl -o rocwmma.deb "https://repo.radeon.com/rocm/apt/7.0.1/pool/main/r/rocwmma-dev/rocwmma-dev_2.0.0.70001-42~24.04_amd64.deb"
-          7z x rocwmma.deb
-          7z x data.tar
-
-      - name: Cache ROCm Installation
-        id: cache-rocm
-        uses: actions/cache@v4
-        with:
-          path: C:\Program Files\AMD\ROCm
-          key: rocm-${{ env.HIPSDK_INSTALLER_VERSION }}-${{ runner.os }}
+          git clone https://github.com/rocm/rocwmma --branch rocm-6.2.4 --depth 1
 
       - name: ccache
-        uses: ggml-org/ccache-action@v1.2.16
+        uses: hendrikmuhs/ccache-action@v1.2.16
         with:
-          key: windows-latest-cmake-hip-${{ env.HIPSDK_INSTALLER_VERSION }}-${{ matrix.name }}-x64
+          key: windows-latest-cmake-hip-${{ matrix.name }}-x64
           evict-old-files: 1d
 
-      - name: Install ROCm
-        if: steps.cache-rocm.outputs.cache-hit != 'true'
+      - name: Install
         id: depends
         run: |
           $ErrorActionPreference = "Stop"
           write-host "Downloading AMD HIP SDK Installer"
-          Invoke-WebRequest -Uri "https://download.amd.com/developer/eula/rocm-hub/AMD-Software-PRO-Edition-${{ env.HIPSDK_INSTALLER_VERSION }}-WinSvr2022-For-HIP.exe" -OutFile "${env:RUNNER_TEMP}\rocm-install.exe"
+          Invoke-WebRequest -Uri "https://download.amd.com/developer/eula/rocm-hub/AMD-Software-PRO-Edition-24.Q3-WinSvr2022-For-HIP.exe" -OutFile "${env:RUNNER_TEMP}\rocm-install.exe"
           write-host "Installing AMD HIP SDK"
-          $proc = Start-Process "${env:RUNNER_TEMP}\rocm-install.exe" -ArgumentList '-install' -NoNewWindow -PassThru
-          $completed = $proc.WaitForExit(600000)
-          if (-not $completed) {
-              Write-Error "ROCm installation timed out after 10 minutes. Killing the process"
-              $proc.Kill()
-              exit 1
-          }
-          if ($proc.ExitCode -ne 0) {
-              Write-Error "ROCm installation failed with exit code $($proc.ExitCode)"
-              exit 1
-          }
+          Start-Process "${env:RUNNER_TEMP}\rocm-install.exe" -ArgumentList '-install' -NoNewWindow -Wait
           write-host "Completed AMD HIP SDK installation"
 
       - name: Verify ROCm
         id: verify
         run: |
-          # Find and test ROCm installation
-          $clangPath = Get-ChildItem 'C:\Program Files\AMD\ROCm\*\bin\clang.exe' | Select-Object -First 1
-          if (-not $clangPath) {
-            Write-Error "ROCm installation not found"
-            exit 1
-          }
-          & $clangPath.FullName --version
+          & 'C:\Program Files\AMD\ROCm\*\bin\clang.exe' --version
 
       - name: Build
         id: cmake_build
@@ -614,7 +573,7 @@ jobs:
           cmake -G "Unix Makefiles" -B build -S . `
             -DCMAKE_C_COMPILER="${env:HIP_PATH}\bin\clang.exe" `
             -DCMAKE_CXX_COMPILER="${env:HIP_PATH}\bin\clang++.exe" `
-            -DCMAKE_CXX_FLAGS="-I$($PWD.Path.Replace('\', '/'))/opt/rocm-7.0.1/include/ -Wno-ignored-attributes -Wno-nested-anon-types" `
+            -DCMAKE_CXX_FLAGS="-I$($PWD.Path.Replace('\', '/'))/rocwmma/library/include/ -Wno-ignored-attributes -Wno-nested-anon-types" `
             -DCMAKE_BUILD_TYPE=Release `
             -DGGML_BACKEND_DL=ON `
             -DGGML_NATIVE=OFF `
@@ -625,17 +584,14 @@ jobs:
             -DLLAMA_CURL=OFF
           cmake --build build --target ggml-hip -j ${env:NUMBER_OF_PROCESSORS}
           md "build\bin\rocblas\library\"
-          md "build\bin\hipblaslt\library"
           cp "${env:HIP_PATH}\bin\hipblas.dll" "build\bin\"
-          cp "${env:HIP_PATH}\bin\hipblaslt.dll" "build\bin\"
           cp "${env:HIP_PATH}\bin\rocblas.dll" "build\bin\"
           cp "${env:HIP_PATH}\bin\rocblas\library\*" "build\bin\rocblas\library\"
-          cp "${env:HIP_PATH}\bin\hipblaslt\library\*" "build\bin\hipblaslt\library\"
 
       - name: Pack artifacts
         id: pack_artifacts
         run: |
-          7z a -snl llama-bin-win-hip-${{ matrix.name }}-x64.zip .\build\bin\*
+          7z a llama-bin-win-hip-${{ matrix.name }}-x64.zip .\build\bin\*
 
       - name: Upload artifacts
         uses: actions/upload-artifact@v4
@@ -644,7 +600,7 @@ jobs:
           name: llama-bin-win-hip-${{ matrix.name }}-x64.zip
 
   ios-xcode-build:
-    runs-on: macos-15
+    runs-on: macos-latest
 
     steps:
       - name: Checkout code
@@ -652,10 +608,6 @@ jobs:
         with:
           fetch-depth: 0
 
-      - name: Setup Xcode
-        run: |
-          sudo xcode-select -s /Applications/Xcode_16.4.app
-
       - name: Build
         id: cmake_build
         run: |
@@ -688,87 +640,13 @@ jobs:
       - name: Pack artifacts
         id: pack_artifacts
         run: |
-          # Zip file is required for Swift Package Manager, which does not support tar.gz for binary targets.
-          # For more details, see https://developer.apple.com/documentation/xcode/distributing-binary-frameworks-as-swift-packages
-          zip -r -y llama-${{ steps.tag.outputs.name }}-xcframework.zip build-apple/llama.xcframework
+          zip --symlinks -r llama-${{ steps.tag.outputs.name }}-xcframework.zip build-apple/llama.xcframework
 
       - name: Upload artifacts
         uses: actions/upload-artifact@v4
         with:
           path: llama-${{ steps.tag.outputs.name }}-xcframework.zip
-          name: llama-${{ steps.tag.outputs.name }}-xcframework.zip
-
-
-  openEuler-cann:
-    strategy:
-      matrix:
-        arch: [x86, aarch64]
-        chip_type: ['910b', '310p']
-        build: ['Release']
-    runs-on: ${{ matrix.arch == 'aarch64' && 'ubuntu-24.04-arm' || 'ubuntu-24.04' }}
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
-
-      - name: Free up disk space
-        uses: ggml-org/free-disk-space@v1.3.1
-        with:
-          tool-cache: true
-
-      - name: Set container image
-        id: cann-image
-        run: |
-          image="ascendai/cann:${{ matrix.chip_type == '910b' &&  '8.3.rc2-910b-openeuler24.03-py3.11' || '8.3.rc2-310p-openeuler24.03-py3.11' }}"
-          echo "image=${image}" >> "${GITHUB_OUTPUT}"
-
-      - name: Pull container image
-        run: docker pull "${{ steps.cann-image.outputs.image }}"
-
-      - name: Build
-        env:
-          BUILD_TYPE: ${{ matrix.build }}
-          SOC_TYPE: ascend${{ matrix.chip_type }}
-        run: |
-          HOST_UID=$(id -u)
-          HOST_GID=$(id -g)
-
-          docker run --rm \
-            -v "${PWD}:/workspace" \
-            -w /workspace \
-            -e SOC_TYPE=${SOC_TYPE} \
-            -e BUILD_TYPE=${BUILD_TYPE} \
-            "${{ steps.cann-image.outputs.image }}" \
-            bash -lc '
-              set -e
-              yum install -y --setopt=install_weak_deps=False --setopt=tsflags=nodocs git gcc gcc-c++ make cmake libcurl-devel
-              yum clean all && rm -rf /var/cache/yum
-              git config --global --add safe.directory "/workspace"
-              export LD_LIBRARY_PATH=${ASCEND_TOOLKIT_HOME}/lib64:${ASCEND_TOOLKIT_HOME}/$(uname -m)-linux/devlib/:${LD_LIBRARY_PATH}
-              cmake -S . -B build \
-                  -DCMAKE_BUILD_TYPE=${BUILD_TYPE} \
-                  -DGGML_CANN=on \
-                  -DSOC_TYPE=${SOC_TYPE}
-              cmake --build build -j $(nproc)
-
-              chown -R '"${HOST_UID}"':'"${HOST_GID}"' /workspace/build
-            '
-
-      - name: Determine tag name
-        id: tag
-        uses: ./.github/actions/get-tag-name
-
-      - name: Pack artifacts
-        run: |
-          cp LICENSE ./build/bin/
-          tar -czvf llama-${{ steps.tag.outputs.name }}-bin-${{ matrix.chip_type }}-openEuler-${{ matrix.arch }}.tar.gz --transform "s,./,llama-${{ steps.tag.outputs.name }}/," -C ./build/bin .
-
-      - name: Upload artifacts
-        uses: actions/upload-artifact@v4
-        with:
-          path: llama-${{ steps.tag.outputs.name }}-bin-${{ matrix.chip_type }}-openEuler-${{ matrix.arch }}.tar.gz
-          name: llama-bin-${{ matrix.chip_type }}-openEuler-${{ matrix.arch }}.tar.gz
+          name: llama-${{ steps.tag.outputs.name }}-xcframework
 
   release:
     if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }}
@@ -791,7 +669,6 @@ jobs:
       - macOS-arm64
       - macOS-x64
       - ios-xcode-build
-      - openEuler-cann
 
     steps:
       - name: Clone
@@ -846,7 +723,6 @@ jobs:
 
           echo "Moving other artifacts..."
           mv -v artifact/*.zip release
-          mv -v artifact/*.tar.gz release
 
       - name: Create release
         id: create_release
@@ -855,37 +731,6 @@ jobs:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
         with:
           tag_name: ${{ steps.tag.outputs.name }}
-          body: |
-            <details open>
-
-            ${{ github.event.head_commit.message }}
-
-            </details>
-
-            **macOS/iOS:**
-            - [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-macos-arm64.tar.gz)
-            - [macOS Intel (x64)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-macos-x64.tar.gz)
-            - [iOS XCFramework](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-xcframework.zip)
-
-            **Linux:**
-            - [Ubuntu x64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-ubuntu-x64.tar.gz)
-            - [Ubuntu x64 (Vulkan)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-ubuntu-vulkan-x64.tar.gz)
-            - [Ubuntu s390x (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-ubuntu-s390x.tar.gz)
-
-            **Windows:**
-            - [Windows x64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-win-cpu-x64.zip)
-            - [Windows arm64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-win-cpu-arm64.zip)
-            - [Windows x64 (CUDA 12)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-win-cuda-12.4-x64.zip) - [CUDA 12.4 DLLs](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/cudart-llama-bin-win-cuda-12.4-x64.zip)
-            - [Windows x64 (CUDA 13)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-win-cuda-13.1-x64.zip) - [CUDA 13.1 DLLs](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/cudart-llama-bin-win-cuda-13.1-x64.zip)
-            - [Windows x64 (Vulkan)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-win-vulkan-x64.zip)
-            - [Windows x64 (SYCL)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-win-sycl-x64.zip)
-            - [Windows x64 (HIP)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-win-hip-radeon-x64.zip)
-
-            **openEuler:**
-            - [openEuler x86 (310p)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-310p-openEuler-x86.tar.gz)
-            - [openEuler x86 (910b)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-910b-openEuler-x86.tar.gz)
-            - [openEuler aarch64 (310p)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-310p-openEuler-aarch64.tar.gz)
-            - [openEuler aarch64 (910b)](https://github.com/ggml-org/llama.cpp/releases/download/${{ steps.tag.outputs.name }}/llama-${{ steps.tag.outputs.name }}-bin-910b-openEuler-aarch64.tar.gz)
 
       - name: Upload release
         id: upload_release
@@ -897,7 +742,7 @@ jobs:
             const fs = require('fs');
             const release_id = '${{ steps.create_release.outputs.id }}';
             for (let file of await fs.readdirSync('./release')) {
-              if (path.extname(file) === '.zip' || file.endsWith('.tar.gz')) {
+              if (path.extname(file) === '.zip') {
                 console.log('uploadReleaseAsset', file);
                 await github.repos.uploadReleaseAsset({
                   owner: context.repo.owner,

.github/workflows/server-webui.yml

@@ -1,225 +0,0 @@
-# Server WebUI build and tests
-name: Server WebUI
-
-on:
-  workflow_dispatch: # allows manual triggering
-    inputs:
-      sha:
-        description: 'Commit SHA1 to build'
-        required: false
-        type: string
-      slow_tests:
-        description: 'Run slow tests'
-        required: true
-        type: boolean
-  push:
-    branches:
-      - master
-    paths: ['.github/workflows/server-webui.yml', 'tools/server/webui/**.*', 'tools/server/tests/**.*', 'tools/server/public/**']
-  pull_request:
-    types: [opened, synchronize, reopened]
-    paths: ['.github/workflows/server-webui.yml', 'tools/server/webui/**.*', 'tools/server/tests/**.*', 'tools/server/public/**']
-
-env:
-  LLAMA_LOG_COLORS: 1
-  LLAMA_LOG_PREFIX: 1
-  LLAMA_LOG_TIMESTAMPS: 1
-  LLAMA_LOG_VERBOSITY: 10
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.ref }}-${{ github.head_ref || github.run_id }}
-  cancel-in-progress: true
-
-jobs:
-  webui-check:
-    name: WebUI Checks
-    runs-on: ubuntu-latest
-    continue-on-error: true
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
-          ref: ${{ github.event.inputs.sha || github.event.pull_request.head.sha || github.sha || github.head_ref || github.ref_name }}
-
-      - name: Setup Node.js
-        id: node
-        uses: actions/setup-node@v4
-        with:
-          node-version: "22"
-          cache: "npm"
-          cache-dependency-path: "tools/server/webui/package-lock.json"
-
-      - name: Install dependencies
-        id: setup
-        if: ${{ steps.node.conclusion == 'success' }}
-        run: npm ci
-        working-directory: tools/server/webui
-
-      - name: Run type checking
-        if: ${{ always() && steps.setup.conclusion == 'success' }}
-        run: npm run check
-        working-directory: tools/server/webui
-
-      - name: Run linting
-        if: ${{ always() && steps.setup.conclusion == 'success' }}
-        run: npm run lint
-        working-directory: tools/server/webui
-
-      - name: Build application
-        if: ${{ always() && steps.setup.conclusion == 'success' }}
-        run: npm run build
-        working-directory: tools/server/webui
-
-      - name: Install Playwright browsers
-        id: playwright
-        if: ${{ always() && steps.setup.conclusion == 'success' }}
-        run: npx playwright install --with-deps
-        working-directory: tools/server/webui
-
-      - name: Build Storybook
-        if: ${{ always() && steps.playwright.conclusion == 'success' }}
-        run: npm run build-storybook
-        working-directory: tools/server/webui
-
-      - name: Run Client tests
-        if: ${{ always() && steps.playwright.conclusion == 'success' }}
-        run: npm run test:client
-        working-directory: tools/server/webui
-
-      - name: Run Unit tests
-        if: ${{ always() && steps.playwright.conclusion == 'success' }}
-        run: npm run test:unit
-        working-directory: tools/server/webui
-
-      - name: Run UI tests
-        if: ${{ always() && steps.playwright.conclusion == 'success' }}
-        run: npm run test:ui -- --testTimeout=60000
-        working-directory: tools/server/webui
-
-      - name: Run E2E tests
-        if: ${{ always() && steps.playwright.conclusion == 'success' }}
-        run: npm run test:e2e
-        working-directory: tools/server/webui
-
-  server-build:
-    runs-on: ubuntu-latest
-
-    strategy:
-      matrix:
-        sanitizer: [ADDRESS, UNDEFINED] # THREAD is broken
-        build_type: [RelWithDebInfo]
-        include:
-          - build_type: Release
-            sanitizer: ""
-      fail-fast: false # While -DLLAMA_SANITIZE_THREAD=ON is broken
-
-    steps:
-      - name: Dependencies
-        id: depends
-        run: |
-          sudo apt-get update
-          sudo apt-get -y install \
-            build-essential \
-            xxd \
-            git \
-            cmake \
-            curl \
-            wget \
-            language-pack-en \
-            libssl-dev
-
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
-          ref: ${{ github.event.inputs.sha || github.event.pull_request.head.sha || github.sha || github.head_ref || github.ref_name }}
-
-      - name: Python setup
-        id: setup_python
-        uses: actions/setup-python@v5
-        with:
-          python-version: '3.11'
-
-      - name: Tests dependencies
-        id: test_dependencies
-        run: |
-          pip install -r tools/server/tests/requirements.txt
-
-      - name: Setup Node.js for WebUI
-        uses: actions/setup-node@v4
-        with:
-          node-version: "22"
-          cache: "npm"
-          cache-dependency-path: "tools/server/webui/package-lock.json"
-
-      - name: Install WebUI dependencies
-        run: npm ci
-        working-directory: tools/server/webui
-
-      - name: Build WebUI
-        run: npm run build
-        working-directory: tools/server/webui
-
-      - name: Build (no OpenMP)
-        id: cmake_build_no_openmp
-        if: ${{ matrix.sanitizer == 'THREAD' }}
-        run: |
-          cmake -B build \
-              -DGGML_NATIVE=OFF \
-              -DLLAMA_CURL=OFF \
-              -DLLAMA_OPENSSL=ON \
-              -DLLAMA_BUILD_SERVER=ON \
-              -DCMAKE_BUILD_TYPE=${{ matrix.build_type }} \
-              -DLLAMA_SANITIZE_${{ matrix.sanitizer }}=ON \
-              -DGGML_OPENMP=OFF ;
-          cmake --build build --config ${{ matrix.build_type }} -j $(nproc) --target llama-server
-
-      - name: Build (sanitizers)
-        id: cmake_build_sanitizers
-        if: ${{ matrix.sanitizer != '' && matrix.sanitizer != 'THREAD' }}
-        run: |
-          cmake -B build \
-              -DGGML_NATIVE=OFF \
-              -DLLAMA_CURL=OFF \
-              -DLLAMA_OPENSSL=ON \
-              -DLLAMA_BUILD_SERVER=ON \
-              -DCMAKE_BUILD_TYPE=${{ matrix.build_type }} \
-              -DLLAMA_SANITIZE_${{ matrix.sanitizer }}=ON ;
-          cmake --build build --config ${{ matrix.build_type }} -j $(nproc) --target llama-server
-
-      - name: Build (sanitizers)
-        id: cmake_build
-        if: ${{ matrix.sanitizer == '' }}
-        run: |
-          cmake -B build \
-              -DGGML_NATIVE=OFF \
-              -DLLAMA_CURL=OFF \
-              -DLLAMA_OPENSSL=ON \
-              -DLLAMA_BUILD_SERVER=ON \
-              -DCMAKE_BUILD_TYPE=${{ matrix.build_type }} ;
-          cmake --build build --config ${{ matrix.build_type }} -j $(nproc) --target llama-server
-
-      - name: Tests
-        id: server_integration_tests
-        if: ${{ matrix.sanitizer == '' }}
-        env:
-          GITHUB_ACTIONS: "true"
-        run: |
-          cd tools/server/tests
-          ./tests.sh
-
-      - name: Tests (sanitizers)
-        id: server_integration_tests_sanitizers
-        if: ${{ matrix.sanitizer != '' }}
-        run: |
-          cd tools/server/tests
-          LLAMA_SANITIZE=1 ./tests.sh
-
-      - name: Slow tests
-        id: server_integration_tests_slow
-        if: ${{ (github.event.schedule || github.event.inputs.slow_tests == 'true') && matrix.build_type == 'Release' }}
-        run: |
-          cd tools/server/tests
-          SLOW_TESTS=1 ./tests.sh

.github/workflows/server.yml

@@ -56,7 +56,7 @@ jobs:
             curl \
             wget \
             language-pack-en \
-            libssl-dev
+            libcurl4-openssl-dev
 
       - name: Clone
         id: checkout
@@ -76,6 +76,109 @@ jobs:
         run: |
           pip install -r tools/server/tests/requirements.txt
 
+      # Setup nodejs (to be used for verifying bundled index.html)
+      - uses: actions/setup-node@v4
+        with:
+          node-version: '22.11.0'
+
+      - name: WebUI - Install dependencies
+        id: webui_lint
+        run: |
+          cd tools/server/webui
+          npm ci
+
+      - name: WebUI - Check code format
+        id: webui_format
+        run: |
+          git config --global --add safe.directory $(realpath .)
+          cd tools/server/webui
+          git status
+
+          npm run format
+          git status
+          modified_files="$(git status -s)"
+          echo "Modified files: ${modified_files}"
+          if [ -n "${modified_files}" ]; then
+            echo "Files do not follow coding style. To fix: npm run format"
+            echo "${modified_files}"
+            exit 1
+          fi
+
+      - name: Verify bundled index.html
+        id: verify_server_index_html
+        run: |
+          git config --global --add safe.directory $(realpath .)
+          cd tools/server/webui
+          git status
+
+          npm run build
+          git status
+          modified_files="$(git status -s)"
+          echo "Modified files: ${modified_files}"
+          if [ -n "${modified_files}" ]; then
+            echo "Repository is dirty or server/webui is not built as expected"
+            echo "Hint: You may need to follow Web UI build guide in server/README.md"
+            echo "${modified_files}"
+            exit 1
+          fi
+
+      - name: Build (no OpenMP)
+        id: cmake_build_no_openmp
+        if: ${{ matrix.sanitizer == 'THREAD' }}
+        run: |
+          cmake -B build \
+              -DGGML_NATIVE=OFF \
+              -DLLAMA_BUILD_SERVER=ON \
+              -DCMAKE_BUILD_TYPE=${{ matrix.build_type }} \
+              -DLLAMA_SANITIZE_${{ matrix.sanitizer }}=ON \
+              -DGGML_OPENMP=OFF ;
+          cmake --build build --config ${{ matrix.build_type }} -j $(nproc) --target llama-server
+
+      - name: Build (sanitizers)
+        id: cmake_build_sanitizers
+        if: ${{ matrix.sanitizer != '' && matrix.sanitizer != 'THREAD' }}
+        run: |
+          cmake -B build \
+              -DGGML_NATIVE=OFF \
+              -DLLAMA_BUILD_SERVER=ON \
+              -DCMAKE_BUILD_TYPE=${{ matrix.build_type }} \
+              -DLLAMA_SANITIZE_${{ matrix.sanitizer }}=ON ;
+          cmake --build build --config ${{ matrix.build_type }} -j $(nproc) --target llama-server
+
+      - name: Build (sanitizers)
+        id: cmake_build
+        if: ${{ matrix.sanitizer == '' }}
+        run: |
+          cmake -B build \
+              -DGGML_NATIVE=OFF \
+              -DLLAMA_BUILD_SERVER=ON \
+              -DCMAKE_BUILD_TYPE=${{ matrix.build_type }} ;
+          cmake --build build --config ${{ matrix.build_type }} -j $(nproc) --target llama-server
+
+      - name: Tests
+        id: server_integration_tests
+        if: ${{ matrix.sanitizer == '' }}
+        env:
+          GITHUB_ACTIONS: "true"
+        run: |
+          cd tools/server/tests
+          ./tests.sh
+
+      - name: Tests (sanitizers)
+        id: server_integration_tests_sanitizers
+        if: ${{ matrix.sanitizer != '' }}
+        run: |
+          cd tools/server/tests
+          LLAMA_SANITIZE=1 ./tests.sh
+
+      - name: Slow tests
+        id: server_integration_tests_slow
+        if: ${{ (github.event.schedule || github.event.inputs.slow_tests == 'true') && matrix.build_type == 'Release' }}
+        run: |
+          cd tools/server/tests
+          SLOW_TESTS=1 ./tests.sh
+
+
   server-windows:
     runs-on: windows-2022
 
@@ -87,10 +190,16 @@ jobs:
           fetch-depth: 0
           ref: ${{ github.event.inputs.sha || github.event.pull_request.head.sha || github.sha || github.head_ref || github.ref_name }}
 
+      - name: libCURL
+        id: get_libcurl
+        uses: ./.github/actions/windows-setup-curl
+
       - name: Build
         id: cmake_build
+        env:
+          CURL_PATH: ${{ steps.get_libcurl.outputs.curl_path }}
         run: |
-          cmake -B build -DLLAMA_CURL=OFF -DLLAMA_BUILD_BORINGSSL=ON
+          cmake -B build -DCURL_LIBRARY="$env:CURL_PATH/lib/libcurl.dll.a" -DCURL_INCLUDE_DIR="$env:CURL_PATH/include"
           cmake --build build --config Release -j ${env:NUMBER_OF_PROCESSORS} --target llama-server
 
       - name: Python setup
@@ -104,6 +213,13 @@ jobs:
         run: |
           pip install -r tools/server/tests/requirements.txt
 
+      - name: Copy Libcurl
+        id: prepare_libcurl
+        env:
+          CURL_PATH: ${{ steps.get_libcurl.outputs.curl_path }}
+        run: |
+          cp $env:CURL_PATH/bin/libcurl-x64.dll ./build/bin/Release/libcurl-x64.dll
+
       - name: Tests
         id: server_integration_tests
         if: ${{ !matrix.disabled_on_pr || !github.event.pull_request }}

.github/workflows/update-ops-docs.yml

@@ -3,12 +3,10 @@ name: Update Operations Documentation
 on:
     push:
         paths:
-            - 'docs/ops.md'
             - 'docs/ops/**'
             - 'scripts/create_ops_docs.py'
     pull_request:
         paths:
-            - 'docs/ops.md'
             - 'docs/ops/**'
             - 'scripts/create_ops_docs.py'
 

.github/workflows/winget.yml

@@ -9,7 +9,6 @@ jobs:
   update:
     name: Update Winget Package
     runs-on: ubuntu-latest
-    if: github.repository_owner == 'ggml-org'
 
     steps:
       - name: Install cargo binstall

.gitignore

@@ -20,41 +20,52 @@
 *.so
 *.swp
 *.tmp
-*.DS_Store
 
 # IDE / OS
 
-/.cache/
-/.ccls-cache/
-/.direnv/
-/.envrc
-/.idea/
-/.swiftpm
-/.vs/
-/.vscode/
-/nppBackup
+.cache/
+.ccls-cache/
+.direnv/
+.DS_Store
+.envrc
+.idea/
+.swiftpm
+.vs/
+.vscode/
+nppBackup
 
 
 # Coverage
 
-/gcovr-report/
-/lcov-report/
+gcovr-report/
+lcov-report/
 
 # Build Artifacts
 
-/tags
-/.build/
-/build*
-/release
-/debug
+tags
+.build/
+build*
+release
+debug
+!build-info.cmake
+!build-info.cpp.in
+!build-info.sh
+!build.zig
+!docs/build.md
 /libllama.so
 /llama-*
 /vulkan-shaders-gen
+android-ndk-*
+arm_neon.h
+cmake-build-*
+CMakeSettings.json
+compile_commands.json
+ggml-metal-embed.metal
+llama-batched-swift
 /rpc-server
-/out/
-/tmp/
-/autogen-*.md
-/common/build-info.cpp
+out/
+tmp/
+autogen-*.md
 
 # Deprecated
 
@@ -63,38 +74,44 @@
 
 # CI
 
-!/.github/workflows/*.yml
+!.github/workflows/*.yml
 
 # Models
 
-/models/*
-/models-mnt
-!/models/.editorconfig
-!/models/ggml-vocab-*.gguf*
-!/models/templates
+models/*
+models-mnt
+!models/.editorconfig
+!models/ggml-vocab-*.gguf*
+!models/templates
 
 # Zig
-/zig-out/
-/zig-cache/
+zig-out/
+zig-cache/
+
+# Logs
+
+ppl-*.txt
+qnt-*.txt
+perf-*.txt
 
 # Examples
 
-/examples/jeopardy/results.txt
-/tools/server/*.css.hpp
-/tools/server/*.html.hpp
-/tools/server/*.js.hpp
-/tools/server/*.mjs.hpp
-/tools/server/*.gz.hpp
-!/build_64.sh
-!/examples/*.bat
-!/examples/*/*.kts
-!/examples/*/*/*.kts
-!/examples/sycl/*.bat
-!/examples/sycl/*.sh
+examples/jeopardy/results.txt
+tools/server/*.css.hpp
+tools/server/*.html.hpp
+tools/server/*.js.hpp
+tools/server/*.mjs.hpp
+tools/server/*.gz.hpp
+!build_64.sh
+!examples/*.bat
+!examples/*/*.kts
+!examples/*/*/*.kts
+!examples/sycl/*.bat
+!examples/sycl/*.sh
 
 # Server Web UI temporary files
-/tools/server/webui/node_modules
-/tools/server/webui/dist
+node_modules
+tools/server/webui/dist
 
 # Python
 
@@ -130,10 +147,3 @@ poetry.toml
 # Local scripts
 /run-vim.sh
 /run-chat.sh
-/.ccache/
-
-# IDE
-/*.code-workspace
-/.windsurf/
-# emscripten
-a.out.*

AGENTS.md

@@ -1,281 +0,0 @@
-# Instructions for llama.cpp
-
-## Repository Overview
-
-llama.cpp is a large-scale C/C++ project for efficient LLM (Large Language Model) inference with minimal setup and dependencies. The project enables running language models on diverse hardware with state-of-the-art performance.
-
-**Key Facts:**
-- **Primary language**: C/C++ with Python utility scripts
-- **Size**: ~200k+ lines of code across 1000+ files
-- **Architecture**: Modular design with main library (`libllama`) and 40+ executable tools/examples
-- **Core dependency**: ggml tensor library (vendored in `ggml/` directory)
-- **Backends supported**: CPU (AVX/NEON/RVV optimized), CUDA, Metal, Vulkan, SYCL, ROCm, MUSA
-- **License**: MIT
-
-## Disclose AI Usage
-
-It is crucial to remind contributors that the project mandates disclosure of any AI usage in pull requests. This requirement stems from the potential for AI-generated code to include suboptimal optimizations and hidden bugs, owing to the inherent overconfidence in AI outputs.
-
-When generating significant portions of code, address this by:
-- Informing the user that AI-generated content may be rejected by maintainers.
-- Clearly marking AI-generated code in commit messages and comments.
-    - Example of commit message: `[AI] Fix a race condition in ...`
-    - Example of code comment: `// [AI] spawn a new thread ...`
-
-These measures apply to:
-- Changes resulting in large portions of code or complex logic.
-- Modifications or additions to public APIs in `llama.h`, `ggml.h`, or `mtmd.h`.
-- Backend-related changes, such as those involving CPU, CUDA, Metal, Vulkan, etc.
-- Modifications to `tools/server`.
-
-Note: These measures can be omitted for small fixes or trivial changes.
-
-## Build Instructions
-
-### Prerequisites
-- CMake 3.14+ (primary build system)
-- C++17 compatible compiler (GCC 13.3+, Clang, MSVC)
-- Optional: ccache for faster compilation
-
-### Basic Build (CPU-only)
-**ALWAYS run these commands in sequence:**
-```bash
-cmake -B build
-cmake --build build --config Release -j $(nproc)
-```
-
-**Build time**: ~10 minutes on 4-core system with ccache enabled, ~25 minutes without ccache.
-
-**Important Notes:**
-- The Makefile is deprecated - always use CMake
-- ccache is automatically detected and used if available
-- Built binaries are placed in `build/bin/`
-- Parallel builds (`-j`) significantly reduce build time
-
-### Backend-Specific Builds
-For CUDA support:
-```bash
-cmake -B build -DGGML_CUDA=ON
-cmake --build build --config Release -j $(nproc)
-```
-
-For Metal (macOS):
-```bash
-cmake -B build -DGGML_METAL=ON
-cmake --build build --config Release -j $(nproc)
-```
-
-**Important Note**: While all backends can be built as long as the correct requirements for that backend are installed, you will not be able to run them without the correct hardware. The only backend that can be run for testing and validation is the CPU backend.
-
-### Debug Builds
-Single-config generators:
-```bash
-cmake -B build -DCMAKE_BUILD_TYPE=Debug
-cmake --build build
-```
-
-Multi-config generators:
-```bash
-cmake -B build -G "Xcode"
-cmake --build build --config Debug
-```
-
-### Common Build Issues
-- **Issue**: Network tests fail in isolated environments
-  **Solution**: Expected behavior - core functionality tests will still pass
-
-## Testing
-
-### Running Tests
-```bash
-ctest --test-dir build --output-on-failure -j $(nproc)
-```
-
-**Test suite**: 38 tests covering tokenizers, grammar parsing, sampling, backends, and integration
-**Expected failures**: 2-3 tests may fail if network access is unavailable (they download models)
-**Test time**: ~30 seconds for passing tests
-
-### Server Unit Tests
-Run server-specific unit tests after building the server:
-```bash
-# Build the server first
-cmake --build build --target llama-server
-
-# Navigate to server tests and run
-cd tools/server/tests
-source ../../../.venv/bin/activate
-./tests.sh
-```
-**Server test dependencies**: The `.venv` environment includes the required dependencies for server unit tests (pytest, aiohttp, etc.). Tests can be run individually or with various options as documented in `tools/server/tests/README.md`.
-
-### Test Categories
-- Tokenizer tests: Various model tokenizers (BERT, GPT-2, LLaMA, etc.)
-- Grammar tests: GBNF parsing and validation
-- Backend tests: Core ggml operations across different backends
-- Integration tests: End-to-end workflows
-
-### Manual Testing Commands
-```bash
-# Test basic inference
-./build/bin/llama-cli --version
-
-# Test model loading (requires model file)
-./build/bin/llama-cli -m path/to/model.gguf -p "Hello" -n 10
-```
-
-## Code Quality and Linting
-
-### C++ Code Formatting
-**ALWAYS format C++ code before committing:**
-```bash
-git clang-format
-```
-
-Configuration is in `.clang-format` with these key rules:
-- 4-space indentation
-- 120 column limit
-- Braces on same line for functions
-- Pointer alignment: `void * ptr` (middle)
-- Reference alignment: `int & ref` (middle)
-
-### Python Code
-**ALWAYS activate the Python environment in `.venv` and use tools from that environment:**
-```bash
-# Activate virtual environment
-source .venv/bin/activate
-```
-
-Configuration files:
-- `.flake8`: flake8 settings (max-line-length=125, excludes examples/tools)
-- `pyrightconfig.json`: pyright type checking configuration
-
-### Pre-commit Hooks
-Run before committing:
-```bash
-pre-commit run --all-files
-```
-
-## Continuous Integration
-
-### GitHub Actions Workflows
-Key workflows that run on every PR:
-- `.github/workflows/build.yml`: Multi-platform builds
-- `.github/workflows/server.yml`: Server functionality tests
-- `.github/workflows/python-lint.yml`: Python code quality
-- `.github/workflows/python-type-check.yml`: Python type checking
-
-### Local CI Validation
-**Run full CI locally before submitting PRs:**
-```bash
-mkdir tmp
-
-# CPU-only build
-bash ./ci/run.sh ./tmp/results ./tmp/mnt
-```
-
-**CI Runtime**: 30-60 minutes depending on backend configuration
-
-### Triggering CI
-Add `ggml-ci` to commit message to trigger heavy CI workloads on the custom CI infrastructure.
-
-## Project Layout and Architecture
-
-### Core Directories
-- **`src/`**: Main llama library implementation (`llama.cpp`, `llama-*.cpp`)
-- **`include/`**: Public API headers, primarily `include/llama.h`
-- **`ggml/`**: Core tensor library (submodule with custom GGML framework)
-- **`examples/`**: 30+ example applications and tools
-- **`tools/`**: Additional development and utility tools (server benchmarks, tests)
-- **`tests/`**: Comprehensive test suite with CTest integration
-- **`docs/`**: Detailed documentation (build guides, API docs, etc.)
-- **`scripts/`**: Utility scripts for CI, data processing, and automation
-- **`common/`**: Shared utility code used across examples
-
-### Key Files
-- **`CMakeLists.txt`**: Primary build configuration
-- **`include/llama.h`**: Main C API header (~2000 lines)
-- **`src/llama.cpp`**: Core library implementation (~8000 lines)
-- **`CONTRIBUTING.md`**: Coding guidelines and PR requirements
-- **`.clang-format`**: C++ formatting rules
-- **`.pre-commit-config.yaml`**: Git hook configuration
-
-### Built Executables (in `build/bin/`)
-Primary tools:
-- **`llama-cli`**: Main inference tool
-- **`llama-server`**: OpenAI-compatible HTTP server
-- **`llama-quantize`**: Model quantization utility
-- **`llama-perplexity`**: Model evaluation tool
-- **`llama-bench`**: Performance benchmarking
-- **`llama-convert-llama2c-to-ggml`**: Model conversion utilities
-
-### Configuration Files
-- **CMake**: `CMakeLists.txt`, `cmake/` directory
-- **Linting**: `.clang-format`, `.clang-tidy`, `.flake8`
-- **CI**: `.github/workflows/`, `ci/run.sh`
-- **Git**: `.gitignore` (includes build artifacts, models, cache)
-
-### Dependencies
-- **System**: OpenMP, libcurl (for model downloading)
-- **Optional**: CUDA SDK, Metal framework, Vulkan SDK, Intel oneAPI
-- **Bundled**: httplib, json (header-only libraries in vendored form)
-
-## Common Validation Steps
-
-### After Making Changes
-1. **Format code**: `git clang-format`
-2. **Build**: `cmake --build build --config Release`
-3. **Test**: `ctest --test-dir build --output-on-failure`
-4. **Server tests** (if modifying server): `cd tools/server/tests && source ../../../.venv/bin/activate && ./tests.sh`
-5. **Manual validation**: Test relevant tools in `build/bin/`
-
-### Performance Validation
-```bash
-# Benchmark inference performance
-./build/bin/llama-bench -m model.gguf
-
-# Evaluate model perplexity
-./build/bin/llama-perplexity -m model.gguf -f dataset.txt
-```
-
-### Backend Validation
-```bash
-# Test backend operations
-./build/bin/test-backend-ops
-```
-
-## Environment Setup
-
-### Required Tools
-- CMake 3.14+ (install via system package manager)
-- Modern C++ compiler with C++17 support
-- Git (for submodule management)
-- Python 3.9+ with virtual environment (`.venv` is provided)
-
-### Optional but Recommended
-- ccache: `apt install ccache` or `brew install ccache`
-- clang-format 15+: Usually included with LLVM/Clang installation
-- pre-commit: `pip install pre-commit`
-
-### Backend-Specific Requirements
-- **CUDA**: NVIDIA CUDA Toolkit 11.2+
-- **Metal**: Xcode command line tools (macOS only)
-- **Vulkan**: Vulkan SDK
-- **SYCL**: Intel oneAPI toolkit
-
-## Important Guidelines
-
-### Code Changes
-- **Minimal dependencies**: Avoid adding new external dependencies
-- **Cross-platform compatibility**: Test on Linux, macOS, Windows when possible
-- **Performance focus**: This is a performance-critical inference library
-- **API stability**: Changes to `include/llama.h` require careful consideration
-- **Disclose AI Usage**: Refer to the "Disclose AI Usage" earlier in this document
-
-### Git Workflow
-- Always create feature branches from `master`
-- **Never** commit build artifacts (`build/`, `.ccache/`, `*.o`, `*.gguf`)
-- Use descriptive commit messages following project conventions
-
-### Trust These Instructions
-Only search for additional information if these instructions are incomplete or found to be incorrect. This document contains validated build and test procedures that work reliably across different environments.
-

CMakeLists.txt

@@ -12,8 +12,6 @@ if (NOT XCODE AND NOT MSVC AND NOT CMAKE_BUILD_TYPE)
     set_property(CACHE CMAKE_BUILD_TYPE PROPERTY STRINGS "Debug" "Release" "MinSizeRel" "RelWithDebInfo")
 endif()
 
-message("CMAKE_BUILD_TYPE=${CMAKE_BUILD_TYPE}")
-
 # Add path to modules
 list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/cmake/")
 
@@ -33,24 +31,10 @@ endif()
 
 option(LLAMA_USE_SYSTEM_GGML "Use system libggml" OFF)
 
-option(LLAMA_WASM_MEM64 "llama: use 64-bit memory in WASM builds" ON)
-
 if (EMSCRIPTEN)
     set(BUILD_SHARED_LIBS_DEFAULT OFF)
 
-    # Use 64-bit memory to support backend_get_memory queries
-    # TODO: analyze performance impact, see https://spidermonkey.dev/blog/2025/01/15/is-memory64-actually-worth-using
-    if (LLAMA_WASM_MEM64)
-      add_compile_options("-sMEMORY64=1")
-      add_link_options("-sMEMORY64=1")
-    endif()
-    add_link_options("-sALLOW_MEMORY_GROWTH=1")
-
-    option(LLAMA_WASM_SINGLE_FILE "llama: embed WASM inside the generated llama.js" OFF)
-    option(LLAMA_BUILD_HTML "llama: build HTML file" ON)
-    if (LLAMA_BUILD_HTML)
-        set(CMAKE_EXECUTABLE_SUFFIX ".html")
-    endif()
+    option(LLAMA_WASM_SINGLE_FILE "llama: embed WASM inside the generated llama.js" ON)
 else()
     if (MINGW)
         set(BUILD_SHARED_LIBS_DEFAULT OFF)
@@ -72,18 +56,6 @@ if (MSVC)
     add_compile_options("$<$<COMPILE_LANGUAGE:CXX>:/bigobj>")
 endif()
 
-if (LLAMA_STANDALONE)
-    # enable parallel builds for msbuild
-    list(APPEND CMAKE_VS_GLOBALS UseMultiToolTask=true)
-    list(APPEND CMAKE_VS_GLOBALS EnforceProcessCountAcrossBuilds=true)
-endif()
-
-if (CMAKE_SYSTEM_NAME STREQUAL "iOS")
-    set(LLAMA_TOOLS_INSTALL_DEFAULT OFF)
-else()
-    set(LLAMA_TOOLS_INSTALL_DEFAULT ${LLAMA_STANDALONE})
-endif()
-
 #
 # option list
 #
@@ -108,12 +80,9 @@ option(LLAMA_BUILD_TESTS    "llama: build tests"          ${LLAMA_STANDALONE})
 option(LLAMA_BUILD_TOOLS    "llama: build tools"          ${LLAMA_STANDALONE})
 option(LLAMA_BUILD_EXAMPLES "llama: build examples"       ${LLAMA_STANDALONE})
 option(LLAMA_BUILD_SERVER   "llama: build server example" ${LLAMA_STANDALONE})
-option(LLAMA_TOOLS_INSTALL  "llama: install tools"        ${LLAMA_TOOLS_INSTALL_DEFAULT})
 
 # 3rd party libs
 option(LLAMA_CURL       "llama: use libcurl to download model from an URL" ON)
-option(LLAMA_HTTPLIB    "llama: if libcurl is disabled, use httplib to download model from an URL" ON)
-option(LLAMA_OPENSSL    "llama: use openssl to support HTTPS" OFF)
 option(LLAMA_LLGUIDANCE "llama-common: include LLGuidance library for structured output in common utils" OFF)
 
 # Required for relocatable CMake package
@@ -199,6 +168,11 @@ if (NOT TARGET ggml AND NOT LLAMA_USE_SYSTEM_GGML)
     # ... otherwise assume ggml is added by a parent CMakeLists.txt
 endif()
 
+if (MINGW)
+    # Target Windows 8 for PrefetchVirtualMemory
+    add_compile_definitions(_WIN32_WINNT=${GGML_WIN_VER})
+endif()
+
 #
 # build the library
 #
@@ -216,9 +190,6 @@ endif()
 
 if (LLAMA_BUILD_COMMON)
     add_subdirectory(common)
-    if (LLAMA_HTTPLIB)
-        add_subdirectory(vendor/cpp-httplib)
-    endif()
 endif()
 
 if (LLAMA_BUILD_COMMON AND LLAMA_BUILD_TESTS AND NOT CMAKE_JS_VERSION)

CODEOWNERS

@@ -1,108 +1,12 @@
 # collaborators can optionally add themselves here to indicate their availability for reviewing related PRs
-# multiplie collaborators per item can be specified
 
-/.devops/*.Dockerfile                   @ngxson
-/.github/actions/                       @CISC
-/.github/workflows/                     @CISC
-/ci/                                    @ggerganov
-/cmake/                                 @ggerganov
-/common/CMakeLists.txt                  @ggerganov
-/common/arg.*                           @ggerganov
-/common/base64.hpp.*                    @ggerganov
-/common/build-info.*                    @ggerganov
-/common/chat.*                          @pwilkin
-/common/chat-peg-parser.*               @aldehir
-/common/common.*                        @ggerganov
-/common/console.*                       @ggerganov
-/common/http.*                          @angt
-/common/llguidance.*                    @ggerganov
-/common/log.*                           @ggerganov
-/common/peg-parser.*                    @aldehir
-/common/sampling.*                      @ggerganov
-/common/speculative.*                   @ggerganov
-/common/unicode.*                       @aldehir
-/convert_*.py                           @CISC
-/examples/batched.swift/                @ggerganov
-/examples/batched/                      @ggerganov
-/examples/convert-llama2c-to-ggml/      @ggerganov
-/examples/deprecation-warning/          @ggerganov
-/examples/diffusion/                    @am17an
-/examples/embedding/                    @ggerganov
-/examples/eval-callback/                @ggerganov
-/examples/export-docs/                  @ggerganov
-/examples/gen-docs/                     @ggerganov
-/examples/gguf/                         @ggerganov
-/examples/llama.android/                @ggerganov @hanyin-arm @naco-siren
-/examples/llama.swiftui/                @ggerganov
-/examples/llama.vim                     @ggerganov
-/examples/lookahead/                    @ggerganov
-/examples/lookup/                       @JohannesGaessler
-/examples/model-conversion/             @danbev
-/examples/parallel/                     @ggerganov
-/examples/passkey/                      @ggerganov
-/examples/retrieval/                    @ggerganov
-/examples/save-load-state/              @ggerganov
-/examples/speculative-simple/           @ggerganov
-/examples/speculative/                  @ggerganov
-/ggml/cmake/                            @ggerganov
-/ggml/include/                          @ggerganov
-/ggml/src/ggml-common.h                 @ggerganov
-/ggml/src/ggml-cpu/                     @ggerganov
-/ggml/src/ggml-cpu/spacemit/            @alex-spacemit
-/ggml/src/ggml-cuda/fattn*              @JohannesGaessler
-/ggml/src/ggml-cuda/mmf.*               @JohannesGaessler @am17an
-/ggml/src/ggml-cuda/mmq.*               @JohannesGaessler
-/ggml/src/ggml-cuda/mmvf.*              @JohannesGaessler
-/ggml/src/ggml-cuda/mmvq.*              @JohannesGaessler
-/ggml/src/ggml-cuda/fattn-wmma*         @IMbackK
-/ggml/src/ggml-hip/                     @IMbackK
-/ggml/src/ggml-cuda/vendors/hip.h       @IMbackK
-/ggml/src/ggml-impl.h                   @ggerganov
-/ggml/src/ggml-metal/                   @ggerganov
-/ggml/src/ggml-opencl/                  @lhez @max-krasnyansky
-/ggml/src/ggml-hexagon/                 @max-krasnyansky @lhez
-/ggml/src/ggml-opt.cpp                  @JohannesGaessler
-/ggml/src/ggml-quants.*                 @ggerganov
-/ggml/src/ggml-rpc/                     @rgerganov
-/ggml/src/ggml-threading.*              @ggerganov
-/ggml/src/ggml-vulkan/                  @0cc4m
-/ggml/src/ggml-webgpu/                  @reeselevine
-/ggml/src/ggml-zdnn/                    @taronaeo @Andreas-Krebbel @AlekseiNikiforovIBM
-/ggml/src/ggml.c                        @ggerganov
-/ggml/src/ggml.cpp                      @ggerganov
-/ggml/src/gguf.cpp                      @JohannesGaessler @Green-Sky
-/gguf-py/                               @CISC
-/media/                                 @ggerganov
-/scripts/gen*                           @ggerganov
-/scripts/get*                           @ggerganov
-/scripts/sync*                          @ggerganov
-/src/                                   @ggerganov
-/src/llama-adapter.*                    @CISC
-/src/llama-arch.*                       @CISC
-/src/llama-chat.*                       @ngxson
-/src/llama-graph.*                      @CISC
-/src/llama-model.*                      @CISC
-/src/llama-vocab.*                      @CISC
-/src/models/                            @CISC
-/tests/                                 @ggerganov
-/tests/test-chat-.*                     @pwilkin
-/tools/batched-bench/                   @ggerganov
-/tools/cli/                             @ngxson
-/tools/completion/                      @ggerganov
-/tools/mtmd/                            @ngxson
-/tools/perplexity/                      @ggerganov
-/tools/quantize/                        @ggerganov
-/tools/rpc/                             @rgerganov
-/tools/server/*                         @ngxson @ggerganov # no subdir
-/tools/server/webui/                    @allozaur
-/tools/tokenize/                        @ggerganov
-/tools/tts/                             @ggerganov
-/vendor/                                @ggerganov
-/AUTHORS                                @ggerganov
-/CMakeLists.txt                         @ggerganov
-/CONTRIBUTING.md                        @ggerganov
-/LICENSE                                @ggerganov
-/README.md                              @ggerganov
-/SECURITY.md                            @ggerganov
-/build-xcframework.sh                   @danbev
-requirements*.txt                       @CISC
+/ci/ @ggerganov
+/.devops/*.Dockerfile @ngxson
+/tools/server/ @ngxson
+/ggml/src/ggml-cuda/fattn* @JohannesGaessler
+/ggml/src/ggml-cuda/mmq.* @JohannesGaessler
+/ggml/src/ggml-cuda/mmv.* @JohannesGaessler
+/ggml/src/ggml-cuda/mmvq.* @JohannesGaessler
+/ggml/src/ggml-opt.cpp @JohannesGaessler
+/ggml/src/gguf.cpp @JohannesGaessler
+/ggml/src/ggml-vulkan/ @0cc4m

CONTRIBUTING.md

@@ -1,12 +1,4 @@
-# Contributors
-
-The project differentiates between 3 levels of contributors:
-
-- Contributors: people who have contributed before (no special privileges)
-- Collaborators (Triage): people with significant contributions, who may be responsible for some parts of the code, and are expected to maintain and review contributions for the code they own
-- Maintainers: responsible for reviewing and merging PRs, after approval from the code owners
-
-# Pull requests (for contributors & collaborators)
+# Pull requests (for contributors)
 
 - llama.cpp uses the ggml tensor library for model evaluation. If you are unfamiliar with ggml, consider taking a look at the [examples in the ggml repository](https://github.com/ggml-org/ggml/tree/master/examples/). [simple](https://github.com/ggml-org/ggml/tree/master/examples/simple) shows the bare minimum for using ggml. [gpt-2](https://github.com/ggml-org/ggml/tree/master/examples/gpt-2) has minimal implementations for language model inference using GPT-2. [mnist](https://github.com/ggml-org/ggml/tree/master/examples/mnist) demonstrates how to train and evaluate a simple image classifier
 - Test your changes:
@@ -15,21 +7,15 @@ The project differentiates between 3 levels of contributors:
     - If you modified the `ggml` source, run the `test-backend-ops` tool to check whether different backend implementations of the `ggml` operators produce consistent results (this requires access to at least two different `ggml` backends)
     - If you modified a `ggml` operator or added a new one, add the corresponding test cases to `test-backend-ops`
 - Create separate PRs for each feature or fix. Avoid combining unrelated changes in a single PR
-- When adding support for a new model or feature, focus on **CPU support only** in the initial PR unless you have a good reason not to. Add support for other backends like CUDA in follow-up PRs
 - Consider allowing write access to your branch for faster reviews, as reviewers can push commits directly
-- If your PR becomes stale, rebase it on top of latest `master` to get maintainers attention
-- Maintainers will rely on your insights and approval when making a final decision to approve and merge a PR
-- Consider adding yourself to [CODEOWNERS](CODEOWNERS) to indicate your availability for reviewing related PRs
-- Using AI to generate PRs is permitted. However, you must (1) explicitly disclose how AI was used and (2) conduct a thorough manual review before publishing the PR. Note that trivial tab autocompletions do not require disclosure.
+- If your PR becomes stale, don't hesitate to ping the maintainers in the comments
 
-# Pull requests (for maintainers)
+# Pull requests (for collaborators)
 
 - Squash-merge PRs
 - Use the following format for the squashed commit title: `<module> : <commit title> (#<issue_number>)`. For example: `utils : fix typo in utils.py (#1234)`
 - Optionally pick a `<module>` from here: https://github.com/ggml-org/llama.cpp/wiki/Modules
-- Let other maintainers merge their own PRs
-- When merging a PR, make sure you have a good understanding of the changes
-- Be mindful of maintenance: most of the work going into a feature happens after the PR is merged. If the PR author is not committed to contribute long-term, someone else needs to take responsibility (you)
+- Consider adding yourself to [CODEOWNERS](CODEOWNERS)
 
 # Coding guidelines
 
@@ -128,21 +114,6 @@ The project differentiates between 3 levels of contributors:
     #endif // FOO
     ```
 
-# Code maintenance
-
-- Existing code should have designated collaborators and/or maintainers specified in the [CODEOWNERS](CODEOWNERS) file reponsible for:
-  - Reviewing and merging related PRs
-  - Fixing related bugs
-  - Providing developer guidance/support
-
-- When adding or modifying a large piece of code:
-  - If you are a collaborator, make sure to add yourself to [CODEOWNERS](CODEOWNERS) to indicate your availability for reviewing related PRs
-  - If you are a contributor, find an existing collaborator who is willing to review and maintain your code long-term
-  - Provide the necessary CI workflow (and hardware) to test your changes (see [ci/README.md](https://github.com/ggml-org/llama.cpp/tree/master/ci))
-
-- New code should follow the guidelines (coding, naming, etc.) outlined in this document. Exceptions are allowed in isolated, backend-specific parts of the code that do not interface directly with the `ggml` interfaces.
-  _(NOTE: for legacy reasons, existing code is not required to follow this guideline)_
-
 # Documentation
 
 - Documentation is a community effort

README.md

@@ -17,13 +17,11 @@ LLM inference in C/C++
 
 ## Hot topics
 
-- **[guide : using the new WebUI of llama.cpp](https://github.com/ggml-org/llama.cpp/discussions/16938)**
-- [guide : running gpt-oss with llama.cpp](https://github.com/ggml-org/llama.cpp/discussions/15396)
-- [[FEEDBACK] Better packaging for llama.cpp to support downstream consumers 🤗](https://github.com/ggml-org/llama.cpp/discussions/15313)
-- Support for the `gpt-oss` model with native MXFP4 format has been added | [PR](https://github.com/ggml-org/llama.cpp/pull/15091) | [Collaboration with NVIDIA](https://blogs.nvidia.com/blog/rtx-ai-garage-openai-oss) | [Comment](https://github.com/ggml-org/llama.cpp/discussions/15095)
+- Hot PRs: [All](https://github.com/ggml-org/llama.cpp/pulls?q=is%3Apr+label%3Ahot+) | [Open](https://github.com/ggml-org/llama.cpp/pulls?q=is%3Apr+label%3Ahot+is%3Aopen)
 - Multimodal support arrived in `llama-server`: [#12898](https://github.com/ggml-org/llama.cpp/pull/12898) | [documentation](./docs/multimodal.md)
 - VS Code extension for FIM completions: https://github.com/ggml-org/llama.vscode
 - Vim/Neovim plugin for FIM completions: https://github.com/ggml-org/llama.vim
+- Introducing GGUF-my-LoRA https://github.com/ggml-org/llama.cpp/discussions/10123
 - Hugging Face Inference Endpoints now support GGUF out of the box! https://github.com/ggml-org/llama.cpp/discussions/9669
 - Hugging Face GGUF editor: [discussion](https://github.com/ggml-org/llama.cpp/discussions/9268) | [tool](https://huggingface.co/spaces/CISCai/gguf-editor)
 
@@ -61,7 +59,6 @@ range of hardware - locally and in the cloud.
 - Plain C/C++ implementation without any dependencies
 - Apple silicon is a first-class citizen - optimized via ARM NEON, Accelerate and Metal frameworks
 - AVX, AVX2, AVX512 and AMX support for x86 architectures
-- RVV, ZVFH, ZFH, ZICBOP and ZIHINTPAUSE support for RISC-V architectures
 - 1.5-bit, 2-bit, 3-bit, 4-bit, 5-bit, 6-bit, and 8-bit integer quantization for faster inference and reduced memory use
 - Custom CUDA kernels for running LLMs on NVIDIA GPUs (support for AMD GPUs via HIP and Moore Threads GPUs via MUSA)
 - Vulkan and SYCL backend support
@@ -84,7 +81,6 @@ Instructions for adding support for new models: [HOWTO-add-model.md](docs/develo
 - [X] [Mistral 7B](https://huggingface.co/mistralai/Mistral-7B-v0.1)
 - [x] [Mixtral MoE](https://huggingface.co/models?search=mistral-ai/Mixtral)
 - [x] [DBRX](https://huggingface.co/databricks/dbrx-instruct)
-- [x] [Jamba](https://huggingface.co/ai21labs)
 - [X] [Falcon](https://huggingface.co/models?search=tiiuae/falcon)
 - [X] [Chinese LLaMA / Alpaca](https://github.com/ymcui/Chinese-LLaMA-Alpaca) and [Chinese LLaMA-2 / Alpaca-2](https://github.com/ymcui/Chinese-LLaMA-Alpaca-2)
 - [X] [Vigogne (French)](https://github.com/bofenghuang/vigogne)
@@ -138,8 +134,6 @@ Instructions for adding support for new models: [HOWTO-add-model.md](docs/develo
 - [X] [Trillion-7B-preview](https://huggingface.co/trillionlabs/Trillion-7B-preview)
 - [x] [Ling models](https://huggingface.co/collections/inclusionAI/ling-67c51c85b34a7ea0aba94c32)
 - [x] [LFM2 models](https://huggingface.co/collections/LiquidAI/lfm2-686d721927015b2ad73eaa38)
-- [x] [Hunyuan models](https://huggingface.co/collections/tencent/hunyuan-dense-model-6890632cda26b19119c9c5e7)
-- [x] [BailingMoeV2 (Ring/Ling 2.0) models](https://huggingface.co/collections/inclusionAI/ling-v2-68bf1dd2fc34c306c1fa6f86)
 
 #### Multimodal
 
@@ -154,7 +148,6 @@ Instructions for adding support for new models: [HOWTO-add-model.md](docs/develo
 - [x] [Bunny](https://github.com/BAAI-DCAI/Bunny)
 - [x] [GLM-EDGE](https://huggingface.co/models?search=glm-edge)
 - [x] [Qwen2-VL](https://huggingface.co/collections/Qwen/qwen2-vl-66cee7455501d7126940800d)
-- [x] [LFM2-VL](https://huggingface.co/collections/LiquidAI/lfm2-vl-68963bbc84a610f7638d5ffa)
 
 </details>
 
@@ -180,7 +173,6 @@ Instructions for adding support for new models: [HOWTO-add-model.md](docs/develo
 - Clojure: [phronmophobic/llama.clj](https://github.com/phronmophobic/llama.clj)
 - React Native: [mybigday/llama.rn](https://github.com/mybigday/llama.rn)
 - Java: [kherud/java-llama.cpp](https://github.com/kherud/java-llama.cpp)
-- Java: [QuasarByte/llama-cpp-jna](https://github.com/QuasarByte/llama-cpp-jna)
 - Zig: [deins/llama.cpp.zig](https://github.com/Deins/llama.cpp.zig)
 - Flutter/Dart: [netdur/llama_cpp_dart](https://github.com/netdur/llama_cpp_dart)
 - Flutter: [xuegao-tzx/Fllama](https://github.com/xuegao-tzx/Fllama)
@@ -189,8 +181,6 @@ Instructions for adding support for new models: [HOWTO-add-model.md](docs/develo
 - Swift [srgtuszy/llama-cpp-swift](https://github.com/srgtuszy/llama-cpp-swift)
 - Swift [ShenghaiWang/SwiftLlama](https://github.com/ShenghaiWang/SwiftLlama)
 - Delphi [Embarcadero/llama-cpp-delphi](https://github.com/Embarcadero/llama-cpp-delphi)
-- Go (no CGo needed): [hybridgroup/yzma](https://github.com/hybridgroup/yzma)
-- Android: [llama.android](/examples/llama.android)
 
 </details>
 
@@ -243,14 +233,13 @@ Instructions for adding support for new models: [HOWTO-add-model.md](docs/develo
 - [crashr/gppm](https://github.com/crashr/gppm) – launch llama.cpp instances utilizing NVIDIA Tesla P40 or P100 GPUs with reduced idle power consumption
 - [gpustack/gguf-parser](https://github.com/gpustack/gguf-parser-go/tree/main/cmd/gguf-parser) - review/check the GGUF file and estimate the memory usage
 - [Styled Lines](https://marketplace.unity.com/packages/tools/generative-ai/styled-lines-llama-cpp-model-292902) (proprietary licensed, async wrapper of inference part for game development in Unity3d with pre-built Mobile and Web platform wrappers and a model example)
-- [unslothai/unsloth](https://github.com/unslothai/unsloth) – 🦥 exports/saves fine-tuned and trained models to GGUF (Apache-2.0)
 
 </details>
 
 <details>
 <summary>Infrastructure</summary>
 
-- [Paddler](https://github.com/intentee/paddler) - Open-source LLMOps platform for hosting and scaling AI in your own infrastructure
+- [Paddler](https://github.com/distantmagic/paddler) - Stateful load balancer custom-tailored for llama.cpp
 - [GPUStack](https://github.com/gpustack/gpustack) - Manage GPU clusters for running LLMs
 - [llama_cpp_canister](https://github.com/onicai/llama_cpp_canister) - llama.cpp as a smart contract on the Internet Computer, using WebAssembly
 - [llama-swap](https://github.com/mostlygeek/llama-swap) - transparent proxy that adds automatic model switching with llama-server
@@ -277,14 +266,11 @@ Instructions for adding support for new models: [HOWTO-add-model.md](docs/develo
 | [MUSA](docs/build.md#musa) | Moore Threads GPU |
 | [CUDA](docs/build.md#cuda) | Nvidia GPU |
 | [HIP](docs/build.md#hip) | AMD GPU |
-| [ZenDNN](docs/build.md#zendnn) | AMD CPU |
 | [Vulkan](docs/build.md#vulkan) | GPU |
 | [CANN](docs/build.md#cann) | Ascend NPU |
 | [OpenCL](docs/backend/OPENCL.md) | Adreno GPU |
-| [IBM zDNN](docs/backend/zDNN.md) | IBM Z & LinuxONE |
 | [WebGPU [In Progress]](docs/build.md#webgpu) | All |
 | [RPC](https://github.com/ggml-org/llama.cpp/tree/master/tools/rpc) | All |
-| [Hexagon [In Progress]](docs/backend/hexagon/README.md) | Snapdragon |
 
 ## Obtaining and quantizing models
 
@@ -314,7 +300,7 @@ The Hugging Face platform provides a variety of online tools for converting, qua
 
 To learn more about model quantization, [read this documentation](tools/quantize/README.md)
 
-## [`llama-cli`](tools/cli)
+## [`llama-cli`](tools/main)
 
 #### A CLI tool for accessing and experimenting with most of `llama.cpp`'s functionality.
 
@@ -348,6 +334,19 @@ To learn more about model quantization, [read this documentation](tools/quantize
 
     </details>
 
+- <details>
+    <summary>Run simple text completion</summary>
+
+    To disable conversation mode explicitly, use `-no-cnv`
+
+    ```bash
+    llama-cli -m model.gguf -p "I believe the meaning of life is" -n 128 -no-cnv
+
+    # I believe the meaning of life is to find your own truth and to live in accordance with it. For me, this means being true to myself and following my passions, even if they don't align with societal expectations. I think that's what I love about yoga – it's not just a physical practice, but a spiritual one too. It's about connecting with yourself, listening to your inner voice, and honoring your own unique journey.
+    ```
+
+    </details>
+
 - <details>
     <summary>Constrain the output with a custom grammar</summary>
 
@@ -516,8 +515,8 @@ To learn more about model quantization, [read this documentation](tools/quantize
 ## Contributing
 
 - Contributors can open PRs
+- Collaborators can push to branches in the `llama.cpp` repo and merge PRs into the `master` branch
 - Collaborators will be invited based on contributions
-- Maintainers can push to branches in the `llama.cpp` repo and merge PRs into the `master` branch
 - Any help with managing issues, PRs and projects is very appreciated!
 - See [good first issues](https://github.com/ggml-org/llama.cpp/issues?q=is%3Aissue+is%3Aopen+label%3A%22good+first+issue%22) for tasks suitable for first contributions
 - Read the [CONTRIBUTING.md](CONTRIBUTING.md) for more information
@@ -526,8 +525,7 @@ To learn more about model quantization, [read this documentation](tools/quantize
 
 ## Other documentation
 
-- [cli](tools/cli/README.md)
-- [completion](tools/completion/README.md)
+- [main (cli)](tools/main/README.md)
 - [server](tools/server/README.md)
 - [GBNF grammars](grammars/README.md)
 
@@ -603,4 +601,3 @@ $ echo "source ~/.llama-completion.bash" >> ~/.bashrc
 - [linenoise.cpp](./tools/run/linenoise.cpp/linenoise.cpp) - C++ library that provides readline-like line editing capabilities, used by `llama-run` - BSD 2-Clause License
 - [curl](https://curl.se/) - Client-side URL transfer library, used by various tools/examples - [CURL License](https://curl.se/docs/copyright.html)
 - [miniaudio.h](https://github.com/mackron/miniaudio) - Single-header audio format decoder, used by multimodal subsystem - Public domain
-- [subprocess.h](https://github.com/sheredom/subprocess.h) - Single-header process launching solution for C and C++ - Public domain

SECURITY.md

@@ -65,9 +65,4 @@ However, If you have discovered a security vulnerability in this project, please
 
 Please disclose it as a private [security advisory](https://github.com/ggml-org/llama.cpp/security/advisories/new).
 
-Please note that using AI to identify vulnerabilities and generate reports is permitted. However, you must (1) explicitly disclose how AI was used and (2) conduct a thorough manual review before submitting the report.
-
 A team of volunteers on a reasonable-effort basis maintains this project. As such, please give us at least 90 days to work on a fix before public exposure.
-
-> [!IMPORTANT]
-> For collaborators: if you are interested in helping out with reviewing privting security disclosures, please see: https://github.com/ggml-org/llama.cpp/discussions/18080

benches/dgx-spark/aime25_openai__gpt-oss-120b-high_temp1.0_20251109_094547.json

@@ -1,6 +0,0 @@
-{
-  "chars": 2296.1916666666666,
-  "chars:std": 986.051306946325,
-  "score": 0.925,
-  "score:std": 0.26339134382131846
-}

benches/dgx-spark/dgx-spark.md

@@ -1,264 +0,0 @@
-## System info
-
-```bash
-uname --all
-Linux spark-17ed 6.11.0-1016-nvidia #16-Ubuntu SMP PREEMPT_DYNAMIC Sun Sep 21 16:52:46 UTC 2025 aarch64 aarch64 aarch64 GNU/Linux
-
-g++ --version
-g++ (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0
-
-nvidia-smi
-Sun Nov  2 10:43:25 2025
-+-----------------------------------------------------------------------------------------+
-| NVIDIA-SMI 580.95.05              Driver Version: 580.95.05      CUDA Version: 13.0     |
-+-----------------------------------------+------------------------+----------------------+
-| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
-| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
-|                                         |                        |               MIG M. |
-|=========================================+========================+======================|
-|   0  NVIDIA GB10                    On  |   0000000F:01:00.0 Off |                  N/A |
-| N/A   35C    P8              4W /  N/A  | Not Supported          |      0%      Default |
-|                                         |                        |                  N/A |
-+-----------------------------------------+------------------------+----------------------+
-```
-
-## ggml-org/gpt-oss-20b-GGUF
-
-Model: https://huggingface.co/ggml-org/gpt-oss-20b-GGUF
-
-- `llama-batched-bench`
-
-
-main: n_kv_max = 270336, n_batch = 2048, n_ubatch = 2048, flash_attn = 1, is_pp_shared = 0, n_gpu_layers = -1, n_threads = 20, n_threads_batch = 20
-
-|    PP |     TG |    B |   N_KV |   T_PP s | S_PP t/s |   T_TG s | S_TG t/s |      T s |    S t/s |
-|-------|--------|------|--------|----------|----------|----------|----------|----------|----------|
-|   512 |     32 |    1 |    544 |    0.374 |  1369.01 |    0.383 |    83.64 |    0.757 |   719.01 |
-|   512 |     32 |    2 |   1088 |    0.274 |  3741.35 |    0.659 |    97.14 |    0.933 |  1166.66 |
-|   512 |     32 |    4 |   2176 |    0.526 |  3896.47 |    0.817 |   156.73 |    1.342 |  1621.08 |
-|   512 |     32 |    8 |   4352 |    1.044 |  3925.10 |    0.987 |   259.44 |    2.030 |  2143.56 |
-|   512 |     32 |   16 |   8704 |    2.076 |  3945.84 |    1.248 |   410.32 |    3.324 |  2618.60 |
-|   512 |     32 |   32 |  17408 |    4.170 |  3929.28 |    1.630 |   628.40 |    5.799 |  3001.76 |
-|  4096 |     32 |    1 |   4128 |    1.083 |  3782.66 |    0.394 |    81.21 |    1.477 |  2795.13 |
-|  4096 |     32 |    2 |   8256 |    2.166 |  3782.72 |    0.725 |    88.28 |    2.891 |  2856.14 |
-|  4096 |     32 |    4 |  16512 |    4.333 |  3780.88 |    0.896 |   142.82 |    5.230 |  3157.38 |
-|  4096 |     32 |    8 |  33024 |    8.618 |  3802.14 |    1.155 |   221.69 |    9.773 |  3379.08 |
-|  4096 |     32 |   16 |  66048 |   17.330 |  3781.73 |    1.598 |   320.34 |   18.928 |  3489.45 |
-|  4096 |     32 |   32 | 132096 |   34.671 |  3780.48 |    2.336 |   438.35 |   37.007 |  3569.51 |
-|  8192 |     32 |    1 |   8224 |    2.233 |  3668.56 |    0.438 |    72.98 |    2.671 |  3078.44 |
-|  8192 |     32 |    2 |  16448 |    4.425 |  3702.95 |    0.756 |    84.66 |    5.181 |  3174.95 |
-|  8192 |     32 |    4 |  32896 |    8.859 |  3698.64 |    0.967 |   132.38 |    9.826 |  3347.72 |
-|  8192 |     32 |    8 |  65792 |   17.714 |  3699.57 |    1.277 |   200.52 |   18.991 |  3464.35 |
-|  8192 |     32 |   16 | 131584 |   35.494 |  3692.84 |    1.841 |   278.12 |   37.335 |  3524.46 |
-|  8192 |     32 |   32 | 263168 |   70.949 |  3694.82 |    2.798 |   365.99 |   73.747 |  3568.53 |
-
-
-- `llama-bench`
-
-| model                          |       size |     params | backend    | ngl | n_ubatch | fa | mmap |            test |                  t/s |
-| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | ---: | --------------: | -------------------: |
-| gpt-oss 20B MXFP4 MoE          |  11.27 GiB |    20.91 B | CUDA       |  99 |     2048 |  1 |    0 |          pp2048 |      3714.25 ± 20.36 |
-| gpt-oss 20B MXFP4 MoE          |  11.27 GiB |    20.91 B | CUDA       |  99 |     2048 |  1 |    0 |            tg32 |         86.58 ± 0.43 |
-| gpt-oss 20B MXFP4 MoE          |  11.27 GiB |    20.91 B | CUDA       |  99 |     2048 |  1 |    0 |  pp2048 @ d4096 |      3445.17 ± 17.85 |
-| gpt-oss 20B MXFP4 MoE          |  11.27 GiB |    20.91 B | CUDA       |  99 |     2048 |  1 |    0 |    tg32 @ d4096 |         81.72 ± 0.53 |
-| gpt-oss 20B MXFP4 MoE          |  11.27 GiB |    20.91 B | CUDA       |  99 |     2048 |  1 |    0 |  pp2048 @ d8192 |      3218.78 ± 11.34 |
-| gpt-oss 20B MXFP4 MoE          |  11.27 GiB |    20.91 B | CUDA       |  99 |     2048 |  1 |    0 |    tg32 @ d8192 |         74.86 ± 0.64 |
-| gpt-oss 20B MXFP4 MoE          |  11.27 GiB |    20.91 B | CUDA       |  99 |     2048 |  1 |    0 | pp2048 @ d16384 |       2732.83 ± 7.17 |
-| gpt-oss 20B MXFP4 MoE          |  11.27 GiB |    20.91 B | CUDA       |  99 |     2048 |  1 |    0 |   tg32 @ d16384 |         71.57 ± 0.51 |
-| gpt-oss 20B MXFP4 MoE          |  11.27 GiB |    20.91 B | CUDA       |  99 |     2048 |  1 |    0 | pp2048 @ d32768 |      2119.75 ± 12.81 |
-| gpt-oss 20B MXFP4 MoE          |  11.27 GiB |    20.91 B | CUDA       |  99 |     2048 |  1 |    0 |   tg32 @ d32768 |         62.33 ± 0.24 |
-
-build: eeee367de (6989)
-
-## ggml-org/gpt-oss-120b-GGUF
-
-Model: https://huggingface.co/ggml-org/gpt-oss-120b-GGUF
-
-- `llama-batched-bench`
-
-
-main: n_kv_max = 270336, n_batch = 2048, n_ubatch = 2048, flash_attn = 1, is_pp_shared = 0, n_gpu_layers = -1, n_threads = 20, n_threads_batch = 20
-
-|    PP |     TG |    B |   N_KV |   T_PP s | S_PP t/s |   T_TG s | S_TG t/s |      T s |    S t/s |
-|-------|--------|------|--------|----------|----------|----------|----------|----------|----------|
-|   512 |     32 |    1 |    544 |    0.571 |   897.18 |    0.543 |    58.96 |    1.113 |   488.60 |
-|   512 |     32 |    2 |   1088 |    0.593 |  1725.37 |    1.041 |    61.45 |    1.635 |   665.48 |
-|   512 |     32 |    4 |   2176 |    1.043 |  1963.15 |    1.334 |    95.95 |    2.377 |   915.36 |
-|   512 |     32 |    8 |   4352 |    2.099 |  1951.63 |    1.717 |   149.07 |    3.816 |  1140.45 |
-|   512 |     32 |   16 |   8704 |    4.207 |  1947.12 |    2.311 |   221.56 |    6.518 |  1335.35 |
-|   512 |     32 |   32 |  17408 |    8.422 |  1945.36 |    3.298 |   310.46 |   11.720 |  1485.27 |
-|  4096 |     32 |    1 |   4128 |    2.138 |  1915.88 |    0.571 |    56.09 |    2.708 |  1524.12 |
-|  4096 |     32 |    2 |   8256 |    4.266 |  1920.25 |    1.137 |    56.27 |    5.404 |  1527.90 |
-|  4096 |     32 |    4 |  16512 |    8.564 |  1913.02 |    1.471 |    86.99 |   10.036 |  1645.29 |
-|  4096 |     32 |    8 |  33024 |   17.092 |  1917.19 |    1.979 |   129.33 |   19.071 |  1731.63 |
-|  4096 |     32 |   16 |  66048 |   34.211 |  1915.65 |    2.850 |   179.66 |   37.061 |  1782.15 |
-|  4096 |     32 |   32 | 132096 |   68.394 |  1916.44 |    4.381 |   233.72 |   72.775 |  1815.13 |
-|  8192 |     32 |    1 |   8224 |    4.349 |  1883.45 |    0.620 |    51.65 |    4.969 |  1655.04 |
-|  8192 |     32 |    2 |  16448 |    8.674 |  1888.83 |    1.178 |    54.33 |    9.852 |  1669.48 |
-|  8192 |     32 |    4 |  32896 |   17.351 |  1888.55 |    1.580 |    81.01 |   18.931 |  1737.68 |
-|  8192 |     32 |    8 |  65792 |   34.743 |  1886.31 |    2.173 |   117.80 |   36.916 |  1782.20 |
-|  8192 |     32 |   16 | 131584 |   69.413 |  1888.29 |    3.297 |   155.28 |   72.710 |  1809.70 |
-|  8192 |     32 |   32 | 263168 |  138.903 |  1887.24 |    5.004 |   204.63 |  143.907 |  1828.73 |
-
-
-- `llama-bench`
-
-| model                          |       size |     params | backend    | ngl | n_ubatch | fa | mmap |            test |                  t/s |
-| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | ---: | --------------: | -------------------: |
-| gpt-oss 120B MXFP4 MoE         |  59.02 GiB |   116.83 B | CUDA       |  99 |     2048 |  1 |    0 |          pp2048 |       1919.36 ± 5.01 |
-| gpt-oss 120B MXFP4 MoE         |  59.02 GiB |   116.83 B | CUDA       |  99 |     2048 |  1 |    0 |            tg32 |         60.40 ± 0.30 |
-| gpt-oss 120B MXFP4 MoE         |  59.02 GiB |   116.83 B | CUDA       |  99 |     2048 |  1 |    0 |  pp2048 @ d4096 |       1825.30 ± 6.37 |
-| gpt-oss 120B MXFP4 MoE         |  59.02 GiB |   116.83 B | CUDA       |  99 |     2048 |  1 |    0 |    tg32 @ d4096 |         56.94 ± 0.29 |
-| gpt-oss 120B MXFP4 MoE         |  59.02 GiB |   116.83 B | CUDA       |  99 |     2048 |  1 |    0 |  pp2048 @ d8192 |       1739.19 ± 6.00 |
-| gpt-oss 120B MXFP4 MoE         |  59.02 GiB |   116.83 B | CUDA       |  99 |     2048 |  1 |    0 |    tg32 @ d8192 |         52.51 ± 0.42 |
-| gpt-oss 120B MXFP4 MoE         |  59.02 GiB |   116.83 B | CUDA       |  99 |     2048 |  1 |    0 | pp2048 @ d16384 |       1536.75 ± 4.27 |
-| gpt-oss 120B MXFP4 MoE         |  59.02 GiB |   116.83 B | CUDA       |  99 |     2048 |  1 |    0 |   tg32 @ d16384 |         49.33 ± 0.27 |
-| gpt-oss 120B MXFP4 MoE         |  59.02 GiB |   116.83 B | CUDA       |  99 |     2048 |  1 |    0 | pp2048 @ d32768 |       1255.85 ± 3.26 |
-| gpt-oss 120B MXFP4 MoE         |  59.02 GiB |   116.83 B | CUDA       |  99 |     2048 |  1 |    0 |   tg32 @ d32768 |         42.99 ± 0.18 |
-
-build: eeee367de (6989)
-
-## ggml-org/Qwen3-Coder-30B-A3B-Instruct-Q8_0-GGUF
-
-Model: https://huggingface.co/ggml-org/Qwen3-Coder-30B-A3B-Instruct-Q8_0-GGUF
-
-- `llama-batched-bench`
-
-
-main: n_kv_max = 270336, n_batch = 2048, n_ubatch = 2048, flash_attn = 1, is_pp_shared = 0, n_gpu_layers = -1, n_threads = 20, n_threads_batch = 20
-
-|    PP |     TG |    B |   N_KV |   T_PP s | S_PP t/s |   T_TG s | S_TG t/s |      T s |    S t/s |
-|-------|--------|------|--------|----------|----------|----------|----------|----------|----------|
-|   512 |     32 |    1 |    544 |    0.398 |  1285.90 |    0.530 |    60.41 |    0.928 |   586.27 |
-|   512 |     32 |    2 |   1088 |    0.386 |  2651.65 |    0.948 |    67.50 |    1.334 |   815.38 |
-|   512 |     32 |    4 |   2176 |    0.666 |  3076.37 |    1.209 |   105.87 |    1.875 |  1160.71 |
-|   512 |     32 |    8 |   4352 |    1.325 |  3091.39 |    1.610 |   158.98 |    2.935 |  1482.65 |
-|   512 |     32 |   16 |   8704 |    2.664 |  3075.58 |    2.150 |   238.19 |    4.813 |  1808.39 |
-|   512 |     32 |   32 |  17408 |    5.336 |  3070.31 |    2.904 |   352.59 |    8.240 |  2112.50 |
-|  4096 |     32 |    1 |   4128 |    1.444 |  2836.81 |    0.581 |    55.09 |    2.025 |  2038.81 |
-|  4096 |     32 |    2 |   8256 |    2.872 |  2852.14 |    1.084 |    59.06 |    3.956 |  2086.99 |
-|  4096 |     32 |    4 |  16512 |    5.744 |  2852.32 |    1.440 |    88.90 |    7.184 |  2298.47 |
-|  4096 |     32 |    8 |  33024 |   11.463 |  2858.68 |    2.068 |   123.78 |   13.531 |  2440.65 |
-|  4096 |     32 |   16 |  66048 |   22.915 |  2859.95 |    3.018 |   169.67 |   25.933 |  2546.90 |
-|  4096 |     32 |   32 | 132096 |   45.956 |  2852.10 |    4.609 |   222.18 |   50.565 |  2612.39 |
-|  8192 |     32 |    1 |   8224 |    3.063 |  2674.72 |    0.693 |    46.20 |    3.755 |  2189.92 |
-|  8192 |     32 |    2 |  16448 |    6.109 |  2681.87 |    1.214 |    52.71 |    7.323 |  2245.98 |
-|  8192 |     32 |    4 |  32896 |   12.197 |  2686.63 |    1.682 |    76.11 |   13.878 |  2370.30 |
-|  8192 |     32 |    8 |  65792 |   24.409 |  2684.94 |    2.556 |   100.17 |   26.965 |  2439.95 |
-|  8192 |     32 |   16 | 131584 |   48.753 |  2688.50 |    3.994 |   128.20 |   52.747 |  2494.64 |
-|  8192 |     32 |   32 | 263168 |   97.508 |  2688.42 |    6.528 |   156.86 |  104.037 |  2529.57 |
-
-
-- `llama-bench`
-
-| model                          |       size |     params | backend    | ngl | n_ubatch | fa | mmap |            test |                  t/s |
-| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | ---: | --------------: | -------------------: |
-| qwen3moe 30B.A3B Q8_0          |  30.25 GiB |    30.53 B | CUDA       |  99 |     2048 |  1 |    0 |          pp2048 |       2925.55 ± 4.25 |
-| qwen3moe 30B.A3B Q8_0          |  30.25 GiB |    30.53 B | CUDA       |  99 |     2048 |  1 |    0 |            tg32 |         62.80 ± 0.27 |
-| qwen3moe 30B.A3B Q8_0          |  30.25 GiB |    30.53 B | CUDA       |  99 |     2048 |  1 |    0 |  pp2048 @ d4096 |       2531.01 ± 6.79 |
-| qwen3moe 30B.A3B Q8_0          |  30.25 GiB |    30.53 B | CUDA       |  99 |     2048 |  1 |    0 |    tg32 @ d4096 |         55.86 ± 0.33 |
-| qwen3moe 30B.A3B Q8_0          |  30.25 GiB |    30.53 B | CUDA       |  99 |     2048 |  1 |    0 |  pp2048 @ d8192 |       2244.39 ± 5.33 |
-| qwen3moe 30B.A3B Q8_0          |  30.25 GiB |    30.53 B | CUDA       |  99 |     2048 |  1 |    0 |    tg32 @ d8192 |         45.95 ± 0.33 |
-| qwen3moe 30B.A3B Q8_0          |  30.25 GiB |    30.53 B | CUDA       |  99 |     2048 |  1 |    0 | pp2048 @ d16384 |       1783.17 ± 3.68 |
-| qwen3moe 30B.A3B Q8_0          |  30.25 GiB |    30.53 B | CUDA       |  99 |     2048 |  1 |    0 |   tg32 @ d16384 |         39.07 ± 0.10 |
-| qwen3moe 30B.A3B Q8_0          |  30.25 GiB |    30.53 B | CUDA       |  99 |     2048 |  1 |    0 | pp2048 @ d32768 |       1241.90 ± 3.13 |
-| qwen3moe 30B.A3B Q8_0          |  30.25 GiB |    30.53 B | CUDA       |  99 |     2048 |  1 |    0 |   tg32 @ d32768 |         29.92 ± 0.06 |
-
-build: eeee367de (6989)
-
-## ggml-org/Qwen2.5-Coder-7B-Q8_0-GGUF
-
-Model: https://huggingface.co/ggml-org/Qwen2.5-Coder-7B-Q8_0-GGUF
-
-- `llama-batched-bench`
-
-
-main: n_kv_max = 270336, n_batch = 2048, n_ubatch = 2048, flash_attn = 1, is_pp_shared = 0, n_gpu_layers = -1, n_threads = 20, n_threads_batch = 20
-
-|    PP |     TG |    B |   N_KV |   T_PP s | S_PP t/s |   T_TG s | S_TG t/s |      T s |    S t/s |
-|-------|--------|------|--------|----------|----------|----------|----------|----------|----------|
-|   512 |     32 |    1 |    544 |    0.211 |  2421.57 |    1.055 |    30.33 |    1.266 |   429.57 |
-|   512 |     32 |    2 |   1088 |    0.419 |  2441.34 |    1.130 |    56.65 |    1.549 |   702.32 |
-|   512 |     32 |    4 |   2176 |    0.873 |  2345.54 |    1.174 |   108.99 |    2.048 |  1062.74 |
-|   512 |     32 |    8 |   4352 |    1.727 |  2371.85 |    1.254 |   204.22 |    2.980 |  1460.19 |
-|   512 |     32 |   16 |   8704 |    3.452 |  2373.22 |    1.492 |   343.16 |    4.944 |  1760.56 |
-|   512 |     32 |   32 |  17408 |    6.916 |  2368.93 |    1.675 |   611.51 |    8.591 |  2026.36 |
-|  4096 |     32 |    1 |   4128 |    1.799 |  2277.26 |    1.084 |    29.51 |    2.883 |  1431.91 |
-|  4096 |     32 |    2 |   8256 |    3.577 |  2290.01 |    1.196 |    53.50 |    4.774 |  1729.51 |
-|  4096 |     32 |    4 |  16512 |    7.172 |  2284.36 |    1.313 |    97.50 |    8.485 |  1946.00 |
-|  4096 |     32 |    8 |  33024 |   14.341 |  2284.96 |    1.520 |   168.46 |   15.860 |  2082.18 |
-|  4096 |     32 |   16 |  66048 |   28.675 |  2285.44 |    1.983 |   258.21 |   30.658 |  2154.33 |
-|  4096 |     32 |   32 | 132096 |   57.354 |  2285.32 |    2.640 |   387.87 |   59.994 |  2201.82 |
-|  8192 |     32 |    1 |   8224 |    3.701 |  2213.75 |    1.119 |    28.59 |    4.820 |  1706.34 |
-|  8192 |     32 |    2 |  16448 |    7.410 |  2211.19 |    1.272 |    50.31 |    8.682 |  1894.56 |
-|  8192 |     32 |    4 |  32896 |   14.802 |  2213.83 |    1.460 |    87.68 |   16.261 |  2022.96 |
-|  8192 |     32 |    8 |  65792 |   29.609 |  2213.35 |    1.781 |   143.74 |   31.390 |  2095.93 |
-|  8192 |     32 |   16 | 131584 |   59.229 |  2212.96 |    2.495 |   205.17 |   61.725 |  2131.79 |
-|  8192 |     32 |   32 | 263168 |  118.449 |  2213.15 |    3.714 |   275.75 |  122.162 |  2154.25 |
-
-
-- `llama-bench`
-
-| model                          |       size |     params | backend    | ngl | n_ubatch | fa | mmap |            test |                  t/s |
-| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | ---: | --------------: | -------------------: |
-| qwen2 7B Q8_0                  |   7.54 GiB |     7.62 B | CUDA       |  99 |     2048 |  1 |    0 |          pp2048 |       2272.74 ± 4.68 |
-| qwen2 7B Q8_0                  |   7.54 GiB |     7.62 B | CUDA       |  99 |     2048 |  1 |    0 |            tg32 |         30.66 ± 0.02 |
-| qwen2 7B Q8_0                  |   7.54 GiB |     7.62 B | CUDA       |  99 |     2048 |  1 |    0 |  pp2048 @ d4096 |       2107.80 ± 9.55 |
-| qwen2 7B Q8_0                  |   7.54 GiB |     7.62 B | CUDA       |  99 |     2048 |  1 |    0 |    tg32 @ d4096 |         29.71 ± 0.05 |
-| qwen2 7B Q8_0                  |   7.54 GiB |     7.62 B | CUDA       |  99 |     2048 |  1 |    0 |  pp2048 @ d8192 |       1937.80 ± 6.75 |
-| qwen2 7B Q8_0                  |   7.54 GiB |     7.62 B | CUDA       |  99 |     2048 |  1 |    0 |    tg32 @ d8192 |         28.86 ± 0.04 |
-| qwen2 7B Q8_0                  |   7.54 GiB |     7.62 B | CUDA       |  99 |     2048 |  1 |    0 | pp2048 @ d16384 |       1641.12 ± 1.78 |
-| qwen2 7B Q8_0                  |   7.54 GiB |     7.62 B | CUDA       |  99 |     2048 |  1 |    0 |   tg32 @ d16384 |         27.24 ± 0.04 |
-| qwen2 7B Q8_0                  |   7.54 GiB |     7.62 B | CUDA       |  99 |     2048 |  1 |    0 | pp2048 @ d32768 |       1296.02 ± 2.67 |
-| qwen2 7B Q8_0                  |   7.54 GiB |     7.62 B | CUDA       |  99 |     2048 |  1 |    0 |   tg32 @ d32768 |         23.78 ± 0.03 |
-
-build: eeee367de (6989)
-
-## ggml-org/gemma-3-4b-it-qat-GGUF
-
-Model: https://huggingface.co/ggml-org/gemma-3-4b-it-qat-GGUF
-
-- `llama-batched-bench`
-
-
-main: n_kv_max = 270336, n_batch = 2048, n_ubatch = 2048, flash_attn = 1, is_pp_shared = 0, n_gpu_layers = -1, n_threads = 20, n_threads_batch = 20
-
-|    PP |     TG |    B |   N_KV |   T_PP s | S_PP t/s |   T_TG s | S_TG t/s |      T s |    S t/s |
-|-------|--------|------|--------|----------|----------|----------|----------|----------|----------|
-|   512 |     32 |    1 |    544 |    0.094 |  5434.73 |    0.394 |    81.21 |    0.488 |  1114.15 |
-|   512 |     32 |    2 |   1088 |    0.168 |  6091.68 |    0.498 |   128.52 |    0.666 |  1633.41 |
-|   512 |     32 |    4 |   2176 |    0.341 |  6010.68 |    0.542 |   236.37 |    0.882 |  2466.43 |
-|   512 |     32 |    8 |   4352 |    0.665 |  6161.46 |    0.678 |   377.74 |    1.342 |  3241.72 |
-|   512 |     32 |   16 |   8704 |    1.323 |  6193.19 |    0.902 |   567.41 |    2.225 |  3911.74 |
-|   512 |     32 |   32 |  17408 |    2.642 |  6202.03 |    1.231 |   832.03 |    3.872 |  4495.36 |
-|  4096 |     32 |    1 |   4128 |    0.701 |  5840.49 |    0.439 |    72.95 |    1.140 |  3621.23 |
-|  4096 |     32 |    2 |   8256 |    1.387 |  5906.82 |    0.574 |   111.48 |    1.961 |  4210.12 |
-|  4096 |     32 |    4 |  16512 |    2.758 |  5940.33 |    0.651 |   196.58 |    3.409 |  4843.33 |
-|  4096 |     32 |    8 |  33024 |    5.491 |  5967.56 |    0.876 |   292.40 |    6.367 |  5187.12 |
-|  4096 |     32 |   16 |  66048 |   10.978 |  5969.58 |    1.275 |   401.69 |   12.253 |  5390.38 |
-|  4096 |     32 |   32 | 132096 |   21.944 |  5972.93 |    1.992 |   514.16 |   23.936 |  5518.73 |
-|  8192 |     32 |    1 |   8224 |    1.402 |  5841.91 |    0.452 |    70.73 |    1.855 |  4434.12 |
-|  8192 |     32 |    2 |  16448 |    2.793 |  5865.34 |    0.637 |   100.55 |    3.430 |  4795.51 |
-|  8192 |     32 |    4 |  32896 |    5.564 |  5889.64 |    0.770 |   166.26 |    6.334 |  5193.95 |
-|  8192 |     32 |    8 |  65792 |   11.114 |  5896.44 |    1.122 |   228.07 |   12.237 |  5376.51 |
-|  8192 |     32 |   16 | 131584 |   22.210 |  5901.38 |    1.789 |   286.15 |   24.000 |  5482.74 |
-|  8192 |     32 |   32 | 263168 |   44.382 |  5906.56 |    3.044 |   336.38 |   47.426 |  5549.02 |
-
-
-- `llama-bench`
-
-| model                          |       size |     params | backend    | ngl | n_ubatch | fa | mmap |            test |                  t/s |
-| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | ---: | --------------: | -------------------: |
-| gemma3 4B Q4_0                 |   2.35 GiB |     3.88 B | CUDA       |  99 |     2048 |  1 |    0 |          pp2048 |      5810.04 ± 21.71 |
-| gemma3 4B Q4_0                 |   2.35 GiB |     3.88 B | CUDA       |  99 |     2048 |  1 |    0 |            tg32 |         84.54 ± 0.18 |
-| gemma3 4B Q4_0                 |   2.35 GiB |     3.88 B | CUDA       |  99 |     2048 |  1 |    0 |  pp2048 @ d4096 |       5288.04 ± 3.54 |
-| gemma3 4B Q4_0                 |   2.35 GiB |     3.88 B | CUDA       |  99 |     2048 |  1 |    0 |    tg32 @ d4096 |         78.82 ± 1.37 |
-| gemma3 4B Q4_0                 |   2.35 GiB |     3.88 B | CUDA       |  99 |     2048 |  1 |    0 |  pp2048 @ d8192 |      4960.43 ± 16.64 |
-| gemma3 4B Q4_0                 |   2.35 GiB |     3.88 B | CUDA       |  99 |     2048 |  1 |    0 |    tg32 @ d8192 |         74.13 ± 0.30 |
-| gemma3 4B Q4_0                 |   2.35 GiB |     3.88 B | CUDA       |  99 |     2048 |  1 |    0 | pp2048 @ d16384 |      4495.92 ± 31.11 |
-| gemma3 4B Q4_0                 |   2.35 GiB |     3.88 B | CUDA       |  99 |     2048 |  1 |    0 |   tg32 @ d16384 |         72.37 ± 0.29 |
-| gemma3 4B Q4_0                 |   2.35 GiB |     3.88 B | CUDA       |  99 |     2048 |  1 |    0 | pp2048 @ d32768 |      3746.90 ± 40.01 |
-| gemma3 4B Q4_0                 |   2.35 GiB |     3.88 B | CUDA       |  99 |     2048 |  1 |    0 |   tg32 @ d32768 |         63.02 ± 0.20 |
-
-build: eeee367de (6989)
-

build-xcframework.sh

@@ -422,7 +422,6 @@ echo "Building for iOS devices..."
 cmake -B build-ios-device -G Xcode \
     "${COMMON_CMAKE_ARGS[@]}" \
     -DCMAKE_OSX_DEPLOYMENT_TARGET=${IOS_MIN_OS_VERSION} \
-    -DCMAKE_SYSTEM_NAME=iOS \
     -DCMAKE_OSX_SYSROOT=iphoneos \
     -DCMAKE_OSX_ARCHITECTURES="arm64" \
     -DCMAKE_XCODE_ATTRIBUTE_SUPPORTED_PLATFORMS=iphoneos \
@@ -454,8 +453,6 @@ cmake -B build-visionos -G Xcode \
     -DCMAKE_C_FLAGS="-D_XOPEN_SOURCE=700 ${COMMON_C_FLAGS}" \
     -DCMAKE_CXX_FLAGS="-D_XOPEN_SOURCE=700 ${COMMON_CXX_FLAGS}" \
     -DLLAMA_CURL=OFF \
-    -DLLAMA_HTTPLIB=OFF \
-    -DLLAMA_BUILD_SERVER=OFF \
     -S .
 cmake --build build-visionos --config Release -- -quiet
 
@@ -470,8 +467,6 @@ cmake -B build-visionos-sim -G Xcode \
     -DCMAKE_C_FLAGS="-D_XOPEN_SOURCE=700 ${COMMON_C_FLAGS}" \
     -DCMAKE_CXX_FLAGS="-D_XOPEN_SOURCE=700 ${COMMON_CXX_FLAGS}" \
     -DLLAMA_CURL=OFF \
-    -DLLAMA_HTTPLIB=OFF \
-    -DLLAMA_BUILD_SERVER=OFF \
     -S .
 cmake --build build-visionos-sim --config Release -- -quiet
 

ci/README-MUSA.md

@@ -1,35 +0,0 @@
-## Running MUSA CI in a Docker Container
-
-Assuming `$PWD` is the root of the `llama.cpp` repository, follow these steps to set up and run MUSA CI in a Docker container:
-
-### 1. Create a local directory to store cached models, configuration files and venv:
-
-```bash
-mkdir -p $HOME/llama.cpp/ci-cache
-```
-
-### 2. Create a local directory to store CI run results:
-
-```bash
-mkdir -p $HOME/llama.cpp/ci-results
-```
-
-### 3. Start a Docker container and run the CI:
-
-```bash
-docker run --privileged -it \
-    -v $HOME/llama.cpp/ci-cache:/ci-cache \
-    -v $HOME/llama.cpp/ci-results:/ci-results \
-    -v $PWD:/ws -w /ws \
-    mthreads/musa:rc4.3.0-devel-ubuntu22.04-amd64
-```
-
-Inside the container, execute the following commands:
-
-```bash
-apt update -y && apt install -y bc cmake ccache git python3.10-venv time unzip wget
-git config --global --add safe.directory /ws
-GG_BUILD_MUSA=1 bash ./ci/run.sh /ci-results /ci-cache
-```
-
-This setup ensures that the CI runs within an isolated Docker environment while maintaining cached files and results across runs.

ci/README.md

@@ -1,10 +1,18 @@
 # CI
 
-This CI implements heavy-duty workflows that run on self-hosted runners. Typically the purpose of these workflows is to
-cover hardware configurations that are not available from Github-hosted runners and/or require more computational
-resource than normally available.
+In addition to [Github Actions](https://github.com/ggml-org/llama.cpp/actions) `llama.cpp` uses a custom CI framework:
 
-It is a good practice, before publishing changes to execute the full CI locally on your machine. For example:
+https://github.com/ggml-org/ci
+
+It monitors the `master` branch for new commits and runs the
+[ci/run.sh](https://github.com/ggml-org/llama.cpp/blob/master/ci/run.sh) script on dedicated cloud instances. This allows us
+to execute heavier workloads compared to just using Github Actions. Also with time, the cloud instances will be scaled
+to cover various hardware architectures, including GPU and Apple Silicon instances.
+
+Collaborators can optionally trigger the CI run by adding the `ggml-ci` keyword to their commit message.
+Only the branches of this repo are monitored for this keyword.
+
+It is a good practice, before publishing changes to execute the full CI locally on your machine:
 
 ```bash
 mkdir tmp
@@ -21,13 +29,40 @@ GG_BUILD_SYCL=1 bash ./ci/run.sh ./tmp/results ./tmp/mnt
 
 # with MUSA support
 GG_BUILD_MUSA=1 bash ./ci/run.sh ./tmp/results ./tmp/mnt
-
-# etc.
 ```
 
-# Adding self-hosted runners
+## Running MUSA CI in a Docker Container
 
-- Add a self-hosted `ggml-ci` workflow to [[.github/workflows/build.yml]] with an appropriate label
-- Request a runner token from `ggml-org` (for example, via a comment in the PR or email)
-- Set-up a machine using the received token ([docs](https://docs.github.com/en/actions/how-tos/manage-runners/self-hosted-runners/add-runners))
-- Optionally update [ci/run.sh](https://github.com/ggml-org/llama.cpp/blob/master/ci/run.sh) to build and run on the target platform by gating the implementation with a `GG_BUILD_...` env
+Assuming `$PWD` is the root of the `llama.cpp` repository, follow these steps to set up and run MUSA CI in a Docker container:
+
+### 1. Create a local directory to store cached models, configuration files and venv:
+
+```bash
+mkdir -p $HOME/llama.cpp/ci-cache
+```
+
+### 2. Create a local directory to store CI run results:
+
+```bash
+mkdir -p $HOME/llama.cpp/ci-results
+```
+
+### 3. Start a Docker container and run the CI:
+
+```bash
+docker run --privileged -it \
+    -v $HOME/llama.cpp/ci-cache:/ci-cache \
+    -v $HOME/llama.cpp/ci-results:/ci-results \
+    -v $PWD:/ws -w /ws \
+    mthreads/musa:rc4.2.0-devel-ubuntu22.04-amd64
+```
+
+Inside the container, execute the following commands:
+
+```bash
+apt update -y && apt install -y bc cmake ccache git python3.10-venv time unzip wget
+git config --global --add safe.directory /ws
+GG_BUILD_MUSA=1 bash ./ci/run.sh /ci-results /ci-cache
+```
+
+This setup ensures that the CI runs within an isolated Docker environment while maintaining cached files and results across runs.

ci/run.sh

@@ -22,9 +22,6 @@
 # # with MUSA support
 # GG_BUILD_MUSA=1 bash ./ci/run.sh ./tmp/results ./tmp/mnt
 #
-# # with KLEIDIAI support
-# GG_BUILD_KLEIDIAI=1 bash ./ci/run.sh ./tmp/results ./tmp/mnt
-#
 
 if [ -z "$2" ]; then
     echo "usage: $0 <output-dir> <mnt-dir>"
@@ -37,18 +34,18 @@ mkdir -p "$2"
 OUT=$(realpath "$1")
 MNT=$(realpath "$2")
 
-rm -f $OUT/*.log
-rm -f $OUT/*.exit
-rm -f $OUT/*.md
+rm -f "$OUT/*.log"
+rm -f "$OUT/*.exit"
+rm -f "$OUT/*.md"
 
 sd=`dirname $0`
 cd $sd/../
 SRC=`pwd`
 
-CMAKE_EXTRA="-DLLAMA_FATAL_WARNINGS=${LLAMA_FATAL_WARNINGS:-ON} -DLLAMA_CURL=ON -DGGML_SCHED_NO_REALLOC=ON"
+CMAKE_EXTRA="-DLLAMA_FATAL_WARNINGS=ON -DLLAMA_CURL=ON"
 
 if [ ! -z ${GG_BUILD_METAL} ]; then
-    CMAKE_EXTRA="${CMAKE_EXTRA} -DGGML_METAL=ON"
+    CMAKE_EXTRA="${CMAKE_EXTRA} -DGGML_METAL=ON -DGGML_METAL_USE_BF16=ON"
 fi
 
 if [ ! -z ${GG_BUILD_CUDA} ]; then
@@ -68,16 +65,6 @@ if [ ! -z ${GG_BUILD_CUDA} ]; then
     fi
 fi
 
-if [ ! -z ${GG_BUILD_ROCM} ]; then
-    CMAKE_EXTRA="${CMAKE_EXTRA} -DGGML_HIP=ON"
-    if [ -z ${GG_BUILD_AMDGPU_TARGETS} ]; then
-        echo "Missing GG_BUILD_AMDGPU_TARGETS, please set it to your GPU architecture (e.g. gfx90a, gfx1100, etc.)"
-        exit 1
-    fi
-
-    CMAKE_EXTRA="${CMAKE_EXTRA} -DGPU_TARGETS=${GG_BUILD_AMDGPU_TARGETS}"
-fi
-
 if [ ! -z ${GG_BUILD_SYCL} ]; then
     if [ -z ${ONEAPI_ROOT} ]; then
         echo "Not detected ONEAPI_ROOT, please install oneAPI base toolkit and enable it by:"
@@ -95,12 +82,6 @@ fi
 
 if [ ! -z ${GG_BUILD_VULKAN} ]; then
     CMAKE_EXTRA="${CMAKE_EXTRA} -DGGML_VULKAN=1"
-
-    # if on Mac, disable METAL
-    if [[ "$OSTYPE" == "darwin"* ]]; then
-        CMAKE_EXTRA="${CMAKE_EXTRA} -DGGML_METAL=OFF -DGGML_BLAS=OFF"
-    fi
-
 fi
 
 if [ ! -z ${GG_BUILD_WEBGPU} ]; then
@@ -112,45 +93,6 @@ if [ ! -z ${GG_BUILD_MUSA} ]; then
     MUSA_ARCH=${MUSA_ARCH:-21}
     CMAKE_EXTRA="${CMAKE_EXTRA} -DGGML_MUSA=ON -DMUSA_ARCHITECTURES=${MUSA_ARCH}"
 fi
-
-if [ ! -z ${GG_BUILD_NO_SVE} ]; then
-    # arm 9 and newer enables sve by default, adjust these flags depending on the cpu used
-    CMAKE_EXTRA="${CMAKE_EXTRA} -DGGML_NATIVE=OFF -DGGML_CPU_ARM_ARCH=armv8.5-a+fp16+i8mm"
-fi
-
-if [ -n "${GG_BUILD_KLEIDIAI}" ]; then
-    echo ">>===== Enabling KleidiAI support"
-
-    CANDIDATES=(
-        "armv9-a+dotprod+i8mm+sve2"
-        "armv9-a+dotprod+i8mm"
-        "armv8.6-a+dotprod+i8mm"
-        "armv8.2-a+dotprod"
-    )
-    CPU=""
-
-    for cpu in "${CANDIDATES[@]}"; do
-        if echo 'int main(){}' | ${CXX:-c++} -march="$cpu" -x c++ - -c -o /dev/null >/dev/null 2>&1; then
-            CPU="$cpu"
-            break
-        fi
-    done
-
-    if [ -z "$CPU" ]; then
-        echo "ERROR: None of the required ARM baselines (armv9/armv8.6/armv8.2 + dotprod) are supported by this compiler."
-        exit 1
-    fi
-
-    echo ">>===== Using ARM baseline: ${CPU}"
-
-    CMAKE_EXTRA="${CMAKE_EXTRA:+$CMAKE_EXTRA } \
-        -DGGML_NATIVE=OFF \
-        -DGGML_CPU_KLEIDIAI=ON \
-        -DGGML_CPU_AARCH64=ON \
-        -DGGML_CPU_ARM_ARCH=${CPU} \
-        -DBUILD_SHARED_LIBS=OFF"
-fi
-
 ## helpers
 
 # download a file if it does not exist or if it is outdated
@@ -164,7 +106,7 @@ function gg_wget {
     cd $out
 
     # should not re-download if file is the same
-    wget -nv -c -N $url
+    wget -nv -N $url
 
     cd $cwd
 }
@@ -208,7 +150,7 @@ function gg_run_ctest_debug {
     (time cmake -DCMAKE_BUILD_TYPE=Debug ${CMAKE_EXTRA} .. ) 2>&1 | tee -a $OUT/${ci}-cmake.log
     (time make -j$(nproc)                                  ) 2>&1 | tee -a $OUT/${ci}-make.log
 
-    (time ctest --output-on-failure -L main -E "test-opt|test-backend-ops" ) 2>&1 | tee -a $OUT/${ci}-ctest.log
+    (time ctest --output-on-failure -L main -E test-opt ) 2>&1 | tee -a $OUT/${ci}-ctest.log
 
     set +e
 }
@@ -258,9 +200,33 @@ function gg_sum_ctest_release {
     gg_printf '```\n'
 }
 
-# test_scripts
+# test_scripts_debug
 
-function gg_run_test_scripts {
+function gg_run_test_scripts_debug {
+    cd ${SRC}
+
+    set -e
+
+    (cd ./tools/gguf-split && time bash tests.sh "$SRC/build-ci-debug/bin" "$MNT/models") 2>&1 | tee -a $OUT/${ci}-scripts.log
+    (cd ./tools/quantize   && time bash tests.sh "$SRC/build-ci-debug/bin" "$MNT/models") 2>&1 | tee -a $OUT/${ci}-scripts.log
+
+    set +e
+}
+
+function gg_sum_test_scripts_debug {
+    gg_printf '### %s\n\n' "${ci}"
+
+    gg_printf 'Runs test scripts in debug mode\n'
+    gg_printf '- status: %s\n' "$(cat $OUT/${ci}.exit)"
+    gg_printf '```\n'
+    gg_printf '%s\n' "$(cat $OUT/${ci}-scripts.log)"
+    gg_printf '```\n'
+    gg_printf '\n'
+}
+
+# test_scripts_release
+
+function gg_run_test_scripts_release {
     cd ${SRC}
 
     set -e
@@ -271,10 +237,10 @@ function gg_run_test_scripts {
     set +e
 }
 
-function gg_sum_test_scripts {
+function gg_sum_test_scripts_release {
     gg_printf '### %s\n\n' "${ci}"
 
-    gg_printf 'Runs test scripts\n'
+    gg_printf 'Runs test scripts in release mode\n'
     gg_printf '- status: %s\n' "$(cat $OUT/${ci}.exit)"
     gg_printf '```\n'
     gg_printf '%s\n' "$(cat $OUT/${ci}-scripts.log)"
@@ -283,9 +249,15 @@ function gg_sum_test_scripts {
 }
 
 function gg_get_model {
-    local gguf_0="$MNT/models/qwen3/0.6B/ggml-model-f16.gguf"
+    local gguf_0="$MNT/models/pythia/1.4B/ggml-model-f16.gguf"
+    local gguf_1="$MNT/models/pythia/2.8B/ggml-model-f16.gguf"
+    local gguf_2="$MNT/models/open-llama/7B-v2/ggml-model-f16.gguf"
     if [[ -s $gguf_0 ]]; then
         echo -n "$gguf_0"
+    elif [[ -s $gguf_1 ]]; then
+        echo -n "$gguf_1"
+    elif [[ -s $gguf_2 ]]; then
+        echo -n "$gguf_2"
     else
         echo >&2 "No model found. Can't run gg_run_ctest_with_model."
         exit 1
@@ -298,9 +270,7 @@ function gg_run_ctest_with_model_debug {
     local model; model=$(gg_get_model)
     cd build-ci-debug
     set -e
-
     (LLAMACPP_TEST_MODELFILE="$model" time ctest --output-on-failure -L model) 2>&1 | tee -a $OUT/${ci}-ctest.log
-
     set +e
     cd ..
 }
@@ -311,15 +281,7 @@ function gg_run_ctest_with_model_release {
     local model; model=$(gg_get_model)
     cd build-ci-release
     set -e
-
     (LLAMACPP_TEST_MODELFILE="$model" time ctest --output-on-failure -L model) 2>&1 | tee -a $OUT/${ci}-ctest.log
-
-    # test memory leaks
-    #if [[ ! -z ${GG_BUILD_METAL} ]]; then
-    #    # TODO: this hangs for some reason ...
-    #    (time leaks -quiet -atExit -- ./bin/test-thread-safety -m $model --parallel 2 -t 2 -p "hello") 2>&1 | tee -a $OUT/${ci}-leaks.log
-    #fi
-
     set +e
     cd ..
 }
@@ -344,22 +306,24 @@ function gg_sum_ctest_with_model_release {
     gg_printf '```\n'
 }
 
-# qwen3_0_6b
+# open_llama_7b_v2
 
-function gg_run_qwen3_0_6b {
+function gg_run_open_llama_7b_v2 {
     cd ${SRC}
 
-    gg_wget models-mnt/qwen3/0.6B/ https://huggingface.co/Qwen/Qwen3-0.6B-Base/raw/main/config.json
-    gg_wget models-mnt/qwen3/0.6B/ https://huggingface.co/Qwen/Qwen3-0.6B-Base/raw/main/tokenizer.json
-    gg_wget models-mnt/qwen3/0.6B/ https://huggingface.co/Qwen/Qwen3-0.6B-Base/raw/main/tokenizer_config.json
-   #gg_wget models-mnt/qwen3/0.6B/ https://huggingface.co/Qwen/Qwen3-0.6B-Base/raw/main/special_tokens_map.json
-    gg_wget models-mnt/qwen3/0.6B/ https://huggingface.co/Qwen/Qwen3-0.6B-Base/resolve/main/model.safetensors
-
+    gg_wget models-mnt/open-llama/7B-v2/ https://huggingface.co/openlm-research/open_llama_7b_v2/raw/main/config.json
+    gg_wget models-mnt/open-llama/7B-v2/ https://huggingface.co/openlm-research/open_llama_7b_v2/resolve/main/tokenizer.model
+    gg_wget models-mnt/open-llama/7B-v2/ https://huggingface.co/openlm-research/open_llama_7b_v2/raw/main/tokenizer_config.json
+    gg_wget models-mnt/open-llama/7B-v2/ https://huggingface.co/openlm-research/open_llama_7b_v2/raw/main/special_tokens_map.json
+    gg_wget models-mnt/open-llama/7B-v2/ https://huggingface.co/openlm-research/open_llama_7b_v2/raw/main/pytorch_model.bin.index.json
+    gg_wget models-mnt/open-llama/7B-v2/ https://huggingface.co/openlm-research/open_llama_7b_v2/resolve/main/pytorch_model-00001-of-00002.bin
+    gg_wget models-mnt/open-llama/7B-v2/ https://huggingface.co/openlm-research/open_llama_7b_v2/resolve/main/pytorch_model-00002-of-00002.bin
+    gg_wget models-mnt/open-llama/7B-v2/ https://huggingface.co/openlm-research/open_llama_7b_v2/raw/main/generation_config.json
 
     gg_wget models-mnt/wikitext/ https://huggingface.co/datasets/ggml-org/ci/resolve/main/wikitext-2-raw-v1.zip
     unzip -o models-mnt/wikitext/wikitext-2-raw-v1.zip -d models-mnt/wikitext/
 
-    path_models="../models-mnt/qwen3/0.6B"
+    path_models="../models-mnt/open-llama/7B-v2"
     path_wiki="../models-mnt/wikitext/wikitext-2-raw"
 
     rm -rf build-ci-release && mkdir build-ci-release && cd build-ci-release
@@ -369,11 +333,9 @@ function gg_run_qwen3_0_6b {
     (time cmake -DCMAKE_BUILD_TYPE=Release ${CMAKE_EXTRA} .. ) 2>&1 | tee -a $OUT/${ci}-cmake.log
     (time make -j$(nproc)                                    ) 2>&1 | tee -a $OUT/${ci}-make.log
 
-    python3 ../convert_hf_to_gguf.py ${path_models} --outfile ${path_models}/ggml-model-f16.gguf  --outtype f16
-    python3 ../convert_hf_to_gguf.py ${path_models} --outfile ${path_models}/ggml-model-bf16.gguf --outtype bf16
+    python3 ../examples/convert_legacy_llama.py ${path_models} --outfile ${path_models}/ggml-model-f16.gguf
 
     model_f16="${path_models}/ggml-model-f16.gguf"
-    model_bf16="${path_models}/ggml-model-bf16.gguf"
     model_q8_0="${path_models}/ggml-model-q8_0.gguf"
     model_q4_0="${path_models}/ggml-model-q4_0.gguf"
     model_q4_1="${path_models}/ggml-model-q4_1.gguf"
@@ -387,53 +349,179 @@ function gg_run_qwen3_0_6b {
 
     wiki_test="${path_wiki}/wiki.test.raw"
 
-    ./bin/llama-quantize ${model_bf16} ${model_q8_0} q8_0 $(nproc)
-    ./bin/llama-quantize ${model_bf16} ${model_q4_0} q4_0 $(nproc)
-    ./bin/llama-quantize ${model_bf16} ${model_q4_1} q4_1 $(nproc)
-    ./bin/llama-quantize ${model_bf16} ${model_q5_0} q5_0 $(nproc)
-    ./bin/llama-quantize ${model_bf16} ${model_q5_1} q5_1 $(nproc)
-    ./bin/llama-quantize ${model_bf16} ${model_q2_k} q2_k $(nproc)
-    ./bin/llama-quantize ${model_bf16} ${model_q3_k} q3_k $(nproc)
-    ./bin/llama-quantize ${model_bf16} ${model_q4_k} q4_k $(nproc)
-    ./bin/llama-quantize ${model_bf16} ${model_q5_k} q5_k $(nproc)
-    ./bin/llama-quantize ${model_bf16} ${model_q6_k} q6_k $(nproc)
+    ./bin/llama-quantize ${model_f16} ${model_q8_0} q8_0
+    ./bin/llama-quantize ${model_f16} ${model_q4_0} q4_0
+    ./bin/llama-quantize ${model_f16} ${model_q4_1} q4_1
+    ./bin/llama-quantize ${model_f16} ${model_q5_0} q5_0
+    ./bin/llama-quantize ${model_f16} ${model_q5_1} q5_1
+    ./bin/llama-quantize ${model_f16} ${model_q2_k} q2_k
+    ./bin/llama-quantize ${model_f16} ${model_q3_k} q3_k
+    ./bin/llama-quantize ${model_f16} ${model_q4_k} q4_k
+    ./bin/llama-quantize ${model_f16} ${model_q5_k} q5_k
+    ./bin/llama-quantize ${model_f16} ${model_q6_k} q6_k
 
-    (time ./bin/llama-fit-params --model ${model_f16} 2>&1 | tee -a $OUT/${ci}-fp-f16.log)
+    (time ./bin/llama-cli -no-cnv --model ${model_f16}  -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-f16.log
+    (time ./bin/llama-cli -no-cnv --model ${model_q8_0} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q8_0.log
+    (time ./bin/llama-cli -no-cnv --model ${model_q4_0} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q4_0.log
+    (time ./bin/llama-cli -no-cnv --model ${model_q4_1} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q4_1.log
+    (time ./bin/llama-cli -no-cnv --model ${model_q5_0} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q5_0.log
+    (time ./bin/llama-cli -no-cnv --model ${model_q5_1} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q5_1.log
+    (time ./bin/llama-cli -no-cnv --model ${model_q2_k} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q2_k.log
+    (time ./bin/llama-cli -no-cnv --model ${model_q3_k} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q3_k.log
+    (time ./bin/llama-cli -no-cnv --model ${model_q4_k} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q4_k.log
+    (time ./bin/llama-cli -no-cnv --model ${model_q5_k} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q5_k.log
+    (time ./bin/llama-cli -no-cnv --model ${model_q6_k} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q6_k.log
 
-    (time ./bin/llama-completion -no-cnv --model ${model_f16}  -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-f16.log
-    (time ./bin/llama-completion -no-cnv --model ${model_bf16} -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-bf16.log
-    (time ./bin/llama-completion -no-cnv --model ${model_q8_0} -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q8_0.log
-    (time ./bin/llama-completion -no-cnv --model ${model_q4_0} -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q4_0.log
-    (time ./bin/llama-completion -no-cnv --model ${model_q4_1} -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q4_1.log
-    (time ./bin/llama-completion -no-cnv --model ${model_q5_0} -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q5_0.log
-    (time ./bin/llama-completion -no-cnv --model ${model_q5_1} -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q5_1.log
-    (time ./bin/llama-completion -no-cnv --model ${model_q2_k} -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q2_k.log
-    (time ./bin/llama-completion -no-cnv --model ${model_q3_k} -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q3_k.log
-    (time ./bin/llama-completion -no-cnv --model ${model_q4_k} -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q4_k.log
-    (time ./bin/llama-completion -no-cnv --model ${model_q5_k} -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q5_k.log
-    (time ./bin/llama-completion -no-cnv --model ${model_q6_k} -ngl 99 -c 1024 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q6_k.log
+    (time ./bin/llama-perplexity --model ${model_f16}  -f ${wiki_test} -t 1 -ngl 99 -c 2048 -b 512 --chunks 4 ) 2>&1 | tee -a $OUT/${ci}-tg-f16.log
+    (time ./bin/llama-perplexity --model ${model_q8_0} -f ${wiki_test} -t 1 -ngl 99 -c 2048 -b 512 --chunks 4 ) 2>&1 | tee -a $OUT/${ci}-tg-q8_0.log
+    (time ./bin/llama-perplexity --model ${model_q4_0} -f ${wiki_test} -t 1 -ngl 99 -c 2048 -b 512 --chunks 4 ) 2>&1 | tee -a $OUT/${ci}-tg-q4_0.log
+    (time ./bin/llama-perplexity --model ${model_q4_1} -f ${wiki_test} -t 1 -ngl 99 -c 2048 -b 512 --chunks 4 ) 2>&1 | tee -a $OUT/${ci}-tg-q4_1.log
+    (time ./bin/llama-perplexity --model ${model_q5_0} -f ${wiki_test} -t 1 -ngl 99 -c 2048 -b 512 --chunks 4 ) 2>&1 | tee -a $OUT/${ci}-tg-q5_0.log
+    (time ./bin/llama-perplexity --model ${model_q5_1} -f ${wiki_test} -t 1 -ngl 99 -c 2048 -b 512 --chunks 4 ) 2>&1 | tee -a $OUT/${ci}-tg-q5_1.log
+    (time ./bin/llama-perplexity --model ${model_q2_k} -f ${wiki_test} -t 1 -ngl 99 -c 2048 -b 512 --chunks 4 ) 2>&1 | tee -a $OUT/${ci}-tg-q2_k.log
+    (time ./bin/llama-perplexity --model ${model_q3_k} -f ${wiki_test} -t 1 -ngl 99 -c 2048 -b 512 --chunks 4 ) 2>&1 | tee -a $OUT/${ci}-tg-q3_k.log
+    (time ./bin/llama-perplexity --model ${model_q4_k} -f ${wiki_test} -t 1 -ngl 99 -c 2048 -b 512 --chunks 4 ) 2>&1 | tee -a $OUT/${ci}-tg-q4_k.log
+    (time ./bin/llama-perplexity --model ${model_q5_k} -f ${wiki_test} -t 1 -ngl 99 -c 2048 -b 512 --chunks 4 ) 2>&1 | tee -a $OUT/${ci}-tg-q5_k.log
+    (time ./bin/llama-perplexity --model ${model_q6_k} -f ${wiki_test} -t 1 -ngl 99 -c 2048 -b 512 --chunks 4 ) 2>&1 | tee -a $OUT/${ci}-tg-q6_k.log
 
-    (time ./bin/llama-perplexity --model ${model_f16}  -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-f16.log
-    if [ -z ${GG_BUILD_NO_BF16} ]; then
-        (time ./bin/llama-perplexity --model ${model_bf16} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-bf16.log
-    fi
-    (time ./bin/llama-perplexity --model ${model_q8_0} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-q8_0.log
-    (time ./bin/llama-perplexity --model ${model_q4_0} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-q4_0.log
-    (time ./bin/llama-perplexity --model ${model_q4_1} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-q4_1.log
-    (time ./bin/llama-perplexity --model ${model_q5_0} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-q5_0.log
-    (time ./bin/llama-perplexity --model ${model_q5_1} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-q5_1.log
-    (time ./bin/llama-perplexity --model ${model_q2_k} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-q2_k.log
-    (time ./bin/llama-perplexity --model ${model_q3_k} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-q3_k.log
-    (time ./bin/llama-perplexity --model ${model_q4_k} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-q4_k.log
-    (time ./bin/llama-perplexity --model ${model_q5_k} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-q5_k.log
-    (time ./bin/llama-perplexity --model ${model_q6_k} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-tg-q6_k.log
+    (time ./bin/llama-imatrix --model ${model_f16} -f ${wiki_test} -t 1 -ngl 99 -c 2048 -b 512 --chunks 4 ) 2>&1 | tee -a $OUT/${ci}-imatrix.log
 
-    (time ./bin/llama-imatrix --model ${model_f16} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-imatrix.log
-
-    (time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 10 -c 1024 -fa off --no-op-offload) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
-    (time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 10 -c 1024 -fa on  --no-op-offload) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
-    (time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 99 -c 1024 -fa off                ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
-    (time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 99 -c 1024 -fa on                 ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
+    (time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 10 -c 0     ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
+    (time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 10 -c 0 -fa ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
+    (time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 99 -c 0     ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
+    (time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 99 -c 0 -fa ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
+
+    function check_ppl {
+        qnt="$1"
+        ppl=$(echo "$2" | grep -oE "[0-9]+\.[0-9]+" | tail -n 1)
+
+        if [ $(echo "$ppl > 20.0" | bc) -eq 1 ]; then
+            printf '  - %s @ %s (FAIL: ppl > 20.0)\n' "$qnt" "$ppl"
+            return 20
+        fi
+
+        printf '  - %s @ %s OK\n' "$qnt" "$ppl"
+        return 0
+    }
+
+    check_ppl "f16"  "$(cat $OUT/${ci}-tg-f16.log  | grep "^\[1\]")" | tee -a $OUT/${ci}-ppl.log
+    check_ppl "q8_0" "$(cat $OUT/${ci}-tg-q8_0.log | grep "^\[1\]")" | tee -a $OUT/${ci}-ppl.log
+    check_ppl "q4_0" "$(cat $OUT/${ci}-tg-q4_0.log | grep "^\[1\]")" | tee -a $OUT/${ci}-ppl.log
+    check_ppl "q4_1" "$(cat $OUT/${ci}-tg-q4_1.log | grep "^\[1\]")" | tee -a $OUT/${ci}-ppl.log
+    check_ppl "q5_0" "$(cat $OUT/${ci}-tg-q5_0.log | grep "^\[1\]")" | tee -a $OUT/${ci}-ppl.log
+    check_ppl "q5_1" "$(cat $OUT/${ci}-tg-q5_1.log | grep "^\[1\]")" | tee -a $OUT/${ci}-ppl.log
+    check_ppl "q2_k" "$(cat $OUT/${ci}-tg-q2_k.log | grep "^\[1\]")" | tee -a $OUT/${ci}-ppl.log
+    check_ppl "q3_k" "$(cat $OUT/${ci}-tg-q3_k.log | grep "^\[1\]")" | tee -a $OUT/${ci}-ppl.log
+    check_ppl "q4_k" "$(cat $OUT/${ci}-tg-q4_k.log | grep "^\[1\]")" | tee -a $OUT/${ci}-ppl.log
+    check_ppl "q5_k" "$(cat $OUT/${ci}-tg-q5_k.log | grep "^\[1\]")" | tee -a $OUT/${ci}-ppl.log
+    check_ppl "q6_k" "$(cat $OUT/${ci}-tg-q6_k.log | grep "^\[1\]")" | tee -a $OUT/${ci}-ppl.log
+
+    cat $OUT/${ci}-imatrix.log | grep "Final" >> $OUT/${ci}-imatrix-sum.log
+
+    set +e
+}
+
+function gg_sum_open_llama_7b_v2 {
+    gg_printf '### %s\n\n' "${ci}"
+
+    gg_printf 'OpenLLaMA 7B-v2:\n'
+    gg_printf '- status: %s\n' "$(cat $OUT/${ci}.exit)"
+    gg_printf '- perplexity:\n%s\n' "$(cat $OUT/${ci}-ppl.log)"
+    gg_printf '- imatrix:\n```\n%s\n```\n' "$(cat $OUT/${ci}-imatrix-sum.log)"
+    gg_printf '- f16: \n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-f16.log)"
+    gg_printf '- q8_0:\n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-q8_0.log)"
+    gg_printf '- q4_0:\n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-q4_0.log)"
+    gg_printf '- q4_1:\n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-q4_1.log)"
+    gg_printf '- q5_0:\n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-q5_0.log)"
+    gg_printf '- q5_1:\n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-q5_1.log)"
+    gg_printf '- q2_k:\n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-q2_k.log)"
+    gg_printf '- q3_k:\n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-q3_k.log)"
+    gg_printf '- q4_k:\n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-q4_k.log)"
+    gg_printf '- q5_k:\n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-q5_k.log)"
+    gg_printf '- q6_k:\n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-q6_k.log)"
+    gg_printf '- save-load-state: \n```\n%s\n```\n' "$(cat $OUT/${ci}-save-load-state.log)"
+}
+
+# pythia_1.4b
+
+function gg_run_pythia_1_4b {
+    cd ${SRC}
+
+    gg_wget models-mnt/pythia/1.4B/ https://huggingface.co/EleutherAI/pythia-1.4b/raw/main/config.json
+    gg_wget models-mnt/pythia/1.4B/ https://huggingface.co/EleutherAI/pythia-1.4b/raw/main/tokenizer.json
+    gg_wget models-mnt/pythia/1.4B/ https://huggingface.co/EleutherAI/pythia-1.4b/raw/main/tokenizer_config.json
+    gg_wget models-mnt/pythia/1.4B/ https://huggingface.co/EleutherAI/pythia-1.4b/raw/main/special_tokens_map.json
+    gg_wget models-mnt/pythia/1.4B/ https://huggingface.co/EleutherAI/pythia-1.4b/resolve/main/pytorch_model.bin
+
+    gg_wget models-mnt/wikitext/ https://huggingface.co/datasets/ggml-org/ci/resolve/main/wikitext-2-raw-v1.zip
+    unzip -o models-mnt/wikitext/wikitext-2-raw-v1.zip -d models-mnt/wikitext/
+    head -n 60 models-mnt/wikitext/wikitext-2-raw/wiki.test.raw > models-mnt/wikitext/wikitext-2-raw/wiki.test-60.raw
+
+    path_models="../models-mnt/pythia/1.4B"
+    path_wiki="../models-mnt/wikitext/wikitext-2-raw"
+
+    rm -rf build-ci-release && mkdir build-ci-release && cd build-ci-release
+
+    set -e
+
+    (time cmake -DCMAKE_BUILD_TYPE=Release ${CMAKE_EXTRA} .. ) 2>&1 | tee -a $OUT/${ci}-cmake.log
+    (time make -j$(nproc)                                    ) 2>&1 | tee -a $OUT/${ci}-make.log
+
+    python3 ../convert_hf_to_gguf.py ${path_models} --outfile ${path_models}/ggml-model-f16.gguf
+
+    model_f16="${path_models}/ggml-model-f16.gguf"
+    model_q8_0="${path_models}/ggml-model-q8_0.gguf"
+    model_q4_0="${path_models}/ggml-model-q4_0.gguf"
+    model_q4_1="${path_models}/ggml-model-q4_1.gguf"
+    model_q5_0="${path_models}/ggml-model-q5_0.gguf"
+    model_q5_1="${path_models}/ggml-model-q5_1.gguf"
+    model_q2_k="${path_models}/ggml-model-q2_k.gguf"
+    model_q3_k="${path_models}/ggml-model-q3_k.gguf"
+    model_q4_k="${path_models}/ggml-model-q4_k.gguf"
+    model_q5_k="${path_models}/ggml-model-q5_k.gguf"
+    model_q6_k="${path_models}/ggml-model-q6_k.gguf"
+
+    wiki_test_60="${path_wiki}/wiki.test-60.raw"
+
+    ./bin/llama-quantize ${model_f16} ${model_q8_0} q8_0
+    ./bin/llama-quantize ${model_f16} ${model_q4_0} q4_0
+    ./bin/llama-quantize ${model_f16} ${model_q4_1} q4_1
+    ./bin/llama-quantize ${model_f16} ${model_q5_0} q5_0
+    ./bin/llama-quantize ${model_f16} ${model_q5_1} q5_1
+    ./bin/llama-quantize ${model_f16} ${model_q2_k} q2_k
+    ./bin/llama-quantize ${model_f16} ${model_q3_k} q3_k
+    ./bin/llama-quantize ${model_f16} ${model_q4_k} q4_k
+    ./bin/llama-quantize ${model_f16} ${model_q5_k} q5_k
+    ./bin/llama-quantize ${model_f16} ${model_q6_k} q6_k
+
+    (time ./bin/llama-cli -no-cnv --model ${model_f16}  -ngl 99 -c 0 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-f16.log
+    (time ./bin/llama-cli -no-cnv --model ${model_q8_0} -ngl 99 -c 0 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q8_0.log
+    (time ./bin/llama-cli -no-cnv --model ${model_q4_0} -ngl 99 -c 0 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q4_0.log
+    (time ./bin/llama-cli -no-cnv --model ${model_q4_1} -ngl 99 -c 0 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q4_1.log
+    (time ./bin/llama-cli -no-cnv --model ${model_q5_0} -ngl 99 -c 0 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q5_0.log
+    (time ./bin/llama-cli -no-cnv --model ${model_q5_1} -ngl 99 -c 0 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q5_1.log
+    (time ./bin/llama-cli -no-cnv --model ${model_q2_k} -ngl 99 -c 0 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q2_k.log
+    (time ./bin/llama-cli -no-cnv --model ${model_q3_k} -ngl 99 -c 0 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q3_k.log
+    (time ./bin/llama-cli -no-cnv --model ${model_q4_k} -ngl 99 -c 0 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q4_k.log
+    (time ./bin/llama-cli -no-cnv --model ${model_q5_k} -ngl 99 -c 0 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q5_k.log
+    (time ./bin/llama-cli -no-cnv --model ${model_q6_k} -ngl 99 -c 0 -s 1234 -n 64 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q6_k.log
+
+    (time ./bin/llama-perplexity --model ${model_f16}  -f ${wiki_test_60} -ngl 99 -c 128 -b 128 --chunks 1 ) 2>&1 | tee -a $OUT/${ci}-tg-f16.log
+    (time ./bin/llama-perplexity --model ${model_q8_0} -f ${wiki_test_60} -ngl 99 -c 128 -b 128 --chunks 1 ) 2>&1 | tee -a $OUT/${ci}-tg-q8_0.log
+    (time ./bin/llama-perplexity --model ${model_q4_0} -f ${wiki_test_60} -ngl 99 -c 128 -b 128 --chunks 1 ) 2>&1 | tee -a $OUT/${ci}-tg-q4_0.log
+    (time ./bin/llama-perplexity --model ${model_q4_1} -f ${wiki_test_60} -ngl 99 -c 128 -b 128 --chunks 1 ) 2>&1 | tee -a $OUT/${ci}-tg-q4_1.log
+    (time ./bin/llama-perplexity --model ${model_q5_0} -f ${wiki_test_60} -ngl 99 -c 128 -b 128 --chunks 1 ) 2>&1 | tee -a $OUT/${ci}-tg-q5_0.log
+    (time ./bin/llama-perplexity --model ${model_q5_1} -f ${wiki_test_60} -ngl 99 -c 128 -b 128 --chunks 1 ) 2>&1 | tee -a $OUT/${ci}-tg-q5_1.log
+    (time ./bin/llama-perplexity --model ${model_q2_k} -f ${wiki_test_60} -ngl 99 -c 128 -b 128 --chunks 1 ) 2>&1 | tee -a $OUT/${ci}-tg-q2_k.log
+    (time ./bin/llama-perplexity --model ${model_q3_k} -f ${wiki_test_60} -ngl 99 -c 128 -b 128 --chunks 1 ) 2>&1 | tee -a $OUT/${ci}-tg-q3_k.log
+    (time ./bin/llama-perplexity --model ${model_q4_k} -f ${wiki_test_60} -ngl 99 -c 128 -b 128 --chunks 1 ) 2>&1 | tee -a $OUT/${ci}-tg-q4_k.log
+    (time ./bin/llama-perplexity --model ${model_q5_k} -f ${wiki_test_60} -ngl 99 -c 128 -b 128 --chunks 1 ) 2>&1 | tee -a $OUT/${ci}-tg-q5_k.log
+    (time ./bin/llama-perplexity --model ${model_q6_k} -f ${wiki_test_60} -ngl 99 -c 128 -b 128 --chunks 1 ) 2>&1 | tee -a $OUT/${ci}-tg-q6_k.log
+
+    (time ./bin/llama-imatrix --model ${model_f16} -f ${wiki_test_60} -ngl 99 -c 128 -b 128 --chunks 1 ) 2>&1 | tee -a $OUT/${ci}-imatrix.log
+
+    (time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 99 -c 0     ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
+    (time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 99 -c 0 -fa ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
 
     function check_ppl {
         qnt="$1"
@@ -449,9 +537,6 @@ function gg_run_qwen3_0_6b {
     }
 
     check_ppl "f16"  "$(cat $OUT/${ci}-tg-f16.log  | grep "^\[1\]")" | tee -a $OUT/${ci}-ppl.log
-    if [ -z ${GG_BUILD_NO_BF16} ]; then
-        check_ppl "bf16" "$(cat $OUT/${ci}-tg-bf16.log | grep "^\[1\]")" | tee -a $OUT/${ci}-ppl.log
-    fi
     check_ppl "q8_0" "$(cat $OUT/${ci}-tg-q8_0.log | grep "^\[1\]")" | tee -a $OUT/${ci}-ppl.log
     check_ppl "q4_0" "$(cat $OUT/${ci}-tg-q4_0.log | grep "^\[1\]")" | tee -a $OUT/${ci}-ppl.log
     check_ppl "q4_1" "$(cat $OUT/${ci}-tg-q4_1.log | grep "^\[1\]")" | tee -a $OUT/${ci}-ppl.log
@@ -468,17 +553,147 @@ function gg_run_qwen3_0_6b {
     set +e
 }
 
-function gg_sum_qwen3_0_6b {
+function gg_sum_pythia_1_4b {
     gg_printf '### %s\n\n' "${ci}"
 
-    gg_printf 'Qwen3 0.6B:\n'
+    gg_printf 'Pythia 1.4B:\n'
     gg_printf '- status: %s\n' "$(cat $OUT/${ci}.exit)"
     gg_printf '- perplexity:\n%s\n' "$(cat $OUT/${ci}-ppl.log)"
     gg_printf '- imatrix:\n```\n%s\n```\n' "$(cat $OUT/${ci}-imatrix-sum.log)"
-    gg_printf '- f16:\n```\n%s\n```\n'  "$(cat $OUT/${ci}-tg-f16.log)"
-    if [ -z ${GG_BUILD_NO_BF16} ]; then
-        gg_printf '- bf16:\n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-bf16.log)"
-    fi
+    gg_printf '- f16: \n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-f16.log)"
+    gg_printf '- q8_0:\n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-q8_0.log)"
+    gg_printf '- q4_0:\n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-q4_0.log)"
+    gg_printf '- q4_1:\n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-q4_1.log)"
+    gg_printf '- q5_0:\n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-q5_0.log)"
+    gg_printf '- q5_1:\n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-q5_1.log)"
+    gg_printf '- q2_k:\n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-q2_k.log)"
+    gg_printf '- q3_k:\n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-q3_k.log)"
+    gg_printf '- q4_k:\n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-q4_k.log)"
+    gg_printf '- q5_k:\n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-q5_k.log)"
+    gg_printf '- q6_k:\n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-q6_k.log)"
+    gg_printf '- save-load-state: \n```\n%s\n```\n' "$(cat $OUT/${ci}-save-load-state.log)"
+}
+
+# pythia_2_8b
+
+function gg_run_pythia_2_8b {
+    cd ${SRC}
+
+    gg_wget models-mnt/pythia/2.8B/ https://huggingface.co/EleutherAI/pythia-2.8b/raw/main/config.json
+    gg_wget models-mnt/pythia/2.8B/ https://huggingface.co/EleutherAI/pythia-2.8b/raw/main/tokenizer.json
+    gg_wget models-mnt/pythia/2.8B/ https://huggingface.co/EleutherAI/pythia-2.8b/raw/main/tokenizer_config.json
+    gg_wget models-mnt/pythia/2.8B/ https://huggingface.co/EleutherAI/pythia-2.8b/raw/main/special_tokens_map.json
+    gg_wget models-mnt/pythia/2.8B/ https://huggingface.co/EleutherAI/pythia-2.8b/resolve/main/pytorch_model.bin
+
+    gg_wget models-mnt/wikitext/ https://huggingface.co/datasets/ggml-org/ci/resolve/main/wikitext-2-raw-v1.zip
+    unzip -o models-mnt/wikitext/wikitext-2-raw-v1.zip -d models-mnt/wikitext/
+
+    path_models="../models-mnt/pythia/2.8B"
+    path_wiki="../models-mnt/wikitext/wikitext-2-raw"
+
+    rm -rf build-ci-release && mkdir build-ci-release && cd build-ci-release
+
+    set -e
+
+    (time cmake -DCMAKE_BUILD_TYPE=Release ${CMAKE_EXTRA} .. ) 2>&1 | tee -a $OUT/${ci}-cmake.log
+    (time make -j$(nproc)                                    ) 2>&1 | tee -a $OUT/${ci}-make.log
+
+    python3 ../convert_hf_to_gguf.py ${path_models} --outfile ${path_models}/ggml-model-f16.gguf
+
+    model_f16="${path_models}/ggml-model-f16.gguf"
+    model_q8_0="${path_models}/ggml-model-q8_0.gguf"
+    model_q4_0="${path_models}/ggml-model-q4_0.gguf"
+    model_q4_1="${path_models}/ggml-model-q4_1.gguf"
+    model_q5_0="${path_models}/ggml-model-q5_0.gguf"
+    model_q5_1="${path_models}/ggml-model-q5_1.gguf"
+    model_q2_k="${path_models}/ggml-model-q2_k.gguf"
+    model_q3_k="${path_models}/ggml-model-q3_k.gguf"
+    model_q4_k="${path_models}/ggml-model-q4_k.gguf"
+    model_q5_k="${path_models}/ggml-model-q5_k.gguf"
+    model_q6_k="${path_models}/ggml-model-q6_k.gguf"
+
+    wiki_test="${path_wiki}/wiki.test.raw"
+
+    ./bin/llama-quantize ${model_f16} ${model_q8_0} q8_0
+    ./bin/llama-quantize ${model_f16} ${model_q4_0} q4_0
+    ./bin/llama-quantize ${model_f16} ${model_q4_1} q4_1
+    ./bin/llama-quantize ${model_f16} ${model_q5_0} q5_0
+    ./bin/llama-quantize ${model_f16} ${model_q5_1} q5_1
+    ./bin/llama-quantize ${model_f16} ${model_q2_k} q2_k
+    ./bin/llama-quantize ${model_f16} ${model_q3_k} q3_k
+    ./bin/llama-quantize ${model_f16} ${model_q4_k} q4_k
+    ./bin/llama-quantize ${model_f16} ${model_q5_k} q5_k
+    ./bin/llama-quantize ${model_f16} ${model_q6_k} q6_k
+
+    (time ./bin/llama-cli -no-cnv --model ${model_f16}  -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-f16.log
+    (time ./bin/llama-cli -no-cnv --model ${model_q8_0} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q8_0.log
+    (time ./bin/llama-cli -no-cnv --model ${model_q4_0} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q4_0.log
+    (time ./bin/llama-cli -no-cnv --model ${model_q4_1} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q4_1.log
+    (time ./bin/llama-cli -no-cnv --model ${model_q5_0} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q5_0.log
+    (time ./bin/llama-cli -no-cnv --model ${model_q5_1} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q5_1.log
+    (time ./bin/llama-cli -no-cnv --model ${model_q2_k} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q2_k.log
+    (time ./bin/llama-cli -no-cnv --model ${model_q3_k} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q3_k.log
+    (time ./bin/llama-cli -no-cnv --model ${model_q4_k} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q4_k.log
+    (time ./bin/llama-cli -no-cnv --model ${model_q5_k} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q5_k.log
+    (time ./bin/llama-cli -no-cnv --model ${model_q6_k} -t 1 -ngl 99 -c 0 -s 1234 -n 256 --ignore-eos -p "I believe the meaning of life is" ) 2>&1 | tee -a $OUT/${ci}-tg-q6_k.log
+
+    (time ./bin/llama-perplexity --model ${model_f16}  -f ${wiki_test} -t 1 -ngl 99 -c 2048 -b 512 --chunks 4 ) 2>&1 | tee -a $OUT/${ci}-tg-f16.log
+    (time ./bin/llama-perplexity --model ${model_q8_0} -f ${wiki_test} -t 1 -ngl 99 -c 2048 -b 512 --chunks 4 ) 2>&1 | tee -a $OUT/${ci}-tg-q8_0.log
+    (time ./bin/llama-perplexity --model ${model_q4_0} -f ${wiki_test} -t 1 -ngl 99 -c 2048 -b 512 --chunks 4 ) 2>&1 | tee -a $OUT/${ci}-tg-q4_0.log
+    (time ./bin/llama-perplexity --model ${model_q4_1} -f ${wiki_test} -t 1 -ngl 99 -c 2048 -b 512 --chunks 4 ) 2>&1 | tee -a $OUT/${ci}-tg-q4_1.log
+    (time ./bin/llama-perplexity --model ${model_q5_0} -f ${wiki_test} -t 1 -ngl 99 -c 2048 -b 512 --chunks 4 ) 2>&1 | tee -a $OUT/${ci}-tg-q5_0.log
+    (time ./bin/llama-perplexity --model ${model_q5_1} -f ${wiki_test} -t 1 -ngl 99 -c 2048 -b 512 --chunks 4 ) 2>&1 | tee -a $OUT/${ci}-tg-q5_1.log
+    (time ./bin/llama-perplexity --model ${model_q2_k} -f ${wiki_test} -t 1 -ngl 99 -c 2048 -b 512 --chunks 4 ) 2>&1 | tee -a $OUT/${ci}-tg-q2_k.log
+    (time ./bin/llama-perplexity --model ${model_q3_k} -f ${wiki_test} -t 1 -ngl 99 -c 2048 -b 512 --chunks 4 ) 2>&1 | tee -a $OUT/${ci}-tg-q3_k.log
+    (time ./bin/llama-perplexity --model ${model_q4_k} -f ${wiki_test} -t 1 -ngl 99 -c 2048 -b 512 --chunks 4 ) 2>&1 | tee -a $OUT/${ci}-tg-q4_k.log
+    (time ./bin/llama-perplexity --model ${model_q5_k} -f ${wiki_test} -t 1 -ngl 99 -c 2048 -b 512 --chunks 4 ) 2>&1 | tee -a $OUT/${ci}-tg-q5_k.log
+    (time ./bin/llama-perplexity --model ${model_q6_k} -f ${wiki_test} -t 1 -ngl 99 -c 2048 -b 512 --chunks 4 ) 2>&1 | tee -a $OUT/${ci}-tg-q6_k.log
+
+    (time ./bin/llama-imatrix --model ${model_f16} -f ${wiki_test} -t 1 -ngl 99 -c 2048 -b 512 --chunks 4 ) 2>&1 | tee -a $OUT/${ci}-imatrix.log
+
+    (time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 10 -c 0     ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
+    (time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 10 -c 0 -fa ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
+    (time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 99 -c 0     ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
+    (time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 99 -c 0 -fa ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
+
+    function check_ppl {
+        qnt="$1"
+        ppl=$(echo "$2" | grep -oE "[0-9]+\.[0-9]+" | tail -n 1)
+
+        if [ $(echo "$ppl > 20.0" | bc) -eq 1 ]; then
+            printf '  - %s @ %s (FAIL: ppl > 20.0)\n' "$qnt" "$ppl"
+            return 20
+        fi
+
+        printf '  - %s @ %s OK\n' "$qnt" "$ppl"
+        return 0
+    }
+
+    check_ppl "f16"  "$(cat $OUT/${ci}-tg-f16.log  | grep "^\[1\]")" | tee -a $OUT/${ci}-ppl.log
+    check_ppl "q8_0" "$(cat $OUT/${ci}-tg-q8_0.log | grep "^\[1\]")" | tee -a $OUT/${ci}-ppl.log
+    check_ppl "q4_0" "$(cat $OUT/${ci}-tg-q4_0.log | grep "^\[1\]")" | tee -a $OUT/${ci}-ppl.log
+    check_ppl "q4_1" "$(cat $OUT/${ci}-tg-q4_1.log | grep "^\[1\]")" | tee -a $OUT/${ci}-ppl.log
+    check_ppl "q5_0" "$(cat $OUT/${ci}-tg-q5_0.log | grep "^\[1\]")" | tee -a $OUT/${ci}-ppl.log
+    check_ppl "q5_1" "$(cat $OUT/${ci}-tg-q5_1.log | grep "^\[1\]")" | tee -a $OUT/${ci}-ppl.log
+   #check_ppl "q2_k" "$(cat $OUT/${ci}-tg-q2_k.log | grep "^\[1\]")" | tee -a $OUT/${ci}-ppl.log # note: ppl > 20.0 for this quant and model
+    check_ppl "q3_k" "$(cat $OUT/${ci}-tg-q3_k.log | grep "^\[1\]")" | tee -a $OUT/${ci}-ppl.log
+    check_ppl "q4_k" "$(cat $OUT/${ci}-tg-q4_k.log | grep "^\[1\]")" | tee -a $OUT/${ci}-ppl.log
+    check_ppl "q5_k" "$(cat $OUT/${ci}-tg-q5_k.log | grep "^\[1\]")" | tee -a $OUT/${ci}-ppl.log
+    check_ppl "q6_k" "$(cat $OUT/${ci}-tg-q6_k.log | grep "^\[1\]")" | tee -a $OUT/${ci}-ppl.log
+
+    cat $OUT/${ci}-imatrix.log | grep "Final" >> $OUT/${ci}-imatrix-sum.log
+
+    set +e
+}
+
+function gg_sum_pythia_2_8b {
+    gg_printf '### %s\n\n' "${ci}"
+
+    gg_printf 'Pythia 2.8B:\n'
+    gg_printf '- status: %s\n' "$(cat $OUT/${ci}.exit)"
+    gg_printf '- perplexity:\n%s\n' "$(cat $OUT/${ci}-ppl.log)"
+    gg_printf '- imatrix:\n```\n%s\n```\n' "$(cat $OUT/${ci}-imatrix-sum.log)"
+    gg_printf '- f16: \n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-f16.log)"
     gg_printf '- q8_0:\n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-q8_0.log)"
     gg_printf '- q4_0:\n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-q4_0.log)"
     gg_printf '- q4_1:\n```\n%s\n```\n' "$(cat $OUT/${ci}-tg-q4_1.log)"
@@ -525,10 +740,8 @@ function gg_run_embd_bge_small {
 
     ./bin/llama-quantize ${model_f16} ${model_q8_0} q8_0
 
-    (time ./bin/llama-fit-params --model ${model_f16} 2>&1 | tee -a $OUT/${ci}-fp-f16.log)
-
-    (time ./bin/llama-embedding --model ${model_f16}  -p "I believe the meaning of life is" -ngl 99 -c 0 --no-op-offload) 2>&1 | tee -a $OUT/${ci}-tg-f16.log
-    (time ./bin/llama-embedding --model ${model_q8_0} -p "I believe the meaning of life is" -ngl 99 -c 0 --no-op-offload) 2>&1 | tee -a $OUT/${ci}-tg-q8_0.log
+    (time ./bin/llama-embedding --model ${model_f16}  -p "I believe the meaning of life is" -ngl 99 -c 0 ) 2>&1 | tee -a $OUT/${ci}-tg-f16.log
+    (time ./bin/llama-embedding --model ${model_q8_0} -p "I believe the meaning of life is" -ngl 99 -c 0 ) 2>&1 | tee -a $OUT/${ci}-tg-q8_0.log
 
     set +e
 }
@@ -552,7 +765,12 @@ function gg_run_rerank_tiny {
     gg_wget models-mnt/rerank-tiny/ https://huggingface.co/jinaai/jina-reranker-v1-tiny-en/raw/main/tokenizer_config.json
     gg_wget models-mnt/rerank-tiny/ https://huggingface.co/jinaai/jina-reranker-v1-tiny-en/raw/main/special_tokens_map.json
     gg_wget models-mnt/rerank-tiny/ https://huggingface.co/jinaai/jina-reranker-v1-tiny-en/resolve/main/pytorch_model.bin
-    gg_wget models-mnt/rerank-tiny/ https://huggingface.co/jinaai/jina-reranker-v1-tiny-en/raw/main/vocab.json
+    gg_wget models-mnt/rerank-tiny/ https://huggingface.co/jinaai/jina-reranker-v1-tiny-en/raw/main/sentence_bert_config.json
+    gg_wget models-mnt/rerank-tiny/ https://huggingface.co/jinaai/jina-reranker-v1-tiny-en/raw/main/vocab.txt
+    gg_wget models-mnt/rerank-tiny/ https://huggingface.co/jinaai/jina-reranker-v1-tiny-en/raw/main/modules.json
+    gg_wget models-mnt/rerank-tiny/ https://huggingface.co/jinaai/jina-reranker-v1-tiny-en/raw/main/config.json
+
+    gg_wget models-mnt/rerank-tiny/1_Pooling https://huggingface.co/jinaai/jina-reranker-v1-tiny-en/raw/main/1_Pooling/config.json
 
     path_models="../models-mnt/rerank-tiny"
 
@@ -567,10 +785,8 @@ function gg_run_rerank_tiny {
 
     model_f16="${path_models}/ggml-model-f16.gguf"
 
-    (time ./bin/llama-fit-params --model ${model_f16} 2>&1 | tee -a $OUT/${ci}-fp-f16.log)
-
     # for this model, the SEP token is "</s>"
-    (time ./bin/llama-embedding --model ${model_f16} -p "what is panda?\thi\nwhat is panda?\tit's a bear\nwhat is panda?\tThe giant panda (Ailuropoda melanoleuca), sometimes called a panda bear or simply panda, is a bear species endemic to China." -ngl 99 -c 0 --pooling rank --embd-normalize -1 --no-op-offload --verbose-prompt) 2>&1 | tee -a $OUT/${ci}-rk-f16.log
+    (time ./bin/llama-embedding --model ${model_f16} -p "what is panda?\thi\nwhat is panda?\tit's a bear\nwhat is panda?\tThe giant panda (Ailuropoda melanoleuca), sometimes called a panda bear or simply panda, is a bear species endemic to China." -ngl 99 -c 0 --pooling rank --embd-normalize -1 --verbose-prompt) 2>&1 | tee -a $OUT/${ci}-rk-f16.log
 
     # sample output
     # rerank score 0:    0.029
@@ -644,8 +860,10 @@ if [ -z ${GG_BUILD_LOW_PERF} ]; then
 fi
 
 ret=0
-
-test $ret -eq 0 && gg_run ctest_debug
+if [ -z ${GG_BUILD_SYCL} ]; then
+    # SYCL build breaks with debug build flags
+    test $ret -eq 0 && gg_run ctest_debug
+fi
 test $ret -eq 0 && gg_run ctest_release
 
 if [ -z ${GG_BUILD_LOW_PERF} ]; then
@@ -653,15 +871,24 @@ if [ -z ${GG_BUILD_LOW_PERF} ]; then
     test $ret -eq 0 && gg_run rerank_tiny
 
     if [ -z ${GG_BUILD_CLOUD} ] || [ ${GG_BUILD_EXTRA_TESTS_0} ]; then
-        test $ret -eq 0 && gg_run test_scripts
+        if [ -z ${GG_BUILD_SYCL} ]; then
+            test $ret -eq 0 && gg_run test_scripts_debug
+        fi
+        test $ret -eq 0 && gg_run test_scripts_release
     fi
 
-    test $ret -eq 0 && gg_run qwen3_0_6b
-
-    test $ret -eq 0 && gg_run ctest_with_model_debug
-    test $ret -eq 0 && gg_run ctest_with_model_release
+    if [ -z ${GG_BUILD_VRAM_GB} ] || [ ${GG_BUILD_VRAM_GB} -ge 8 ]; then
+        if [ -z ${GG_BUILD_CUDA} ] && [ -z ${GG_BUILD_VULKAN} ]; then
+            test $ret -eq 0 && gg_run pythia_1_4b
+        else
+            test $ret -eq 0 && gg_run pythia_2_8b
+            #test $ret -eq 0 && gg_run open_llama_7b_v2
+        fi
+        if [ -z ${GG_BUILD_SYCL} ]; then
+            test $ret -eq 0 && gg_run ctest_with_model_debug
+        fi
+        test $ret -eq 0 && gg_run ctest_with_model_release
+    fi
 fi
 
-cat $OUT/README.md
-
 exit $ret

cmake/build-info.cmake

@@ -39,10 +39,26 @@ if(Git_FOUND)
     endif()
 endif()
 
-set(BUILD_COMPILER "${CMAKE_C_COMPILER_ID} ${CMAKE_C_COMPILER_VERSION}")
-
-if(CMAKE_VS_PLATFORM_NAME)
-    set(BUILD_TARGET ${CMAKE_VS_PLATFORM_NAME})
+if(MSVC)
+    set(BUILD_COMPILER "${CMAKE_C_COMPILER_ID} ${CMAKE_C_COMPILER_VERSION}")
+    if (CMAKE_VS_PLATFORM_NAME)
+        set(BUILD_TARGET ${CMAKE_VS_PLATFORM_NAME})
+    else()
+        set(BUILD_TARGET "${CMAKE_SYSTEM_NAME} ${CMAKE_SYSTEM_PROCESSOR}")
+    endif()
 else()
-    set(BUILD_TARGET "${CMAKE_SYSTEM_NAME} ${CMAKE_SYSTEM_PROCESSOR}")
+    execute_process(
+        COMMAND ${CMAKE_C_COMPILER} --version
+        OUTPUT_VARIABLE OUT
+        OUTPUT_STRIP_TRAILING_WHITESPACE
+    )
+    string(REGEX REPLACE " *\n.*" "" OUT "${OUT}")
+    set(BUILD_COMPILER ${OUT})
+
+    execute_process(
+        COMMAND ${CMAKE_C_COMPILER} -dumpmachine
+        OUTPUT_VARIABLE OUT
+        OUTPUT_STRIP_TRAILING_WHITESPACE
+    )
+    set(BUILD_TARGET ${OUT})
 endif()

cmake/riscv64-spacemit-linux-gnu-gcc.cmake

@@ -1,29 +0,0 @@
-set(CMAKE_SYSTEM_NAME Linux)
-set(CMAKE_SYSTEM_PROCESSOR riscv64)
-set(CMAKE_SYSTEM_VERSION 1)
-
-if (CMAKE_HOST_SYSTEM_PROCESSOR MATCHES "^(riscv)")
-    message(STATUS "HOST SYSTEM ${CMAKE_HOST_SYSTEM_PROCESSOR}")
-else()
-    set(GNU_MACHINE riscv64-unknown-linux-gnu CACHE STRING "GNU compiler triple")
-    if (DEFINED ENV{RISCV_ROOT_PATH})
-        file(TO_CMAKE_PATH $ENV{RISCV_ROOT_PATH} RISCV_ROOT_PATH)
-    else()
-        message(FATAL_ERROR "RISCV_ROOT_PATH env must be defined")
-    endif()
-
-    set(RISCV_ROOT_PATH ${RISCV_ROOT_PATH} CACHE STRING "root path to riscv toolchain")
-    set(CMAKE_C_COMPILER ${RISCV_ROOT_PATH}/bin/riscv64-unknown-linux-gnu-gcc)
-    set(CMAKE_CXX_COMPILER ${RISCV_ROOT_PATH}/bin/riscv64-unknown-linux-gnu-g++)
-    set(CMAKE_STRIP ${RISCV_ROOT_PATH}/bin/riscv64-unknown-linux-gnu-strip)
-    set(CMAKE_FIND_ROOT_PATH "${RISCV_ROOT_PATH}/riscv64-unknown-linux-gnu")
-    set(CMAKE_SYSROOT "${RISCV_ROOT_PATH}/sysroot")
-endif()
-
-set(CMAKE_FIND_ROOT_PATH_MODE_PROGRAM NEVER)
-set(CMAKE_FIND_ROOT_PATH_MODE_LIBRARY ONLY)
-set(CMAKE_FIND_ROOT_PATH_MODE_INCLUDE ONLY)
-set(CMAKE_FIND_ROOT_PATH_MODE_PACKAGE ONLY)
-set(CMAKE_C_FLAGS "-march=rv64gcv_zfh_zba_zicbop -mabi=lp64d ${CMAKE_C_FLAGS}")
-set(CMAKE_CXX_FLAGS "-march=rv64gcv_zfh_zba_zicbop -mabi=lp64d ${CXX_FLAGS}")
-set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -latomic")

common/CMakeLists.txt

@@ -50,19 +50,12 @@ add_library(${TARGET} STATIC
     base64.hpp
     chat-parser.cpp
     chat-parser.h
-    chat-parser-xml-toolcall.h
-    chat-parser-xml-toolcall.cpp
-    chat-peg-parser.cpp
-    chat-peg-parser.h
     chat.cpp
     chat.h
     common.cpp
     common.h
     console.cpp
     console.h
-    download.cpp
-    download.h
-    http.h
     json-partial.cpp
     json-partial.h
     json-schema-to-grammar.cpp
@@ -71,32 +64,22 @@ add_library(${TARGET} STATIC
     log.h
     ngram-cache.cpp
     ngram-cache.h
-    peg-parser.cpp
-    peg-parser.h
-    preset.cpp
-    preset.h
     regex-partial.cpp
     regex-partial.h
     sampling.cpp
     sampling.h
     speculative.cpp
     speculative.h
-    unicode.cpp
-    unicode.h
     )
 
-target_include_directories(${TARGET} PUBLIC . ../vendor)
-target_compile_features   (${TARGET} PUBLIC cxx_std_17)
-
 if (BUILD_SHARED_LIBS)
     set_target_properties(${TARGET} PROPERTIES POSITION_INDEPENDENT_CODE ON)
 endif()
 
-# TODO: use list(APPEND LLAMA_COMMON_EXTRA_LIBS ...)
 set(LLAMA_COMMON_EXTRA_LIBS build_info)
 
+# Use curl to download model url
 if (LLAMA_CURL)
-    # Use curl to download model url
     find_package(CURL)
     if (NOT CURL_FOUND)
         message(FATAL_ERROR "Could NOT find CURL. Hint: to disable this feature, set -DLLAMA_CURL=OFF")
@@ -104,11 +87,7 @@ if (LLAMA_CURL)
     target_compile_definitions(${TARGET} PUBLIC LLAMA_USE_CURL)
     include_directories(${CURL_INCLUDE_DIRS})
     set(LLAMA_COMMON_EXTRA_LIBS ${LLAMA_COMMON_EXTRA_LIBS} ${CURL_LIBRARIES})
-elseif (LLAMA_HTTPLIB)
-    # otherwise, use cpp-httplib
-    target_compile_definitions(${TARGET} PUBLIC LLAMA_USE_HTTPLIB)
-    set(LLAMA_COMMON_EXTRA_LIBS ${LLAMA_COMMON_EXTRA_LIBS} cpp-httplib)
-endif()
+endif ()
 
 if (LLAMA_LLGUIDANCE)
     include(ExternalProject)
@@ -154,7 +133,9 @@ if (LLAMA_LLGUIDANCE)
     set(LLAMA_COMMON_EXTRA_LIBS ${LLAMA_COMMON_EXTRA_LIBS} llguidance ${LLGUIDANCE_PLATFORM_LIBS})
 endif ()
 
-target_link_libraries(${TARGET} PRIVATE ${LLAMA_COMMON_EXTRA_LIBS} PUBLIC llama Threads::Threads)
+target_include_directories(${TARGET} PUBLIC . ../vendor)
+target_compile_features   (${TARGET} PUBLIC cxx_std_17)
+target_link_libraries     (${TARGET} PRIVATE ${LLAMA_COMMON_EXTRA_LIBS} PUBLIC llama Threads::Threads)
 
 
 #

common/arg.h

@@ -3,13 +3,8 @@
 #include "common.h"
 
 #include <set>
-#include <map>
 #include <string>
 #include <vector>
-#include <cstring>
-
-// pseudo-env variable to identify preset-only arguments
-#define COMMON_ARG_PRESET_LOAD_ON_STARTUP "__PRESET_LOAD_ON_STARTUP"
 
 //
 // CLI argument parsing
@@ -19,20 +14,15 @@ struct common_arg {
     std::set<enum llama_example> examples = {LLAMA_EXAMPLE_COMMON};
     std::set<enum llama_example> excludes = {};
     std::vector<const char *> args;
-    std::vector<const char *> args_neg;  // for negated args like --no-xxx
     const char * value_hint   = nullptr; // help text or example for arg value
     const char * value_hint_2 = nullptr; // for second arg value
     const char * env          = nullptr;
     std::string help;
     bool is_sparam = false; // is current arg a sampling param?
-    bool is_preset_only = false; // is current arg preset-only (not treated as CLI arg)
     void (*handler_void)   (common_params & params) = nullptr;
     void (*handler_string) (common_params & params, const std::string &) = nullptr;
     void (*handler_str_str)(common_params & params, const std::string &, const std::string &) = nullptr;
     void (*handler_int)    (common_params & params, int) = nullptr;
-    void (*handler_bool)   (common_params & params, bool) = nullptr;
-
-    common_arg() = default;
 
     common_arg(
         const std::initializer_list<const char *> & args,
@@ -54,13 +44,6 @@ struct common_arg {
         void (*handler)(common_params & params)
     ) : args(args), help(help), handler_void(handler) {}
 
-    common_arg(
-        const std::initializer_list<const char *> & args,
-        const std::initializer_list<const char *> & args_neg,
-        const std::string & help,
-        void (*handler)(common_params & params, bool)
-    ) : args(args), args_neg(args_neg), help(help), handler_bool(handler) {}
-
     // support 2 values for arg
     common_arg(
         const std::initializer_list<const char *> & args,
@@ -74,38 +57,13 @@ struct common_arg {
     common_arg & set_excludes(std::initializer_list<enum llama_example> excludes);
     common_arg & set_env(const char * env);
     common_arg & set_sparam();
-    common_arg & set_preset_only();
     bool in_example(enum llama_example ex);
     bool is_exclude(enum llama_example ex);
-    bool get_value_from_env(std::string & output) const;
-    bool has_value_from_env() const;
-    std::string to_string() const;
-
-    // for using as key in std::map
-    bool operator<(const common_arg& other) const {
-        if (args.empty() || other.args.empty()) {
-            return false;
-        }
-        return strcmp(args[0], other.args[0]) < 0;
-    }
-    bool operator==(const common_arg& other) const {
-        if (args.empty() || other.args.empty()) {
-            return false;
-        }
-        return strcmp(args[0], other.args[0]) == 0;
-    }
-
-    // get all args and env vars (including negated args/env)
-    std::vector<std::string> get_args() const;
-    std::vector<std::string> get_env() const;
+    bool get_value_from_env(std::string & output);
+    bool has_value_from_env();
+    std::string to_string();
 };
 
-namespace common_arg_utils {
-    bool is_truthy(const std::string & value);
-    bool is_falsey(const std::string & value);
-    bool is_autoy(const std::string & value);
-}
-
 struct common_params_context {
     enum llama_example ex = LLAMA_EXAMPLE_COMMON;
     common_params & params;
@@ -118,16 +76,9 @@ struct common_params_context {
 // if one argument has invalid value, it will automatically display usage of the specific argument (and not the full usage message)
 bool common_params_parse(int argc, char ** argv, common_params & params, llama_example ex, void(*print_usage)(int, char **) = nullptr);
 
-// parse input arguments from CLI into a map
-bool common_params_to_map(int argc, char ** argv, llama_example ex, std::map<common_arg, std::string> & out_map);
-
-// populate preset-only arguments
-// these arguments are not treated as command line arguments
-// see: https://github.com/ggml-org/llama.cpp/issues/18163
-void common_params_add_preset_options(std::vector<common_arg> & args);
-
-// initialize argument parser context - used by test-arg-parser and preset
+// function to be used by test-arg-parser
 common_params_context common_params_parser_init(common_params & params, llama_example ex, void(*print_usage)(int, char **) = nullptr);
+bool common_has_curl();
 
 struct common_remote_params {
     std::vector<std::string> headers;

common/chat-parser-xml-toolcall.cpp

@@ -1,879 +0,0 @@
-#include "chat.h"
-#include "chat-parser.h"
-#include "common.h"
-#include "json-partial.h"
-#include "json-schema-to-grammar.h"
-#include "log.h"
-#include "regex-partial.h"
-
-using json = nlohmann::ordered_json;
-
-class xml_toolcall_syntax_exception : public std::runtime_error {
-  public:
-    xml_toolcall_syntax_exception(const std::string & message) : std::runtime_error(message) {}
-};
-
-template<typename T>
-inline void sort_uniq(std::vector<T> &vec) {
-    std::sort(vec.begin(), vec.end());
-    vec.erase(std::unique(vec.begin(), vec.end()), vec.end());
-}
-
-template<typename T>
-inline bool all_space(const T &str) {
-    return std::all_of(str.begin(), str.end(), [](unsigned char ch) { return std::isspace(ch); });
-}
-
-static size_t utf8_truncate_safe(const std::string_view s) {
-    size_t len = s.size();
-    if (len == 0) return 0;
-    size_t i = len;
-    for (size_t back = 0; back < 4 && i > 0; ++back) {
-        --i;
-        unsigned char c = s[i];
-        if ((c & 0x80) == 0) {
-            return len;
-        } else if ((c & 0xC0) == 0xC0) {
-            size_t expected_len = 0;
-            if ((c & 0xE0) == 0xC0) expected_len = 2;
-            else if ((c & 0xF0) == 0xE0) expected_len = 3;
-            else if ((c & 0xF8) == 0xF0) expected_len = 4;
-            else return i;
-            if (len - i >= expected_len) {
-                return len;
-            } else {
-                return i;
-            }
-        }
-    }
-    return len - std::min(len, size_t(3));
-}
-
-inline void utf8_truncate_safe_resize(std::string &s) {
-    s.resize(utf8_truncate_safe(s));
-}
-
-inline std::string_view utf8_truncate_safe_view(const std::string_view s) {
-    return s.substr(0, utf8_truncate_safe(s));
-}
-
-static std::optional<common_chat_msg_parser::find_regex_result> try_find_2_literal_splited_by_spaces(common_chat_msg_parser & builder, const std::string & literal1, const std::string & literal2) {
-    if (literal1.size() == 0) return builder.try_find_literal(literal2);
-    const auto saved_pos = builder.pos();
-    while (auto res = builder.try_find_literal(literal1)) {
-        builder.consume_spaces();
-        const auto match_len = std::min(literal2.size(), builder.input().size() - builder.pos());
-        if (builder.input().compare(builder.pos(), match_len, literal2, 0, match_len) == 0) {
-            if (res->prelude.size() != res->groups[0].begin - saved_pos) {
-                res->prelude = builder.str({saved_pos, res->groups[0].begin});
-            }
-            builder.move_to(builder.pos() + match_len);
-            res->groups[0].end = builder.pos();
-            GGML_ASSERT(res->groups[0].begin != res->groups[0].end);
-            return res;
-        }
-        builder.move_to(res->groups[0].begin + 1);
-    }
-    builder.move_to(saved_pos);
-    return std::nullopt;
-}
-
-/**
- * make a GBNF that accept any strings except those containing any of the forbidden strings.
- */
-std::string make_gbnf_excluding(std::vector<std::string> forbids) {
-    constexpr auto charclass_escape = [](unsigned char c) -> std::string {
-        if (c == '\\' || c == ']' || c == '^' || c == '-') {
-            std::string s = "\\";
-            s.push_back((char)c);
-            return s;
-        }
-        if (isprint(c)) {
-            return std::string(1, (char)c);
-        }
-        char buf[16];
-        snprintf(buf, 15, "\\x%02X", c);
-        return std::string(buf);
-    };
-    constexpr auto build_expr = [charclass_escape](auto self, const std::vector<std::string>& forbids, int l, int r, int depth) -> std::string {
-        std::vector<std::pair<unsigned char, std::pair<int,int>>> children;
-        int i = l;
-        while (i < r) {
-            const std::string &s = forbids[i];
-            if ((int)s.size() == depth) {
-                ++i;
-                continue;
-            }
-            unsigned char c = (unsigned char)s[depth];
-            int j = i;
-            while (j < r && (int)forbids[j].size() > depth &&
-                   (unsigned char)forbids[j][depth] == c) {
-                ++j;
-            }
-            children.push_back({c, {i, j}});
-            i = j;
-        }
-        std::vector<std::string> alts;
-        if (!children.empty()) {
-            std::string cls;
-            for (auto &ch : children) cls += charclass_escape(ch.first);
-            alts.push_back(std::string("[^") + cls + "]");
-        }
-        for (auto &ch : children) {
-            std::string childExpr = self(self, forbids, ch.second.first, ch.second.second, depth+1);
-            if (!childExpr.empty()) {
-                std::string quoted_ch = "\"";
-                if (ch.first == '\\') quoted_ch += "\\\\";
-                else if (ch.first == '"') quoted_ch += "\\\"";
-                else if (isprint(ch.first)) quoted_ch.push_back(ch.first);
-                else {
-                    char buf[16];
-                    snprintf(buf, 15, "\\x%02X", ch.first);
-                    quoted_ch += buf;
-                }
-                quoted_ch += "\"";
-                std::string branch = quoted_ch + std::string(" ") + childExpr;
-                alts.push_back(branch);
-            }
-        }
-        if (alts.empty()) return "";
-        std::ostringstream oss;
-        oss << "( ";
-        for (size_t k = 0; k < alts.size(); ++k) {
-            if (k) oss << " | ";
-            oss << alts[k];
-        }
-        oss << " )";
-        return oss.str();
-    };
-    if (forbids.empty()) return "( . )*";
-    sort(forbids.begin(), forbids.end());
-    std::string expr = build_expr(build_expr, forbids, 0, forbids.size(), 0);
-    if (expr.empty()) {
-        std::string cls;
-        for (auto &s : forbids) if (!s.empty()) cls += charclass_escape((unsigned char)s[0]);
-        expr = std::string("( [^") + cls + "] )";
-    }
-    if (forbids.size() == 1)
-        return expr + "*";
-    else
-        return std::string("( ") + expr + " )*";
-}
-
-/**
- * Build grammar for xml-style tool call
- * form.scope_start and form.scope_end can be empty.
- * Requires data.format for model-specific hacks.
- */
-void build_grammar_xml_tool_call(common_chat_params & data, const json & tools, const struct xml_tool_call_format & form) {
-    GGML_ASSERT(!form.tool_start.empty());
-    GGML_ASSERT(!form.tool_sep.empty());
-    GGML_ASSERT(!form.key_start.empty());
-    GGML_ASSERT(!form.val_end.empty());
-    GGML_ASSERT(!form.tool_end.empty());
-
-    std::string key_val_sep = form.key_val_sep;
-    if (form.key_val_sep2) {
-        key_val_sep += "\n";
-        key_val_sep += *form.key_val_sep2;
-    }
-    GGML_ASSERT(!key_val_sep.empty());
-
-    if (tools.is_array() && !tools.empty()) {
-        data.grammar = build_grammar([&](const common_grammar_builder &builder) {
-            auto string_arg_val = form.last_val_end ?
-                    builder.add_rule("string-arg-val", make_gbnf_excluding({form.val_end, *form.last_val_end})) :
-                    builder.add_rule("string-arg-val", make_gbnf_excluding({form.val_end}));
-
-            std::vector<std::string> tool_rules;
-            for (const auto & tool : tools) {
-                if (!tool.contains("type") || tool.at("type") != "function" || !tool.contains("function")) {
-                    LOG_WRN("Skipping tool without function: %s", tool.dump(2).c_str());
-                    continue;
-                }
-                const auto & function = tool.at("function");
-                if (!function.contains("name") || !function.at("name").is_string()) {
-                    LOG_WRN("Skipping invalid function (invalid name): %s", function.dump(2).c_str());
-                    continue;
-                }
-                if (!function.contains("parameters") || !function.at("parameters").is_object()) {
-                    LOG_WRN("Skipping invalid function (invalid parameters): %s", function.dump(2).c_str());
-                    continue;
-                }
-                std::string name = function.at("name");
-                auto parameters = function.at("parameters");
-                builder.resolve_refs(parameters);
-
-                struct parameter_rule {
-                    std::string symbol_name;
-                    bool is_required;
-                };
-                std::vector<parameter_rule> arg_rules;
-                if (!parameters.contains("properties") || !parameters.at("properties").is_object()) {
-                    LOG_WRN("Skipping invalid function (invalid properties): %s", function.dump(2).c_str());
-                    continue;
-                } else {
-                    std::vector<std::string> requiredParameters;
-                    if (parameters.contains("required")) {
-                        try { parameters.at("required").get_to(requiredParameters); }
-                        catch (const std::runtime_error&) {
-                            LOG_WRN("Invalid function required parameters, ignoring: %s", function.at("required").dump(2).c_str());
-                        }
-                    }
-                    sort_uniq(requiredParameters);
-                    for (const auto & [key, value] : parameters.at("properties").items()) {
-                        std::string quoted_key = key;
-                        bool required = std::binary_search(requiredParameters.begin(), requiredParameters.end(), key);
-                        if (form.key_start.back() == '"' && key_val_sep[0] == '"') {
-                            quoted_key = gbnf_format_literal(key);
-                            quoted_key = quoted_key.substr(1, quoted_key.size() - 2);
-                        }
-                        arg_rules.push_back(parameter_rule {builder.add_rule("func-" + name + "-kv-" + key,
-                            gbnf_format_literal(form.key_start) + " " +
-                            gbnf_format_literal(quoted_key) + " " +
-                            gbnf_format_literal(key_val_sep) + " " +
-                            ((value.contains("type") && value["type"].is_string() && value["type"] == "string" && (!form.raw_argval || *form.raw_argval)) ?
-                                    (form.raw_argval ?
-                                            string_arg_val :
-                                            "( " + string_arg_val + " | " + builder.add_schema(name + "-arg-" + key, value) + " )"
-                                    ) :
-                                    builder.add_schema(name + "-arg-" + key, value)
-                            )
-                        ), required});
-                    }
-                }
-
-                auto next_arg_with_sep = builder.add_rule(name + "-last-arg-end", form.last_val_end ? gbnf_format_literal(*form.last_val_end) : gbnf_format_literal(form.val_end));
-                decltype(next_arg_with_sep) next_arg = "\"\"";
-                for (auto i = arg_rules.size() - 1; /* i >= 0 && */ i < arg_rules.size(); --i) {
-                    std::string include_this_arg = arg_rules[i].symbol_name + " " + next_arg_with_sep;
-                    next_arg = builder.add_rule(name + "-arg-after-" + std::to_string(i), arg_rules[i].is_required ?
-                            include_this_arg : "( " + include_this_arg + " ) | " + next_arg
-                    );
-                    include_this_arg = gbnf_format_literal(form.val_end) + " " + include_this_arg;
-                    next_arg_with_sep = builder.add_rule(name + "-arg-after-" + std::to_string(i) + "-with-sep", arg_rules[i].is_required ?
-                            include_this_arg : "( " + include_this_arg + " ) | " + next_arg_with_sep
-                    );
-                }
-
-                std::string quoted_name = name;
-                if (form.tool_start.back() == '"' && form.tool_sep[0] == '"') {
-                    quoted_name = gbnf_format_literal(name);
-                    quoted_name = quoted_name.substr(1, quoted_name.size() - 2);
-                }
-                quoted_name = gbnf_format_literal(quoted_name);
-                // Kimi-K2 uses functions.{{ tool_call['function']['name'] }}:{{ loop.index }} as function name
-                if (data.format == COMMON_CHAT_FORMAT_KIMI_K2) {
-                    quoted_name = "\"functions.\" " + quoted_name + " \":\" [0-9]+";
-                }
-                tool_rules.push_back(builder.add_rule(name + "-call",
-                        gbnf_format_literal(form.tool_start) + " " +
-                        quoted_name + " " +
-                        gbnf_format_literal(form.tool_sep) + " " +
-                        next_arg
-                ));
-            }
-
-            auto tool_call_once = builder.add_rule("root-tool-call-once", string_join(tool_rules, " | "));
-            auto tool_call_more = builder.add_rule("root-tool-call-more", gbnf_format_literal(form.tool_end) + " " + tool_call_once);
-            auto call_end = builder.add_rule("root-call-end", form.last_tool_end ? gbnf_format_literal(*form.last_tool_end) : gbnf_format_literal(form.tool_end));
-            auto tool_call_multiple_with_end = builder.add_rule("root-tool-call-multiple-with-end", tool_call_once + " " + tool_call_more + "* " + call_end);
-            builder.add_rule("root",
-                (form.scope_start.empty() ? "" : gbnf_format_literal(form.scope_start) + " ") +
-                tool_call_multiple_with_end  + "?" +
-                (form.scope_end.empty() ? "" : " " + gbnf_format_literal(form.scope_end))
-            );
-        });
-
-        // grammar trigger for tool call
-        data.grammar_triggers.push_back({ COMMON_GRAMMAR_TRIGGER_TYPE_WORD, form.scope_start + form.tool_start });
-    }
-}
-
-/**
- * Parse XML-Style tool call for given xml_tool_call_format. Return false for invalid syntax and get the position untouched.
- * Throws xml_toolcall_syntax_exception if there is invalid syntax and cannot recover the original status for common_chat_msg_parser.
- * form.scope_start, form.tool_sep and form.scope_end can be empty.
- */
-inline bool parse_xml_tool_calls(common_chat_msg_parser & builder, const struct xml_tool_call_format & form) {
-    GGML_ASSERT(!form.tool_start.empty());
-    GGML_ASSERT(!form.key_start.empty());
-    GGML_ASSERT(!form.key_val_sep.empty());
-    GGML_ASSERT(!form.val_end.empty());
-    GGML_ASSERT(!form.tool_end.empty());
-
-    // Helper to choose return false or throw error
-    constexpr auto return_error = [](common_chat_msg_parser & builder, auto &start_pos, const bool &recovery) {
-        LOG_DBG("Failed to parse XML-Style tool call at position: %s\n", gbnf_format_literal(builder.consume_rest().substr(0, 20)).c_str());
-        if (recovery) {
-            builder.move_to(start_pos);
-            return false;
-        } else throw xml_toolcall_syntax_exception("Tool call parsing failed with unrecoverable errors. Try using a grammar to constrain the model’s output.");
-    };
-    // Drop substring from needle to end from a JSON
-    constexpr auto partial_json = [](std::string &json_str, std::string_view needle = "XML_TOOL_CALL_PARTIAL_FLAG") {
-        auto pos = json_str.rfind(needle);
-        if (pos == std::string::npos) {
-            return false;
-        }
-        for (auto i = pos + needle.size(); i < json_str.size(); ++i) {
-            unsigned char ch = static_cast<unsigned char>(json_str[i]);
-            if (ch != '\'' && ch != '"' && ch != '}' && ch != ':' && !std::isspace(ch)) {
-                return false;
-            }
-        }
-        if (pos != 0 && json_str[pos - 1] == '"') {
-            --pos;
-        }
-        json_str.resize(pos);
-        return true;
-    };
-    // Helper to generate a partial argument JSON
-    constexpr auto gen_partial_json = [partial_json](auto set_partial_arg, auto &arguments, auto &builder, auto &function_name) {
-        auto rest = builder.consume_rest();
-        utf8_truncate_safe_resize(rest);
-        set_partial_arg(rest, "XML_TOOL_CALL_PARTIAL_FLAG");
-        auto tool_str = arguments.dump();
-        if (partial_json(tool_str)) {
-            if (builder.add_tool_call(function_name, "", tool_str)) {
-                return;
-            }
-        }
-        LOG_DBG("Failed to parse partial XML-Style tool call, fallback to non-partial: %s\n", tool_str.c_str());
-    };
-    // Helper to find a close (because there may be form.last_val_end or form.last_tool_end)
-    constexpr auto try_find_close = [](
-            common_chat_msg_parser & builder,
-            const std::string & end,
-            const std::optional<std::string> & alt_end,
-            const std::string & end_next,
-            const std::optional<std::string> & alt_end_next
-    ) {
-        auto saved_pos = builder.pos();
-        auto tc = builder.try_find_literal(end);
-        auto val_end_size = end.size();
-        if (alt_end) {
-            auto pos_1 = builder.pos();
-            builder.move_to(saved_pos);
-            auto tc2 = try_find_2_literal_splited_by_spaces(builder, *alt_end, end_next);
-            if (alt_end_next) {
-                builder.move_to(saved_pos);
-                auto tc3 = try_find_2_literal_splited_by_spaces(builder, *alt_end, *alt_end_next);
-                if (tc3 && (!tc2 || tc2->prelude.size() > tc3->prelude.size())) {
-                    tc2 = tc3;
-                }
-            }
-            if (tc2 && (!tc || tc->prelude.size() > tc2->prelude.size())) {
-                tc = tc2;
-                tc->groups[0].end = std::min(builder.input().size(), tc->groups[0].begin + alt_end->size());
-                builder.move_to(tc->groups[0].end);
-                val_end_size = alt_end->size();
-            } else {
-                builder.move_to(pos_1);
-            }
-        }
-        return std::make_pair(val_end_size, tc);
-    };
-    // Helper to find a val_end or last_val_end, returns matched pattern size
-    const auto try_find_val_end = [try_find_close, &builder, &form]() {
-        return try_find_close(builder, form.val_end, form.last_val_end, form.tool_end, form.last_tool_end);
-    };
-    // Helper to find a tool_end or last_tool_end, returns matched pattern size
-    const auto try_find_tool_end = [try_find_close, &builder, &form]() {
-        return try_find_close(builder, form.tool_end, form.last_tool_end, form.scope_end, std::nullopt);
-    };
-
-    bool recovery = true;
-    const auto start_pos = builder.pos();
-    if (!all_space(form.scope_start)) {
-        if (auto tc = builder.try_find_literal(form.scope_start)) {
-            if (all_space(tc->prelude)) {
-                if (form.scope_start.size() != tc->groups[0].end - tc->groups[0].begin)
-                    throw common_chat_msg_partial_exception("Partial literal: " + gbnf_format_literal(form.scope_start));
-            } else {
-                builder.move_to(start_pos);
-                return false;
-            }
-        } else return false;
-    }
-    while (auto tc = builder.try_find_literal(form.tool_start)) {
-        if (!all_space(tc->prelude)) {
-            LOG_DBG("XML-Style tool call: Expected %s, but found %s, trying to match next pattern\n",
-                    gbnf_format_literal(form.tool_start).c_str(),
-                    gbnf_format_literal(tc->prelude).c_str()
-            );
-            builder.move_to(tc->groups[0].begin - tc->prelude.size());
-            break;
-        }
-
-        // Find tool name
-        auto func_name = builder.try_find_literal(all_space(form.tool_sep) ? form.key_start : form.tool_sep);
-        if (!func_name) {
-            auto [sz, tc] = try_find_tool_end();
-            func_name = tc;
-        }
-        if (!func_name) {
-            // Partial tool name not supported
-            throw common_chat_msg_partial_exception("incomplete tool_call");
-        }
-        // If the model generate multiple tool call and the first tool call has no argument
-        if (func_name->prelude.find(form.tool_end) != std::string::npos || (form.last_tool_end ? func_name->prelude.find(*form.last_tool_end) != std::string::npos : false)) {
-            builder.move_to(func_name->groups[0].begin - func_name->prelude.size());
-            auto [sz, tc] = try_find_tool_end();
-            func_name = tc;
-        }
-
-        // Parse tool name
-        builder.move_to(all_space(form.tool_sep) ? func_name->groups[0].begin : func_name->groups[0].end);
-        std::string function_name = string_strip(func_name->prelude);
-        // Kimi-K2 uses functions.{{ tool_call['function']['name'] }}:{{ loop.index }} as function name
-        if (builder.syntax().format == COMMON_CHAT_FORMAT_KIMI_K2) {
-            if (string_starts_with(function_name, "functions.")) {
-                static const std::regex re(":\\d+$");
-                if (std::regex_search(function_name, re)) {
-                    function_name = function_name.substr(10, function_name.rfind(":") - 10);
-                }
-            }
-        }
-
-        // Argument JSON
-        json arguments = json::object();
-
-        // Helper to generate a partial argument JSON
-        const auto gen_partial_args = [&](auto set_partial_arg) {
-            gen_partial_json(set_partial_arg, arguments, builder, function_name);
-        };
-
-        // Parse all arg_key/arg_value pairs
-        while (auto tc = builder.try_find_literal(form.key_start)) {
-            if (!all_space(tc->prelude)) {
-                LOG_DBG("XML-Style tool call: Expected %s, but found %s, trying to match next pattern\n",
-                        gbnf_format_literal(form.key_start).c_str(),
-                        gbnf_format_literal(tc->prelude).c_str()
-                );
-                builder.move_to(tc->groups[0].begin - tc->prelude.size());
-                break;
-            }
-            if (tc->groups[0].end - tc->groups[0].begin != form.key_start.size()) {
-                auto tool_call_arg = arguments.dump();
-                if (tool_call_arg.size() != 0 && tool_call_arg[tool_call_arg.size() - 1] == '}') {
-                    tool_call_arg.resize(tool_call_arg.size() - 1);
-                }
-                builder.add_tool_call(function_name, "", tool_call_arg);
-                throw common_chat_msg_partial_exception("Partial literal: " + gbnf_format_literal(form.key_start));
-            }
-
-            // Parse arg_key
-            auto key_res = builder.try_find_literal(form.key_val_sep);
-            if (!key_res) {
-                gen_partial_args([&](auto &rest, auto &needle) {arguments[rest + needle] = "";});
-                throw common_chat_msg_partial_exception("Expected " + gbnf_format_literal(form.key_val_sep) + " after " + gbnf_format_literal(form.key_start));
-            }
-            if (key_res->groups[0].end - key_res->groups[0].begin != form.key_val_sep.size()) {
-                gen_partial_args([&](auto &, auto &needle) {arguments[key_res->prelude + needle] = "";});
-                throw common_chat_msg_partial_exception("Partial literal: " + gbnf_format_literal(form.key_val_sep));
-            }
-            auto &key = key_res->prelude;
-            recovery = false;
-
-            // Parse arg_value
-            if (form.key_val_sep2) {
-                if (auto tc = builder.try_find_literal(*form.key_val_sep2)) {
-                    if (!all_space(tc->prelude)) {
-                        LOG_DBG("Failed to parse XML-Style tool call: Unexcepted %s between %s and %s\n",
-                                gbnf_format_literal(tc->prelude).c_str(),
-                                gbnf_format_literal(form.key_val_sep).c_str(),
-                                gbnf_format_literal(*form.key_val_sep2).c_str()
-                        );
-                        return return_error(builder, start_pos, false);
-                    }
-                    if (tc->groups[0].end - tc->groups[0].begin != form.key_val_sep2->size()) {
-                        gen_partial_args([&](auto &, auto &needle) {arguments[key] = needle;});
-                        throw common_chat_msg_partial_exception("Partial literal: " + gbnf_format_literal(*form.key_val_sep2));
-                    }
-                } else {
-                    gen_partial_args([&](auto &, auto &needle) {arguments[key] = needle;});
-                    throw common_chat_msg_partial_exception("Expected " + gbnf_format_literal(*form.key_val_sep2) + " after " + gbnf_format_literal(form.key_val_sep));
-                }
-            }
-            auto val_start = builder.pos();
-
-            // Test if arg_val is a partial JSON
-            std::optional<common_json> value_json = std::nullopt;
-            if (!form.raw_argval || !*form.raw_argval) {
-                try { value_json = builder.try_consume_json(); }
-                catch (const std::runtime_error&) { builder.move_to(val_start); }
-                // TODO: Delete this when json_partial adds top-level support for null/true/false
-                if (builder.pos() == val_start) {
-                    const static std::regex number_regex(R"([0-9-][0-9]*(\.\d*)?([eE][+-]?\d*)?)");
-                    builder.consume_spaces();
-                    std::string_view sv = utf8_truncate_safe_view(builder.input());
-                    sv.remove_prefix(builder.pos());
-                    std::string rest = "a";
-                    if (sv.size() < 6) rest = sv;
-                    if (string_starts_with("null", rest) || string_starts_with("true", rest) || string_starts_with("false", rest) || std::regex_match(sv.begin(), sv.end(), number_regex)) {
-                        value_json = {123, {"123", "123"}};
-                        builder.consume_rest();
-                    } else {
-                        builder.move_to(val_start);
-                    }
-                }
-            }
-
-            // If it is a JSON and followed by </arg_value>, parse as json
-            // cannot support streaming because it may be a plain text starting with JSON
-            if (value_json) {
-                auto json_end = builder.pos();
-                builder.consume_spaces();
-                if (builder.pos() == builder.input().size()) {
-                    if (form.raw_argval && !*form.raw_argval && (value_json->json.is_string() || value_json->json.is_object() || value_json->json.is_array())) {
-                        arguments[key] = value_json->json;
-                        auto json_str = arguments.dump();
-                        if (!value_json->healing_marker.json_dump_marker.empty()) {
-                            GGML_ASSERT(std::string::npos != json_str.rfind(value_json->healing_marker.json_dump_marker));
-                            json_str.resize(json_str.rfind(value_json->healing_marker.json_dump_marker));
-                        } else {
-                            GGML_ASSERT(json_str.back() == '}');
-                            json_str.resize(json_str.size() - 1);
-                        }
-                        builder.add_tool_call(function_name, "", json_str);
-                    } else {
-                        gen_partial_args([&](auto &, auto &needle) {arguments[key] = needle;});
-                    }
-                    LOG_DBG("Possible JSON arg_value: %s\n", value_json->json.dump().c_str());
-                    throw common_chat_msg_partial_exception("JSON arg_value detected. Waiting for more tokens for validations.");
-                }
-                builder.move_to(json_end);
-                auto [val_end_size, tc] = try_find_val_end();
-                if (tc && all_space(tc->prelude) && value_json->healing_marker.marker.empty()) {
-                    if (tc->groups[0].end - tc->groups[0].begin != val_end_size) {
-                        gen_partial_args([&](auto &, auto &needle) {arguments[key] = needle;});
-                        LOG_DBG("Possible terminated JSON arg_value: %s\n", value_json->json.dump().c_str());
-                        throw common_chat_msg_partial_exception("Partial literal: " + gbnf_format_literal(form.val_end) + (form.last_val_end ? gbnf_format_literal(*form.last_val_end) : ""));
-                    } else arguments[key] = value_json->json;
-                } else builder.move_to(val_start);
-            }
-
-            // If not, parse as plain text
-            if (val_start == builder.pos()) {
-                if (auto [val_end_size, value_plain] = try_find_val_end(); value_plain) {
-                    auto &value_str = value_plain->prelude;
-                    if (form.trim_raw_argval) value_str = string_strip(value_str);
-                    if (value_plain->groups[0].end - value_plain->groups[0].begin != val_end_size) {
-                        gen_partial_args([&](auto &, auto &needle) {arguments[key] = value_str + needle;});
-                        throw common_chat_msg_partial_exception(
-                                "Expected " + gbnf_format_literal(form.val_end) +
-                                " after " + gbnf_format_literal(form.key_val_sep) +
-                                (form.key_val_sep2 ? " " + gbnf_format_literal(*form.key_val_sep2) : "")
-                        );
-                    }
-                    arguments[key] = value_str;
-                } else {
-                    if (form.trim_raw_argval) {
-                        gen_partial_args([&](auto &rest, auto &needle) {arguments[key] = string_strip(rest) + needle;});
-                    } else {
-                        gen_partial_args([&](auto &rest, auto &needle) {arguments[key] = rest + needle;});
-                    }
-                    throw common_chat_msg_partial_exception(
-                            "Expected " + gbnf_format_literal(form.val_end) +
-                            " after " + gbnf_format_literal(form.key_val_sep) +
-                            (form.key_val_sep2 ? " " + gbnf_format_literal(*form.key_val_sep2) : "")
-                    );
-                }
-            }
-        }
-
-        // Consume closing tag
-        if (auto [tool_end_size, tc] = try_find_tool_end(); tc) {
-            if (!all_space(tc->prelude)) {
-                LOG_DBG("Failed to parse XML-Style tool call: Expected %s, but found %s\n",
-                        gbnf_format_literal(form.tool_end).c_str(),
-                        gbnf_format_literal(tc->prelude).c_str()
-                );
-                return return_error(builder, start_pos, recovery);
-            }
-            if (tc->groups[0].end - tc->groups[0].begin == tool_end_size) {
-                // Add the parsed tool call
-                if (!builder.add_tool_call(function_name, "", arguments.dump())) {
-                    throw common_chat_msg_partial_exception("Failed to add XML-Style tool call");
-                }
-                recovery = false;
-                continue;
-            }
-        }
-
-        auto tool_call_arg = arguments.dump();
-        if (tool_call_arg.size() != 0 && tool_call_arg[tool_call_arg.size() - 1] == '}') {
-            tool_call_arg.resize(tool_call_arg.size() - 1);
-        }
-        builder.add_tool_call(function_name, "", tool_call_arg);
-        throw common_chat_msg_partial_exception("Expected " + gbnf_format_literal(form.tool_end) + " after " + gbnf_format_literal(form.val_end));
-    }
-    if (auto tc = builder.try_find_literal(form.scope_end)) {
-        if (!all_space(tc->prelude)) {
-            LOG_DBG("Failed to parse XML-Style tool call: Expected %s, but found %s\n",
-                    gbnf_format_literal(form.scope_end).c_str(),
-                    gbnf_format_literal(tc->prelude).c_str()
-            );
-            return return_error(builder, start_pos, recovery);
-        }
-    } else {
-        if (all_space(form.scope_end)) return true;
-        builder.consume_spaces();
-        if (builder.pos() == builder.input().size())
-            throw common_chat_msg_partial_exception("incomplete tool calls");
-        LOG_DBG("Failed to parse XML-Style tool call: Expected %s, but found %s\n",
-                gbnf_format_literal(form.scope_end).c_str(),
-                gbnf_format_literal(builder.consume_rest()).c_str()
-        );
-        return return_error(builder, start_pos, recovery);
-    }
-
-    return true;
-}
-
-/**
- * Parse XML-Style tool call for given xml_tool_call_format. Return false for invalid syntax and get the position untouched.
- * May cause std::runtime_error if there is invalid syntax because partial valid tool call is already sent out to client.
- * form.scope_start, form.tool_sep and form.scope_end can be empty.
- */
-bool common_chat_msg_parser::try_consume_xml_tool_calls(const struct xml_tool_call_format & form) {
-    auto pos = pos_;
-    auto tsize = result_.tool_calls.size();
-    try { return parse_xml_tool_calls(*this, form); }
-    catch (const xml_toolcall_syntax_exception&) {}
-    move_to(pos);
-    result_.tool_calls.resize(tsize);
-    return false;
-}
-
-/**
- * Parse content uses reasoning and XML-Style tool call
- * TODO: Note that form.allow_toolcall_in_think is not tested yet. If anyone confirms it works, this comment can be removed.
- */
-inline void parse_msg_with_xml_tool_calls(common_chat_msg_parser & builder, const struct xml_tool_call_format & form, const std::string & start_think = "<think>", const std::string & end_think = "</think>") {
-    constexpr auto rstrip = [](std::string &s) {
-        s.resize(std::distance(s.begin(), std::find_if(s.rbegin(), s.rend(), [](unsigned char ch) { return !std::isspace(ch); }).base()));
-    };
-    // Erase substring from l to r, along with additional spaces nearby
-    constexpr auto erase_spaces = [](auto &str, size_t l, size_t r) {
-        while (/* l > -1 && */ --l < str.size() && std::isspace(static_cast<unsigned char>(str[l])));
-        ++l;
-        while (++r < str.size() && std::isspace(static_cast<unsigned char>(str[r])));
-        if (l < r) str[l] = '\n';
-        if (l + 1 < r) str[l + 1] = '\n';
-        if (l != 0) l += 2;
-        str.erase(l, r - l);
-        return l;
-    };
-    constexpr auto trim_suffix = [](std::string &content, std::initializer_list<std::string_view> list) {
-        auto best_match = content.size();
-        for (auto pattern: list) {
-            if (pattern.size() == 0) continue;
-            for (auto match_idx = content.size() - std::min(pattern.size(), content.size()); content.size() > match_idx; match_idx++) {
-                auto match_len = content.size() - match_idx;
-                if (content.compare(match_idx, match_len, pattern.data(), match_len) == 0 && best_match > match_idx) {
-                    best_match = match_idx;
-                }
-            }
-        }
-        if (content.size() > best_match) {
-            content.erase(best_match);
-        }
-    };
-    const auto trim_potential_partial_word = [&start_think, &end_think, &form, trim_suffix](std::string &content) {
-        return trim_suffix(content, {
-            start_think, end_think, form.scope_start, form.tool_start, form.tool_sep, form.key_start,
-            form.key_val_sep, form.key_val_sep2 ? form.key_val_sep2->c_str() : "",
-            form.val_end, form.last_val_end ? form.last_val_end->c_str() : "",
-            form.tool_end, form.last_tool_end ? form.last_tool_end->c_str() : "",
-            form.scope_end
-        });
-    };
-
-
-    // Trim leading spaces without affecting keyword matching
-    static const common_regex spaces_regex("\\s*");
-    {
-        auto tc = builder.consume_regex(spaces_regex);
-        auto spaces = builder.str(tc.groups[0]);
-        auto s1 = spaces.size();
-        trim_potential_partial_word(spaces);
-        auto s2 = spaces.size();
-        builder.move_to(builder.pos() - (s1 - s2));
-    }
-
-    // Parse content
-    bool reasoning_unclosed = builder.syntax().thinking_forced_open;
-    std::string unclosed_reasoning_content("");
-    for (;;) {
-        auto tc = try_find_2_literal_splited_by_spaces(builder, form.scope_start, form.tool_start);
-        std::string content;
-        std::string tool_call_start;
-
-        if (tc) {
-            content = std::move(tc->prelude);
-            tool_call_start = builder.str(tc->groups[0]);
-            LOG_DBG("Matched tool start: %s\n", gbnf_format_literal(tool_call_start).c_str());
-        } else {
-            content = builder.consume_rest();
-            utf8_truncate_safe_resize(content);
-        }
-
-        // Handle unclosed think block
-        if (reasoning_unclosed) {
-            if (auto pos = content.find(end_think); pos == std::string::npos && builder.pos() != builder.input().size()) {
-                unclosed_reasoning_content += content;
-                if (!(form.allow_toolcall_in_think && tc)) {
-                    unclosed_reasoning_content += tool_call_start;
-                    continue;
-                }
-            } else {
-                reasoning_unclosed = false;
-                std::string reasoning_content;
-                if (pos == std::string::npos) {
-                    reasoning_content = std::move(content);
-                } else {
-                    reasoning_content = content.substr(0, pos);
-                    content.erase(0, pos + end_think.size());
-                }
-                if (builder.pos() == builder.input().size() && all_space(content)) {
-                    rstrip(reasoning_content);
-                    trim_potential_partial_word(reasoning_content);
-                    rstrip(reasoning_content);
-                    if (reasoning_content.empty()) {
-                        rstrip(unclosed_reasoning_content);
-                        trim_potential_partial_word(unclosed_reasoning_content);
-                        rstrip(unclosed_reasoning_content);
-                        if (unclosed_reasoning_content.empty()) continue;
-                    }
-                }
-                if (builder.syntax().reasoning_format == COMMON_REASONING_FORMAT_NONE || builder.syntax().reasoning_in_content) {
-                    builder.add_content(start_think);
-                    builder.add_content(unclosed_reasoning_content);
-                    builder.add_content(reasoning_content);
-                    if (builder.pos() != builder.input().size() || !all_space(content))
-                        builder.add_content(end_think);
-                } else {
-                    builder.add_reasoning_content(unclosed_reasoning_content);
-                    builder.add_reasoning_content(reasoning_content);
-                }
-                unclosed_reasoning_content.clear();
-            }
-        }
-
-        // Handle multiple think block
-        bool toolcall_in_think = false;
-        for (auto think_start = content.find(start_think); think_start != std::string::npos; think_start = content.find(start_think, think_start)) {
-            if (auto think_end = content.find(end_think, think_start + start_think.size()); think_end != std::string::npos) {
-                if (builder.syntax().reasoning_format != COMMON_REASONING_FORMAT_NONE && !builder.syntax().reasoning_in_content) {
-                    auto reasoning_content = content.substr(think_start + start_think.size(), think_end - think_start - start_think.size());
-                    builder.add_reasoning_content(reasoning_content);
-                    think_start = erase_spaces(content, think_start, think_end + end_think.size() - 1);
-                } else {
-                    think_start = think_end + end_think.size() - 1;
-                }
-            } else {
-                // This <tool_call> start is in thinking block, skip this tool call
-                // This <tool_call> start is in thinking block
-                if (form.allow_toolcall_in_think) {
-                    unclosed_reasoning_content = content.substr(think_start + start_think.size());
-                } else {
-                    unclosed_reasoning_content = content.substr(think_start + start_think.size()) + tool_call_start;
-                }
-                reasoning_unclosed = true;
-                content.resize(think_start);
-                toolcall_in_think = true;
-            }
-        }
-
-        if (builder.syntax().reasoning_format != COMMON_REASONING_FORMAT_NONE && !builder.syntax().reasoning_in_content) {
-            rstrip(content);
-            // Handle unclosed </think> token from content: delete all </think> token
-            if (auto pos = content.rfind(end_think); pos != std::string::npos) {
-                while (pos != std::string::npos) {
-                    pos = erase_spaces(content, pos, pos + end_think.size() - 1);
-                    pos = content.rfind(end_think, pos);
-                }
-            }
-            // Strip if needed
-            if (content.size() > 0 && std::isspace(static_cast<unsigned char>(content[0]))) {
-                content = string_strip(content);
-            }
-        }
-
-        // remove potential partial suffix
-        if (builder.pos() == builder.input().size()) {
-            if (unclosed_reasoning_content.empty()) {
-                rstrip(content);
-                trim_potential_partial_word(content);
-                rstrip(content);
-            } else {
-                rstrip(unclosed_reasoning_content);
-                trim_potential_partial_word(unclosed_reasoning_content);
-                rstrip(unclosed_reasoning_content);
-            }
-        }
-
-        // consume unclosed_reasoning_content if allow_toolcall_in_think is set
-        if (form.allow_toolcall_in_think && !unclosed_reasoning_content.empty()) {
-            if (builder.syntax().reasoning_format != COMMON_REASONING_FORMAT_NONE && !builder.syntax().reasoning_in_content) {
-                builder.add_reasoning_content(unclosed_reasoning_content);
-            } else {
-                if (content.empty()) {
-                    content = start_think + unclosed_reasoning_content;
-                } else {
-                    content += "\n\n" + start_think;
-                    content += unclosed_reasoning_content;
-                }
-            }
-            unclosed_reasoning_content.clear();
-        }
-
-        // Add content
-        if (!content.empty()) {
-            // If there are multiple content blocks
-            if (builder.syntax().reasoning_format != COMMON_REASONING_FORMAT_NONE && !builder.syntax().reasoning_in_content && builder.result().content.size() != 0) {
-                builder.add_content("\n\n");
-            }
-            builder.add_content(content);
-        }
-
-        // This <tool_call> start is in thinking block and toolcall_in_think not set, skip this tool call
-        if (toolcall_in_think && !form.allow_toolcall_in_think) {
-            continue;
-        }
-
-        // There is no tool call and all content is parsed
-        if (!tc) {
-            GGML_ASSERT(builder.pos() == builder.input().size());
-            GGML_ASSERT(unclosed_reasoning_content.empty());
-            if (!form.allow_toolcall_in_think) GGML_ASSERT(!reasoning_unclosed);
-            break;
-        }
-
-        builder.move_to(tc->groups[0].begin);
-        if (builder.try_consume_xml_tool_calls(form)) {
-            auto end_of_tool = builder.pos();
-            builder.consume_spaces();
-            if (builder.pos() != builder.input().size()) {
-                builder.move_to(end_of_tool);
-                if (!builder.result().content.empty()) {
-                    builder.add_content("\n\n");
-                }
-            }
-        } else {
-            static const common_regex next_char_regex(".");
-            auto c = builder.str(builder.consume_regex(next_char_regex).groups[0]);
-            rstrip(c);
-            builder.add_content(c);
-        }
-    }
-}
-
-/**
- * Parse content uses reasoning and XML-Style tool call
- */
-void common_chat_msg_parser::consume_reasoning_with_xml_tool_calls(const struct xml_tool_call_format & form, const std::string & start_think, const std::string & end_think) {
-    parse_msg_with_xml_tool_calls(*this, form, start_think, end_think);
-}

common/chat-parser-xml-toolcall.h

@@ -1,45 +0,0 @@
-#pragma once
-
-#include "chat.h"
-
-#include <nlohmann/json.hpp>
-
-#include <optional>
-#include <string>
-#include <vector>
-
-
-// Sample config:
-// MiniMax-M2 (left): <minimax:tool_call>\n<invoke name="tool-name">\n<parameter name="key">value</parameter>\n...</invoke>\n...</minimax:tool_call>
-// GLM 4.5   (right): <tool_call>function_name\n<arg_key>key</arg_key>\n<arg_value>value</arg_value>\n</tool_call>
-struct xml_tool_call_format {
-    std::string scope_start; // <minimax:tool_call>\n  // \n                      // can be empty
-    std::string tool_start;  // <invoke name=\"        // <tool_call>
-    std::string tool_sep;    // \">\n                  // \n                      // can be empty only for parse_xml_tool_calls
-    std::string key_start;   // <parameter name=\"     // <arg_key>
-    std::string key_val_sep; // \">                    // </arg_key>\n<arg_value>
-    std::string val_end;     // </parameter>\n         // </arg_value>\n
-    std::string tool_end;    // </invoke>\n            // </tool_call>\n
-    std::string scope_end;   // </minimax:tool_call>   //                         // can be empty
-    // Set this if there can be dynamic spaces inside key_val_sep.
-    // e.g. key_val_sep=</arg_key> key_val_sep2=<arg_value> for GLM4.5
-    std::optional<std::string> key_val_sep2 = std::nullopt;
-    // Set true if argval should only be raw string. e.g. Hello "world" hi
-    // Set false if argval should only be json string. e.g. "Hello \"world\" hi"
-    // Defaults to std::nullopt, both will be allowed.
-    std::optional<bool> raw_argval = std::nullopt;
-    std::optional<std::string> last_val_end = std::nullopt;
-    std::optional<std::string> last_tool_end = std::nullopt;
-    bool trim_raw_argval = false;
-    bool allow_toolcall_in_think = false;
-};
-
-// make a GBNF that accept any strings except those containing any of the forbidden strings.
-std::string make_gbnf_excluding(std::vector<std::string> forbids);
-
-/**
- * Build grammar for xml-style tool call
- * form.scope_start and form.scope_end can be empty.
- * Requires data.format for model-specific hacks.
- */
-void build_grammar_xml_tool_call(common_chat_params & data, const nlohmann::ordered_json & tools, const struct xml_tool_call_format & form);

common/chat-parser.h

@@ -1,7 +1,6 @@
 #pragma once
 
 #include "chat.h"
-#include "chat-parser-xml-toolcall.h"
 #include "json-partial.h"
 #include "regex-partial.h"
 
@@ -65,9 +64,6 @@ class common_chat_msg_parser {
     // Adds an array of tool calls using their "name", "id" and "arguments" fields.
     bool add_tool_calls(const nlohmann::ordered_json & arr);
 
-    // Adds a tool call using the short form: { "tool_name": { "arg1": val, "arg2": val } }
-    bool add_tool_call_short_form(const nlohmann::ordered_json & tool_call);
-
     void finish();
 
     bool consume_spaces();
@@ -120,14 +116,5 @@ class common_chat_msg_parser {
         const std::vector<std::vector<std::string>> & content_paths = {}
     );
 
-    /**
-     * Parse XML-Style tool call for given xml_tool_call_format. Return false for invalid syntax and get the position untouched.
-     * form.scope_start, form.tool_sep and form.scope_end can be empty.
-     */
-    bool try_consume_xml_tool_calls(const struct xml_tool_call_format & form);
-
-    // Parse content uses reasoning and XML-Style tool call
-    void consume_reasoning_with_xml_tool_calls(const struct xml_tool_call_format & form, const std::string & start_think = "<think>", const std::string & end_think = "</think>");
-
     void clear_tools();
 };

common/chat-peg-parser.cpp

@@ -1,124 +0,0 @@
-#include "chat-peg-parser.h"
-
-#include <nlohmann/json.hpp>
-
-using json = nlohmann::json;
-
-static std::string_view trim_trailing_space(std::string_view sv, int max = -1) {
-    int count = 0;
-    while (!sv.empty() && std::isspace(static_cast<unsigned char>(sv.back()))) {
-        if (max != -1 && count <= max) {
-            break;
-        }
-        sv.remove_suffix(1);
-        count++;
-    }
-    return sv;
-}
-
-void common_chat_peg_mapper::from_ast(const common_peg_ast_arena & arena, const common_peg_parse_result & result) {
-    arena.visit(result, [this](const common_peg_ast_node & node) {
-        map(node);
-    });
-}
-
-void common_chat_peg_mapper::map(const common_peg_ast_node & node) {
-    bool is_reasoning = node.tag == common_chat_peg_builder::REASONING;
-    bool is_content = node.tag == common_chat_peg_builder::CONTENT;
-
-    if (is_reasoning) {
-        result.reasoning_content = std::string(trim_trailing_space(node.text));
-    }
-
-    if (is_content) {
-        result.content = std::string(trim_trailing_space(node.text));
-    }
-}
-
-void common_chat_peg_native_mapper::map(const common_peg_ast_node & node) {
-    common_chat_peg_mapper::map(node);
-
-    bool is_tool_open = node.tag == common_chat_peg_native_builder::TOOL_OPEN;
-    bool is_tool_name = node.tag == common_chat_peg_native_builder::TOOL_NAME;
-    bool is_tool_id = node.tag == common_chat_peg_native_builder::TOOL_ID;
-    bool is_tool_args = node.tag == common_chat_peg_native_builder::TOOL_ARGS;
-
-    if (is_tool_open) {
-        result.tool_calls.emplace_back();
-        current_tool = &result.tool_calls.back();
-    }
-
-    if (is_tool_id && current_tool) {
-        current_tool->id = std::string(trim_trailing_space(node.text));
-    }
-
-    if (is_tool_name && current_tool) {
-        current_tool->name = std::string(trim_trailing_space(node.text));
-    }
-
-    if (is_tool_args && current_tool) {
-        current_tool->arguments = std::string(trim_trailing_space(node.text));
-    }
-}
-
-void common_chat_peg_constructed_mapper::map(const common_peg_ast_node & node) {
-    common_chat_peg_mapper::map(node);
-
-    bool is_tool_open = node.tag == common_chat_peg_constructed_builder::TOOL_OPEN;
-    bool is_tool_name = node.tag == common_chat_peg_constructed_builder::TOOL_NAME;
-    bool is_tool_close = node.tag == common_chat_peg_constructed_builder::TOOL_CLOSE;
-    bool is_arg_open = node.tag == common_chat_peg_constructed_builder::TOOL_ARG_OPEN;
-    bool is_arg_close = node.tag == common_chat_peg_constructed_builder::TOOL_ARG_CLOSE;
-    bool is_arg_name = node.tag == common_chat_peg_constructed_builder::TOOL_ARG_NAME;
-    bool is_arg_string = node.tag == common_chat_peg_constructed_builder::TOOL_ARG_STRING_VALUE;
-    bool is_arg_json = node.tag == common_chat_peg_constructed_builder::TOOL_ARG_JSON_VALUE;
-
-    if (is_tool_open) {
-        result.tool_calls.emplace_back();
-        current_tool = &result.tool_calls.back();
-        arg_count = 0;
-    }
-
-    if (is_tool_name) {
-        current_tool->name = std::string(node.text);
-        current_tool->arguments = "{";
-    }
-
-    if (is_arg_open) {
-        needs_closing_quote = false;
-    }
-
-    if (is_arg_name && current_tool) {
-        if (arg_count > 0) {
-            current_tool->arguments += ",";
-        }
-        current_tool->arguments += json(trim_trailing_space(node.text)).dump() + ":";
-        ++arg_count;
-    }
-
-    if (is_arg_string && current_tool) {
-        // Serialize to JSON, but exclude the end quote
-        std::string dumped = json(trim_trailing_space(node.text)).dump();
-        current_tool->arguments += dumped.substr(0, dumped.size() - 1);
-        needs_closing_quote = true;
-    }
-
-    if (is_arg_close && current_tool) {
-        if (needs_closing_quote) {
-            current_tool->arguments += "\"";
-            needs_closing_quote = false;
-        }
-    }
-
-    if (is_arg_json && current_tool) {
-        current_tool->arguments += std::string(trim_trailing_space(node.text));
-    }
-
-    if (is_tool_close && current_tool) {
-        if (needs_closing_quote) {
-            current_tool->arguments += "\"";
-            needs_closing_quote = false;
-        }
-        current_tool->arguments += "}";
-    }
-}

common/chat-peg-parser.h

@@ -1,105 +0,0 @@
-#pragma once
-
-#include "chat.h"
-#include "peg-parser.h"
-
-class common_chat_peg_builder : public common_peg_parser_builder {
-  public:
-    static constexpr const char * REASONING_BLOCK = "reasoning-block";
-    static constexpr const char * REASONING = "reasoning";
-    static constexpr const char * CONTENT = "content";
-
-    common_peg_parser reasoning_block(const common_peg_parser & p) { return tag(REASONING_BLOCK, p); }
-    common_peg_parser reasoning(const common_peg_parser & p) { return tag(REASONING, p); }
-    common_peg_parser content(const common_peg_parser & p) { return tag(CONTENT, p); }
-};
-
-inline common_peg_arena build_chat_peg_parser(const std::function<common_peg_parser(common_chat_peg_builder & builder)> & fn) {
-    common_chat_peg_builder builder;
-    builder.set_root(fn(builder));
-    return builder.build();
-}
-
-class common_chat_peg_mapper {
-  public:
-    common_chat_msg & result;
-
-    common_chat_peg_mapper(common_chat_msg & msg) : result(msg) {}
-
-    virtual void from_ast(const common_peg_ast_arena & arena, const common_peg_parse_result & result);
-    virtual void map(const common_peg_ast_node & node);
-};
-
-class common_chat_peg_native_builder : public common_chat_peg_builder {
-  public:
-    static constexpr const char * TOOL = "tool";
-    static constexpr const char * TOOL_OPEN = "tool-open";
-    static constexpr const char * TOOL_CLOSE = "tool-close";
-    static constexpr const char * TOOL_ID = "tool-id";
-    static constexpr const char * TOOL_NAME = "tool-name";
-    static constexpr const char * TOOL_ARGS = "tool-args";
-
-    common_peg_parser tool(const common_peg_parser & p) { return tag(TOOL, p); }
-    common_peg_parser tool_open(const common_peg_parser & p) { return atomic(tag(TOOL_OPEN, p)); }
-    common_peg_parser tool_close(const common_peg_parser & p) { return atomic(tag(TOOL_CLOSE, p)); }
-    common_peg_parser tool_id(const common_peg_parser & p) { return atomic(tag(TOOL_ID, p)); }
-    common_peg_parser tool_name(const common_peg_parser & p) { return atomic(tag(TOOL_NAME, p)); }
-    common_peg_parser tool_args(const common_peg_parser & p) { return tag(TOOL_ARGS, p); }
-};
-
-class common_chat_peg_native_mapper : public common_chat_peg_mapper {
-    common_chat_tool_call * current_tool;
-
-  public:
-    common_chat_peg_native_mapper(common_chat_msg & msg) : common_chat_peg_mapper(msg) {}
-
-    void map(const common_peg_ast_node & node) override;
-};
-
-inline common_peg_arena build_chat_peg_native_parser(const std::function<common_peg_parser(common_chat_peg_native_builder & builder)> & fn) {
-    common_chat_peg_native_builder builder;
-    builder.set_root(fn(builder));
-    return builder.build();
-}
-
-class common_chat_peg_constructed_builder : public common_chat_peg_builder {
-  public:
-    static constexpr const char * TOOL = "tool";
-    static constexpr const char * TOOL_OPEN = "tool-open";
-    static constexpr const char * TOOL_CLOSE = "tool-close";
-    static constexpr const char * TOOL_NAME = "tool-name";
-    static constexpr const char * TOOL_ARG = "tool-arg";
-    static constexpr const char * TOOL_ARG_OPEN = "tool-arg-open";
-    static constexpr const char * TOOL_ARG_CLOSE = "tool-arg-close";
-    static constexpr const char * TOOL_ARG_NAME = "tool-arg-name";
-    static constexpr const char * TOOL_ARG_STRING_VALUE = "tool-arg-string-value";
-    static constexpr const char * TOOL_ARG_JSON_VALUE = "tool-arg-json-value";
-
-    common_peg_parser tool(const common_peg_parser & p) { return tag(TOOL, p); }
-    common_peg_parser tool_open(const common_peg_parser & p) { return atomic(tag(TOOL_OPEN, p)); }
-    common_peg_parser tool_close(const common_peg_parser & p) { return atomic(tag(TOOL_CLOSE, p)); }
-    common_peg_parser tool_name(const common_peg_parser & p) { return atomic(tag(TOOL_NAME, p)); }
-    common_peg_parser tool_arg(const common_peg_parser & p) { return tag(TOOL_ARG, p); }
-    common_peg_parser tool_arg_open(const common_peg_parser & p) { return atomic(tag(TOOL_ARG_OPEN, p)); }
-    common_peg_parser tool_arg_close(const common_peg_parser & p) { return atomic(tag(TOOL_ARG_CLOSE, p)); }
-    common_peg_parser tool_arg_name(const common_peg_parser & p) { return atomic(tag(TOOL_ARG_NAME, p)); }
-    common_peg_parser tool_arg_string_value(const common_peg_parser & p) { return tag(TOOL_ARG_STRING_VALUE, p); }
-    common_peg_parser tool_arg_json_value(const common_peg_parser & p) { return tag(TOOL_ARG_JSON_VALUE, p); }
-};
-
-class common_chat_peg_constructed_mapper : public common_chat_peg_mapper {
-    common_chat_tool_call * current_tool;
-    int arg_count = 0;
-    bool needs_closing_quote = false;
-
-  public:
-    common_chat_peg_constructed_mapper(common_chat_msg & msg) : common_chat_peg_mapper(msg) {}
-
-    void map(const common_peg_ast_node & node) override;
-};
-
-inline common_peg_arena build_chat_peg_constructed_parser(const std::function<common_peg_parser(common_chat_peg_constructed_builder & builder)> & fn) {
-    common_chat_peg_constructed_builder builder;
-    builder.set_root(fn(builder));
-    return builder.build();
-}

common/chat.h

@@ -3,7 +3,6 @@
 #pragma once
 
 #include "common.h"
-#include "peg-parser.h"
 #include <functional>
 #include <chrono>
 #include <string>
@@ -34,8 +33,8 @@ struct common_chat_msg_content_part {
 struct common_chat_msg {
     std::string role;
     std::string content;
-    std::vector<common_chat_msg_content_part> content_parts;
-    std::vector<common_chat_tool_call> tool_calls;
+    std::vector<common_chat_msg_content_part> content_parts = {};
+    std::vector<common_chat_tool_call> tool_calls = {};
     std::string reasoning_content;
     std::string tool_name;
     std::string tool_call_id;
@@ -45,7 +44,7 @@ struct common_chat_msg {
     bool empty() const {
         return content.empty() && content_parts.empty() && tool_calls.empty() && reasoning_content.empty() && tool_name.empty() && tool_call_id.empty();
     }
-    void set_tool_call_ids(std::vector<std::string> & ids_cache, const std::function<std::string()> & gen_tool_call_id) {
+    void ensure_tool_call_ids_set(std::vector<std::string> & ids_cache, const std::function<std::string()> & gen_tool_call_id) {
         for (auto i = 0u; i < tool_calls.size(); i++) {
             if (ids_cache.size() <= i) {
                 auto id = tool_calls[i].id;
@@ -77,7 +76,7 @@ struct common_chat_msg_diff {
     size_t tool_call_index = std::string::npos;
     common_chat_tool_call tool_call_delta;
 
-    static std::vector<common_chat_msg_diff> compute_diffs(const common_chat_msg & msg_prv, const common_chat_msg & msg_new);
+    static std::vector<common_chat_msg_diff> compute_diffs(const common_chat_msg & previous_msg, const common_chat_msg & new_msg);
 
     bool operator==(const common_chat_msg_diff & other) const {
         return content_delta == other.content_delta
@@ -102,33 +101,14 @@ enum common_chat_format {
     COMMON_CHAT_FORMAT_CONTENT_ONLY,
     COMMON_CHAT_FORMAT_GENERIC,
     COMMON_CHAT_FORMAT_MISTRAL_NEMO,
-    COMMON_CHAT_FORMAT_MAGISTRAL,
     COMMON_CHAT_FORMAT_LLAMA_3_X,
     COMMON_CHAT_FORMAT_LLAMA_3_X_WITH_BUILTIN_TOOLS,
     COMMON_CHAT_FORMAT_DEEPSEEK_R1,
     COMMON_CHAT_FORMAT_FIREFUNCTION_V2,
     COMMON_CHAT_FORMAT_FUNCTIONARY_V3_2,
     COMMON_CHAT_FORMAT_FUNCTIONARY_V3_1_LLAMA_3_1,
-    COMMON_CHAT_FORMAT_DEEPSEEK_V3_1,
     COMMON_CHAT_FORMAT_HERMES_2_PRO,
     COMMON_CHAT_FORMAT_COMMAND_R7B,
-    COMMON_CHAT_FORMAT_GRANITE,
-    COMMON_CHAT_FORMAT_GPT_OSS,
-    COMMON_CHAT_FORMAT_SEED_OSS,
-    COMMON_CHAT_FORMAT_NEMOTRON_V2,
-    COMMON_CHAT_FORMAT_APERTUS,
-    COMMON_CHAT_FORMAT_LFM2_WITH_JSON_TOOLS,
-    COMMON_CHAT_FORMAT_GLM_4_5,
-    COMMON_CHAT_FORMAT_MINIMAX_M2,
-    COMMON_CHAT_FORMAT_KIMI_K2,
-    COMMON_CHAT_FORMAT_QWEN3_CODER_XML,
-    COMMON_CHAT_FORMAT_APRIEL_1_5,
-    COMMON_CHAT_FORMAT_XIAOMI_MIMO,
-
-    // These are intended to be parsed by the PEG parser
-    COMMON_CHAT_FORMAT_PEG_SIMPLE,
-    COMMON_CHAT_FORMAT_PEG_NATIVE,
-    COMMON_CHAT_FORMAT_PEG_CONSTRUCTED,
 
     COMMON_CHAT_FORMAT_COUNT, // Not a format, just the # formats
 };
@@ -147,8 +127,6 @@ struct common_chat_templates_inputs {
     bool enable_thinking = true;
     std::chrono::system_clock::time_point now = std::chrono::system_clock::now();
     std::map<std::string, std::string> chat_template_kwargs;
-    bool add_bos = false;
-    bool add_eos = false;
 };
 
 struct common_chat_params {
@@ -160,7 +138,6 @@ struct common_chat_params {
     std::vector<common_grammar_trigger> grammar_triggers;
     std::vector<std::string>            preserved_tokens;
     std::vector<std::string>            additional_stops;
-    std::string                         parser;
 };
 
 struct common_chat_syntax {
@@ -170,7 +147,6 @@ struct common_chat_syntax {
     bool                     reasoning_in_content  = false;
     bool                     thinking_forced_open  = false;
     bool                     parse_tool_calls      = true;
-    common_peg_arena         parser                = {};
 };
 
 // Check if the template supplied via "--chat-template" is supported or not. Returns true if it's valid
@@ -207,19 +183,14 @@ std::string common_chat_format_single(
 // Returns an example of formatted chat
 std::string common_chat_format_example(
     const struct common_chat_templates * tmpls,
-    bool use_jinja,
-    const std::map<std::string, std::string> & chat_template_kwargs);
+    bool use_jinja);
 
 const char*               common_chat_format_name(common_chat_format format);
 const char*               common_reasoning_format_name(common_reasoning_format format);
-common_reasoning_format   common_reasoning_format_from_name(const std::string & format);
 common_chat_msg           common_chat_parse(const std::string & input, bool is_partial, const common_chat_syntax & syntax);
-common_chat_msg           common_chat_peg_parse(const common_peg_arena & parser, const std::string & input, bool is_partial, const common_chat_syntax & syntax);
 
 common_chat_tool_choice common_chat_tool_choice_parse_oaicompat(const std::string & tool_choice);
 
-bool common_chat_templates_support_enable_thinking(const common_chat_templates * chat_templates);
-
 // Parses a JSON array of messages in OpenAI's chat completion API format.
 // T can be std::string containing JSON or nlohmann::ordered_json
 template <class T> std::vector<common_chat_msg> common_chat_msgs_parse_oaicompat(const T & messages);

common/common.cpp

@@ -8,14 +8,12 @@
 #include "common.h"
 #include "log.h"
 #include "llama.h"
-#include "sampling.h"
 
 #include <algorithm>
 #include <cinttypes>
 #include <climits>
 #include <cmath>
 #include <codecvt>
-#include <chrono>
 #include <cstdarg>
 #include <cstring>
 #include <ctime>
@@ -27,6 +25,7 @@
 #include <sstream>
 #include <string>
 #include <thread>
+#include <unordered_map>
 #include <unordered_set>
 #include <vector>
 
@@ -42,7 +41,6 @@
 #endif
 #include <locale>
 #include <windows.h>
-#include <string.h>
 #include <fcntl.h>
 #include <io.h>
 #else
@@ -51,23 +49,10 @@
 #include <unistd.h>
 #endif
 
-#if defined(__linux__)
-#include <sys/types.h>
-#include <pwd.h>
-#endif
-
 #if defined(_MSC_VER)
 #pragma warning(disable: 4244 4267) // possible loss of data
 #endif
 
-common_time_meas::common_time_meas(int64_t & t_acc, bool disable) : t_start_us(disable ? -1 : ggml_time_us()), t_acc(t_acc) {}
-
-common_time_meas::~common_time_meas() {
-    if (t_start_us >= 0) {
-        t_acc += ggml_time_us() - t_start_us;
-    }
-}
-
 //
 // CPU utils
 //
@@ -363,7 +348,11 @@ bool parse_cpu_mask(const std::string & mask, bool (&boolmask)[GGML_MAX_N_THREAD
 }
 
 void common_init() {
-    llama_log_set(common_log_default_callback, NULL);
+    llama_log_set([](ggml_log_level level, const char * text, void * /*user_data*/) {
+        if (LOG_DEFAULT_LLAMA <= common_log_verbosity_thold) {
+            common_log_add(common_log_main(), level, "%s", text);
+        }
+    }, NULL);
 
 #ifdef NDEBUG
     const char * build_type = "";
@@ -568,6 +557,13 @@ std::string string_from(const struct llama_context * ctx, const std::vector<llam
 
         auto detokenized = common_token_to_piece(ctx, token);
 
+        detokenized.erase(
+            std::remove_if(
+                detokenized.begin(),
+                detokenized.end(),
+                [](const unsigned char c) { return !std::isprint(c); }),
+            detokenized.end());
+
         buf << "'" << detokenized << "'"
             << ":" << std::to_string(token);
     }
@@ -592,6 +588,13 @@ std::string string_from(const struct llama_context * ctx, const struct llama_bat
 
         auto detokenized = common_token_to_piece(ctx, batch.token[i]);
 
+        detokenized.erase(
+                std::remove_if(
+                    detokenized.begin(),
+                    detokenized.end(),
+                    [](const unsigned char c) { return !std::isprint(c); }),
+                detokenized.end());
+
         buf << "\n"          << std::to_string(i)
             << ", token '"   << detokenized << "'"
             << ", pos "      << std::to_string(batch.pos[i])
@@ -694,7 +697,7 @@ bool string_parse_kv_override(const char * data, std::vector<llama_model_kv_over
 
 // Validate if a filename is safe to use
 // To validate a full path, split the path by the OS-specific path separator, and validate each part with this function
-bool fs_validate_filename(const std::string & filename, bool allow_subdirs) {
+bool fs_validate_filename(const std::string & filename) {
     if (!filename.length()) {
         // Empty filename invalid
         return false;
@@ -754,14 +757,10 @@ bool fs_validate_filename(const std::string & filename, bool allow_subdirs) {
             || (c >= 0xD800 && c <= 0xDFFF) // UTF-16 surrogate pairs
             || c == 0xFFFD // Replacement Character (UTF-8)
             || c == 0xFEFF // Byte Order Mark (BOM)
-            || c == ':' || c == '*' // Illegal characters
+            || c == '/' || c == '\\' || c == ':' || c == '*' // Illegal characters
             || c == '?' || c == '"' || c == '<' || c == '>' || c == '|') {
             return false;
         }
-        if (!allow_subdirs && (c == '/' || c == '\\')) {
-            // Subdirectories not allowed, reject path separators
-            return false;
-        }
     }
 
     // Reject any leading or trailing ' ', or any trailing '.', these are stripped on Windows and will cause a different filename
@@ -786,29 +785,11 @@ bool fs_validate_filename(const std::string & filename, bool allow_subdirs) {
 #include <iostream>
 
 
-#ifdef _WIN32
-static std::wstring utf8_to_wstring(const std::string & str) {
-    if (str.empty()) {
-        return std::wstring();
-    }
-
-    int size = MultiByteToWideChar(CP_UTF8, 0, str.c_str(), (int)str.size(), NULL, 0);
-
-    if (size <= 0) {
-        return std::wstring();
-    }
-
-    std::wstring wstr(size, 0);
-    MultiByteToWideChar(CP_UTF8, 0, str.c_str(), (int)str.size(), &wstr[0], size);
-
-    return wstr;
-}
-#endif
-
 // returns true if successful, false otherwise
 bool fs_create_directory_with_parents(const std::string & path) {
 #ifdef _WIN32
-    std::wstring wpath = utf8_to_wstring(path);
+    std::wstring_convert<std::codecvt_utf8<wchar_t>> converter;
+    std::wstring wpath = converter.from_bytes(path);
 
     // if the path already exists, check whether it's a directory
     const DWORD attributes = GetFileAttributesW(wpath.c_str());
@@ -881,11 +862,6 @@ bool fs_create_directory_with_parents(const std::string & path) {
 #endif // _WIN32
 }
 
-bool fs_is_directory(const std::string & path) {
-    std::filesystem::path dir(path);
-    return std::filesystem::exists(dir) && std::filesystem::is_directory(dir);
-}
-
 std::string fs_get_cache_directory() {
     std::string cache_directory = "";
     auto ensure_trailing_slash = [](std::string p) {
@@ -901,27 +877,13 @@ std::string fs_get_cache_directory() {
 #if defined(__linux__) || defined(__FreeBSD__) || defined(_AIX) || defined(__OpenBSD__)
         if (std::getenv("XDG_CACHE_HOME")) {
             cache_directory = std::getenv("XDG_CACHE_HOME");
-        } else if (std::getenv("HOME")) {
-            cache_directory = std::getenv("HOME") + std::string("/.cache/");
         } else {
-#if defined(__linux__)
-            /* no $HOME is defined, fallback to getpwuid */
-            struct passwd *pw = getpwuid(getuid());
-            if ((!pw) || (!pw->pw_dir)) {
-                throw std::runtime_error("Failed to find $HOME directory");
-            }
-
-            cache_directory = std::string(pw->pw_dir) + std::string("/.cache/");
-#else /* defined(__linux__) */
-            throw std::runtime_error("Failed to find $HOME directory");
-#endif /* defined(__linux__) */
+            cache_directory = std::getenv("HOME") + std::string("/.cache/");
         }
 #elif defined(__APPLE__)
         cache_directory = std::getenv("HOME") + std::string("/Library/Caches/");
 #elif defined(_WIN32)
         cache_directory = std::getenv("LOCALAPPDATA");
-#elif defined(__EMSCRIPTEN__)
-        GGML_ABORT("not implemented on this platform");
 #else
 #  error Unknown architecture
 #endif
@@ -941,260 +903,32 @@ std::string fs_get_cache_file(const std::string & filename) {
     return cache_directory + filename;
 }
 
-std::vector<common_file_info> fs_list(const std::string & path, bool include_directories) {
-    std::vector<common_file_info> files;
-    if (path.empty()) return files;
-
-    std::filesystem::path dir(path);
-    if (!std::filesystem::exists(dir) || !std::filesystem::is_directory(dir)) {
-        return files;
-    }
-
-    for (const auto & entry : std::filesystem::directory_iterator(dir)) {
-        try {
-            // Only include regular files (skip directories)
-            const auto & p = entry.path();
-            if (std::filesystem::is_regular_file(p)) {
-                common_file_info info;
-                info.path   = p.string();
-                info.name   = p.filename().string();
-                info.is_dir = false;
-                try {
-                    info.size = static_cast<size_t>(std::filesystem::file_size(p));
-                } catch (const std::filesystem::filesystem_error &) {
-                    info.size = 0;
-                }
-                files.push_back(std::move(info));
-            } else if (include_directories && std::filesystem::is_directory(p)) {
-                common_file_info info;
-                info.path   = p.string();
-                info.name   = p.filename().string();
-                info.size   = 0; // Directories have no size
-                info.is_dir = true;
-                files.push_back(std::move(info));
-            }
-        } catch (const std::filesystem::filesystem_error &) {
-            // skip entries we cannot inspect
-            continue;
-        }
-    }
-
-    return files;
-}
-
-//
-// TTY utils
-//
-
-bool tty_can_use_colors() {
-    // Check NO_COLOR environment variable (https://no-color.org/)
-    if (const char * no_color = std::getenv("NO_COLOR")) {
-        if (no_color[0] != '\0') {
-            return false;
-        }
-    }
-
-    // Check TERM environment variable
-    if (const char * term = std::getenv("TERM")) {
-        if (std::strcmp(term, "dumb") == 0) {
-            return false;
-        }
-    }
-
-    // Check if stdout and stderr are connected to a terminal
-    // We check both because log messages can go to either
-    bool stdout_is_tty = isatty(fileno(stdout));
-    bool stderr_is_tty = isatty(fileno(stderr));
-
-    return stdout_is_tty || stderr_is_tty;
-}
 
 //
 // Model utils
 //
 
-// TODO: move to common/sampling
-static void common_init_sampler_from_model(
-    const llama_model * model,
-    common_params_sampling & sparams) {
-
-    const uint64_t config = sparams.user_sampling_config;
-
-    auto get_int32 = [&](const char * key, int32_t & dst, uint64_t user_config) {
-        if (config & user_config) {
-            return;
-        }
-
-        char buf[64] = {0};
-        if (llama_model_meta_val_str(model, key, buf, sizeof(buf)) > 0) {
-            char * end = nullptr;
-            int32_t v = strtol(buf, &end, 10);
-            if (end && end != buf) {
-                dst = v;
-            }
-        }
-    };
-
-    auto get_float = [&](const char * key, float & dst, uint64_t user_config) {
-        if (config & user_config) {
-            return;
-        }
-
-        char buf[128] = {0};
-        if (llama_model_meta_val_str(model, key, buf, sizeof(buf)) > 0) {
-            char * end = nullptr;
-            float v = strtof(buf, &end);
-            if (end && end != buf) {
-                dst = v;
-            }
-        }
-    };
-
-    // Sampling sequence
-    if (!(config & common_params_sampling_config::COMMON_PARAMS_SAMPLING_CONFIG_SAMPLERS)) {
-        char buf[512] = {0};
-        if (llama_model_meta_val_str(model, llama_model_meta_key_str(LLAMA_MODEL_META_KEY_SAMPLING_SEQUENCE), buf, sizeof(buf)) > 0) {
-            const std::vector<std::string> sampler_names = string_split<std::string>(std::string(buf), ';');
-            if (!sampler_names.empty()) {
-                sparams.samplers = common_sampler_types_from_names(sampler_names, true);
-            }
-        }
-    }
-
-    get_int32(llama_model_meta_key_str(LLAMA_MODEL_META_KEY_SAMPLING_TOP_K),           sparams.top_k,           common_params_sampling_config::COMMON_PARAMS_SAMPLING_CONFIG_TOP_K);
-    get_float(llama_model_meta_key_str(LLAMA_MODEL_META_KEY_SAMPLING_TOP_P),           sparams.top_p,           common_params_sampling_config::COMMON_PARAMS_SAMPLING_CONFIG_TOP_P);
-    get_float(llama_model_meta_key_str(LLAMA_MODEL_META_KEY_SAMPLING_MIN_P),           sparams.min_p,           common_params_sampling_config::COMMON_PARAMS_SAMPLING_CONFIG_MIN_P);
-    get_float(llama_model_meta_key_str(LLAMA_MODEL_META_KEY_SAMPLING_XTC_PROBABILITY), sparams.xtc_probability, common_params_sampling_config::COMMON_PARAMS_SAMPLING_CONFIG_XTC_PROBABILITY);
-    get_float(llama_model_meta_key_str(LLAMA_MODEL_META_KEY_SAMPLING_XTC_THRESHOLD),   sparams.xtc_threshold,   common_params_sampling_config::COMMON_PARAMS_SAMPLING_CONFIG_XTC_THRESHOLD);
-    get_float(llama_model_meta_key_str(LLAMA_MODEL_META_KEY_SAMPLING_TEMP),            sparams.temp,            common_params_sampling_config::COMMON_PARAMS_SAMPLING_CONFIG_TEMP);
-    get_int32(llama_model_meta_key_str(LLAMA_MODEL_META_KEY_SAMPLING_PENALTY_LAST_N),  sparams.penalty_last_n,  common_params_sampling_config::COMMON_PARAMS_SAMPLING_CONFIG_PENALTY_LAST_N);
-    get_float(llama_model_meta_key_str(LLAMA_MODEL_META_KEY_SAMPLING_PENALTY_REPEAT),  sparams.penalty_repeat,  common_params_sampling_config::COMMON_PARAMS_SAMPLING_CONFIG_PENALTY_REPEAT);
-    get_int32(llama_model_meta_key_str(LLAMA_MODEL_META_KEY_SAMPLING_MIROSTAT),        sparams.mirostat,        common_params_sampling_config::COMMON_PARAMS_SAMPLING_CONFIG_MIROSTAT);
-    get_float(llama_model_meta_key_str(LLAMA_MODEL_META_KEY_SAMPLING_MIROSTAT_TAU),    sparams.mirostat_tau,    common_params_sampling_config::COMMON_PARAMS_SAMPLING_CONFIG_MIROSTAT_TAU);
-    get_float(llama_model_meta_key_str(LLAMA_MODEL_META_KEY_SAMPLING_MIROSTAT_ETA),    sparams.mirostat_eta,    common_params_sampling_config::COMMON_PARAMS_SAMPLING_CONFIG_MIROSTAT_ETA);
-}
-
-struct common_init_result::impl {
-    impl() = default;
-    ~impl() = default;
-
-    // note: the order in which model, context, etc. are declared matters because their destructors will be called bottom-to-top
-
-    llama_model_ptr   model;
-    llama_context_ptr context;
-
-    std::vector<llama_adapter_lora_ptr> lora;
-
-    std::vector<common_sampler_ptr> samplers;
-};
-
-common_init_result::common_init_result(common_params & params) :
-    pimpl(new impl{}) {
+struct common_init_result common_init_from_params(common_params & params) {
+    common_init_result iparams;
     auto mparams = common_model_params_to_llama(params);
-    auto cparams = common_context_params_to_llama(params);
-
-    if (params.fit_params) {
-        LOG_INF("%s: fitting params to device memory, for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on\n", __func__);
-        llama_params_fit(params.model.path.c_str(), &mparams, &cparams,
-            params.tensor_split, params.tensor_buft_overrides.data(), params.fit_params_target, params.fit_params_min_ctx,
-            params.verbosity >= 4 ? GGML_LOG_LEVEL_DEBUG : GGML_LOG_LEVEL_ERROR);
-    }
 
     llama_model * model = llama_model_load_from_file(params.model.path.c_str(), mparams);
     if (model == NULL) {
-        return;
+        LOG_ERR("%s: failed to load model '%s'\n", __func__, params.model.path.c_str());
+        return iparams;
     }
 
-    pimpl->model.reset(model);
-
     const llama_vocab * vocab = llama_model_get_vocab(model);
 
-    // updates params.sampling
-    // TODO: fix naming
-    common_init_sampler_from_model(model, params.sampling);
-
-    if (params.sampling.ignore_eos && llama_vocab_eos(vocab) == LLAMA_TOKEN_NULL) {
-        LOG_WRN("%s: warning: vocab does not have an EOS token, ignoring --ignore-eos\n", __func__);
-        params.sampling.ignore_eos = false;
-    }
-
-    // initialize once
-    for (llama_token i = 0; i < llama_vocab_n_tokens(vocab); i++) {
-        if (llama_vocab_is_eog(vocab, i)) {
-            LOG_INF("%s: added %s logit bias = %f\n", __func__, common_token_to_piece(vocab, i).c_str(), -INFINITY);
-            params.sampling.logit_bias_eog.push_back({i, -INFINITY});
-        }
-    }
-
-    if (params.sampling.ignore_eos) {
-        // add EOG biases to the active set of logit biases
-        params.sampling.logit_bias.insert(
-                params.sampling.logit_bias.end(),
-                params.sampling.logit_bias_eog.begin(), params.sampling.logit_bias_eog.end());
-    }
-
-    //if (params.sampling.penalty_last_n == -1) {
-    //    LOG_INF("%s: setting penalty_last_n to ctx_size = %d\n", __func__, llama_n_ctx(lctx));
-    //    params.sampling.penalty_last_n = llama_n_ctx(lctx);
-    //}
-
-    //if (params.sampling.dry_penalty_last_n == -1) {
-    //    LOG_INF("%s: setting dry_penalty_last_n to ctx_size = %d\n", __func__, llama_n_ctx(lctx));
-    //    params.sampling.dry_penalty_last_n = llama_n_ctx(lctx);
-    //}
-
-    pimpl->samplers.resize(cparams.n_seq_max);
-
-    for (int i = 0; i < (int) cparams.n_seq_max; ++i) {
-        pimpl->samplers[i].reset(common_sampler_init(model, params.sampling));
-    }
+    auto cparams = common_context_params_to_llama(params);
 
     llama_context * lctx = llama_init_from_model(model, cparams);
     if (lctx == NULL) {
         LOG_ERR("%s: failed to create context with model '%s'\n", __func__, params.model.path.c_str());
-        return;
+        llama_model_free(model);
+        return iparams;
     }
 
-    pimpl->context.reset(lctx);
-}
-
-llama_model * common_init_result::model() {
-    return pimpl->model.get();
-}
-
-llama_context * common_init_result::context() {
-    return pimpl->context.get();
-}
-
-common_sampler * common_init_result::sampler(llama_seq_id seq_id) {
-    return pimpl->samplers[seq_id].get();
-}
-
-std::vector<llama_adapter_lora_ptr> & common_init_result::lora() {
-    return pimpl->lora;
-}
-
-void common_init_result::free_context() {
-    pimpl->context.reset();
-}
-
-common_init_result_ptr common_init_from_params(common_params & params) {
-    common_init_result_ptr res(new common_init_result(params));
-
-    llama_model * model = res->model();
-    if (model == NULL) {
-        LOG_ERR("%s: failed to load model '%s'\n", __func__, params.model.path.c_str());
-        return res;
-    }
-
-    llama_context * lctx = res->context();
-    if (lctx == NULL) {
-        LOG_ERR("%s: failed to create context with model '%s'\n", __func__, params.model.path.c_str());
-        return res;
-    }
-
-    const llama_vocab * vocab = llama_model_get_vocab(model);
-
     if (params.ctx_shift && !llama_memory_can_shift(llama_get_memory(lctx))) {
         LOG_WRN("%s: KV cache shifting is not supported for this context, disabling KV cache shifting\n", __func__);
         params.ctx_shift = false;
@@ -1206,7 +940,10 @@ common_init_result_ptr common_init_from_params(common_params & params) {
 
         const auto cvec = common_control_vector_load(params.control_vectors);
         if (cvec.n_embd == -1) {
-            return res;
+            llama_free(lctx);
+            llama_model_free(model);
+
+            return iparams;
         }
 
         int err = llama_apply_adapter_cvec(
@@ -1217,7 +954,10 @@ common_init_result_ptr common_init_from_params(common_params & params) {
                 params.control_vector_layer_start,
                 params.control_vector_layer_end);
         if (err) {
-            return res;
+            llama_free(lctx);
+            llama_model_free(model);
+
+            return iparams;
         }
     }
 
@@ -1231,17 +971,22 @@ common_init_result_ptr common_init_from_params(common_params & params) {
 
         bool has_eos = llama_vocab_eos(vocab) != LLAMA_TOKEN_NULL;
         bool has_sep = llama_vocab_sep(vocab) != LLAMA_TOKEN_NULL;
-        bool has_rerank_prompt = llama_model_chat_template(model, "rerank") != NULL;
 
-        if (!has_eos && !has_sep && !has_rerank_prompt) {
-            LOG_WRN("%s: warning: vocab does not have an EOS token, SEP token, or rerank prompt. Reranking will not work\n", __func__);
+        if (!has_eos && !has_sep) {
+            LOG_WRN("%s: warning: vocab does not have an EOS token or SEP token, reranking will not work\n", __func__);
             ok = false;
         } else if (!has_eos) {
             LOG_WRN("%s: warning: vocab does not have an EOS token, using SEP token as fallback\n", __func__);
+        } else if (!has_sep) {
+            LOG_WRN("%s: warning: vocab does not have a SEP token, reranking will not work\n", __func__);
+            ok = false;
         }
 
         if (!ok) {
-            return res;
+            llama_free(lctx);
+            llama_model_free(model);
+
+            return iparams;
         }
     }
 
@@ -1251,22 +996,49 @@ common_init_result_ptr common_init_from_params(common_params & params) {
         lora.reset(llama_adapter_lora_init(model, la.path.c_str()));
         if (lora == nullptr) {
             LOG_ERR("%s: failed to apply lora adapter '%s'\n", __func__, la.path.c_str());
-            return res;
+            llama_free(lctx);
+            llama_model_free(model);
+            return iparams;
         }
 
-        char buf[1024];
         la.ptr = lora.get();
-        llama_adapter_meta_val_str(la.ptr, "adapter.lora.task_name", buf, sizeof(buf));
-        la.task_name = buf;
-        llama_adapter_meta_val_str(la.ptr, "adapter.lora.prompt_prefix", buf, sizeof(buf));
-        la.prompt_prefix = buf;
-        res->lora().emplace_back(std::move(lora)); // copy to list of loaded adapters
+        iparams.lora.emplace_back(std::move(lora)); // copy to list of loaded adapters
     }
 
     if (!params.lora_init_without_apply) {
         common_set_adapter_lora(lctx, params.lora_adapters);
     }
 
+    if (params.sampling.ignore_eos && llama_vocab_eos(vocab) == LLAMA_TOKEN_NULL) {
+        LOG_WRN("%s: warning: vocab does not have an EOS token, ignoring --ignore-eos\n", __func__);
+        params.sampling.ignore_eos = false;
+    }
+
+    // initialize once
+    for (llama_token i = 0; i < llama_vocab_n_tokens(vocab); i++) {
+        if (llama_vocab_is_eog(vocab, i)) {
+            LOG_INF("%s: added %s logit bias = %f\n", __func__, common_token_to_piece(lctx, i).c_str(), -INFINITY);
+            params.sampling.logit_bias_eog.push_back({i, -INFINITY});
+        }
+    }
+
+    if (params.sampling.ignore_eos) {
+        // add EOG biases to the active set of logit biases
+        params.sampling.logit_bias.insert(
+                params.sampling.logit_bias.end(),
+                params.sampling.logit_bias_eog.begin(), params.sampling.logit_bias_eog.end());
+    }
+
+    if (params.sampling.penalty_last_n == -1) {
+        LOG_INF("%s: setting penalty_last_n to ctx_size = %d\n", __func__, llama_n_ctx(lctx));
+        params.sampling.penalty_last_n = llama_n_ctx(lctx);
+    }
+
+    if (params.sampling.dry_penalty_last_n == -1) {
+        LOG_INF("%s: setting dry_penalty_last_n to ctx_size = %d\n", __func__, llama_n_ctx(lctx));
+        params.sampling.dry_penalty_last_n = llama_n_ctx(lctx);
+    }
+
     if (params.warmup) {
         LOG_WRN("%s: warming up the model with an empty run - please wait ... (--no-warmup to disable)\n", __func__);
 
@@ -1305,10 +1077,11 @@ common_init_result_ptr common_init_from_params(common_params & params) {
         llama_set_warmup(lctx, false);
     }
 
-    return res;
-}
+    iparams.model.reset(model);
+    iparams.context.reset(lctx);
 
-common_init_result::~common_init_result() = default;
+    return iparams;
+}
 
 std::string get_model_endpoint() {
     const char * model_endpoint_env = getenv("MODEL_ENDPOINT");
@@ -1318,9 +1091,7 @@ std::string get_model_endpoint() {
     std::string model_endpoint = "https://huggingface.co/";
     if (endpoint_env) {
         model_endpoint = endpoint_env;
-        if (model_endpoint.back() != '/') {
-            model_endpoint += '/';
-        }
+        if (model_endpoint.back() != '/') model_endpoint += '/';
     }
     return model_endpoint;
 }
@@ -1351,8 +1122,6 @@ struct llama_model_params common_model_params_to_llama(common_params & params) {
     mparams.use_mmap        = params.use_mmap;
     mparams.use_mlock       = params.use_mlock;
     mparams.check_tensors   = params.check_tensors;
-    mparams.use_extra_bufts = !params.no_extra_bufts;
-    mparams.no_host         = params.no_host;
 
     if (params.kv_overrides.empty()) {
         mparams.kv_overrides = NULL;
@@ -1395,10 +1164,11 @@ struct llama_context_params common_context_params_to_llama(const common_params &
     cparams.yarn_orig_ctx     = params.yarn_orig_ctx;
     cparams.pooling_type      = params.pooling_type;
     cparams.attention_type    = params.attention_type;
-    cparams.flash_attn_type   = params.flash_attn_type;
+    cparams.defrag_thold      = params.defrag_thold;
     cparams.cb_eval           = params.cb_eval;
     cparams.cb_eval_user_data = params.cb_eval_user_data;
     cparams.offload_kqv       = !params.no_kv_offload;
+    cparams.flash_attn        = params.flash_attn;
     cparams.no_perf           = params.no_perf;
     cparams.op_offload        = !params.no_op_offload;
     cparams.swa_full          = params.swa_full;
@@ -1794,56 +1564,3 @@ ggml_opt_dataset_t common_opt_dataset_init(struct llama_context * ctx, const std
 
     return result;
 }
-
-ggml_opt_optimizer_params common_opt_lr_pars(void * userdata) {
-    ggml_opt_optimizer_params result = ggml_opt_get_default_optimizer_params(nullptr);
-    const lr_opt &            d      = *(lr_opt *) userdata;
-    result.adamw.alpha = result.sgd.alpha = d.get_lr(d.epoch);
-    result.sgd.wd = result.adamw.wd = d.wd;
-    return result;
-}
-
-// TODO make all command line args case-insensitive
-static inline bool eq_case_insensitive(char const* a, char const* b) {
-    return !
-#if defined(_MSC_VER)
-        _stricmp
-#else
-        strcasecmp
-#endif // defined(_MSC_VER)
-        (a, b);
-}
-
-enum ggml_opt_optimizer_type common_opt_get_optimizer(const char * n) {
-    if (eq_case_insensitive("adamw", n)) {
-        return GGML_OPT_OPTIMIZER_TYPE_ADAMW;
-    }
-    if (eq_case_insensitive("sgd", n)) {
-        return GGML_OPT_OPTIMIZER_TYPE_SGD;
-    }
-    return GGML_OPT_OPTIMIZER_TYPE_COUNT;
-}
-
-// TODO simplify to use just log and exp
-static float const k_log_2 = std::log(2.f);
-
-void lr_opt::init() {
-    if (lr_min > 0 && lr_min < lr0) {
-        float nhalf = std::log(lr0 / lr_min) / k_log_2;
-        float e     = epochs;
-        if (decay_epochs > 0 && decay_epochs < e) {
-            e = decay_epochs;
-        } else {
-            decay_epochs = e;
-        }
-        scale_epoch = nhalf / e;
-    }
-}
-
-float lr_opt::get_lr(float epoch) const {
-    float r = lr_min <= 0 ? lr0 :
-        epoch >= decay_epochs ? lr_min :
-        lr0 * std::pow(0.5f, epoch * scale_epoch);
-    LOG_INF("epoch %.2g lr=%.2g\n", epoch, r);
-    return r;
-}

common/common.h

@@ -2,19 +2,14 @@
 
 #pragma once
 
-#include "ggml-opt.h"
 #include "llama-cpp.h"
 
 #include <set>
-#include <sstream>
 #include <string>
 #include <string_view>
 #include <vector>
 #include <map>
-
-#if defined(_WIN32) && !defined(_WIN32_WINNT)
-#define _WIN32_WINNT 0x0A00
-#endif
+#include <sstream>
 
 #ifdef _WIN32
 #define DIRECTORY_SEPARATOR '\\'
@@ -30,22 +25,12 @@
     fprintf(stderr, "%s: built with %s for %s\n", __func__, LLAMA_COMPILER, LLAMA_BUILD_TARGET);    \
 } while(0)
 
-struct common_time_meas {
-    common_time_meas(int64_t & t_acc, bool disable = false);
-    ~common_time_meas();
-
-    const int64_t t_start_us;
-
-    int64_t & t_acc;
-};
+#define DEFAULT_MODEL_PATH "models/7B/ggml-model-f16.gguf"
 
 struct common_adapter_lora_info {
     std::string path;
     float scale;
 
-    std::string task_name;
-    std::string prompt_prefix;
-
     struct llama_adapter_lora * ptr;
 };
 
@@ -82,8 +67,7 @@ int32_t cpu_get_num_math();
 enum llama_example {
     LLAMA_EXAMPLE_COMMON,
     LLAMA_EXAMPLE_SPECULATIVE,
-    LLAMA_EXAMPLE_COMPLETION,
-    LLAMA_EXAMPLE_CLI,
+    LLAMA_EXAMPLE_MAIN,
     LLAMA_EXAMPLE_EMBEDDING,
     LLAMA_EXAMPLE_PERPLEXITY,
     LLAMA_EXAMPLE_RETRIEVAL,
@@ -98,8 +82,6 @@ enum llama_example {
     LLAMA_EXAMPLE_PARALLEL,
     LLAMA_EXAMPLE_TTS,
     LLAMA_EXAMPLE_DIFFUSION,
-    LLAMA_EXAMPLE_FINETUNE,
-    LLAMA_EXAMPLE_FIT_PARAMS,
 
     LLAMA_EXAMPLE_COUNT,
 };
@@ -144,22 +126,6 @@ struct common_grammar_trigger {
     llama_token token = LLAMA_TOKEN_NULL;
 };
 
-enum common_params_sampling_config : uint64_t {
-    COMMON_PARAMS_SAMPLING_CONFIG_SAMPLERS        = 1 << 0,
-    COMMON_PARAMS_SAMPLING_CONFIG_TOP_K           = 1 << 1,
-    COMMON_PARAMS_SAMPLING_CONFIG_TOP_P           = 1 << 2,
-    COMMON_PARAMS_SAMPLING_CONFIG_MIN_P           = 1 << 3,
-    COMMON_PARAMS_SAMPLING_CONFIG_XTC_PROBABILITY = 1 << 4,
-    COMMON_PARAMS_SAMPLING_CONFIG_XTC_THRESHOLD   = 1 << 5,
-    COMMON_PARAMS_SAMPLING_CONFIG_TEMP            = 1 << 6,
-    COMMON_PARAMS_SAMPLING_CONFIG_PENALTY_LAST_N  = 1 << 7,
-    COMMON_PARAMS_SAMPLING_CONFIG_PENALTY_REPEAT  = 1 << 8,
-    COMMON_PARAMS_SAMPLING_CONFIG_MIROSTAT        = 1 << 9,
-    COMMON_PARAMS_SAMPLING_CONFIG_MIROSTAT_TAU    = 1 << 10,
-    COMMON_PARAMS_SAMPLING_CONFIG_MIROSTAT_ETA    = 1 << 11,
-};
-
-
 // sampling parameters
 struct common_params_sampling {
     uint32_t seed = LLAMA_DEFAULT_SEED; // the seed used to initialize llama_sampler
@@ -192,10 +158,9 @@ struct common_params_sampling {
     bool    no_perf            = false; // disable performance metrics
     bool    timing_per_token   = false;
 
-    uint64_t user_sampling_config = 0; // bitfield to track user-specified samplers
-
     std::vector<std::string> dry_sequence_breakers = {"\n", ":", "\"", "*"};     // default sequence breakers for DRY
 
+
     std::vector<enum common_sampler_type> samplers = {
         COMMON_SAMPLER_TYPE_PENALTIES,
         COMMON_SAMPLER_TYPE_DRY,
@@ -216,21 +181,15 @@ struct common_params_sampling {
     std::vector<llama_logit_bias> logit_bias;     // logit biases to apply
     std::vector<llama_logit_bias> logit_bias_eog; // pre-calculated logit biases for EOG tokens
 
-    bool has_logit_bias() const {
-        return !logit_bias.empty();
-    }
-
     // print the parameters into a string
     std::string print() const;
 };
 
 struct common_params_model {
-    std::string path        = ""; // model local path                                       // NOLINT
-    std::string url         = ""; // model url to download                                  // NOLINT
-    std::string hf_repo     = ""; // HF repo                                                // NOLINT
-    std::string hf_file     = ""; // HF file                                                // NOLINT
-    std::string docker_repo = ""; // Docker repo                                            // NOLINT
-    std::string name        = ""; // in format <user>/<model>[:<tag>] (tag is optional)     // NOLINT
+    std::string path    = ""; // model local path                                           // NOLINT
+    std::string url     = ""; // model url to download                                      // NOLINT
+    std::string hf_repo = ""; // HF repo                                                    // NOLINT
+    std::string hf_file = ""; // HF file                                                    // NOLINT
 };
 
 struct common_params_speculative {
@@ -242,8 +201,6 @@ struct common_params_speculative {
     int32_t n_gpu_layers =    -1; // number of layers to store in VRAM for the draft model (-1 - use default)
     float   p_split      =  0.1f; // speculative decoding split probability
     float   p_min        = 0.75f; // minimum speculative decoding probability (greedy)
-    std::vector<std::pair<std::string, std::string>> replacements; // main to speculative model replacements
-    std::vector<llama_model_tensor_buft_override> tensor_buft_overrides;
 
     ggml_type cache_type_k = GGML_TYPE_F16; // KV cache data type for the K
     ggml_type cache_type_v = GGML_TYPE_F16; // KV cache data type for the V
@@ -263,52 +220,22 @@ struct common_params_vocoder {
 };
 
 struct common_params_diffusion {
-    int32_t steps         = 128;
-    bool    visual_mode   = false;
-
-    float   eps           = 0;        // epsilon for timesteps
-    int32_t block_length  = 0;        // block length for generation
-
-    int32_t algorithm     = 4;        // default algorithm: low-confidence
-    float   alg_temp      = 0.0f;     // algorithm temperature
-
-    float   cfg_scale     = 0;        // classifier-free guidance scale
-    bool    add_gumbel_noise = false; // add gumbel noise to the logits if temp > 0.0
+    int32_t steps       = 64;     // number of diffusion steps
+    float   eps         = 1e-3f;  // epsilon for timesteps
+    int32_t algorithm   = 0;      // diffusion algorithm (0=ORIGIN, 1=MASKGIT_PLUS, 2=TOPK_MARGIN, 3=ENTROPY)
+    float   alg_temp    = 0.0f;   // algorithm temperature
+    bool    visual_mode = false;  // show progressive diffusion on screen
 };
 
-// reasoning API response format (not to be confused as chat template's reasoning format)
 enum common_reasoning_format {
     COMMON_REASONING_FORMAT_NONE,
-    COMMON_REASONING_FORMAT_AUTO,            // Same as deepseek, using `message.reasoning_content`
     COMMON_REASONING_FORMAT_DEEPSEEK_LEGACY, // Extract thinking tag contents and return as `message.reasoning_content`, or leave inline in <think> tags in stream mode
     COMMON_REASONING_FORMAT_DEEPSEEK,        // Extract thinking tag contents and return as `message.reasoning_content`, including in streaming deltas.
-    // do not extend this enum unless you absolutely have to
-    // in most cases, use COMMON_REASONING_FORMAT_AUTO
-    // see: https://github.com/ggml-org/llama.cpp/pull/15408
 };
 
-
-struct lr_opt {
-    float    lr0          = 1e-5; // learning rate at first epoch
-    float    lr_min       = -1;
-    float    decay_epochs = -1;   // if >0, the learning rate starts at lr0 and decays to lr_min after this many epochs
-    float    scale_epoch  = 0;
-    float    wd           = 0;
-    unsigned epochs       = 2;
-
-    unsigned epoch; // set by optimizer outer (epochs) loop
-    // learning rate decay - constant LR per epoch only for now
-    float get_lr(float e) const;
-    float get_lr() const { return get_lr(epoch); }
-    // must call after arg parse, before get_lr
-    void init();
-};
-
-struct ggml_opt_optimizer_params common_opt_lr_pars(void * userdata);
-
 struct common_params {
-    int32_t n_predict             =    -1; // max. number of new tokens to predict, -1 == no limit
-    int32_t n_ctx                 =     0; // context size, 0 == context the model was trained with
+    int32_t n_predict             =    -1; // new tokens to predict
+    int32_t n_ctx                 =  4096; // context size
     int32_t n_batch               =  2048; // logical batch size for prompt processing (must be >=32 to use BLAS)
     int32_t n_ubatch              =   512; // physical batch size for prompt processing (must be >=32 to use BLAS)
     int32_t n_keep                =     0; // number of tokens to keep from initial prompt
@@ -321,20 +248,18 @@ struct common_params {
     float   rope_freq_base        =  0.0f; // RoPE base frequency
     float   rope_freq_scale       =  0.0f; // RoPE frequency scaling factor
     float   yarn_ext_factor       = -1.0f; // YaRN extrapolation mix factor
-    float   yarn_attn_factor      = -1.0f; // YaRN magnitude scaling factor
-    float   yarn_beta_fast        = -1.0f; // YaRN low correction dim
-    float   yarn_beta_slow        = -1.0f; // YaRN high correction dim
+    float   yarn_attn_factor      =  1.0f; // YaRN magnitude scaling factor
+    float   yarn_beta_fast        = 32.0f; // YaRN low correction dim
+    float   yarn_beta_slow        =  1.0f; // YaRN high correction dim
     int32_t yarn_orig_ctx         =     0; // YaRN original context length
+    float   defrag_thold          =  0.1f; // KV cache defragmentation threshold
 
     // offload params
     std::vector<ggml_backend_dev_t> devices; // devices to use for offloading
 
-    int32_t n_gpu_layers       = -1;               // number of layers to store in VRAM (-1 - use default)
-    int32_t main_gpu           = 0;                // the GPU that is used for scratch and small tensors
-    float   tensor_split[128]  = {0};              // how split tensors should be distributed across GPUs
-    bool    fit_params         = true;             // whether to fit unset model/context parameters to free device memory
-    size_t  fit_params_target  = 1024 * 1024*1024; // margin per device in bytes for fitting parameters to free memory
-    int32_t fit_params_min_ctx = 4096;             // minimum context size to set when trying to reduce memory use
+    int32_t n_gpu_layers      = -1;  // number of layers to store in VRAM (-1 - use default)
+    int32_t main_gpu          = 0;   // the GPU that is used for scratch and small tensors
+    float   tensor_split[128] = {0}; // how split tensors should be distributed across GPUs
 
     enum llama_split_mode split_mode = LLAMA_SPLIT_MODE_LAYER; // how to split the model across GPUs
 
@@ -349,7 +274,6 @@ struct common_params {
     enum llama_rope_scaling_type rope_scaling_type = LLAMA_ROPE_SCALING_TYPE_UNSPECIFIED;
     enum llama_pooling_type      pooling_type      = LLAMA_POOLING_TYPE_UNSPECIFIED; // pooling type for embeddings
     enum llama_attention_type    attention_type    = LLAMA_ATTENTION_TYPE_UNSPECIFIED; // attention type for embeddings
-    enum llama_flash_attn_type   flash_attn_type   = LLAMA_FLASH_ATTN_TYPE_AUTO; // whether to use Flash Attention
 
     struct common_params_sampling    sampling;
     struct common_params_speculative speculative;
@@ -380,7 +304,7 @@ struct common_params {
 
     std::vector<common_control_vector_load_info> control_vectors; // control vector with user defined scale
 
-    int32_t verbosity                  = 3;  // LOG_LEVEL_INFO
+    int32_t verbosity                  = 0;
     int32_t control_vector_layer_start = -1; // layer range for control vector
     int32_t control_vector_layer_end   = -1; // layer range for control vector
     bool    offline                    = false;
@@ -413,9 +337,9 @@ struct common_params {
     bool multiline_input   = false; // reverse the usage of `\`
     bool simple_io         = false; // improves compatibility with subprocesses and limited consoles
     bool cont_batching     = true;  // insert new sequences for decoding on-the-fly
+    bool flash_attn        = false; // flash attention
     bool no_perf           = false; // disable performance metrics
-    bool show_timings      = true;  // show timing information on CLI
-    bool ctx_shift         = false; // context shift on infinite text generation
+    bool ctx_shift         = true;  // context shift on inifinite text generation
     bool swa_full          = false; // use full-size SWA cache (https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
     bool kv_unified        = false; // enable unified KV cache
 
@@ -428,8 +352,6 @@ struct common_params {
     bool warmup            = true;  // warmup run
     bool check_tensors     = false; // validate tensor data
     bool no_op_offload     = false; // globally disable offload host tensor operations to device
-    bool no_extra_bufts    = false; // disable extra buffer types (used for weight repacking)
-    bool no_host           = false; // bypass host buffer allowing extra buffers to be used
 
     bool single_turn       = false; // single turn chat conversation
 
@@ -443,13 +365,6 @@ struct common_params {
     bool mmproj_use_gpu = true;     // use GPU for multimodal model
     bool no_mmproj = false;         // explicitly disable multimodal model
     std::vector<std::string> image; // path to image file(s)
-    int image_min_tokens = -1;
-    int image_max_tokens = -1;
-
-    // finetune
-    struct lr_opt lr;
-    enum ggml_opt_optimizer_type optimizer = GGML_OPT_OPTIMIZER_TYPE_ADAMW;
-    float val_split = 0.05f; // fraction of the data used for the validation set
 
     // embedding
     bool embedding         = false; // get only sentence embedding
@@ -459,24 +374,21 @@ struct common_params {
     std::string cls_sep    = "\t";  // separator of classification sequences
 
     // server params
-    int32_t port              = 8080;         // server listens on this network port
-    int32_t timeout_read      = 600;          // http read timeout in seconds
-    int32_t timeout_write     = timeout_read; // http write timeout in seconds
-    int32_t n_threads_http    = -1;           // number of threads to process HTTP requests (TODO: support threadpool)
-    int32_t n_cache_reuse     = 0;            // min chunk size to reuse from the cache via KV shifting
-    int32_t n_ctx_checkpoints = 8;            // max number of context checkpoints per slot
-    int32_t cache_ram_mib     = 8192;         // -1 = no limit, 0 - disable, 1 = 1 MiB, etc.
+    int32_t port           = 8080;         // server listens on this network port
+    int32_t timeout_read   = 600;          // http read timeout in seconds
+    int32_t timeout_write  = timeout_read; // http write timeout in seconds
+    int32_t n_threads_http = -1;           // number of threads to process HTTP requests (TODO: support threadpool)
+    int32_t n_cache_reuse  = 0;            // min chunk size to reuse from the cache via KV shifting
 
     std::string hostname      = "127.0.0.1";
     std::string public_path   = "";                                                                         // NOLINT
     std::string api_prefix    = "";                                                                         // NOLINT
     std::string chat_template = "";                                                                         // NOLINT
-    bool use_jinja = true;                                                                                  // NOLINT
+    bool use_jinja = false;                                                                                 // NOLINT
     bool enable_chat_template = true;
     common_reasoning_format reasoning_format = COMMON_REASONING_FORMAT_DEEPSEEK;
     int reasoning_budget = -1;
-    bool prefill_assistant = true; // if true, any trailing assistant message will be prefilled into the response
-    int sleep_idle_seconds = -1;   // if >0, server will sleep after this many seconds of idle time
+    bool prefill_assistant = true;                                                                          // if true, any trailing assistant message will be prefilled into the response
 
     std::vector<std::string> api_keys;
 
@@ -485,31 +397,20 @@ struct common_params {
 
     std::map<std::string, std::string> default_template_kwargs;
 
-    // webui configs
-    bool webui = true;
-    std::string webui_config_json;
-
     // "advanced" endpoints are disabled by default for better security
-    bool endpoint_slots   = true;
+    bool webui            = true;
+    bool endpoint_slots   = false;
     bool endpoint_props   = false; // only control POST requests, not GET
     bool endpoint_metrics = false;
 
-    // router server configs
-    std::string models_dir    = ""; // directory containing models for the router server
-    std::string models_preset = ""; // directory containing model presets for the router server
-    int models_max = 4;             // maximum number of models to load simultaneously
-    bool models_autoload = true;    // automatically load models when requested via the router server
-
     bool log_json = false;
 
     std::string slot_save_path;
-    std::string media_path; // path to directory for loading media files
 
-    float slot_prompt_similarity = 0.1f;
+    float slot_prompt_similarity = 0.5f;
 
     // batched-bench params
-    bool is_pp_shared   = false;
-    bool is_tg_separate = false;
+    bool is_pp_shared = false;
 
     std::vector<int32_t> n_pp;
     std::vector<int32_t> n_tg;
@@ -530,7 +431,6 @@ struct common_params {
     int32_t n_out_freq  = 10; // output the imatrix every n_out_freq iterations
     int32_t n_save_freq =  0; // save the imatrix every n_save_freq iterations
     int32_t i_chunk     =  0; // start processing from this chunk
-    int8_t  imat_dat    =  0; // whether the legacy imatrix.dat format should be output (gguf <= 0 < dat)
 
     bool process_output  = false; // collect data for the output tensor
     bool compute_ppl     = true;  // whether to compute perplexity
@@ -556,10 +456,6 @@ struct common_params {
     // return false from callback to abort model loading or true to continue
     llama_progress_callback load_progress_callback = NULL;
     void *                  load_progress_callback_user_data = NULL;
-
-    bool has_speculative() const {
-        return !speculative.model.path.empty() || !speculative.model.hf_repo.empty();
-    }
 };
 
 // call once at the start of a program if it uses libcommon
@@ -654,55 +550,25 @@ std::string string_from(const struct llama_context * ctx, const struct llama_bat
 // Filesystem utils
 //
 
-bool fs_validate_filename(const std::string & filename, bool allow_subdirs = false);
+bool fs_validate_filename(const std::string & filename);
 bool fs_create_directory_with_parents(const std::string & path);
-bool fs_is_directory(const std::string & path);
 
 std::string fs_get_cache_directory();
 std::string fs_get_cache_file(const std::string & filename);
 
-struct common_file_info {
-    std::string path;
-    std::string name;
-    size_t      size = 0; // in bytes
-    bool        is_dir = false;
-};
-std::vector<common_file_info> fs_list(const std::string & path, bool include_directories);
-
-//
-// TTY utils
-//
-
-// Auto-detect if colors can be enabled based on terminal and environment
-bool tty_can_use_colors();
-
 //
 // Model utils
 //
 
-struct common_sampler;
-
-// note: defines the model, context, samplers, ets. lifetimes
+// note: defines object's lifetime
 struct common_init_result {
-    common_init_result(common_params & params);
-    ~common_init_result();
+    llama_model_ptr   model;
+    llama_context_ptr context;
 
-    llama_model * model();
-    llama_context * context();
-    common_sampler * sampler(llama_seq_id seq_id);
-
-    std::vector<llama_adapter_lora_ptr> & lora();
-
-    void free_context();
-
-private:
-    struct impl;
-    std::unique_ptr<impl> pimpl;
+    std::vector<llama_adapter_lora_ptr> lora;
 };
 
-using common_init_result_ptr = std::unique_ptr<common_init_result>;
-
-common_init_result_ptr common_init_from_params(common_params & params);
+struct common_init_result     common_init_from_params(common_params & params);
 
 struct llama_model_params     common_model_params_to_llama  (      common_params & params);
 struct llama_context_params   common_context_params_to_llama(const common_params & params);
@@ -821,25 +687,8 @@ const char * const LLM_KV_SPLIT_TENSORS_COUNT = "split.tensors.count";
 
 }
 
-//
-// MoE utils
-//
-
-const char * const LLM_FFN_EXPS_REGEX = "\\.ffn_(up|down|gate)_(ch|)exps";
-
-static std::string llm_ffn_exps_block_regex(int idx) {
-    return string_format("blk\\.%d%s", idx, LLM_FFN_EXPS_REGEX);
-}
-
-static llama_model_tensor_buft_override llm_ffn_exps_cpu_override() {
-    return { LLM_FFN_EXPS_REGEX, ggml_backend_cpu_buffer_type() };
-}
-
 //
 // training utils
 //
 
 ggml_opt_dataset_t common_opt_dataset_init(struct llama_context * ctx, const std::vector<llama_token> & tokens, int64_t stride);
-
-// "adamw" or "sgd" (case insensitive)
-enum ggml_opt_optimizer_type common_opt_get_optimizer(const char *);

common/console.cpp

@@ -1,16 +1,6 @@
 #include "console.h"
-#include "log.h"
 #include <vector>
 #include <iostream>
-#include <cassert>
-#include <cstddef>
-#include <cctype>
-#include <cwctype>
-#include <cstdint>
-#include <condition_variable>
-#include <mutex>
-#include <thread>
-#include <stdarg.h>
 
 #if defined(_WIN32)
 #define WIN32_LEAN_AND_MEAN
@@ -40,44 +30,26 @@
 #define ANSI_COLOR_BLUE    "\x1b[34m"
 #define ANSI_COLOR_MAGENTA "\x1b[35m"
 #define ANSI_COLOR_CYAN    "\x1b[36m"
-#define ANSI_COLOR_GRAY    "\x1b[90m"
 #define ANSI_COLOR_RESET   "\x1b[0m"
 #define ANSI_BOLD          "\x1b[1m"
 
 namespace console {
 
-#if defined (_WIN32)
-    namespace {
-        // Use private-use unicode values to represent special keys that are not reported
-        // as characters (e.g. arrows on Windows). These values should never clash with
-        // real input and let the rest of the code handle navigation uniformly.
-        static constexpr char32_t KEY_ARROW_LEFT       = 0xE000;
-        static constexpr char32_t KEY_ARROW_RIGHT      = 0xE001;
-        static constexpr char32_t KEY_ARROW_UP         = 0xE002;
-        static constexpr char32_t KEY_ARROW_DOWN       = 0xE003;
-        static constexpr char32_t KEY_HOME             = 0xE004;
-        static constexpr char32_t KEY_END              = 0xE005;
-        static constexpr char32_t KEY_CTRL_ARROW_LEFT  = 0xE006;
-        static constexpr char32_t KEY_CTRL_ARROW_RIGHT = 0xE007;
-        static constexpr char32_t KEY_DELETE           = 0xE008;
-    }
-
     //
     // Console state
     //
-#endif
 
-    static bool         advanced_display = false;
-    static bool         simple_io        = true;
-    static display_type current_display  = DISPLAY_TYPE_RESET;
+    static bool      advanced_display = false;
+    static bool      simple_io        = true;
+    static display_t current_display  = reset;
 
-    static FILE*        out              = stdout;
+    static FILE*     out              = stdout;
 
 #if defined (_WIN32)
-    static void*        hConsole;
+    static void*     hConsole;
 #else
-    static FILE*        tty              = nullptr;
-    static termios      initial_state;
+    static FILE*     tty              = nullptr;
+    static termios   initial_state;
 #endif
 
     //
@@ -148,7 +120,7 @@ namespace console {
 
     void cleanup() {
         // Reset console display
-        set_display(DISPLAY_TYPE_RESET);
+        set_display(reset);
 
 #if !defined(_WIN32)
         // Restore settings on POSIX systems
@@ -168,26 +140,20 @@ namespace console {
     //
 
     // Keep track of current display and only emit ANSI code if it changes
-    void set_display(display_type display) {
+    void set_display(display_t display) {
         if (advanced_display && current_display != display) {
-            common_log_flush(common_log_main());
+            fflush(stdout);
             switch(display) {
-                case DISPLAY_TYPE_RESET:
+                case reset:
                     fprintf(out, ANSI_COLOR_RESET);
                     break;
-                case DISPLAY_TYPE_INFO:
-                    fprintf(out, ANSI_COLOR_MAGENTA);
-                    break;
-                case DISPLAY_TYPE_PROMPT:
+                case prompt:
                     fprintf(out, ANSI_COLOR_YELLOW);
                     break;
-                case DISPLAY_TYPE_REASONING:
-                    fprintf(out, ANSI_COLOR_GRAY);
-                    break;
-                case DISPLAY_TYPE_USER_INPUT:
+                case user_input:
                     fprintf(out, ANSI_BOLD ANSI_COLOR_GREEN);
                     break;
-                case DISPLAY_TYPE_ERROR:
+                case error:
                     fprintf(out, ANSI_BOLD ANSI_COLOR_RED);
             }
             current_display = display;
@@ -210,18 +176,7 @@ namespace console {
             if (record.EventType == KEY_EVENT && record.Event.KeyEvent.bKeyDown) {
                 wchar_t wc = record.Event.KeyEvent.uChar.UnicodeChar;
                 if (wc == 0) {
-                    const DWORD ctrl_mask = LEFT_CTRL_PRESSED | RIGHT_CTRL_PRESSED;
-                    const bool ctrl_pressed = (record.Event.KeyEvent.dwControlKeyState & ctrl_mask) != 0;
-                    switch (record.Event.KeyEvent.wVirtualKeyCode) {
-                        case VK_LEFT:   return ctrl_pressed ? KEY_CTRL_ARROW_LEFT  : KEY_ARROW_LEFT;
-                        case VK_RIGHT:  return ctrl_pressed ? KEY_CTRL_ARROW_RIGHT : KEY_ARROW_RIGHT;
-                        case VK_UP:     return KEY_ARROW_UP;
-                        case VK_DOWN:   return KEY_ARROW_DOWN;
-                        case VK_HOME:   return KEY_HOME;
-                        case VK_END:    return KEY_END;
-                        case VK_DELETE: return KEY_DELETE;
-                        default:        continue;
-                    }
+                    continue;
                 }
 
                 if ((wc >= 0xD800) && (wc <= 0xDBFF)) { // Check if wc is a high surrogate
@@ -360,52 +315,6 @@ namespace console {
 #endif
     }
 
-    static char32_t decode_utf8(const std::string & input, size_t pos, size_t & advance) {
-        unsigned char c = static_cast<unsigned char>(input[pos]);
-        if ((c & 0x80u) == 0u) {
-            advance = 1;
-            return c;
-        }
-        if ((c & 0xE0u) == 0xC0u && pos + 1 < input.size()) {
-            unsigned char c1 = static_cast<unsigned char>(input[pos + 1]);
-            if ((c1 & 0xC0u) != 0x80u) {
-                advance = 1;
-                return 0xFFFD;
-            }
-            advance = 2;
-            return ((c & 0x1Fu) << 6) | (static_cast<unsigned char>(input[pos + 1]) & 0x3Fu);
-        }
-        if ((c & 0xF0u) == 0xE0u && pos + 2 < input.size()) {
-            unsigned char c1 = static_cast<unsigned char>(input[pos + 1]);
-            unsigned char c2 = static_cast<unsigned char>(input[pos + 2]);
-            if ((c1 & 0xC0u) != 0x80u || (c2 & 0xC0u) != 0x80u) {
-                advance = 1;
-                return 0xFFFD;
-            }
-            advance = 3;
-            return ((c & 0x0Fu) << 12) |
-                   ((static_cast<unsigned char>(input[pos + 1]) & 0x3Fu) << 6) |
-                   (static_cast<unsigned char>(input[pos + 2]) & 0x3Fu);
-        }
-        if ((c & 0xF8u) == 0xF0u && pos + 3 < input.size()) {
-            unsigned char c1 = static_cast<unsigned char>(input[pos + 1]);
-            unsigned char c2 = static_cast<unsigned char>(input[pos + 2]);
-            unsigned char c3 = static_cast<unsigned char>(input[pos + 3]);
-            if ((c1 & 0xC0u) != 0x80u || (c2 & 0xC0u) != 0x80u || (c3 & 0xC0u) != 0x80u) {
-                advance = 1;
-                return 0xFFFD;
-            }
-            advance = 4;
-            return ((c & 0x07u) << 18) |
-                   ((static_cast<unsigned char>(input[pos + 1]) & 0x3Fu) << 12) |
-                   ((static_cast<unsigned char>(input[pos + 2]) & 0x3Fu) << 6) |
-                   (static_cast<unsigned char>(input[pos + 3]) & 0x3Fu);
-        }
-
-        advance = 1;
-        return 0xFFFD; // replacement character for invalid input
-    }
-
     static void append_utf8(char32_t ch, std::string & out) {
         if (ch <= 0x7F) {
             out.push_back(static_cast<unsigned char>(ch));
@@ -427,319 +336,22 @@ namespace console {
     }
 
     // Helper function to remove the last UTF-8 character from a string
-    static size_t prev_utf8_char_pos(const std::string & line, size_t pos) {
-        if (pos == 0) return 0;
-        pos--;
-        while (pos > 0 && (line[pos] & 0xC0) == 0x80) {
-            pos--;
-        }
-        return pos;
-    }
-
-    static size_t next_utf8_char_pos(const std::string & line, size_t pos) {
-        if (pos >= line.length()) return line.length();
-        pos++;
-        while (pos < line.length() && (line[pos] & 0xC0) == 0x80) {
-            pos++;
-        }
-        return pos;
-    }
-
-    static void move_cursor(int delta);
-    static void move_word_left(size_t & char_pos, size_t & byte_pos, const std::vector<int> & widths, const std::string & line);
-    static void move_word_right(size_t & char_pos, size_t & byte_pos, const std::vector<int> & widths, const std::string & line);
-    static void move_to_line_start(size_t & char_pos, size_t & byte_pos, const std::vector<int> & widths);
-    static void move_to_line_end(size_t & char_pos, size_t & byte_pos, const std::vector<int> & widths, const std::string & line);
-
-    static void delete_at_cursor(std::string & line, std::vector<int> & widths, size_t & char_pos, size_t & byte_pos) {
-        if (char_pos >= widths.size()) {
+    static void pop_back_utf8_char(std::string & line) {
+        if (line.empty()) {
             return;
         }
 
-        size_t next_pos = next_utf8_char_pos(line, byte_pos);
-        int w = widths[char_pos];
-        size_t char_len = next_pos - byte_pos;
+        size_t pos = line.length() - 1;
 
-        line.erase(byte_pos, char_len);
-        widths.erase(widths.begin() + char_pos);
-
-        size_t p = byte_pos;
-        int tail_width = 0;
-        for (size_t i = char_pos; i < widths.size(); ++i) {
-            size_t following = next_utf8_char_pos(line, p);
-            put_codepoint(line.c_str() + p, following - p, widths[i]);
-            tail_width += widths[i];
-            p = following;
+        // Find the start of the last UTF-8 character (checking up to 4 bytes back)
+        for (size_t i = 0; i < 3 && pos > 0; ++i, --pos) {
+            if ((line[pos] & 0xC0) != 0x80) {
+                break; // Found the start of the character
+            }
         }
-
-        for (int i = 0; i < w; ++i) {
-            fputc(' ', out);
-        }
-
-        move_cursor(-(tail_width + w));
+        line.erase(pos);
     }
 
-    static void clear_current_line(const std::vector<int> & widths) {
-        int total_width = 0;
-        for (int w : widths) {
-            total_width += (w > 0 ? w : 1);
-        }
-
-        if (total_width > 0) {
-            std::string spaces(total_width, ' ');
-            fwrite(spaces.c_str(), 1, total_width, out);
-            move_cursor(-total_width);
-        }
-    }
-
-    static void set_line_contents(std::string new_line, std::string & line, std::vector<int> & widths, size_t & char_pos,
-                                  size_t & byte_pos) {
-        move_to_line_start(char_pos, byte_pos, widths);
-        clear_current_line(widths);
-
-        line = std::move(new_line);
-        widths.clear();
-        byte_pos = 0;
-        char_pos = 0;
-
-        size_t idx = 0;
-        while (idx < line.size()) {
-            size_t advance = 0;
-            char32_t cp = decode_utf8(line, idx, advance);
-            int expected_width = estimateWidth(cp);
-            int real_width = put_codepoint(line.c_str() + idx, advance, expected_width);
-            if (real_width < 0) real_width = 0;
-            widths.push_back(real_width);
-            idx += advance;
-            ++char_pos;
-            byte_pos = idx;
-        }
-    }
-
-    static void move_to_line_start(size_t & char_pos, size_t & byte_pos, const std::vector<int> & widths) {
-        int back_width = 0;
-        for (size_t i = 0; i < char_pos; ++i) {
-            back_width += widths[i];
-        }
-        move_cursor(-back_width);
-        char_pos = 0;
-        byte_pos = 0;
-    }
-
-    static void move_to_line_end(size_t & char_pos, size_t & byte_pos, const std::vector<int> & widths, const std::string & line) {
-        int forward_width = 0;
-        for (size_t i = char_pos; i < widths.size(); ++i) {
-            forward_width += widths[i];
-        }
-        move_cursor(forward_width);
-        char_pos = widths.size();
-        byte_pos = line.length();
-    }
-
-    static bool has_ctrl_modifier(const std::string & params) {
-        size_t start = 0;
-        while (start < params.size()) {
-            size_t end = params.find(';', start);
-            size_t len = (end == std::string::npos) ? params.size() - start : end - start;
-            if (len > 0) {
-                int value = 0;
-                for (size_t i = 0; i < len; ++i) {
-                    char ch = params[start + i];
-                    if (!std::isdigit(static_cast<unsigned char>(ch))) {
-                        value = -1;
-                        break;
-                    }
-                    value = value * 10 + (ch - '0');
-                }
-                if (value == 5) {
-                    return true;
-                }
-            }
-
-            if (end == std::string::npos) {
-                break;
-            }
-            start = end + 1;
-        }
-        return false;
-    }
-
-    static bool is_space_codepoint(char32_t cp) {
-        return std::iswspace(static_cast<wint_t>(cp)) != 0;
-    }
-
-    static void move_word_left(size_t & char_pos, size_t & byte_pos, const std::vector<int> & widths, const std::string & line) {
-        if (char_pos == 0) {
-            return;
-        }
-
-        size_t new_char_pos = char_pos;
-        size_t new_byte_pos = byte_pos;
-        int move_width = 0;
-
-        while (new_char_pos > 0) {
-            size_t prev_byte = prev_utf8_char_pos(line, new_byte_pos);
-            size_t advance = 0;
-            char32_t cp = decode_utf8(line, prev_byte, advance);
-            if (!is_space_codepoint(cp)) {
-                break;
-            }
-            move_width += widths[new_char_pos - 1];
-            new_char_pos--;
-            new_byte_pos = prev_byte;
-        }
-
-        while (new_char_pos > 0) {
-            size_t prev_byte = prev_utf8_char_pos(line, new_byte_pos);
-            size_t advance = 0;
-            char32_t cp = decode_utf8(line, prev_byte, advance);
-            if (is_space_codepoint(cp)) {
-                break;
-            }
-            move_width += widths[new_char_pos - 1];
-            new_char_pos--;
-            new_byte_pos = prev_byte;
-        }
-
-        move_cursor(-move_width);
-        char_pos = new_char_pos;
-        byte_pos = new_byte_pos;
-    }
-
-    static void move_word_right(size_t & char_pos, size_t & byte_pos, const std::vector<int> & widths, const std::string & line) {
-        if (char_pos >= widths.size()) {
-            return;
-        }
-
-        size_t new_char_pos = char_pos;
-        size_t new_byte_pos = byte_pos;
-        int move_width = 0;
-
-        while (new_char_pos < widths.size()) {
-            size_t advance = 0;
-            char32_t cp = decode_utf8(line, new_byte_pos, advance);
-            if (!is_space_codepoint(cp)) {
-                break;
-            }
-            move_width += widths[new_char_pos];
-            new_char_pos++;
-            new_byte_pos += advance;
-        }
-
-        while (new_char_pos < widths.size()) {
-            size_t advance = 0;
-            char32_t cp = decode_utf8(line, new_byte_pos, advance);
-            if (is_space_codepoint(cp)) {
-                break;
-            }
-            move_width += widths[new_char_pos];
-            new_char_pos++;
-            new_byte_pos += advance;
-        }
-
-        while (new_char_pos < widths.size()) {
-            size_t advance = 0;
-            char32_t cp = decode_utf8(line, new_byte_pos, advance);
-            if (!is_space_codepoint(cp)) {
-                break;
-            }
-            move_width += widths[new_char_pos];
-            new_char_pos++;
-            new_byte_pos += advance;
-        }
-
-        move_cursor(move_width);
-        char_pos = new_char_pos;
-        byte_pos = new_byte_pos;
-    }
-
-    static void move_cursor(int delta) {
-        if (delta == 0) return;
-#if defined(_WIN32)
-        if (hConsole != NULL) {
-            CONSOLE_SCREEN_BUFFER_INFO bufferInfo;
-            GetConsoleScreenBufferInfo(hConsole, &bufferInfo);
-            COORD newCursorPosition = bufferInfo.dwCursorPosition;
-            int width = bufferInfo.dwSize.X;
-            int newX = newCursorPosition.X + delta;
-            int newY = newCursorPosition.Y;
-
-            while (newX >= width) {
-                newX -= width;
-                newY++;
-            }
-            while (newX < 0) {
-                newX += width;
-                newY--;
-            }
-
-            newCursorPosition.X = newX;
-            newCursorPosition.Y = newY;
-            SetConsoleCursorPosition(hConsole, newCursorPosition);
-        }
-#else
-        if (delta < 0) {
-            for (int i = 0; i < -delta; i++) fprintf(out, "\b");
-        } else {
-            for (int i = 0; i < delta; i++) fprintf(out, "\033[C");
-        }
-#endif
-    }
-
-    struct history_t {
-        std::vector<std::string> entries;
-        size_t viewing_idx = SIZE_MAX;
-        std::string backup_line; // current line before viewing history
-        void add(const std::string & line) {
-            if (line.empty()) {
-                return;
-            }
-            // avoid duplicates with the last entry
-            if (entries.empty() || entries.back() != line) {
-                entries.push_back(line);
-            }
-            // also clear viewing state
-            end_viewing();
-        }
-        bool prev(std::string & cur_line) {
-            if (entries.empty()) {
-                return false;
-            }
-            if (viewing_idx == SIZE_MAX) {
-                return false;
-            }
-            if (viewing_idx > 0) {
-                viewing_idx--;
-            }
-            cur_line = entries[viewing_idx];
-            return true;
-        }
-        bool next(std::string & cur_line) {
-            if (entries.empty() || viewing_idx == SIZE_MAX) {
-                return false;
-            }
-            viewing_idx++;
-            if (viewing_idx >= entries.size()) {
-                cur_line = backup_line;
-                end_viewing();
-            } else {
-                cur_line = entries[viewing_idx];
-            }
-            return true;
-        }
-        void begin_viewing(const std::string & line) {
-            backup_line = line;
-            viewing_idx = entries.size();
-        }
-        void end_viewing() {
-            viewing_idx = SIZE_MAX;
-            backup_line.clear();
-        }
-        bool is_viewing() const {
-            return viewing_idx != SIZE_MAX;
-        }
-    } history;
-
     static bool readline_advanced(std::string & line, bool multiline_input) {
         if (out != stdout) {
             fflush(stdout);
@@ -750,33 +362,8 @@ namespace console {
         bool is_special_char = false;
         bool end_of_stream = false;
 
-        size_t byte_pos = 0; // current byte index
-        size_t char_pos = 0; // current character index (one char can be multiple bytes)
-
         char32_t input_char;
         while (true) {
-            assert(char_pos <= byte_pos);
-            assert(char_pos <= widths.size());
-            auto history_prev = [&]() {
-                if (!history.is_viewing()) {
-                    history.begin_viewing(line);
-                }
-                std::string new_line;
-                if (!history.prev(new_line)) {
-                    return;
-                }
-                set_line_contents(new_line, line, widths, char_pos, byte_pos);
-            };
-            auto history_next = [&]() {
-                if (history.is_viewing()) {
-                    std::string new_line;
-                    if (!history.next(new_line)) {
-                        return;
-                    }
-                    set_line_contents(new_line, line, widths, char_pos, byte_pos);
-                }
-            };
-
             fflush(out); // Ensure all output is displayed before waiting for input
             input_char = getchar32();
 
@@ -784,83 +371,20 @@ namespace console {
                 break;
             }
 
-            if (input_char == (char32_t) WEOF || input_char == 0x04 /* Ctrl+D */) {
+            if (input_char == (char32_t) WEOF || input_char == 0x04 /* Ctrl+D*/) {
                 end_of_stream = true;
                 break;
             }
 
             if (is_special_char) {
+                set_display(user_input);
                 replace_last(line.back());
                 is_special_char = false;
             }
 
             if (input_char == '\033') { // Escape sequence
                 char32_t code = getchar32();
-                if (code == '[') {
-                    std::string params;
-                    while (true) {
-                        code = getchar32();
-                        if ((code >= 'A' && code <= 'Z') || (code >= 'a' && code <= 'z') || code == '~' || code == (char32_t) WEOF) {
-                            break;
-                        }
-                        params.push_back(static_cast<char>(code));
-                    }
-
-                    const bool ctrl_modifier = has_ctrl_modifier(params);
-
-                    if (code == 'D') { // left
-                        if (ctrl_modifier) {
-                            move_word_left(char_pos, byte_pos, widths, line);
-                        } else if (char_pos > 0) {
-                            int w = widths[char_pos - 1];
-                            move_cursor(-w);
-                            char_pos--;
-                            byte_pos = prev_utf8_char_pos(line, byte_pos);
-                        }
-                    } else if (code == 'C') { // right
-                        if (ctrl_modifier) {
-                            move_word_right(char_pos, byte_pos, widths, line);
-                        } else if (char_pos < widths.size()) {
-                            int w = widths[char_pos];
-                            move_cursor(w);
-                            char_pos++;
-                            byte_pos = next_utf8_char_pos(line, byte_pos);
-                        }
-                    } else if (code == 'H') { // home
-                        move_to_line_start(char_pos, byte_pos, widths);
-                    } else if (code == 'F') { // end
-                        move_to_line_end(char_pos, byte_pos, widths, line);
-                    } else if (code == 'A' || code == 'B') {
-                        // up/down
-                        if (code == 'A') {
-                            history_prev();
-                            is_special_char = false;
-                        } else if (code == 'B') {
-                            history_next();
-                            is_special_char = false;
-                        }
-                    } else if ((code == '~' || (code >= 'A' && code <= 'Z') || (code >= 'a' && code <= 'z')) && !params.empty()) {
-                        std::string digits;
-                        for (char ch : params) {
-                            if (ch == ';') {
-                                break;
-                            }
-                            if (std::isdigit(static_cast<unsigned char>(ch))) {
-                                digits.push_back(ch);
-                            }
-                        }
-
-                        if (code == '~') {
-                            if (digits == "1" || digits == "7") { // home
-                                move_to_line_start(char_pos, byte_pos, widths);
-                            } else if (digits == "4" || digits == "8") { // end
-                                move_to_line_end(char_pos, byte_pos, widths, line);
-                            } else if (digits == "3") { // delete
-                                delete_at_cursor(line, widths, char_pos, byte_pos);
-                            }
-                        }
-                    }
-                } else if (code == 0x1B) {
+                if (code == '[' || code == 0x1B) {
                     // Discard the rest of the escape sequence
                     while ((code = getchar32()) != (char32_t) WEOF) {
                         if ((code >= 'A' && code <= 'Z') || (code >= 'a' && code <= 'z') || code == '~') {
@@ -868,110 +392,32 @@ namespace console {
                         }
                     }
                 }
-#if defined(_WIN32)
-            } else if (input_char == KEY_ARROW_LEFT) {
-                if (char_pos > 0) {
-                    int w = widths[char_pos - 1];
-                    move_cursor(-w);
-                    char_pos--;
-                    byte_pos = prev_utf8_char_pos(line, byte_pos);
-                }
-            } else if (input_char == KEY_ARROW_RIGHT) {
-                if (char_pos < widths.size()) {
-                    int w = widths[char_pos];
-                    move_cursor(w);
-                    char_pos++;
-                    byte_pos = next_utf8_char_pos(line, byte_pos);
-                }
-            } else if (input_char == KEY_CTRL_ARROW_LEFT) {
-                move_word_left(char_pos, byte_pos, widths, line);
-            } else if (input_char == KEY_CTRL_ARROW_RIGHT) {
-                move_word_right(char_pos, byte_pos, widths, line);
-            } else if (input_char == KEY_HOME) {
-                move_to_line_start(char_pos, byte_pos, widths);
-            } else if (input_char == KEY_END) {
-                move_to_line_end(char_pos, byte_pos, widths, line);
-            } else if (input_char == KEY_DELETE) {
-                delete_at_cursor(line, widths, char_pos, byte_pos);
-            } else if (input_char == KEY_ARROW_UP || input_char == KEY_ARROW_DOWN) {
-                if (input_char == KEY_ARROW_UP) {
-                    history_prev();
-                    is_special_char = false;
-                } else if (input_char == KEY_ARROW_DOWN) {
-                    history_next();
-                    is_special_char = false;
-                }
-#endif
             } else if (input_char == 0x08 || input_char == 0x7F) { // Backspace
-                if (char_pos > 0) {
-                    int w = widths[char_pos - 1];
-                    move_cursor(-w);
-                    char_pos--;
-                    size_t prev_pos = prev_utf8_char_pos(line, byte_pos);
-                    size_t char_len = byte_pos - prev_pos;
-                    byte_pos = prev_pos;
-
-                    // remove the character
-                    line.erase(byte_pos, char_len);
-                    widths.erase(widths.begin() + char_pos);
-
-                    // redraw tail
-                    size_t p = byte_pos;
-                    int tail_width = 0;
-                    for (size_t i = char_pos; i < widths.size(); ++i) {
-                        size_t next_p = next_utf8_char_pos(line, p);
-                        put_codepoint(line.c_str() + p, next_p - p, widths[i]);
-                        tail_width += widths[i];
-                        p = next_p;
-                    }
-
-                    // clear display
-                    for (int i = 0; i < w; ++i) {
-                        fputc(' ', out);
-                    }
-                    move_cursor(-(tail_width + w));
+                if (!widths.empty()) {
+                    int count;
+                    do {
+                        count = widths.back();
+                        widths.pop_back();
+                        // Move cursor back, print space, and move cursor back again
+                        for (int i = 0; i < count; i++) {
+                            replace_last(' ');
+                            pop_cursor();
+                        }
+                        pop_back_utf8_char(line);
+                    } while (count == 0 && !widths.empty());
                 }
             } else {
-                // insert character
-                std::string new_char_str;
-                append_utf8(input_char, new_char_str);
-                int w = estimateWidth(input_char);
-
-                if (char_pos == widths.size()) {
-                    // insert at the end
-                    line += new_char_str;
-                    int real_w = put_codepoint(new_char_str.c_str(), new_char_str.length(), w);
-                    if (real_w < 0) real_w = 0;
-                    widths.push_back(real_w);
-                    byte_pos += new_char_str.length();
-                    char_pos++;
-                } else {
-                    // insert in middle
-                    line.insert(byte_pos, new_char_str);
-
-                    int real_w = put_codepoint(new_char_str.c_str(), new_char_str.length(), w);
-                    if (real_w < 0) real_w = 0;
-
-                    widths.insert(widths.begin() + char_pos, real_w);
-
-                    // print the tail
-                    size_t p = byte_pos + new_char_str.length();
-                    int tail_width = 0;
-                    for (size_t i = char_pos + 1; i < widths.size(); ++i) {
-                        size_t next_p = next_utf8_char_pos(line, p);
-                        put_codepoint(line.c_str() + p, next_p - p, widths[i]);
-                        tail_width += widths[i];
-                        p = next_p;
-                    }
-
-                    move_cursor(-tail_width);
-
-                    byte_pos += new_char_str.length();
-                    char_pos++;
+                int offset = line.length();
+                append_utf8(input_char, line);
+                int width = put_codepoint(line.c_str() + offset, line.length() - offset, estimateWidth(input_char));
+                if (width < 0) {
+                    width = 0;
                 }
+                widths.push_back(width);
             }
 
             if (!line.empty() && (line.back() == '\\' || line.back() == '/')) {
+                set_display(prompt);
                 replace_last(line.back());
                 is_special_char = true;
             }
@@ -1005,15 +451,6 @@ namespace console {
             }
         }
 
-        if (!end_of_stream && !line.empty()) {
-            // remove the trailing newline for history storage
-            if (!line.empty() && line.back() == '\n') {
-                line.pop_back();
-            }
-            // TODO: maybe support multiline history entries?
-            history.add(line);
-        }
-
         fflush(out);
         return has_more;
     }
@@ -1056,82 +493,12 @@ namespace console {
     }
 
     bool readline(std::string & line, bool multiline_input) {
+        set_display(user_input);
+
         if (simple_io) {
             return readline_simple(line, multiline_input);
         }
         return readline_advanced(line, multiline_input);
     }
 
-    namespace spinner {
-        static const char LOADING_CHARS[] = {'|', '/', '-', '\\'};
-        static std::condition_variable cv_stop;
-        static std::thread th;
-        static size_t frame = 0; // only modified by one thread
-        static bool running = false;
-        static std::mutex mtx;
-        static auto wait_time = std::chrono::milliseconds(100);
-        static void draw_next_frame() {
-            // don't need lock because only one thread modifies running
-            frame = (frame + 1) % sizeof(LOADING_CHARS);
-            replace_last(LOADING_CHARS[frame]);
-            fflush(out);
-        }
-        void start() {
-            std::unique_lock<std::mutex> lock(mtx);
-            if (simple_io || running) {
-                return;
-            }
-            common_log_flush(common_log_main());
-            fprintf(out, "%c", LOADING_CHARS[0]);
-            fflush(out);
-            frame = 1;
-            running = true;
-            th = std::thread([]() {
-                std::unique_lock<std::mutex> lock(mtx);
-                while (true) {
-                    if (cv_stop.wait_for(lock, wait_time, []{ return !running; })) {
-                        break;
-                    }
-                    draw_next_frame();
-                }
-            });
-        }
-        void stop() {
-            {
-                std::unique_lock<std::mutex> lock(mtx);
-                if (simple_io || !running) {
-                    return;
-                }
-                running = false;
-                cv_stop.notify_all();
-            }
-            if (th.joinable()) {
-                th.join();
-            }
-            replace_last(' ');
-            pop_cursor();
-            fflush(out);
-        }
-    }
-
-    void log(const char * fmt, ...) {
-        va_list args;
-        va_start(args, fmt);
-        vfprintf(out, fmt, args);
-        va_end(args);
-    }
-
-    void error(const char * fmt, ...) {
-        va_list args;
-        va_start(args, fmt);
-        display_type cur = current_display;
-        set_display(DISPLAY_TYPE_ERROR);
-        vfprintf(out, fmt, args);
-        set_display(cur); // restore previous color
-        va_end(args);
-    }
-
-    void flush() {
-        fflush(out);
-    }
 }

common/console.h

@@ -2,40 +2,18 @@
 
 #pragma once
 
-#include "common.h"
-
 #include <string>
 
-enum display_type {
-    DISPLAY_TYPE_RESET = 0,
-    DISPLAY_TYPE_INFO,
-    DISPLAY_TYPE_PROMPT,
-    DISPLAY_TYPE_REASONING,
-    DISPLAY_TYPE_USER_INPUT,
-    DISPLAY_TYPE_ERROR
-};
-
 namespace console {
+    enum display_t {
+        reset = 0,
+        prompt,
+        user_input,
+        error
+    };
+
     void init(bool use_simple_io, bool use_advanced_display);
     void cleanup();
-    void set_display(display_type display);
+    void set_display(display_t display);
     bool readline(std::string & line, bool multiline_input);
-
-    namespace spinner {
-        void start();
-        void stop();
-    }
-
-    // note: the logging API below output directly to stdout
-    // it can negatively impact performance if used on inference thread
-    // only use in in a dedicated CLI thread
-    // for logging in inference thread, use log.h instead
-
-    LLAMA_COMMON_ATTRIBUTE_FORMAT(1, 2)
-    void log(const char * fmt, ...);
-
-    LLAMA_COMMON_ATTRIBUTE_FORMAT(1, 2)
-    void error(const char * fmt, ...);
-
-    void flush();
 }

common/download.h

@@ -1,57 +0,0 @@
-#pragma once
-
-#include <string>
-
-struct common_params_model;
-
-//
-// download functionalities
-//
-
-struct common_cached_model_info {
-    std::string manifest_path;
-    std::string user;
-    std::string model;
-    std::string tag;
-    size_t      size = 0; // GGUF size in bytes
-    // return string representation like "user/model:tag"
-    // if tag is "latest", it will be omitted
-    std::string to_string() const {
-        return user + "/" + model + (tag == "latest" ? "" : ":" + tag);
-    }
-};
-
-struct common_hf_file_res {
-    std::string repo; // repo name with ":tag" removed
-    std::string ggufFile;
-    std::string mmprojFile;
-};
-
-/**
- * Allow getting the HF file from the HF repo with tag (like ollama), for example:
- * - bartowski/Llama-3.2-3B-Instruct-GGUF:q4
- * - bartowski/Llama-3.2-3B-Instruct-GGUF:Q4_K_M
- * - bartowski/Llama-3.2-3B-Instruct-GGUF:q5_k_s
- * Tag is optional, default to "latest" (meaning it checks for Q4_K_M first, then Q4, then if not found, return the first GGUF file in repo)
- *
- * Return pair of <repo, file> (with "repo" already having tag removed)
- *
- * Note: we use the Ollama-compatible HF API, but not using the blobId. Instead, we use the special "ggufFile" field which returns the value for "hf_file". This is done to be backward-compatible with existing cache files.
- */
-common_hf_file_res common_get_hf_file(
-    const std::string & hf_repo_with_tag,
-    const std::string & bearer_token,
-    bool offline);
-
-// returns true if download succeeded
-bool common_download_model(
-    const common_params_model & model,
-    const std::string & bearer_token,
-    bool offline);
-
-// returns list of cached models
-std::vector<common_cached_model_info> common_list_cached_models();
-
-// resolve and download model from Docker registry
-// return local path to downloaded model file
-std::string common_docker_resolve_model(const std::string & docker);

common/http.h

@@ -1,73 +0,0 @@
-#pragma once
-
-#include <cpp-httplib/httplib.h>
-
-struct common_http_url {
-    std::string scheme;
-    std::string user;
-    std::string password;
-    std::string host;
-    std::string path;
-};
-
-static common_http_url common_http_parse_url(const std::string & url) {
-    common_http_url parts;
-    auto scheme_end = url.find("://");
-
-    if (scheme_end == std::string::npos) {
-        throw std::runtime_error("invalid URL: no scheme");
-    }
-    parts.scheme = url.substr(0, scheme_end);
-
-    if (parts.scheme != "http" && parts.scheme != "https") {
-        throw std::runtime_error("unsupported URL scheme: " + parts.scheme);
-    }
-
-    auto rest = url.substr(scheme_end + 3);
-    auto at_pos = rest.find('@');
-
-    if (at_pos != std::string::npos) {
-        auto auth = rest.substr(0, at_pos);
-        auto colon_pos = auth.find(':');
-        if (colon_pos != std::string::npos) {
-            parts.user = auth.substr(0, colon_pos);
-            parts.password = auth.substr(colon_pos + 1);
-        } else {
-            parts.user = auth;
-        }
-        rest = rest.substr(at_pos + 1);
-    }
-
-    auto slash_pos = rest.find('/');
-
-    if (slash_pos != std::string::npos) {
-        parts.host = rest.substr(0, slash_pos);
-        parts.path = rest.substr(slash_pos);
-    } else {
-        parts.host = rest;
-        parts.path = "/";
-    }
-    return parts;
-}
-
-static std::pair<httplib::Client, common_http_url> common_http_client(const std::string & url) {
-    common_http_url parts = common_http_parse_url(url);
-
-    if (parts.host.empty()) {
-        throw std::runtime_error("error: invalid URL format");
-    }
-
-    httplib::Client cli(parts.scheme + "://" + parts.host);
-
-    if (!parts.user.empty()) {
-        cli.set_basic_auth(parts.user, parts.password);
-    }
-
-    cli.set_follow_location(true);
-
-    return { std::move(cli), std::move(parts) };
-}
-
-static std::string common_http_show_masked_url(const common_http_url & parts) {
-    return parts.scheme + "://" + (parts.user.empty() ? "" : "****:****@") + parts.host + parts.path;
-}

common/json-partial.cpp

@@ -5,7 +5,6 @@
 #include <nlohmann/json.hpp>
 
 #include <string>
-#include <regex>
 
 using json = nlohmann::ordered_json;
 
@@ -169,47 +168,6 @@ bool common_json_parse(
                 }
             }
 
-            // Matches a potentially partial unicode escape sequence, e.g. \u, \uX, \uXX, \uXXX, \uXXXX
-            static const std::regex partial_unicode_regex(R"(\\u(?:[0-9a-fA-F](?:[0-9a-fA-F](?:[0-9a-fA-F](?:[0-9a-fA-F])?)?)?)?$)");
-
-            auto is_high_surrogate = [&](const std::string & s) {
-                // Check if a partial of a high surrogate (U+D800-U+DBFF)
-                return s.length() >= 4 &&
-                    s[0] == '\\' && s[1] == 'u' &&
-                    std::tolower(s[2]) == 'd' &&
-                    (s[3] == '8' || s[3] == '9' || std::tolower(s[3]) == 'a' || std::tolower(s[3]) == 'b');
-            };
-
-            // Initialize the unicode marker to a low surrogate to handle the edge case
-            // where a high surrogate (U+D800-U+DBFF) is immediately followed by a
-            // backslash (\)
-            std::string unicode_marker_padding = "udc00";
-            std::smatch last_unicode_seq;
-
-            if (std::regex_search(str, last_unicode_seq, partial_unicode_regex)) {
-                std::smatch second_last_seq;
-                std::string prelude = str.substr(0, last_unicode_seq.position());
-
-                // Pad the escape sequence with 0s until it forms a complete sequence of 6 characters
-                unicode_marker_padding = std::string(6 - last_unicode_seq.length(), '0');
-
-                if (is_high_surrogate(last_unicode_seq.str())) {
-                    // If the sequence is a partial match for a high surrogate, add a low surrogate (U+DC00-U+UDFF)
-                    unicode_marker_padding += "\\udc00";
-                } else if (std::regex_search(prelude, second_last_seq, partial_unicode_regex)) {
-                    if (is_high_surrogate(second_last_seq.str())) {
-                        // If this follows a high surrogate, pad it to be a low surrogate
-                        if (last_unicode_seq.length() == 2) {
-                            unicode_marker_padding = "dc00";
-                        } else if (last_unicode_seq.length() == 3) {
-                            unicode_marker_padding = "c00";
-                        } else {
-                            // The original unicode_marker_padding is already padded with 0s
-                        }
-                    }
-                }
-            }
-
             const auto & magic_seed = out.healing_marker.marker = healing_marker;//"$llama.cpp.json$";
 
             if (err_loc.stack.back().type == COMMON_JSON_STACK_ELEMENT_KEY) {
@@ -228,9 +186,6 @@ bool common_json_parse(
                 } else if (str[str.length() - 1] == '\\' && can_parse(str + "\\\"" + closing)) {
                     // Was inside an object value string after an escape
                     str += (out.healing_marker.json_dump_marker = "\\" + magic_seed) + "\"" + closing;
-                } else if (can_parse(str + unicode_marker_padding + "\"" + closing)) {
-                    // Was inside an object value string after a partial unicode escape
-                    str += (out.healing_marker.json_dump_marker = unicode_marker_padding + magic_seed) + "\"" + closing;
                 } else {
                     // find last :
                     auto last_pos = str.find_last_of(':');
@@ -250,9 +205,6 @@ bool common_json_parse(
                 } else if (str[str.length() - 1] == '\\' && can_parse(str + "\\\"" + closing)) {
                     // Was inside an array value string after an escape
                     str += (out.healing_marker.json_dump_marker = "\\" + magic_seed) + "\"" + closing;
-                } else if (can_parse(str + unicode_marker_padding + "\"" + closing)) {
-                    // Was inside an array value string after a partial unicode escape
-                    str += (out.healing_marker.json_dump_marker = unicode_marker_padding + magic_seed) + "\"" + closing;
                 } else if (!was_maybe_number() && can_parse(str + ", 1" + closing)) {
                     // Had just finished a value
                     str += (out.healing_marker.json_dump_marker = ",\"" + magic_seed) + "\"" + closing;
@@ -278,9 +230,6 @@ bool common_json_parse(
                 } else if (str[str.length() - 1] == '\\' && can_parse(str + "\\\": 1" + closing)) {
                     // Was inside an object key string after an escape
                     str += (out.healing_marker.json_dump_marker = "\\" + magic_seed) + "\": 1" + closing;
-                } else if (can_parse(str + unicode_marker_padding + "\": 1" + closing)) {
-                    // Was inside an object key string after a partial unicode escape
-                    str += (out.healing_marker.json_dump_marker = unicode_marker_padding + magic_seed) + "\": 1" + closing;
                 } else {
                     auto last_pos = str.find_last_of(':');
                     if (last_pos == std::string::npos) {
@@ -297,25 +246,8 @@ bool common_json_parse(
             it = temptative_end;
             return true;
         }
-        // handle unclosed top-level primitive
-        if (err_loc.position != 0 && !healing_marker.empty() && err_loc.stack.empty()) {
-            std::string str(it, temptative_end);
-            const auto & magic_seed = out.healing_marker.marker = healing_marker;
-            if (can_parse(str + "\"")) {
-                // Was inside an string
-                str += (out.healing_marker.json_dump_marker = magic_seed) + "\"";
-            } else if (str[str.length() - 1] == '\\' && can_parse(str + "\\\"")) {
-                // Was inside an string after an escape
-                str += (out.healing_marker.json_dump_marker = "\\" + magic_seed) + "\"";
-            } else {
-                // TODO: handle more unclosed top-level primitive if the stack was empty but we got an error (e.g. "tru", "\"", etc...)
-                // fprintf(stderr, "Closing: TODO\n");
-                return false;
-            }
-            out.json = json::parse(str);
-            it = temptative_end;
-            return true;
-        }
+        // TODO: handle unclosed top-level primitive if the stack was empty but we got an error (e.g. "tru", "\"", etc...)
+        // fprintf(stderr, "Closing: TODO\n");
         return false;
     }
     out.json = json::parse(it, end);

common/json-schema-to-grammar.cpp

@@ -41,9 +41,9 @@ static std::string build_repetition(const std::string & item_rule, int min_items
     return result;
 }
 
-static void _build_min_max_int(int64_t min_value, int64_t max_value, std::stringstream & out, int decimals_left = 16, bool top_level = true) {
-    auto has_min = min_value != std::numeric_limits<int64_t>::min();
-    auto has_max = max_value != std::numeric_limits<int64_t>::max();
+static void _build_min_max_int(int min_value, int max_value, std::stringstream & out, int decimals_left = 16, bool top_level = true) {
+    auto has_min = min_value != std::numeric_limits<int>::min();
+    auto has_max = max_value != std::numeric_limits<int>::max();
 
     auto digit_range = [&](char from, char to) {
         out << "[";
@@ -159,7 +159,7 @@ static void _build_min_max_int(int64_t min_value, int64_t max_value, std::string
     if (has_min) {
         if (min_value < 0) {
             out << "\"-\" (";
-            _build_min_max_int(std::numeric_limits<int64_t>::min(), -min_value, out, decimals_left, /* top_level= */ false);
+            _build_min_max_int(std::numeric_limits<int>::min(), -min_value, out, decimals_left, /* top_level= */ false);
             out << ") | [0] | [1-9] ";
             more_digits(0, decimals_left - 1);
         } else if (min_value == 0) {
@@ -194,7 +194,7 @@ static void _build_min_max_int(int64_t min_value, int64_t max_value, std::string
             }
             digit_range(c, c);
             out << " (";
-            _build_min_max_int(std::stoll(min_s.substr(1)), std::numeric_limits<int64_t>::max(), out, less_decimals, /* top_level= */ false);
+            _build_min_max_int(std::stoi(min_s.substr(1)), std::numeric_limits<int>::max(), out, less_decimals, /* top_level= */ false);
             out << ")";
             if (c < '9') {
                 out << " | ";
@@ -216,7 +216,7 @@ static void _build_min_max_int(int64_t min_value, int64_t max_value, std::string
             _build_min_max_int(0, max_value, out, decimals_left, /* top_level= */ true);
         } else {
             out << "\"-\" (";
-            _build_min_max_int(-max_value, std::numeric_limits<int64_t>::max(), out, decimals_left, /* top_level= */ false);
+            _build_min_max_int(-max_value, std::numeric_limits<int>::max(), out, decimals_left, /* top_level= */ false);
             out << ")";
         }
         return;
@@ -257,21 +257,20 @@ std::unordered_map<std::string, BuiltinRule> STRING_FORMAT_RULES = {
 };
 
 static bool is_reserved_name(const std::string & name) {
-    static const std::unordered_set<std::string> RESERVED_NAMES = [] {
-        std::unordered_set<std::string> s;
-        s.insert("root");
-        for (const auto & p : PRIMITIVE_RULES) s.insert(p.first);
-        for (const auto & p : STRING_FORMAT_RULES) s.insert(p.first);
-        return s;
-    }();
+    static std::unordered_set<std::string> RESERVED_NAMES;
+    if (RESERVED_NAMES.empty()) {
+        RESERVED_NAMES.insert("root");
+        for (const auto &p : PRIMITIVE_RULES) RESERVED_NAMES.insert(p.first);
+        for (const auto &p : STRING_FORMAT_RULES) RESERVED_NAMES.insert(p.first);
+    }
     return RESERVED_NAMES.find(name) != RESERVED_NAMES.end();
 }
 
 std::regex INVALID_RULE_CHARS_RE("[^a-zA-Z0-9-]+");
-std::regex GRAMMAR_LITERAL_ESCAPE_RE("[\r\n\"\\\\]");
+std::regex GRAMMAR_LITERAL_ESCAPE_RE("[\r\n\"]");
 std::regex GRAMMAR_RANGE_LITERAL_ESCAPE_RE("[\r\n\"\\]\\-\\\\]");
 std::unordered_map<char, std::string> GRAMMAR_LITERAL_ESCAPES = {
-    {'\r', "\\r"}, {'\n', "\\n"}, {'"', "\\\""}, {'-', "\\-"}, {']', "\\]"}, {'\\', "\\\\"}
+    {'\r', "\\r"}, {'\n', "\\n"}, {'"', "\\\""}, {'-', "\\-"}, {']', "\\]"}
 };
 
 std::unordered_set<char> NON_LITERAL_SET = {'|', '.', '(', ')', '[', ']', '{', '}', '*', '+', '?'};
@@ -303,11 +302,8 @@ static std::string format_literal(const std::string & literal) {
     return "\"" + escaped + "\"";
 }
 
-std::string gbnf_format_literal(const std::string & literal) { return format_literal(literal); }
-
-class common_schema_converter {
+class SchemaConverter {
 private:
-    friend class common_schema_info;
     friend std::string build_grammar(const std::function<void(const common_grammar_builder &)> & cb, const common_grammar_options & options);
     std::function<json(const std::string &)> _fetch_json;
     bool _dotall;
@@ -604,10 +600,7 @@ private:
     }
 
     std::string _resolve_ref(const std::string & ref) {
-        auto it = ref.find('#');
-        std::string ref_fragment = it != std::string::npos ? ref.substr(it + 1) : ref;
-        static const std::regex nonalphanumeric_regex(R"([^a-zA-Z0-9-]+)");
-        std::string ref_name = "ref" + std::regex_replace(ref_fragment, nonalphanumeric_regex, "-");
+        std::string ref_name = ref.substr(ref.find_last_of('/') + 1);
         if (_rules.find(ref_name) == _rules.end() && _refs_being_resolved.find(ref) == _refs_being_resolved.end()) {
             _refs_being_resolved.insert(ref);
             json resolved = _refs[ref];
@@ -730,7 +723,7 @@ private:
     }
 
 public:
-    common_schema_converter(
+    SchemaConverter(
         const std::function<json(const std::string &)> & fetch_json,
         bool dotall)
           : _fetch_json(fetch_json), _dotall(dotall)
@@ -780,24 +773,11 @@ public:
                         std::vector<std::string> tokens = string_split(pointer, "/");
                         for (size_t i = 1; i < tokens.size(); ++i) {
                             std::string sel = tokens[i];
-                            if (target.is_object() && target.contains(sel)) {
-                                target = target[sel];
-                            } else if (target.is_array()) {
-                                size_t sel_index;
-                                try {
-                                    sel_index = std::stoul(sel);
-                                } catch (const std::invalid_argument & e) {
-                                    sel_index = target.size();
-                                }
-                                if (sel_index >= target.size()) {
-                                    _errors.push_back("Error resolving ref " + ref + ": " + sel + " not in " + target.dump());
-                                    return;
-                                }
-                                target = target[sel_index];
-                            } else {
+                            if (target.is_null() || !target.contains(sel)) {
                                 _errors.push_back("Error resolving ref " + ref + ": " + sel + " not in " + target.dump());
                                 return;
                             }
+                            target = target[sel];
                         }
                         _refs[ref] = target;
                     }
@@ -863,10 +843,9 @@ public:
                 _build_object_rule(
                     properties, required, name,
                     schema.contains("additionalProperties") ? schema["additionalProperties"] : json()));
-        } else if ((schema_type.is_null() || schema_type == "object" || schema_type == "string") && schema.contains("allOf")) {
+        } else if ((schema_type.is_null() || schema_type == "object") && schema.contains("allOf")) {
             std::unordered_set<std::string> required;
             std::vector<std::pair<std::string, json>> properties;
-            std::map<std::string, size_t> enum_values;
             std::string hybrid_name = name;
             std::function<void(const json &, bool)> add_component = [&](const json & comp_schema, bool is_required) {
                 if (comp_schema.contains("$ref")) {
@@ -878,14 +857,6 @@ public:
                             required.insert(prop.key());
                         }
                     }
-                } else if (comp_schema.contains("enum")) {
-                    for (const auto & v : comp_schema["enum"]) {
-                        const auto rule = _generate_constant_rule(v);
-                        if (enum_values.find(rule) == enum_values.end()) {
-                            enum_values[rule] = 0;
-                        }
-                        enum_values[rule] += 1;
-                    }
                 } else {
                   // todo warning
                 }
@@ -899,17 +870,6 @@ public:
                     add_component(t, true);
                 }
             }
-            if (!enum_values.empty()) {
-                std::vector<std::string> enum_intersection;
-                for (const auto & p : enum_values) {
-                    if (p.second == schema["allOf"].size()) {
-                        enum_intersection.push_back(p.first);
-                    }
-                }
-                if (!enum_intersection.empty()) {
-                    return _add_rule(rule_name, "(" + string_join(enum_intersection, " | ") + ") space");
-                }
-            }
             return _add_rule(rule_name, _build_object_rule(properties, required, hybrid_name, json()));
         } else if ((schema_type.is_null() || schema_type == "array") && (schema.contains("items") || schema.contains("prefixItems"))) {
             json items = schema.contains("items") ? schema["items"] : schema["prefixItems"];
@@ -944,17 +904,17 @@ public:
             int max_len = schema.contains("maxLength") ? schema["maxLength"].get<int>() : std::numeric_limits<int>::max();
             return _add_rule(rule_name, "\"\\\"\" " + build_repetition(char_rule, min_len, max_len) + " \"\\\"\" space");
         } else if (schema_type == "integer" && (schema.contains("minimum") || schema.contains("exclusiveMinimum") || schema.contains("maximum") || schema.contains("exclusiveMaximum"))) {
-            int64_t min_value = std::numeric_limits<int64_t>::min();
-            int64_t max_value = std::numeric_limits<int64_t>::max();
+            int min_value = std::numeric_limits<int>::min();
+            int max_value = std::numeric_limits<int>::max();
             if (schema.contains("minimum")) {
-                min_value = schema["minimum"].get<int64_t>();
+                min_value = schema["minimum"].get<int>();
             } else if (schema.contains("exclusiveMinimum")) {
-                min_value = schema["exclusiveMinimum"].get<int64_t>() + 1;
+                min_value = schema["exclusiveMinimum"].get<int>() + 1;
             }
             if (schema.contains("maximum")) {
-                max_value = schema["maximum"].get<int64_t>();
+                max_value = schema["maximum"].get<int>();
             } else if (schema.contains("exclusiveMaximum")) {
-                max_value = schema["exclusiveMaximum"].get<int64_t>() - 1;
+                max_value = schema["exclusiveMaximum"].get<int>() - 1;
             }
             std::stringstream out;
             out << "(";
@@ -975,7 +935,7 @@ public:
 
     void check_errors() {
         if (!_errors.empty()) {
-            throw std::invalid_argument("JSON schema conversion failed:\n" + string_join(_errors, "\n"));
+            throw std::runtime_error("JSON schema conversion failed:\n" + string_join(_errors, "\n"));
         }
         if (!_warnings.empty()) {
             fprintf(stderr, "WARNING: JSON schema conversion was incomplete: %s\n", string_join(_warnings, "; ").c_str());
@@ -991,134 +951,6 @@ public:
     }
 };
 
-// common_schema_info implementation (pimpl)
-
-common_schema_info::common_schema_info()
-    : impl_(std::make_unique<common_schema_converter>(
-        [](const std::string &) { return json(); },
-        false)) {}
-
-common_schema_info::~common_schema_info() = default;
-
-common_schema_info::common_schema_info(common_schema_info &&) noexcept = default;
-common_schema_info & common_schema_info::operator=(common_schema_info &&) noexcept = default;
-
-void common_schema_info::resolve_refs(nlohmann::ordered_json & schema) {
-    impl_->resolve_refs(schema, "");
-}
-
-// Determines if a JSON schema can resolve to a string type through any path.
-// Some models emit raw string values rather than JSON-encoded strings for string parameters.
-// If any branch of the schema (via oneOf, anyOf, $ref, etc.) permits a string, this returns
-// true, allowing callers to handle the value as a raw string for simplicity.
-bool common_schema_info::resolves_to_string(const nlohmann::ordered_json & schema) {
-    std::unordered_set<std::string> visited_refs;
-
-    std::function<bool(const json &)> check = [&](const json & s) -> bool {
-        if (!s.is_object()) {
-            return false;
-        }
-
-        // Handle $ref
-        if (s.contains("$ref")) {
-            const std::string & ref = s["$ref"];
-            if (visited_refs.find(ref) != visited_refs.end()) {
-                // Circular reference, assume not a string to be safe
-                return false;
-            }
-            visited_refs.insert(ref);
-            auto it = impl_->_refs.find(ref);
-            if (it != impl_->_refs.end()) {
-                return check(it->second);
-            }
-            return false;
-        }
-
-        // Check type field
-        if (s.contains("type")) {
-            const json & schema_type = s["type"];
-            if (schema_type.is_string()) {
-                if (schema_type == "string") {
-                    return true;
-                }
-            } else if (schema_type.is_array()) {
-                // Type can be an array like ["string", "null"]
-                for (const auto & t : schema_type) {
-                    if (t == "string") {
-                        return true;
-                    }
-                }
-            }
-        }
-
-        // Check oneOf/anyOf - if any alternative can be a string
-        if (s.contains("oneOf")) {
-            for (const auto & alt : s["oneOf"]) {
-                if (check(alt)) {
-                    return true;
-                }
-            }
-        }
-        if (s.contains("anyOf")) {
-            for (const auto & alt : s["anyOf"]) {
-                if (check(alt)) {
-                    return true;
-                }
-            }
-        }
-
-        // Check allOf - all components must be compatible with string type
-        if (s.contains("allOf")) {
-            bool all_string = true;
-            for (const auto & component : s["allOf"]) {
-                if (!check(component)) {
-                    all_string = false;
-                    break;
-                }
-            }
-            if (all_string) {
-                return true;
-            }
-        }
-
-        // Check const - if the constant value is a string
-        if (s.contains("const")) {
-            if (s["const"].is_string()) {
-                return true;
-            }
-        }
-
-        // Check enum - if any enum value is a string
-        if (s.contains("enum")) {
-            for (const auto & val : s["enum"]) {
-                if (val.is_string()) {
-                    return true;
-                }
-            }
-        }
-
-        // String-specific keywords imply string type
-        if (s.contains("pattern") || s.contains("minLength") || s.contains("maxLength")) {
-            return true;
-        }
-
-        // Check format - many formats imply string
-        if (s.contains("format")) {
-            const std::string & fmt = s["format"];
-            if (fmt == "date" || fmt == "time" || fmt == "date-time" ||
-                fmt == "uri" || fmt == "email" || fmt == "hostname" ||
-                fmt == "ipv4" || fmt == "ipv6" || fmt == "uuid" ||
-                fmt.find("uuid") == 0) {
-                return true;
-            }
-        }
-
-        return false;
-    };
-
-    return check(schema);
-}
-
 std::string json_schema_to_grammar(const json & schema, bool force_gbnf) {
 #ifdef LLAMA_USE_LLGUIDANCE
     if (!force_gbnf) {
@@ -1135,7 +967,7 @@ std::string json_schema_to_grammar(const json & schema, bool force_gbnf) {
 }
 
 std::string build_grammar(const std::function<void(const common_grammar_builder &)> & cb, const common_grammar_options & options) {
-    common_schema_converter converter([&](const std::string &) { return json(); }, options.dotall);
+    SchemaConverter converter([&](const std::string &) { return json(); }, options.dotall);
     common_grammar_builder builder {
         /* .add_rule = */ [&](const std::string & name, const std::string & rule) {
             return converter._add_rule(name, rule);

common/json-schema-to-grammar.h

@@ -3,31 +3,11 @@
 #include <nlohmann/json_fwd.hpp>
 
 #include <functional>
-#include <memory>
 #include <string>
 
 std::string json_schema_to_grammar(const nlohmann::ordered_json & schema,
                                    bool force_gbnf = false);
 
-class common_schema_converter;
-
-// Probes a JSON schema to extract information about its structure and type constraints.
-class common_schema_info {
-    std::unique_ptr<common_schema_converter> impl_;
-
-  public:
-    common_schema_info();
-    ~common_schema_info();
-
-    common_schema_info(const common_schema_info &) = delete;
-    common_schema_info & operator=(const common_schema_info &) = delete;
-    common_schema_info(common_schema_info &&) noexcept;
-    common_schema_info & operator=(common_schema_info &&) noexcept;
-
-    void resolve_refs(nlohmann::ordered_json & schema);
-    bool resolves_to_string(const nlohmann::ordered_json & schema);
-};
-
 struct common_grammar_builder {
     std::function<std::string(const std::string &, const std::string &)> add_rule;
     std::function<std::string(const std::string &, const nlohmann::ordered_json &)> add_schema;
@@ -38,6 +18,4 @@ struct common_grammar_options {
     bool dotall = false;
 };
 
-std::string gbnf_format_literal(const std::string & literal);
-
 std::string build_grammar(const std::function<void(const common_grammar_builder &)> & cb, const common_grammar_options & options = {});

common/log.cpp

@@ -1,26 +1,14 @@
-#include "common.h"
 #include "log.h"
 
 #include <chrono>
 #include <condition_variable>
 #include <cstdarg>
 #include <cstdio>
-#include <cstdlib>
-#include <cstring>
 #include <mutex>
 #include <sstream>
 #include <thread>
 #include <vector>
 
-#if defined(_WIN32)
-#    include <io.h>
-#    include <windows.h>
-#    define isatty _isatty
-#    define fileno _fileno
-#else
-#    include <unistd.h>
-#endif // defined(_WIN32)
-
 int common_log_verbosity_thold = LOG_DEFAULT_LLAMA;
 
 void common_log_set_verbosity_thold(int verbosity) {
@@ -365,11 +353,6 @@ struct common_log * common_log_init() {
 
 struct common_log * common_log_main() {
     static struct common_log log;
-    static std::once_flag    init_flag;
-    std::call_once(init_flag, [&]() {
-        // Set default to auto-detect colors
-        log.set_colors(tty_can_use_colors());
-    });
 
     return &log;
 }
@@ -397,19 +380,8 @@ void common_log_set_file(struct common_log * log, const char * file) {
     log->set_file(file);
 }
 
-void common_log_set_colors(struct common_log * log, log_colors colors) {
-    if (colors == LOG_COLORS_AUTO) {
-        log->set_colors(tty_can_use_colors());
-        return;
-    }
-
-    if (colors == LOG_COLORS_DISABLED) {
-        log->set_colors(false);
-        return;
-    }
-
-    GGML_ASSERT(colors == LOG_COLORS_ENABLED);
-    log->set_colors(true);
+void common_log_set_colors(struct common_log * log, bool colors) {
+    log->set_colors(colors);
 }
 
 void common_log_set_prefix(struct common_log * log, bool prefix) {
@@ -419,28 +391,3 @@ void common_log_set_prefix(struct common_log * log, bool prefix) {
 void common_log_set_timestamps(struct common_log * log, bool timestamps) {
     log->set_timestamps(timestamps);
 }
-
-void common_log_flush(struct common_log * log) {
-    log->pause();
-    log->resume();
-}
-
-static int common_get_verbosity(enum ggml_log_level level) {
-    switch (level) {
-        case GGML_LOG_LEVEL_DEBUG: return LOG_LEVEL_DEBUG;
-        case GGML_LOG_LEVEL_INFO:  return LOG_LEVEL_INFO;
-        case GGML_LOG_LEVEL_WARN:  return LOG_LEVEL_WARN;
-        case GGML_LOG_LEVEL_ERROR: return LOG_LEVEL_ERROR;
-        case GGML_LOG_LEVEL_CONT:  return LOG_LEVEL_INFO; // same as INFO
-        case GGML_LOG_LEVEL_NONE:
-        default:
-            return LOG_LEVEL_OUTPUT;
-    }
-}
-
-void common_log_default_callback(enum ggml_log_level level, const char * text, void * /*user_data*/) {
-    auto verbosity = common_get_verbosity(level);
-    if (verbosity <= common_log_verbosity_thold) {
-        common_log_add(common_log_main(), level, "%s", text);
-    }
-}

common/log.h

@@ -21,20 +21,8 @@
 #    define LOG_ATTRIBUTE_FORMAT(...) __attribute__((format(printf, __VA_ARGS__)))
 #endif
 
-#define LOG_LEVEL_DEBUG  4
-#define LOG_LEVEL_INFO   3
-#define LOG_LEVEL_WARN   2
-#define LOG_LEVEL_ERROR  1
-#define LOG_LEVEL_OUTPUT 0 // output data from tools
-
-#define LOG_DEFAULT_DEBUG LOG_LEVEL_DEBUG
-#define LOG_DEFAULT_LLAMA LOG_LEVEL_INFO
-
-enum log_colors {
-    LOG_COLORS_AUTO     = -1,
-    LOG_COLORS_DISABLED = 0,
-    LOG_COLORS_ENABLED  = 1,
-};
+#define LOG_DEFAULT_DEBUG 1
+#define LOG_DEFAULT_LLAMA 0
 
 // needed by the LOG_TMPL macro to avoid computing log arguments if the verbosity lower
 // set via common_log_set_verbosity()
@@ -42,8 +30,6 @@ extern int common_log_verbosity_thold;
 
 void common_log_set_verbosity_thold(int verbosity); // not thread-safe
 
-void common_log_default_callback(enum ggml_log_level level, const char * text, void * user_data);
-
 // the common_log uses an internal worker thread to print/write log messages
 // when the worker thread is paused, incoming log messages are discarded
 struct common_log;
@@ -73,18 +59,16 @@ void common_log_add(struct common_log * log, enum ggml_log_level level, const ch
 //   0.00.090.578 I llm_load_tensors: offloading 32 repeating layers to GPU
 //   0.00.090.579 I llm_load_tensors: offloading non-repeating layers to GPU
 //
+// I - info    (stdout, V = 0)
+// W - warning (stderr, V = 0)
+// E - error   (stderr, V = 0)
 // D - debug   (stderr, V = LOG_DEFAULT_DEBUG)
-// I - info    (stdout, V = LOG_DEFAULT_INFO)
-// W - warning (stderr, V = LOG_DEFAULT_WARN)
-// E - error   (stderr, V = LOG_DEFAULT_ERROR)
-// O - output  (stdout, V = LOG_DEFAULT_OUTPUT)
 //
 
-void common_log_set_file      (struct common_log * log, const char * file); // not thread-safe
-void common_log_set_colors    (struct common_log * log, log_colors colors); // not thread-safe
-void common_log_set_prefix    (struct common_log * log, bool prefix);       // whether to output prefix to each log
-void common_log_set_timestamps(struct common_log * log, bool timestamps);   // whether to output timestamps in the prefix
-void common_log_flush         (struct common_log * log);                    // flush all pending log messages
+void common_log_set_file      (struct common_log * log, const char * file);       // not thread-safe
+void common_log_set_colors    (struct common_log * log,       bool   colors);     // not thread-safe
+void common_log_set_prefix    (struct common_log * log,       bool   prefix);     // whether to output prefix to each log
+void common_log_set_timestamps(struct common_log * log,       bool   timestamps); // whether to output timestamps in the prefix
 
 // helper macros for logging
 // use these to avoid computing log arguments if the verbosity of the log is higher than the threshold
@@ -103,14 +87,14 @@ void common_log_flush         (struct common_log * log);                    // f
         } \
     } while (0)
 
-#define LOG(...)             LOG_TMPL(GGML_LOG_LEVEL_NONE, LOG_LEVEL_OUTPUT, __VA_ARGS__)
-#define LOGV(verbosity, ...) LOG_TMPL(GGML_LOG_LEVEL_NONE, verbosity,        __VA_ARGS__)
+#define LOG(...)             LOG_TMPL(GGML_LOG_LEVEL_NONE, 0,         __VA_ARGS__)
+#define LOGV(verbosity, ...) LOG_TMPL(GGML_LOG_LEVEL_NONE, verbosity, __VA_ARGS__)
 
-#define LOG_DBG(...) LOG_TMPL(GGML_LOG_LEVEL_DEBUG, LOG_LEVEL_DEBUG,  __VA_ARGS__)
-#define LOG_INF(...) LOG_TMPL(GGML_LOG_LEVEL_INFO,  LOG_LEVEL_INFO,   __VA_ARGS__)
-#define LOG_WRN(...) LOG_TMPL(GGML_LOG_LEVEL_WARN,  LOG_LEVEL_WARN,   __VA_ARGS__)
-#define LOG_ERR(...) LOG_TMPL(GGML_LOG_LEVEL_ERROR, LOG_LEVEL_ERROR,  __VA_ARGS__)
-#define LOG_CNT(...) LOG_TMPL(GGML_LOG_LEVEL_CONT,  LOG_LEVEL_INFO,   __VA_ARGS__) // same as INFO
+#define LOG_INF(...) LOG_TMPL(GGML_LOG_LEVEL_INFO,  0,                 __VA_ARGS__)
+#define LOG_WRN(...) LOG_TMPL(GGML_LOG_LEVEL_WARN,  0,                 __VA_ARGS__)
+#define LOG_ERR(...) LOG_TMPL(GGML_LOG_LEVEL_ERROR, 0,                 __VA_ARGS__)
+#define LOG_DBG(...) LOG_TMPL(GGML_LOG_LEVEL_DEBUG, LOG_DEFAULT_DEBUG, __VA_ARGS__)
+#define LOG_CNT(...) LOG_TMPL(GGML_LOG_LEVEL_CONT,  0,                 __VA_ARGS__)
 
 #define LOG_INFV(verbosity, ...) LOG_TMPL(GGML_LOG_LEVEL_INFO,  verbosity, __VA_ARGS__)
 #define LOG_WRNV(verbosity, ...) LOG_TMPL(GGML_LOG_LEVEL_WARN,  verbosity, __VA_ARGS__)

common/peg-parser.h

@@ -1,459 +0,0 @@
-#pragma once
-
-#include <nlohmann/json_fwd.hpp>
-
-#include <memory>
-#include <unordered_map>
-#include <string>
-#include <string_view>
-#include <functional>
-#include <vector>
-#include <variant>
-
-struct common_grammar_builder;
-
-class common_peg_parser_builder;
-
-using common_peg_parser_id = size_t;
-constexpr common_peg_parser_id COMMON_PEG_INVALID_PARSER_ID = static_cast<common_peg_parser_id>(-1);
-
-using common_peg_ast_id = size_t;
-constexpr common_peg_ast_id COMMON_PEG_INVALID_AST_ID = static_cast<common_peg_ast_id>(-1);
-
-// Lightweight wrapper around common_peg_parser_id for convenience
-class common_peg_parser {
-    common_peg_parser_id id_;
-    common_peg_parser_builder & builder_;
-
-  public:
-    common_peg_parser(const common_peg_parser & other) : id_(other.id_), builder_(other.builder_) {}
-    common_peg_parser(common_peg_parser_id id, common_peg_parser_builder & builder) : id_(id), builder_(builder) {}
-
-    common_peg_parser & operator=(const common_peg_parser & other);
-    common_peg_parser & operator+=(const common_peg_parser & other);
-    common_peg_parser & operator|=(const common_peg_parser & other);
-
-    operator common_peg_parser_id() const { return id_; }
-    common_peg_parser_id id() const { return id_; }
-
-    common_peg_parser_builder & builder() const { return builder_; }
-
-    // Creates a sequence
-    common_peg_parser operator+(const common_peg_parser & other) const;
-
-    // Creates a sequence separated by spaces.
-    common_peg_parser operator<<(const common_peg_parser & other) const;
-
-    // Creates a choice
-    common_peg_parser operator|(const common_peg_parser & other) const;
-
-    common_peg_parser operator+(const char * str) const;
-    common_peg_parser operator+(const std::string & str) const;
-    common_peg_parser operator<<(const char * str) const;
-    common_peg_parser operator<<(const std::string & str) const;
-    common_peg_parser operator|(const char * str) const;
-    common_peg_parser operator|(const std::string & str) const;
-};
-
-common_peg_parser operator+(const char * str, const common_peg_parser & p);
-common_peg_parser operator+(const std::string & str, const common_peg_parser & p);
-common_peg_parser operator<<(const char * str, const common_peg_parser & p);
-common_peg_parser operator<<(const std::string & str, const common_peg_parser & p);
-common_peg_parser operator|(const char * str, const common_peg_parser & p);
-common_peg_parser operator|(const std::string & str, const common_peg_parser & p);
-
-enum common_peg_parse_result_type {
-    COMMON_PEG_PARSE_RESULT_FAIL            = 0,
-    COMMON_PEG_PARSE_RESULT_SUCCESS         = 1,
-    COMMON_PEG_PARSE_RESULT_NEED_MORE_INPUT = 2,
-};
-
-const char * common_peg_parse_result_type_name(common_peg_parse_result_type type);
-
-struct common_peg_ast_node {
-    common_peg_ast_id id;
-    std::string rule;
-    std::string tag;
-    size_t start;
-    size_t end;
-    std::string_view text;
-    std::vector<common_peg_ast_id> children;
-
-    bool is_partial = false;
-};
-
-struct common_peg_parse_result;
-
-using common_peg_ast_visitor = std::function<void(const common_peg_ast_node & node)>;
-
-class common_peg_ast_arena {
-    std::vector<common_peg_ast_node> nodes_;
-  public:
-    common_peg_ast_id add_node(
-        const std::string & rule,
-        const std::string & tag,
-        size_t start,
-        size_t end,
-        std::string_view text,
-        std::vector<common_peg_ast_id> children,
-        bool is_partial = false
-    ) {
-        common_peg_ast_id id = nodes_.size();
-        nodes_.push_back({id, rule, tag, start, end, text, std::move(children), is_partial});
-        return id;
-    }
-
-    const common_peg_ast_node & get(common_peg_ast_id id) const { return nodes_.at(id); }
-
-    size_t size() const { return nodes_.size(); }
-
-    void clear() { nodes_.clear(); }
-
-    void visit(common_peg_ast_id id, const common_peg_ast_visitor & visitor) const;
-    void visit(const common_peg_parse_result & result, const common_peg_ast_visitor & visitor) const;
-};
-
-struct common_peg_parse_result {
-    common_peg_parse_result_type type = COMMON_PEG_PARSE_RESULT_FAIL;
-    size_t start = 0;
-    size_t end = 0;
-
-    std::vector<common_peg_ast_id> nodes;
-
-    common_peg_parse_result() = default;
-
-    common_peg_parse_result(common_peg_parse_result_type type, size_t start)
-        : type(type), start(start), end(start) {}
-
-    common_peg_parse_result(common_peg_parse_result_type type, size_t start, size_t end)
-        : type(type), start(start), end(end) {}
-
-    common_peg_parse_result(common_peg_parse_result_type type, size_t start, size_t end, std::vector<common_peg_ast_id> nodes)
-        : type(type), start(start), end(end), nodes(std::move(nodes)) {}
-
-    bool fail() const { return type == COMMON_PEG_PARSE_RESULT_FAIL; }
-    bool need_more_input() const { return type == COMMON_PEG_PARSE_RESULT_NEED_MORE_INPUT; }
-    bool success() const { return type == COMMON_PEG_PARSE_RESULT_SUCCESS; }
-};
-
-struct common_peg_parse_context {
-    std::string input;
-    bool is_partial;
-    common_peg_ast_arena ast;
-
-    int parse_depth;
-
-    common_peg_parse_context()
-        : is_partial(false), parse_depth(0) {}
-
-    common_peg_parse_context(const std::string & input)
-        : input(input), is_partial(false), parse_depth(0) {}
-
-    common_peg_parse_context(const std::string & input, bool is_partial)
-        : input(input), is_partial(is_partial), parse_depth(0) {}
-};
-
-class common_peg_arena;
-
-// Parser variants
-struct common_peg_epsilon_parser {};
-
-struct common_peg_start_parser {};
-
-struct common_peg_end_parser {};
-
-struct common_peg_literal_parser {
-    std::string literal;
-};
-
-struct common_peg_sequence_parser {
-    std::vector<common_peg_parser_id> children;
-};
-
-struct common_peg_choice_parser {
-    std::vector<common_peg_parser_id> children;
-};
-
-struct common_peg_repetition_parser {
-    common_peg_parser_id child;
-    int min_count;
-    int max_count;  // -1 for unbounded
-};
-
-struct common_peg_and_parser {
-    common_peg_parser_id child;
-};
-
-struct common_peg_not_parser {
-    common_peg_parser_id child;
-};
-
-struct common_peg_any_parser {};
-
-struct common_peg_space_parser {};
-
-struct common_peg_chars_parser {
-    struct char_range {
-        uint32_t start;
-        uint32_t end;
-        bool contains(uint32_t codepoint) const { return codepoint >= start && codepoint <= end; }
-    };
-
-    std::string pattern;
-    std::vector<char_range> ranges;
-    bool negated;
-    int min_count;
-    int max_count;  // -1 for unbounded
-};
-
-struct common_peg_json_string_parser {};
-
-struct common_peg_until_parser {
-    std::vector<std::string> delimiters;
-};
-
-struct common_peg_schema_parser {
-    common_peg_parser_id child;
-    std::string name;
-    std::shared_ptr<nlohmann::ordered_json> schema;
-
-    // Indicates if the GBNF should accept a raw string that matches the schema.
-    bool raw;
-};
-
-struct common_peg_rule_parser {
-    std::string name;
-    common_peg_parser_id child;
-    bool trigger;
-};
-
-struct common_peg_ref_parser {
-    std::string name;
-};
-
-struct common_peg_atomic_parser {
-    common_peg_parser_id child;
-};
-
-struct common_peg_tag_parser {
-    common_peg_parser_id child;
-    std::string tag;
-};
-
-// Variant holding all parser types
-using common_peg_parser_variant = std::variant<
-    common_peg_epsilon_parser,
-    common_peg_start_parser,
-    common_peg_end_parser,
-    common_peg_literal_parser,
-    common_peg_sequence_parser,
-    common_peg_choice_parser,
-    common_peg_repetition_parser,
-    common_peg_and_parser,
-    common_peg_not_parser,
-    common_peg_any_parser,
-    common_peg_space_parser,
-    common_peg_chars_parser,
-    common_peg_json_string_parser,
-    common_peg_until_parser,
-    common_peg_schema_parser,
-    common_peg_rule_parser,
-    common_peg_ref_parser,
-    common_peg_atomic_parser,
-    common_peg_tag_parser
->;
-
-class common_peg_arena {
-    std::vector<common_peg_parser_variant> parsers_;
-    std::unordered_map<std::string, common_peg_parser_id> rules_;
-    common_peg_parser_id root_ = COMMON_PEG_INVALID_PARSER_ID;
-
-  public:
-    const common_peg_parser_variant & get(common_peg_parser_id id) const { return parsers_.at(id); }
-    common_peg_parser_variant & get(common_peg_parser_id id) { return parsers_.at(id); }
-
-    size_t size() const { return parsers_.size(); }
-    bool empty() const { return parsers_.empty(); }
-
-    common_peg_parser_id get_rule(const std::string & name) const;
-    bool has_rule(const std::string & name) const { return rules_.find(name) != rules_.end(); }
-
-    common_peg_parser_id root() const { return root_; }
-    void set_root(common_peg_parser_id id) { root_ = id; }
-
-    common_peg_parse_result parse(common_peg_parse_context & ctx, size_t start = 0) const;
-    common_peg_parse_result parse(common_peg_parser_id id, common_peg_parse_context & ctx, size_t start) const;
-
-    void resolve_refs();
-
-    void build_grammar(const common_grammar_builder & builder, bool lazy = false) const;
-
-    std::string dump(common_peg_parser_id id) const;
-
-    nlohmann::json to_json() const;
-    static common_peg_arena from_json(const nlohmann::json & j);
-
-    std::string save() const;
-    void load(const std::string & data);
-
-    friend class common_peg_parser_builder;
-
-  private:
-    common_peg_parser_id add_parser(common_peg_parser_variant parser);
-    void add_rule(const std::string & name, common_peg_parser_id id);
-
-    common_peg_parser_id resolve_ref(common_peg_parser_id id);
-};
-
-class common_peg_parser_builder {
-    common_peg_arena arena_;
-
-    common_peg_parser wrap(common_peg_parser_id id) { return common_peg_parser(id, *this); }
-    common_peg_parser add(const common_peg_parser_variant & p) { return wrap(arena_.add_parser(p)); }
-
-  public:
-    common_peg_parser_builder();
-
-    // Match nothing, always succeed.
-    //   S -> ε
-    common_peg_parser eps() { return add(common_peg_epsilon_parser{}); }
-
-    // Matches the start of the input.
-    //   S -> ^
-    common_peg_parser start() { return add(common_peg_start_parser{}); }
-
-    // Matches the end of the input.
-    //   S -> $
-    common_peg_parser end() { return add(common_peg_end_parser{}); }
-
-    // Matches an exact literal string.
-    //   S -> "hello"
-    common_peg_parser literal(const std::string & literal) { return add(common_peg_literal_parser{literal}); }
-
-    // Matches a sequence of parsers in order, all must succeed.
-    //   S -> A B C
-    common_peg_parser sequence() { return add(common_peg_sequence_parser{}); }
-    common_peg_parser sequence(const std::vector<common_peg_parser_id> & parsers);
-    common_peg_parser sequence(const std::vector<common_peg_parser> & parsers);
-    common_peg_parser sequence(std::initializer_list<common_peg_parser> parsers);
-
-    // Matches the first parser that succeeds from a list of alternatives.
-    //   S -> A | B | C
-    common_peg_parser choice() { return add(common_peg_choice_parser{}); }
-    common_peg_parser choice(const std::vector<common_peg_parser_id> & parsers);
-    common_peg_parser choice(const std::vector<common_peg_parser> & parsers);
-    common_peg_parser choice(std::initializer_list<common_peg_parser> parsers);
-
-    // Matches one or more repetitions of a parser.
-    //   S -> A+
-    common_peg_parser one_or_more(const common_peg_parser & p) { return repeat(p, 1, -1); }
-
-    // Matches zero or more repetitions of a parser, always succeeds.
-    //   S -> A*
-    common_peg_parser zero_or_more(const common_peg_parser & p) { return repeat(p, 0, -1); }
-
-    // Matches zero or one occurrence of a parser, always succeeds.
-    //   S -> A?
-    common_peg_parser optional(const common_peg_parser & p) { return repeat(p, 0, 1); }
-
-    // Positive lookahead: succeeds if child parser succeeds, consumes no input.
-    //   S -> &A
-    common_peg_parser peek(const common_peg_parser & p) { return add(common_peg_and_parser{p}); }
-
-    // Negative lookahead: succeeds if child parser fails, consumes no input.
-    //   S -> !A
-    common_peg_parser negate(const common_peg_parser & p) { return add(common_peg_not_parser{p}); }
-
-    // Matches any single character.
-    //   S -> .
-    common_peg_parser any() { return add(common_peg_any_parser{}); }
-
-    // Matches between min and max repetitions of characters from a character class.
-    //   S -> [a-z]{m,n}
-    //
-    // Use -1 for max to represent unbounded repetition (equivalent to {m,})
-    common_peg_parser chars(const std::string & classes, int min = 1, int max = -1);
-
-    // Creates a lightweight reference to a named rule (resolved during build()).
-    // Use this for forward references in recursive grammars.
-    //   expr_ref -> expr
-    common_peg_parser ref(const std::string & name) { return add(common_peg_ref_parser{name}); }
-
-    // Matches zero or more whitespace characters (space, tab, newline).
-    //   S -> [ \t\n]*
-    common_peg_parser space() { return add(common_peg_space_parser{}); }
-
-    // Matches all characters until a delimiter is found (delimiter not consumed).
-    //   S -> (!delim .)*
-    common_peg_parser until(const std::string & delimiter) { return add(common_peg_until_parser{{delimiter}}); }
-
-    // Matches all characters until one of the delimiters in the list is found (delimiter not consumed).
-    //   S -> (!delim .)*
-    common_peg_parser until_one_of(const std::vector<std::string> & delimiters) { return add(common_peg_until_parser{delimiters}); }
-
-    // Matches everything
-    //   S -> .*
-    common_peg_parser rest() { return until_one_of({}); }
-
-    // Matches between min and max repetitions of a parser (inclusive).
-    //   S -> A{m,n}
-    // Use -1 for max to represent unbounded repetition (equivalent to {m,})
-    common_peg_parser repeat(const common_peg_parser & p, int min, int max) { return add(common_peg_repetition_parser{p, min,max}); }
-
-    // Matches exactly n repetitions of a parser.
-    //   S -> A{n}
-    common_peg_parser repeat(const common_peg_parser & p, int n) { return repeat(p, n, n); }
-
-    // Creates a complete JSON parser supporting objects, arrays, strings, numbers, booleans, and null.
-    //   value -> object | array | string | number | true | false | null
-    common_peg_parser json();
-    common_peg_parser json_object();
-    common_peg_parser json_string();
-    common_peg_parser json_array();
-    common_peg_parser json_number();
-    common_peg_parser json_bool();
-    common_peg_parser json_null();
-
-    // Matches JSON string content without the surrounding quotes.
-    // Useful for extracting content within a JSON string.
-    common_peg_parser json_string_content();
-
-    // Matches a JSON object member with a key and associated parser as the
-    // value.
-    common_peg_parser json_member(const std::string & key, const common_peg_parser & p);
-
-    // Wraps a parser with JSON schema metadata for grammar generation.
-    // Used internally to convert JSON schemas to GBNF grammar rules.
-    common_peg_parser schema(const common_peg_parser & p, const std::string & name, const nlohmann::ordered_json & schema, bool raw = false);
-
-    // Creates a named rule, stores it in the grammar, and returns a ref.
-    // If trigger=true, marks this rule as an entry point for lazy grammar generation.
-    //   auto json = p.rule("json", json_obj | json_arr | ...)
-    common_peg_parser rule(const std::string & name, const common_peg_parser & p, bool trigger = false);
-
-    // Creates a named rule using a builder function, and returns a ref.
-    // If trigger=true, marks this rule as an entry point for lazy grammar generation.
-    //   auto json = p.rule("json", [&]() { return json_object() | json_array() | ... })
-    common_peg_parser rule(const std::string & name, const std::function<common_peg_parser()> & builder, bool trigger = false);
-
-    // Creates a trigger rule. When generating a lazy grammar from the parser,
-    // only trigger rules and descendents are emitted.
-    common_peg_parser trigger_rule(const std::string & name, const common_peg_parser & p) { return rule(name, p, true); }
-    common_peg_parser trigger_rule(const std::string & name, const std::function<common_peg_parser()> & builder) { return rule(name, builder, true); }
-
-    // Creates an atomic parser. Atomic parsers do not create an AST node if
-    // the child results in a partial parse, i.e. NEEDS_MORE_INPUT. This is
-    // intended for situations where partial output is undesirable.
-    common_peg_parser atomic(const common_peg_parser & p) { return add(common_peg_atomic_parser{p}); }
-
-    // Tags create nodes in the generated AST for semantic purposes.
-    // Unlike rules, you can tag multiple nodes with the same tag.
-    common_peg_parser tag(const std::string & tag, const common_peg_parser & p) { return add(common_peg_tag_parser{p.id(), tag}); }
-
-    void set_root(const common_peg_parser & p);
-
-    common_peg_arena build();
-};
-
-// Helper function for building parsers
-common_peg_arena build_peg_parser(const std::function<common_peg_parser(common_peg_parser_builder & builder)> & fn);

common/preset.cpp

@@ -1,398 +0,0 @@
-#include "arg.h"
-#include "preset.h"
-#include "peg-parser.h"
-#include "log.h"
-#include "download.h"
-
-#include <fstream>
-#include <sstream>
-#include <filesystem>
-
-static std::string rm_leading_dashes(const std::string & str) {
-    size_t pos = 0;
-    while (pos < str.size() && str[pos] == '-') {
-        ++pos;
-    }
-    return str.substr(pos);
-}
-
-std::vector<std::string> common_preset::to_args(const std::string & bin_path) const {
-    std::vector<std::string> args;
-
-    if (!bin_path.empty()) {
-        args.push_back(bin_path);
-    }
-
-    for (const auto & [opt, value] : options) {
-        if (opt.is_preset_only) {
-            continue; // skip preset-only options (they are not CLI args)
-        }
-
-        // use the last arg as the main arg (i.e. --long-form)
-        args.push_back(opt.args.back());
-
-        // handle value(s)
-        if (opt.value_hint == nullptr && opt.value_hint_2 == nullptr) {
-            // flag option, no value
-            if (common_arg_utils::is_falsey(value)) {
-                // use negative arg if available
-                if (!opt.args_neg.empty()) {
-                    args.back() = opt.args_neg.back();
-                } else {
-                    // otherwise, skip the flag
-                    // TODO: maybe throw an error instead?
-                    args.pop_back();
-                }
-            }
-        }
-        if (opt.value_hint != nullptr) {
-            // single value
-            args.push_back(value);
-        }
-        if (opt.value_hint != nullptr && opt.value_hint_2 != nullptr) {
-            throw std::runtime_error(string_format(
-                "common_preset::to_args(): option '%s' has two values, which is not supported yet",
-                opt.args.back()
-            ));
-        }
-    }
-
-    return args;
-}
-
-std::string common_preset::to_ini() const {
-    std::ostringstream ss;
-
-    ss << "[" << name << "]\n";
-    for (const auto & [opt, value] : options) {
-        auto espaced_value = value;
-        string_replace_all(espaced_value, "\n", "\\\n");
-        ss << rm_leading_dashes(opt.args.back()) << " = ";
-        ss << espaced_value << "\n";
-    }
-    ss << "\n";
-
-    return ss.str();
-}
-
-void common_preset::set_option(const common_preset_context & ctx, const std::string & env, const std::string & value) {
-    // try if option exists, update it
-    for (auto & [opt, val] : options) {
-        if (opt.env && env == opt.env) {
-            val = value;
-            return;
-        }
-    }
-    // if option does not exist, we need to add it
-    if (ctx.key_to_opt.find(env) == ctx.key_to_opt.end()) {
-        throw std::runtime_error(string_format(
-            "%s: option with env '%s' not found in ctx_params",
-            __func__, env.c_str()
-        ));
-    }
-    options[ctx.key_to_opt.at(env)] = value;
-}
-
-void common_preset::unset_option(const std::string & env) {
-    for (auto it = options.begin(); it != options.end(); ) {
-        const common_arg & opt = it->first;
-        if (opt.env && env == opt.env) {
-            it = options.erase(it);
-            return;
-        } else {
-            ++it;
-        }
-    }
-}
-
-bool common_preset::get_option(const std::string & env, std::string & value) const {
-    for (const auto & [opt, val] : options) {
-        if (opt.env && env == opt.env) {
-            value = val;
-            return true;
-        }
-    }
-    return false;
-}
-
-void common_preset::merge(const common_preset & other) {
-    for (const auto & [opt, val] : other.options) {
-        options[opt] = val; // overwrite existing options
-    }
-}
-
-static std::map<std::string, std::map<std::string, std::string>> parse_ini_from_file(const std::string & path) {
-    std::map<std::string, std::map<std::string, std::string>> parsed;
-
-    if (!std::filesystem::exists(path)) {
-        throw std::runtime_error("preset file does not exist: " + path);
-    }
-
-    std::ifstream file(path);
-    if (!file.good()) {
-        throw std::runtime_error("failed to open server preset file: " + path);
-    }
-
-    std::string contents((std::istreambuf_iterator<char>(file)), std::istreambuf_iterator<char>());
-
-    static const auto parser = build_peg_parser([](auto & p) {
-        // newline ::= "\r\n" / "\n" / "\r"
-        auto newline = p.rule("newline", p.literal("\r\n") | p.literal("\n") | p.literal("\r"));
-
-        // ws ::= [ \t]*
-        auto ws = p.rule("ws", p.chars("[ \t]", 0, -1));
-
-        // comment ::= [;#] (!newline .)*
-        auto comment = p.rule("comment", p.chars("[;#]", 1, 1) + p.zero_or_more(p.negate(newline) + p.any()));
-
-        // eol ::= ws comment? (newline / EOF)
-        auto eol = p.rule("eol", ws + p.optional(comment) + (newline | p.end()));
-
-        // ident ::= [a-zA-Z_] [a-zA-Z0-9_.-]*
-        auto ident = p.rule("ident", p.chars("[a-zA-Z_]", 1, 1) + p.chars("[a-zA-Z0-9_.-]", 0, -1));
-
-        // value ::= (!eol-start .)*
-        auto eol_start = p.rule("eol-start", ws + (p.chars("[;#]", 1, 1) | newline | p.end()));
-        auto value = p.rule("value", p.zero_or_more(p.negate(eol_start) + p.any()));
-
-        // header-line ::= "[" ws ident ws "]" eol
-        auto header_line = p.rule("header-line", "[" + ws + p.tag("section-name", p.chars("[^]]")) + ws + "]" + eol);
-
-        // kv-line ::= ident ws "=" ws value eol
-        auto kv_line = p.rule("kv-line", p.tag("key", ident) + ws + "=" + ws + p.tag("value", value) + eol);
-
-        // comment-line ::= ws comment (newline / EOF)
-        auto comment_line = p.rule("comment-line", ws + comment + (newline | p.end()));
-
-        // blank-line ::= ws (newline / EOF)
-        auto blank_line = p.rule("blank-line", ws + (newline | p.end()));
-
-        // line ::= header-line / kv-line / comment-line / blank-line
-        auto line = p.rule("line", header_line | kv_line | comment_line | blank_line);
-
-        // ini ::= line* EOF
-        auto ini = p.rule("ini", p.zero_or_more(line) + p.end());
-
-        return ini;
-    });
-
-    common_peg_parse_context ctx(contents);
-    const auto result = parser.parse(ctx);
-    if (!result.success()) {
-        throw std::runtime_error("failed to parse server config file: " + path);
-    }
-
-    std::string current_section = COMMON_PRESET_DEFAULT_NAME;
-    std::string current_key;
-
-    ctx.ast.visit(result, [&](const auto & node) {
-        if (node.tag == "section-name") {
-            const std::string section = std::string(node.text);
-            current_section = section;
-            parsed[current_section] = {};
-        } else if (node.tag == "key") {
-            const std::string key = std::string(node.text);
-            current_key = key;
-        } else if (node.tag == "value" && !current_key.empty() && !current_section.empty()) {
-            parsed[current_section][current_key] = std::string(node.text);
-            current_key.clear();
-        }
-    });
-
-    return parsed;
-}
-
-static std::map<std::string, common_arg> get_map_key_opt(common_params_context & ctx_params) {
-    std::map<std::string, common_arg> mapping;
-    for (const auto & opt : ctx_params.options) {
-        for (const auto & env : opt.get_env()) {
-            mapping[env] = opt;
-        }
-        for (const auto & arg : opt.get_args()) {
-            mapping[rm_leading_dashes(arg)] = opt;
-        }
-    }
-    return mapping;
-}
-
-static bool is_bool_arg(const common_arg & arg) {
-    return !arg.args_neg.empty();
-}
-
-static std::string parse_bool_arg(const common_arg & arg, const std::string & key, const std::string & value) {
-    // if this is a negated arg, we need to reverse the value
-    for (const auto & neg_arg : arg.args_neg) {
-        if (rm_leading_dashes(neg_arg) == key) {
-            return common_arg_utils::is_truthy(value) ? "false" : "true";
-        }
-    }
-    // otherwise, not negated
-    return value;
-}
-
-common_preset_context::common_preset_context(llama_example ex)
-        : ctx_params(common_params_parser_init(default_params, ex)) {
-    common_params_add_preset_options(ctx_params.options);
-    key_to_opt = get_map_key_opt(ctx_params);
-}
-
-common_presets common_preset_context::load_from_ini(const std::string & path, common_preset & global) const {
-    common_presets out;
-    auto ini_data = parse_ini_from_file(path);
-
-    for (auto section : ini_data) {
-        common_preset preset;
-        if (section.first.empty()) {
-            preset.name = COMMON_PRESET_DEFAULT_NAME;
-        } else {
-            preset.name = section.first;
-        }
-        LOG_DBG("loading preset: %s\n", preset.name.c_str());
-        for (const auto & [key, value] : section.second) {
-            LOG_DBG("option: %s = %s\n", key.c_str(), value.c_str());
-            if (key_to_opt.find(key) != key_to_opt.end()) {
-                const auto & opt = key_to_opt.at(key);
-                if (is_bool_arg(opt)) {
-                    preset.options[opt] = parse_bool_arg(opt, key, value);
-                } else {
-                    preset.options[opt] = value;
-                }
-                LOG_DBG("accepted option: %s = %s\n", key.c_str(), preset.options[opt].c_str());
-            } else {
-                // TODO: maybe warn about unknown key?
-            }
-        }
-
-        if (preset.name == "*") {
-            // handle global preset
-            global = preset;
-        } else {
-            out[preset.name] = preset;
-        }
-    }
-
-    return out;
-}
-
-common_presets common_preset_context::load_from_cache() const {
-    common_presets out;
-
-    auto cached_models = common_list_cached_models();
-    for (const auto & model : cached_models) {
-        common_preset preset;
-        preset.name = model.to_string();
-        preset.set_option(*this, "LLAMA_ARG_HF_REPO", model.to_string());
-        out[preset.name] = preset;
-    }
-
-    return out;
-}
-
-struct local_model {
-    std::string name;
-    std::string path;
-    std::string path_mmproj;
-};
-
-common_presets common_preset_context::load_from_models_dir(const std::string & models_dir) const {
-    if (!std::filesystem::exists(models_dir) || !std::filesystem::is_directory(models_dir)) {
-        throw std::runtime_error(string_format("error: '%s' does not exist or is not a directory\n", models_dir.c_str()));
-    }
-
-    std::vector<local_model> models;
-    auto scan_subdir = [&models](const std::string & subdir_path, const std::string & name) {
-        auto files = fs_list(subdir_path, false);
-        common_file_info model_file;
-        common_file_info first_shard_file;
-        common_file_info mmproj_file;
-        for (const auto & file : files) {
-            if (string_ends_with(file.name, ".gguf")) {
-                if (file.name.find("mmproj") != std::string::npos) {
-                    mmproj_file = file;
-                } else if (file.name.find("-00001-of-") != std::string::npos) {
-                    first_shard_file = file;
-                } else {
-                    model_file = file;
-                }
-            }
-        }
-        // single file model
-        local_model model{
-            /* name        */ name,
-            /* path        */ first_shard_file.path.empty() ? model_file.path : first_shard_file.path,
-            /* path_mmproj */ mmproj_file.path // can be empty
-        };
-        if (!model.path.empty()) {
-            models.push_back(model);
-        }
-    };
-
-    auto files = fs_list(models_dir, true);
-    for (const auto & file : files) {
-        if (file.is_dir) {
-            scan_subdir(file.path, file.name);
-        } else if (string_ends_with(file.name, ".gguf")) {
-            // single file model
-            std::string name = file.name;
-            string_replace_all(name, ".gguf", "");
-            local_model model{
-                /* name        */ name,
-                /* path        */ file.path,
-                /* path_mmproj */ ""
-            };
-            models.push_back(model);
-        }
-    }
-
-    // convert local models to presets
-    common_presets out;
-    for (const auto & model : models) {
-        common_preset preset;
-        preset.name = model.name;
-        preset.set_option(*this, "LLAMA_ARG_MODEL", model.path);
-        if (!model.path_mmproj.empty()) {
-            preset.set_option(*this, "LLAMA_ARG_MMPROJ", model.path_mmproj);
-        }
-        out[preset.name] = preset;
-    }
-
-    return out;
-}
-
-common_preset common_preset_context::load_from_args(int argc, char ** argv) const {
-    common_preset preset;
-    preset.name = COMMON_PRESET_DEFAULT_NAME;
-
-    bool ok = common_params_to_map(argc, argv, ctx_params.ex, preset.options);
-    if (!ok) {
-        throw std::runtime_error("failed to parse CLI arguments into preset");
-    }
-
-    return preset;
-}
-
-common_presets common_preset_context::cascade(const common_presets & base, const common_presets & added) const {
-    common_presets out = base; // copy
-    for (const auto & [name, preset_added] : added) {
-        if (out.find(name) != out.end()) {
-            // if exists, merge
-            common_preset & target = out[name];
-            target.merge(preset_added);
-        } else {
-            // otherwise, add directly
-            out[name] = preset_added;
-        }
-    }
-    return out;
-}
-
-common_presets common_preset_context::cascade(const common_preset & base, const common_presets & presets) const {
-    common_presets out;
-    for (const auto & [name, preset] : presets) {
-        common_preset tmp = base; // copy
-        tmp.name = name;
-        tmp.merge(preset);
-        out[name] = std::move(tmp);
-    }
-    return out;
-}

common/preset.h

@@ -1,74 +0,0 @@
-#pragma once
-
-#include "common.h"
-#include "arg.h"
-
-#include <string>
-#include <vector>
-#include <map>
-
-//
-// INI preset parser and writer
-//
-
-constexpr const char * COMMON_PRESET_DEFAULT_NAME = "default";
-
-struct common_preset_context;
-
-struct common_preset {
-    std::string name;
-
-    // options are stored as common_arg to string mapping, representing CLI arg and its value
-    std::map<common_arg, std::string> options;
-
-    // convert preset to CLI argument list
-    std::vector<std::string> to_args(const std::string & bin_path = "") const;
-
-    // convert preset to INI format string
-    std::string to_ini() const;
-
-    // TODO: maybe implement to_env() if needed
-
-    // modify preset options where argument is identified by its env variable
-    void set_option(const common_preset_context & ctx, const std::string & env, const std::string & value);
-
-    // unset option by its env variable
-    void unset_option(const std::string & env);
-
-    // get option value by its env variable, return false if not found
-    bool get_option(const std::string & env, std::string & value) const;
-
-    // merge another preset into this one, overwriting existing options
-    void merge(const common_preset & other);
-};
-
-// interface for multiple presets in one file
-using common_presets = std::map<std::string, common_preset>;
-
-// context for loading and editing presets
-struct common_preset_context {
-    common_params default_params; // unused for now
-    common_params_context ctx_params;
-    std::map<std::string, common_arg> key_to_opt;
-    common_preset_context(llama_example ex);
-
-    // load presets from INI file
-    common_presets load_from_ini(const std::string & path, common_preset & global) const;
-
-    // generate presets from cached models
-    common_presets load_from_cache() const;
-
-    // generate presets from local models directory
-    // for the directory structure, see "Using multiple models" in server/README.md
-    common_presets load_from_models_dir(const std::string & models_dir) const;
-
-    // generate one preset from CLI arguments
-    common_preset load_from_args(int argc, char ** argv) const;
-
-    // cascade multiple presets if exist on both: base < added
-    // if preset does not exist in base, it will be added without modification
-    common_presets cascade(const common_presets & base, const common_presets & added) const;
-
-    // apply presets over a base preset (same idea as CSS cascading)
-    common_presets cascade(const common_preset & base, const common_presets & presets) const;
-};

common/sampling.cpp

@@ -3,10 +3,9 @@
 #include "common.h"
 #include "log.h"
 
-#include <algorithm>
 #include <cmath>
-#include <cstring>
 #include <unordered_map>
+#include <algorithm>
 
 // the ring buffer works similarly to std::deque, but with a fixed capacity
 // TODO: deduplicate with llama-impl.h
@@ -113,12 +112,6 @@ struct common_sampler {
 
     llama_token_data_array cur_p;
 
-    void reset() {
-        prev.clear();
-
-        llama_sampler_reset(chain);
-    }
-
     void set_logits(struct llama_context * ctx, int idx) {
         const auto * logits = llama_get_logits_ith(ctx, idx);
 
@@ -135,12 +128,6 @@ struct common_sampler {
 
         cur_p = { cur.data(), cur.size(), -1, false };
     }
-
-    common_time_meas tm() {
-        return common_time_meas(t_total_us, params.no_perf);
-    }
-
-    mutable int64_t t_total_us = 0;
 };
 
 std::string common_params_sampling::print() const {
@@ -166,11 +153,7 @@ struct common_sampler * common_sampler_init(const struct llama_model * model, co
 
     lparams.no_perf = params.no_perf;
 
-    llama_sampler * grmr = nullptr;
-    llama_sampler * chain = llama_sampler_chain_init(lparams);
-
-    std::vector<llama_sampler *> samplers;
-
+    struct llama_sampler * grmr;
     if (params.grammar.compare(0, 11, "%llguidance") == 0) {
 #ifdef LLAMA_USE_LLGUIDANCE
         grmr = llama_sampler_init_llg(vocab, "lark", params.grammar.c_str());
@@ -220,20 +203,30 @@ struct common_sampler * common_sampler_init(const struct llama_model * model, co
             trigger_patterns_c.push_back(regex.c_str());
         }
 
-        if (!params.grammar.empty()) {
-             if (params.grammar_lazy) {
-                 grmr = llama_sampler_init_grammar_lazy_patterns(vocab, params.grammar.c_str(), "root",
-                         trigger_patterns_c.data(), trigger_patterns_c.size(),
-                         trigger_tokens.data(), trigger_tokens.size());
-             } else {
-                 grmr = llama_sampler_init_grammar(vocab, params.grammar.c_str(), "root");
-             }
+        grmr = params.grammar_lazy
+             ? llama_sampler_init_grammar_lazy_patterns(vocab, params.grammar.c_str(), "root",
+                                                        trigger_patterns_c.data(), trigger_patterns_c.size(),
+                                                        trigger_tokens.data(), trigger_tokens.size())
+             :      llama_sampler_init_grammar(vocab, params.grammar.c_str(), "root");
+        if (!grmr) {
+            return nullptr;
         }
     }
 
-    if (params.has_logit_bias()) {
-        samplers.push_back(llama_sampler_init_logit_bias(llama_vocab_n_tokens(vocab), params.logit_bias.size(), params.logit_bias.data()));
-    }
+    auto * result = new common_sampler {
+        /* .params = */ params,
+        /* .grmr   = */ grmr,
+        /* .chain  = */ llama_sampler_chain_init(lparams),
+        /* .prev   = */ ring_buffer<llama_token>(std::max(32, params.n_prev)),
+        /* .cur    = */ {},
+        /* .cur_p  = */ {},
+    };
+
+    llama_sampler_chain_add(result->chain,
+            llama_sampler_init_logit_bias(
+                llama_vocab_n_tokens(vocab),
+                params.logit_bias.size(),
+                params.logit_bias.data()));
 
     if (params.mirostat == 0) {
         for (const auto & cnstr : params.samplers) {
@@ -246,71 +239,58 @@ struct common_sampler * common_sampler_init(const struct llama_model * model, co
                             c_breakers.push_back(str.c_str());
                         }
 
-                        samplers.push_back(llama_sampler_init_dry    (vocab, llama_model_n_ctx_train(model), params.dry_multiplier, params.dry_base, params.dry_allowed_length, params.dry_penalty_last_n, c_breakers.data(), c_breakers.size()));
+                        llama_sampler_chain_add(result->chain, llama_sampler_init_dry      (vocab, llama_model_n_ctx_train(model), params.dry_multiplier, params.dry_base, params.dry_allowed_length, params.dry_penalty_last_n, c_breakers.data(), c_breakers.size()));
                     }
                     break;
                 case COMMON_SAMPLER_TYPE_TOP_K:
-                    samplers.push_back(llama_sampler_init_top_k      (params.top_k));
+                    llama_sampler_chain_add(result->chain, llama_sampler_init_top_k       (params.top_k));
                     break;
                 case COMMON_SAMPLER_TYPE_TOP_P:
-                    samplers.push_back(llama_sampler_init_top_p      (params.top_p, params.min_keep));
+                    llama_sampler_chain_add(result->chain, llama_sampler_init_top_p       (params.top_p, params.min_keep));
                     break;
                 case COMMON_SAMPLER_TYPE_TOP_N_SIGMA:
-                    samplers.push_back(llama_sampler_init_top_n_sigma(params.top_n_sigma));
+                    llama_sampler_chain_add(result->chain, llama_sampler_init_top_n_sigma (params.top_n_sigma));
                     break;
                 case COMMON_SAMPLER_TYPE_MIN_P:
-                    samplers.push_back(llama_sampler_init_min_p      (params.min_p, params.min_keep));
+                    llama_sampler_chain_add(result->chain, llama_sampler_init_min_p       (params.min_p, params.min_keep));
                     break;
                 case COMMON_SAMPLER_TYPE_XTC:
-                    samplers.push_back(llama_sampler_init_xtc        (params.xtc_probability, params.xtc_threshold, params.min_keep, params.seed));
+                    llama_sampler_chain_add(result->chain, llama_sampler_init_xtc         (params.xtc_probability, params.xtc_threshold, params.min_keep, params.seed));
                     break;
                 case COMMON_SAMPLER_TYPE_TYPICAL_P:
-                    samplers.push_back(llama_sampler_init_typical    (params.typ_p, params.min_keep));
+                    llama_sampler_chain_add(result->chain, llama_sampler_init_typical     (params.typ_p, params.min_keep));
                     break;
                 case COMMON_SAMPLER_TYPE_TEMPERATURE:
-                    samplers.push_back(llama_sampler_init_temp_ext   (params.temp, params.dynatemp_range, params.dynatemp_exponent));
+                    llama_sampler_chain_add(result->chain, llama_sampler_init_temp_ext    (params.temp, params.dynatemp_range, params.dynatemp_exponent));
                     break;
                 case COMMON_SAMPLER_TYPE_INFILL:
-                    samplers.push_back(llama_sampler_init_infill     (vocab));
+                    llama_sampler_chain_add(result->chain, llama_sampler_init_infill      (vocab));
                     break;
                 case COMMON_SAMPLER_TYPE_PENALTIES:
-                    samplers.push_back(llama_sampler_init_penalties  (params.penalty_last_n, params.penalty_repeat, params.penalty_freq, params.penalty_present));
+                    llama_sampler_chain_add(result->chain, llama_sampler_init_penalties   (params.penalty_last_n, params.penalty_repeat, params.penalty_freq, params.penalty_present));
                     break;
                 default:
                     GGML_ASSERT(false && "unknown sampler type");
             }
         }
-
-        samplers.push_back(llama_sampler_init_dist(params.seed));
+        llama_sampler_chain_add(result->chain, llama_sampler_init_dist(params.seed));
     } else if (params.mirostat == 1) {
-        samplers.push_back(llama_sampler_init_temp(params.temp));
-        samplers.push_back(llama_sampler_init_mirostat(llama_vocab_n_tokens(vocab), params.seed, params.mirostat_tau, params.mirostat_eta, 100));
+        llama_sampler_chain_add(result->chain, llama_sampler_init_temp(params.temp));
+        llama_sampler_chain_add(result->chain, llama_sampler_init_mirostat(llama_vocab_n_tokens(vocab), params.seed, params.mirostat_tau, params.mirostat_eta, 100));
     } else if (params.mirostat == 2) {
-        samplers.push_back(llama_sampler_init_temp(params.temp));
-        samplers.push_back(llama_sampler_init_mirostat_v2(params.seed, params.mirostat_tau, params.mirostat_eta));
+        llama_sampler_chain_add(result->chain, llama_sampler_init_temp(params.temp));
+        llama_sampler_chain_add(result->chain, llama_sampler_init_mirostat_v2(params.seed, params.mirostat_tau, params.mirostat_eta));
     } else {
         GGML_ASSERT(false && "unknown mirostat version");
     }
 
-    for (auto * smpl : samplers) {
-        llama_sampler_chain_add(chain, smpl);
-    }
-
-    auto * result = new common_sampler {
-        /* .params  = */ params,
-        /* .grmr    = */ grmr,
-        /* .chain   = */ chain,
-        /* .prev    = */ ring_buffer<llama_token>(std::max(32, params.n_prev)),
-        /* .cur     = */ {},
-        /* .cur_p   = */ {},
-    };
-
     return result;
 }
 
 void common_sampler_free(struct common_sampler * gsmpl) {
     if (gsmpl) {
         llama_sampler_free(gsmpl->grmr);
+
         llama_sampler_free(gsmpl->chain);
 
         delete gsmpl;
@@ -318,9 +298,7 @@ void common_sampler_free(struct common_sampler * gsmpl) {
 }
 
 void common_sampler_accept(struct common_sampler * gsmpl, llama_token token, bool accept_grammar) {
-    const auto tm = gsmpl->tm();
-
-    if (gsmpl->grmr && accept_grammar) {
+    if (accept_grammar) {
         llama_sampler_accept(gsmpl->grmr, token);
     }
 
@@ -330,96 +308,55 @@ void common_sampler_accept(struct common_sampler * gsmpl, llama_token token, boo
 }
 
 void common_sampler_reset(struct common_sampler * gsmpl) {
-    gsmpl->reset();
+    llama_sampler_reset(gsmpl->grmr);
+
+    llama_sampler_reset(gsmpl->chain);
 }
 
 struct common_sampler * common_sampler_clone(common_sampler * gsmpl) {
     return new common_sampler {
-        /* .params  = */ gsmpl->params,
-        /* .grmr    = */ llama_sampler_clone(gsmpl->grmr),
-        /* .chain   = */ llama_sampler_clone(gsmpl->chain),
-        /* .prev    = */ gsmpl->prev,
-        /* .cur     = */ gsmpl->cur,
-        /* .cur_p   = */ gsmpl->cur_p,
+        /* .params = */ gsmpl->params,
+        /* .grmr   = */ llama_sampler_clone(gsmpl->grmr),
+        /* .chain  = */ llama_sampler_clone(gsmpl->chain),
+        /* .prev   = */ gsmpl->prev,
+        /* .cur    = */ gsmpl->cur,
+        /* .cur_p  = */ gsmpl->cur_p,
     };
 }
 
 void common_perf_print(const struct llama_context * ctx, const struct common_sampler * gsmpl) {
     // TODO: measure grammar performance
 
-    const double t_sampling_ms = gsmpl ? 1e-3*gsmpl->t_total_us : 0;
-
-    llama_perf_sampler_data data_smpl;
-    llama_perf_context_data data_ctx;
-
-    memset(&data_smpl, 0, sizeof(data_smpl));
-    memset(&data_ctx,  0, sizeof(data_ctx));
-
     if (gsmpl) {
-        auto & data = data_smpl;
-
-        data = llama_perf_sampler(gsmpl->chain);
-
-        // note: the sampling time includes the samplers time + extra time spent in common/sampling
-        LOG_INF("%s:    sampling time = %10.2f ms\n", __func__, t_sampling_ms);
-        LOG_INF("%s:    samplers time = %10.2f ms / %5d tokens\n", __func__, data.t_sample_ms, data.n_sample);
+        llama_perf_sampler_print(gsmpl->chain);
     }
-
     if (ctx) {
-        auto & data = data_ctx;
-
-        data = llama_perf_context(ctx);
-
-        const double t_end_ms = 1e-3 * ggml_time_us();
-
-        const double t_total_ms = t_end_ms - data.t_start_ms;
-        const double t_unacc_ms = t_total_ms - (t_sampling_ms + data.t_p_eval_ms + data.t_eval_ms);
-        const double t_unacc_pc = 100.0 * t_unacc_ms /  t_total_ms;
-
-        LOG_INF("%s:        load time = %10.2f ms\n", __func__, data.t_load_ms);
-        LOG_INF("%s: prompt eval time = %10.2f ms / %5d tokens (%8.2f ms per token, %8.2f tokens per second)\n",
-                __func__, data.t_p_eval_ms, data.n_p_eval, data.t_p_eval_ms / data.n_p_eval, 1e3 / data.t_p_eval_ms * data.n_p_eval);
-        LOG_INF("%s:        eval time = %10.2f ms / %5d runs   (%8.2f ms per token, %8.2f tokens per second)\n",
-                __func__, data.t_eval_ms, data.n_eval, data.t_eval_ms / data.n_eval, 1e3 / data.t_eval_ms * data.n_eval);
-        LOG_INF("%s:       total time = %10.2f ms / %5d tokens\n", __func__, (t_end_ms - data.t_start_ms), (data.n_p_eval + data.n_eval));
-        LOG_INF("%s: unaccounted time = %10.2f ms / %5.1f %%      (total - sampling - prompt eval - eval) / (total)\n", __func__, t_unacc_ms, t_unacc_pc);
-        LOG_INF("%s:    graphs reused = %10d\n", __func__, data.n_reused);
-
-        llama_memory_breakdown_print(ctx);
+        llama_perf_context_print(ctx);
     }
 }
 
-struct llama_sampler * common_sampler_get(const struct common_sampler * gsmpl) {
-    return gsmpl->chain;
-}
-
 llama_token common_sampler_sample(struct common_sampler * gsmpl, struct llama_context * ctx, int idx, bool grammar_first) {
-    llama_synchronize(ctx);
-
-    // start measuring sampling time after the llama_context synchronization in order to not measure any ongoing async operations
-    const auto tm = gsmpl->tm();
-
-    llama_token id = LLAMA_TOKEN_NULL;
+    gsmpl->set_logits(ctx, idx);
 
     auto & grmr  = gsmpl->grmr;
     auto & chain = gsmpl->chain;
     auto & cur_p = gsmpl->cur_p; // initialized by set_logits
 
-    gsmpl->set_logits(ctx, idx);
-
     if (grammar_first) {
         llama_sampler_apply(grmr, &cur_p);
     }
 
     llama_sampler_apply(chain, &cur_p);
 
-    id = cur_p.data[cur_p.selected].id;
+    GGML_ASSERT(cur_p.selected != -1 && "no selected token during sampling - check your sampling configuration");
+
+    const llama_token id = cur_p.data[cur_p.selected].id;
 
     if (grammar_first) {
         return id;
     }
 
-    // check if it the sampled token fits the grammar (grammar-based rejection sampling)
+    // check if it the sampled token fits the grammar
     {
         llama_token_data       single_token_data       = { id, 1.0f, 0.0f };
         llama_token_data_array single_token_data_array = { &single_token_data, 1, -1, false };
@@ -439,11 +376,9 @@ llama_token common_sampler_sample(struct common_sampler * gsmpl, struct llama_co
     llama_sampler_apply(grmr,  &cur_p);
     llama_sampler_apply(chain, &cur_p);
 
-    GGML_ASSERT(cur_p.selected != -1 && "no selected token during sampling - check your sampling configuration");
+    GGML_ASSERT(cur_p.selected != -1 && "no selected token during re-sampling - check your sampling configuration");
 
-    id = cur_p.data[cur_p.selected].id;
-
-    return id;
+    return cur_p.data[cur_p.selected].id;
 }
 
 std::vector<llama_token> common_sampler_sample_and_accept_n(struct common_sampler * gsmpl, struct llama_context * ctx, const std::vector<int> & idxs, const llama_tokens & draft, bool grammar_first) {
@@ -491,31 +426,8 @@ uint32_t common_sampler_get_seed(const struct common_sampler * gsmpl) {
 
 // helpers
 
-llama_token_data_array * common_sampler_get_candidates(struct common_sampler * gsmpl, bool do_sort) {
-    const auto tm = gsmpl->tm();
-
-    auto * res = &gsmpl->cur_p;
-
-    if (do_sort && !res->sorted) {
-        // remember the selected token before sorting
-        const llama_token id = res->data[res->selected].id;
-
-        std::sort(res->data, res->data + res->size, [](const llama_token_data & a, const llama_token_data & b) {
-            return a.p > b.p;
-        });
-
-        // restore the selected token after sorting
-        for (size_t i = 0; i < res->size; ++i) {
-            if (res->data[i].id == id) {
-                res->selected = i;
-                break;
-            }
-        }
-
-        res->sorted = true;
-    }
-
-    return res;
+llama_token_data_array * common_sampler_get_candidates(struct common_sampler * gsmpl) {
+    return &gsmpl->cur_p;
 }
 
 llama_token common_sampler_last(const struct common_sampler * gsmpl) {
@@ -527,8 +439,7 @@ std::string common_sampler_print(const struct common_sampler * gsmpl) {
 
     for (int i = 0; i < llama_sampler_chain_n(gsmpl->chain); i++) {
         const auto * smpl = llama_sampler_chain_get(gsmpl->chain, i);
-        result += std::string("-> ");
-        result += std::string(llama_sampler_name(smpl)) + " ";
+        result += std::string("-> ") + llama_sampler_name(smpl) + " ";
     }
 
     return result;

common/sampling.h

@@ -48,8 +48,6 @@ struct common_sampler * common_sampler_clone (struct common_sampler * gsmpl);
 // arguments can be nullptr to skip printing
 void common_perf_print(const struct llama_context * ctx, const struct common_sampler * gsmpl);
 
-struct llama_sampler * common_sampler_get(const struct common_sampler * gsmpl);
-
 // extended sampling implementation:
 //
 // - set logits
@@ -88,9 +86,7 @@ uint32_t common_sampler_get_seed(const struct common_sampler * gsmpl);
 // helpers
 
 // access the internal list of current candidate tokens
-// if do_sort == true, the candidates are guaranteed to be sorted afterwards (in descending order of probability)
-// the .sorted flag of the result indicates whether the returned candidates are sorted
-llama_token_data_array * common_sampler_get_candidates(struct common_sampler * gsmpl, bool do_sort);
+llama_token_data_array * common_sampler_get_candidates(struct common_sampler * gsmpl);
 
 // get the last accepted token
 llama_token common_sampler_last(const struct common_sampler * gsmpl);
@@ -109,9 +105,3 @@ std::vector<enum common_sampler_type> common_sampler_types_from_chars(const std:
 
 llama_sampler * llama_sampler_init_llg(const llama_vocab * vocab,
                 const char * grammar_kind, const char * grammar_data);
-
-struct common_sampler_deleter {
-    void operator()(common_sampler * s) { common_sampler_free(s); }
-};
-
-typedef std::unique_ptr<common_sampler, common_sampler_deleter> common_sampler_ptr;

common/speculative.cpp

@@ -1,39 +1,30 @@
 #include "speculative.h"
 
-#include "ggml.h"
-#include "llama.h"
 #include "log.h"
 #include "common.h"
 #include "sampling.h"
 
 #include <cstring>
 #include <algorithm>
-#include <map>
 
 #define SPEC_VOCAB_MAX_SIZE_DIFFERENCE  128
 #define SPEC_VOCAB_CHECK_START_TOKEN_ID 5
 
 struct common_speculative {
-    struct llama_context * ctx_tgt; // only used for retokenizing from ctx_dft
-    struct llama_context * ctx_dft;
+    struct llama_context * ctx;
     struct common_sampler * smpl;
 
     llama_batch batch;
-    llama_tokens prompt_dft;
-    bool vocab_dft_compatible = true; // whether retokenization is needed
-    std::map<std::string, std::string> tgt_dft_replacements = {};
+    llama_tokens prompt;
 };
 
 struct common_speculative * common_speculative_init(
-        struct llama_context * ctx_tgt,
         struct llama_context * ctx_dft) {
     auto * result = new common_speculative {
-        /* .ctx_tgt    = */ ctx_tgt,
-        /* .ctx_dft    = */ ctx_dft,
-        /* .smpl       = */ nullptr,
-        /* .batch      = */ llama_batch_init(llama_n_batch(ctx_dft), 0, 1),
-        /* .prompt_dft = */ {},
-        /* .vocab_dft_compatible = */ false,
+        /* .ctx    = */ ctx_dft,
+        /* .smpl   = */ nullptr,
+        /* .batch  = */ llama_batch_init(llama_n_batch(ctx_dft), 0, 1),
+        /* .prompt = */ {},
     };
 
     // TODO: optimize or pass from outside?
@@ -68,9 +59,6 @@ struct common_speculative * common_speculative_init(
     }
 #endif
 
-    result->vocab_dft_compatible = common_speculative_are_compatible(ctx_tgt, ctx_dft);
-    LOG_DBG("vocab_dft_compatible = %d\n", result->vocab_dft_compatible);
-
     return result;
 }
 
@@ -87,8 +75,8 @@ void common_speculative_free(struct common_speculative * spec) {
 }
 
 bool common_speculative_are_compatible(
-    const struct llama_context * ctx_tgt,
-    const struct llama_context * ctx_dft) {
+        const struct llama_context * ctx_tgt,
+        const struct llama_context * ctx_dft) {
     const struct llama_model * model_tgt = llama_get_model(ctx_tgt);
     const struct llama_model * model_dft = llama_get_model(ctx_dft);
 
@@ -102,32 +90,31 @@ bool common_speculative_are_compatible(
     LOG_DBG("%s: vocab_type dft: %d\n", __func__, vocab_type_dft);
 
     if (vocab_type_tgt != vocab_type_dft) {
-        LOG_DBG("%s: draft model vocab type must match target model to use speculation but ", __func__);
-        LOG_DBG("vocab_type_dft = %d while vocab_type_tgt = %d\n", vocab_type_dft, vocab_type_tgt);
+        LOG_ERR("%s: draft model vocab type must match target model to use speculation but "
+                     "vocab_type_dft = %d while vocab_type_tgt = %d\n", __func__, vocab_type_dft, vocab_type_tgt);
         return false;
     }
 
-    if (
-        llama_vocab_get_add_bos(vocab_tgt) != llama_vocab_get_add_bos(vocab_dft) ||
+    if (llama_vocab_get_add_bos(vocab_tgt) != llama_vocab_get_add_bos(vocab_dft) ||
         llama_vocab_get_add_eos(vocab_tgt) != llama_vocab_get_add_eos(vocab_dft) ||
         llama_vocab_bos(vocab_tgt) != llama_vocab_bos(vocab_dft) ||
-        llama_vocab_eos(vocab_tgt) != llama_vocab_eos(vocab_dft)
-    ) {
-        LOG_DBG("%s: draft model special tokens must match target model to use speculation\n", __func__);
+        llama_vocab_eos(vocab_tgt) != llama_vocab_eos(vocab_dft)) {
+        LOG_ERR("%s: draft vocab special tokens must match target vocab to use speculation\n", __func__);
+        LOG_ERR("%s: tgt: bos = %d (%d), eos = %d (%d)\n", __func__, llama_vocab_bos(vocab_tgt), llama_vocab_get_add_bos(vocab_tgt), llama_vocab_eos(vocab_tgt), llama_vocab_get_add_eos(vocab_tgt));
+        LOG_ERR("%s: dft: bos = %d (%d), eos = %d (%d)\n", __func__, llama_vocab_bos(vocab_dft), llama_vocab_get_add_bos(vocab_dft), llama_vocab_eos(vocab_dft), llama_vocab_get_add_eos(vocab_dft));
         return false;
     }
 
     {
         const int n_vocab_tgt = llama_vocab_n_tokens(vocab_tgt);
         const int n_vocab_dft = llama_vocab_n_tokens(vocab_dft);
-        const int vocab_diff  = n_vocab_tgt > n_vocab_dft
-            ? n_vocab_tgt - n_vocab_dft
-            : n_vocab_dft - n_vocab_tgt;
+
+        const int vocab_diff = std::abs(n_vocab_tgt - n_vocab_dft);
 
         if (vocab_diff > SPEC_VOCAB_MAX_SIZE_DIFFERENCE) {
-            LOG_DBG("%s: draft model vocab must closely match target model to use speculation but ", __func__);
-            LOG_DBG("target vocab size %d does not match draft vocab size %d - difference %d, max allowed %d\n",
-                    n_vocab_tgt, llama_vocab_n_tokens(vocab_dft), vocab_diff, SPEC_VOCAB_MAX_SIZE_DIFFERENCE);
+            LOG_ERR("%s: draft model vocab must closely match target model to use speculation but "
+                         "target vocab size %d does not match draft vocab size %d - difference %d, max allowed %d\n",
+                    __func__, n_vocab_tgt, llama_vocab_n_tokens(vocab_dft), vocab_diff, SPEC_VOCAB_MAX_SIZE_DIFFERENCE);
             return false;
         }
 
@@ -135,8 +122,8 @@ bool common_speculative_are_compatible(
             const char * token_text_tgt = llama_vocab_get_text(vocab_tgt, i);
             const char * token_text_dft = llama_vocab_get_text(vocab_dft, i);
             if (std::strcmp(token_text_tgt, token_text_dft) != 0) {
-                LOG_DBG("%s: draft model vocab must match target model to use speculation but ", __func__);
-                LOG_DBG("token %d content differs - target '%s', draft '%s'\n", i,
+                LOG_ERR("%s: draft vocab vocab must match target vocab to use speculation but "
+                             "token %d content differs - target '%s', draft '%s'\n", __func__, i,
                         common_token_to_piece(ctx_tgt, i).c_str(),
                         common_token_to_piece(ctx_dft, i).c_str());
                 return false;
@@ -147,93 +134,32 @@ bool common_speculative_are_compatible(
     return true;
 }
 
-void common_speculative_add_replacement_tgt_dft(
-        struct common_speculative * spec,
-        const char *source, const char *dest) {
-    spec->tgt_dft_replacements[source] = dest;
-}
-
-static std::string replace_to_dft(
-        struct common_speculative * spec,
-        const std::string& input) {
-    std::string result = input;
-    for (const auto & pair : spec->tgt_dft_replacements) {
-        size_t pos = result.find(pair.first);
-        while (pos != std::string::npos) {
-            result.replace(pos, pair.first.length(), pair.second);
-            pos = result.find(pair.first, pos + pair.second.length());
-        }
-    }
-    return result;
-}
-
-static std::string replace_to_tgt(
-        struct common_speculative * spec,
-        const std::string& input) {
-    std::string result = input;
-    for (const auto& pair : spec->tgt_dft_replacements) {
-        size_t pos = result.find(pair.second);
-        while (pos != std::string::npos) {
-            result.replace(pos, pair.second.length(), pair.first);
-            pos = result.find(pair.second, pos + pair.first.length());
-        }
-    }
-    return result;
-}
-
-
 llama_tokens common_speculative_gen_draft(
         struct common_speculative * spec,
         struct common_speculative_params params,
-        const llama_tokens & prompt_tgt_main_model, // specified in target model vocab
+        const llama_tokens & prompt_tgt,
         llama_token id_last) {
     auto & batch  = spec->batch;
-    auto & ctx_tgt = spec->ctx_tgt;
-    auto & ctx_dft = spec->ctx_dft;
+    auto & ctx    = spec->ctx;
     auto & smpl   = spec->smpl;
-    auto & prompt_dft = spec->prompt_dft;
+    auto & prompt = spec->prompt;
 
-    auto * mem_dft = llama_get_memory(ctx_dft);
+    auto * mem = llama_get_memory(ctx);
 
     int reuse_i = 0;
     int reuse_n = 0;
 
-    const int n_ctx = llama_n_ctx(ctx_dft) - params.n_draft;
-
-    llama_tokens prompt_tgt_draft_model;
-    if (!spec->vocab_dft_compatible) {
-        std::string text;
-        text = common_detokenize(ctx_tgt, prompt_tgt_main_model, true);
-        text = replace_to_dft(spec, text);
-        LOG_DBG("%s: main->draft detokenized string: '%s'\n", __func__, text.c_str());
-        prompt_tgt_draft_model = common_tokenize(ctx_dft, text, false, true);
-
-        // convert id_last to draft vocab. llama_detokenize is called directly to avoid an allocation
-        const auto * model_tgt = llama_get_model(ctx_tgt);
-        const auto * vocab_tgt = llama_model_get_vocab(model_tgt);
-
-        int32_t n_chars = llama_detokenize(vocab_tgt, &id_last, 1, nullptr, 0, false, false);
-        GGML_ASSERT(n_chars < 0 && "failed to detokenize id_last");
-        text.resize(-n_chars);
-        llama_detokenize(vocab_tgt, &id_last, 1, text.data(), text.size(), false, false);
-        text = replace_to_dft(spec, text);
-
-        LOG_DBG("main->draft detokenized id_last(%d): '%s'\n", id_last, text.c_str());
-        id_last = common_tokenize(ctx_dft, text, false, true)[0];
-    }
-    // prompt_tgt's tokens will always be compatible with ctx_dft
-    const llama_tokens &prompt_tgt =
-        spec->vocab_dft_compatible ? prompt_tgt_main_model : prompt_tgt_draft_model;
+    const int n_ctx = llama_n_ctx(ctx) - params.n_draft;
 
     const int i_start = std::max<int>(0, (int) prompt_tgt.size() - n_ctx);
 
     // reuse as much as possible from the old draft context
     // ideally, the draft context should be as big as the target context and we will always reuse the entire prompt
-    for (int i = 0; i < (int) prompt_dft.size(); ++i) {
+    for (int i = 0; i < (int) prompt.size(); ++i) {
         int cur = 0;
         while (i_start + cur < (int) prompt_tgt.size() &&
-               i       + cur < (int) prompt_dft.size() &&
-               prompt_tgt[i_start + cur] == prompt_dft[i + cur]) {
+               i       + cur < (int) prompt.size() &&
+               prompt_tgt[i_start + cur] == prompt[i + cur]) {
             cur++;
         }
 
@@ -243,20 +169,21 @@ llama_tokens common_speculative_gen_draft(
         }
     }
 
-    LOG_DBG("%s: reuse_i = %d, reuse_n = %d, prompt = %d\n", __func__, reuse_i, reuse_n, (int) prompt_dft.size());
+    LOG_DBG("%s: reuse_i = %d, reuse_n = %d, prompt = %d\n", __func__, reuse_i, reuse_n, (int) prompt.size());
 
     llama_tokens result;
     result.reserve(params.n_draft);
 
     if (reuse_n == 0) {
-        llama_memory_clear(mem_dft, false);
-        prompt_dft.clear();
+        llama_memory_clear(mem, false);
+
+        prompt.clear();
     } else {
         // this happens when a previous draft has been discarded (for example, due to being too small), but the
         // target model agreed with it. in this case, we simply pass back the previous results to save compute
-        if (reuse_i + reuse_n < (int) prompt_dft.size() && prompt_dft[reuse_i + reuse_n] == id_last) {
-            for (int i = reuse_i + reuse_n + 1; i < (int) prompt_dft.size(); ++i) {
-                result.push_back(prompt_dft[i]);
+        if (reuse_i + reuse_n < (int) prompt.size() && prompt[reuse_i + reuse_n] == id_last) {
+            for (int i = reuse_i + reuse_n + 1; i < (int) prompt.size(); ++i) {
+                result.push_back(prompt[i]);
 
                 if (params.n_draft <= (int) result.size()) {
                     break;
@@ -267,15 +194,16 @@ llama_tokens common_speculative_gen_draft(
         }
 
         if (reuse_i > 0) {
-            llama_memory_seq_rm (mem_dft, 0, 0, reuse_i);
-            llama_memory_seq_add(mem_dft, 0, reuse_i, -1, -reuse_i);
+            llama_memory_seq_rm (mem, 0, 0, reuse_i);
+            llama_memory_seq_add(mem, 0, reuse_i, -1, -reuse_i);
 
-            prompt_dft.erase(prompt_dft.begin(), prompt_dft.begin() + reuse_i);
+            prompt.erase(prompt.begin(), prompt.begin() + reuse_i);
         }
 
-        if (reuse_n < (int) prompt_dft.size()) {
-            llama_memory_seq_rm (mem_dft, 0, reuse_n, -1);
-            prompt_dft.erase(prompt_dft.begin() + reuse_n, prompt_dft.end());
+        if (reuse_n < (int) prompt.size()) {
+            llama_memory_seq_rm (mem, 0, reuse_n, -1);
+
+            prompt.erase(prompt.begin() + reuse_n, prompt.end());
         }
     }
 
@@ -286,28 +214,28 @@ llama_tokens common_speculative_gen_draft(
         //LOG_DBG("i = %d, i_start = %d, reuse_n = %d, i - i_start = %d, id = %6d\n", i, i_start, reuse_n, i - i_start, prompt_tgt[i]);
         common_batch_add(batch, prompt_tgt[i], i - i_start, { 0 }, false);
 
-        prompt_dft.push_back(prompt_tgt[i]);
+        prompt.push_back(prompt_tgt[i]);
     }
 
     // we should rarely end-up here during normal decoding
     if (batch.n_tokens > 0) {
         //LOG_DBG("%s: draft prompt batch: %s\n", __func__, string_from(ctx, batch).c_str());
 
-        llama_decode(ctx_dft, batch);
+        llama_decode(ctx, batch);
     }
 
-    const llama_pos n_past = prompt_dft.size();
+    const llama_pos n_past = prompt.size();
 
     LOG_DBG("%s: n_past = %d\n", __func__, n_past);
 
     common_batch_clear(batch);
     common_batch_add  (batch, id_last, n_past, { 0 }, true);
 
-    prompt_dft.push_back(id_last);
+    prompt.push_back(id_last);
 
-    LOG_DBG("%s: draft prompt: %s\n", __func__, string_from(ctx_dft, prompt_dft).c_str());
+    //LOG_DBG("%s: draft prompt: %s\n", __func__, string_from(ctx, prompt).c_str());
 
-    llama_decode(ctx_dft, batch);
+    llama_decode(ctx, batch);
 
     common_sampler_reset(smpl);
 
@@ -315,13 +243,13 @@ llama_tokens common_speculative_gen_draft(
     for (int i = 0; i < params.n_draft; ++i) {
         common_batch_clear(batch);
 
-        common_sampler_sample(smpl, ctx_dft, 0, true);
+        common_sampler_sample(smpl, ctx, 0, true);
 
-        const auto * cur_p = common_sampler_get_candidates(smpl, true);
+        const auto * cur_p = common_sampler_get_candidates(smpl);
 
         for (int k = 0; k < std::min(3, (int) cur_p->size); ++k) {
             LOG_DBG(" - draft candidate %3d, pos %3d: %6d (%8.3f) '%s'\n",
-                    k, i, cur_p->data[k].id, cur_p->data[k].p, common_token_to_piece(ctx_dft, cur_p->data[k].id).c_str());
+                    k, i, cur_p->data[k].id, cur_p->data[k].p, common_token_to_piece(ctx, cur_p->data[k].id).c_str());
         }
 
         // add drafted token for each sequence
@@ -343,19 +271,10 @@ llama_tokens common_speculative_gen_draft(
         common_batch_add(batch, id, n_past + i + 1, { 0 }, true);
 
         // evaluate the drafted tokens on the draft model
-        llama_decode(ctx_dft, batch);
+        llama_decode(ctx, batch);
 
-        prompt_dft.push_back(id);
+        prompt.push_back(id);
     }
 
-    if (!spec->vocab_dft_compatible) {
-        std::string detokenized = common_detokenize(ctx_dft, result, true);
-        detokenized = replace_to_tgt(spec, detokenized);
-        LOG_DBG("draft->main detokenized string: '%s'\n", detokenized.c_str());
-        result = common_tokenize(ctx_tgt, detokenized, false, true);
-        if (result.size() > (size_t)params.n_draft) {
-            result.resize(params.n_draft);
-        }
-    }
     return result;
 }

common/speculative.h

@@ -12,10 +12,7 @@ struct common_speculative_params {
     float p_min = 0.75f; // min probability required to accept a token in the draft
 };
 
-struct common_speculative * common_speculative_init(
-        struct llama_context * ctx_tgt,
-        struct llama_context * ctx_dft
-);
+struct common_speculative * common_speculative_init(struct llama_context * ctx_dft);
 
 void common_speculative_free(struct common_speculative * spec);
 
@@ -23,10 +20,6 @@ bool common_speculative_are_compatible(
         const struct llama_context * ctx_tgt,
         const struct llama_context * ctx_dft);
 
-void common_speculative_add_replacement_tgt_dft(
-        struct common_speculative * spec,
-        const char *source, const char *dest);
-
 // sample up to n_draft tokens and add them to the batch using the draft model
 llama_tokens common_speculative_gen_draft(
                struct common_speculative * spec,

common/unicode.cpp

@@ -1,64 +0,0 @@
-#include "unicode.h"
-
-// implementation adopted from src/unicode.cpp
-
-size_t utf8_sequence_length(unsigned char first_byte) {
-    const size_t lookup[] = { 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 3, 4 };
-    uint8_t highbits = static_cast<uint8_t>(first_byte) >> 4;
-    return lookup[highbits];
-}
-
-utf8_parse_result parse_utf8_codepoint(std::string_view input, size_t offset) {
-    if (offset >= input.size()) {
-        return utf8_parse_result(utf8_parse_result::INCOMPLETE);
-    }
-
-    // ASCII fast path
-    if (!(input[offset] & 0x80)) {
-        return utf8_parse_result(utf8_parse_result::SUCCESS, input[offset], 1);
-    }
-
-    // Invalid: continuation byte as first byte
-    if (!(input[offset] & 0x40)) {
-        return utf8_parse_result(utf8_parse_result::INVALID);
-    }
-
-    // 2-byte sequence
-    if (!(input[offset] & 0x20)) {
-        if (offset + 1 >= input.size()) {
-            return utf8_parse_result(utf8_parse_result::INCOMPLETE);
-        }
-        if ((input[offset + 1] & 0xc0) != 0x80) {
-            return utf8_parse_result(utf8_parse_result::INVALID);
-        }
-        auto result = ((input[offset] & 0x1f) << 6) | (input[offset + 1] & 0x3f);
-        return utf8_parse_result(utf8_parse_result::SUCCESS, result, 2);
-    }
-
-    // 3-byte sequence
-    if (!(input[offset] & 0x10)) {
-        if (offset + 2 >= input.size()) {
-            return utf8_parse_result(utf8_parse_result::INCOMPLETE);
-        }
-        if ((input[offset + 1] & 0xc0) != 0x80 || (input[offset + 2] & 0xc0) != 0x80) {
-            return utf8_parse_result(utf8_parse_result::INVALID);
-        }
-        auto result = ((input[offset] & 0x0f) << 12) | ((input[offset + 1] & 0x3f) << 6) | (input[offset + 2] & 0x3f);
-        return utf8_parse_result(utf8_parse_result::SUCCESS, result, 3);
-    }
-
-    // 4-byte sequence
-    if (!(input[offset] & 0x08)) {
-        if (offset + 3 >= input.size()) {
-            return utf8_parse_result(utf8_parse_result::INCOMPLETE);
-        }
-        if ((input[offset + 1] & 0xc0) != 0x80 || (input[offset + 2] & 0xc0) != 0x80 || (input[offset + 3] & 0xc0) != 0x80) {
-            return utf8_parse_result(utf8_parse_result::INVALID);
-        }
-        auto result = ((input[offset] & 0x07) << 18) | ((input[offset + 1] & 0x3f) << 12) | ((input[offset + 2] & 0x3f) << 6) | (input[offset + 3] & 0x3f);
-        return utf8_parse_result(utf8_parse_result::SUCCESS, result, 4);
-    }
-
-    // Invalid first byte
-    return utf8_parse_result(utf8_parse_result::INVALID);
-}

common/unicode.h

@@ -1,22 +0,0 @@
-#pragma once
-
-#include <cstdint>
-#include <string_view>
-
-// UTF-8 parsing utilities for streaming-aware unicode support
-
-struct utf8_parse_result {
-    uint32_t codepoint;      // Decoded codepoint (only valid if status == SUCCESS)
-    size_t bytes_consumed;   // How many bytes this codepoint uses (1-4)
-    enum status { SUCCESS, INCOMPLETE, INVALID } status;
-
-    utf8_parse_result(enum status s, uint32_t cp = 0, size_t bytes = 0)
-        : codepoint(cp), bytes_consumed(bytes), status(s) {}
-};
-
-// Determine the expected length of a UTF-8 sequence from its first byte
-// Returns 0 for invalid first bytes
-size_t utf8_sequence_length(unsigned char first_byte);
-
-// Parse a single UTF-8 codepoint from input
-utf8_parse_result parse_utf8_codepoint(std::string_view input, size_t offset);

convert_hf_to_gguf_update.py

@@ -59,10 +59,6 @@ parser.add_argument(
     "--full", action="store_true",
     help="download full list of models - make sure you have access to all of them",
 )
-parser.add_argument(
-    "--check-missing", action="store_true",
-    help="only check for missing pre-tokenizer hashes",
-)
 parser.add_argument(
     "hf_token",
     help="optional HF token",
@@ -74,10 +70,6 @@ hf_token = args.hf_token if args.hf_token is not None else hf_token
 if hf_token is None:
     logger.warning("HF token not found. You can provide it as an argument or set it in ~/.cache/huggingface/token")
 
-if args.check_missing and args.full:
-    logger.warning("Downloading full list of models requested, ignoring --check-missing!")
-    args.check_missing = False
-
 # TODO: this string has to exercise as much pre-tokenizer functionality as possible
 #       will be updated with time - contributions welcome
 CHK_TXT = '\n \n\n \n\n\n \t \t\t \t\n  \n   \n    \n     \n🚀 (normal) 😶‍🌫️ (multiple emojis concatenated) ✅ 🦙🦙 3 33 333 3333 33333 333333 3333333 33333333 3.3 3..3 3...3 កាន់តែពិសេសអាច😁 ?我想在apple工作1314151天～ ------======= нещо на Български \'\'\'\'\'\'```````\"\"\"\"......!!!!!!?????? I\'ve been \'told he\'s there, \'RE you sure? \'M not sure I\'ll make it, \'D you like some tea? We\'Ve a\'lL'
@@ -138,13 +130,6 @@ models = [
     {"name": "midm-2.0",         "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/K-intelligence/Midm-2.0-Base-Instruct", },
     {"name": "lfm2",             "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/LiquidAI/LFM2-Tokenizer"},
     {"name": "exaone4",          "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/LGAI-EXAONE/EXAONE-4.0-32B", },
-    {"name": "mellum",           "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/JetBrains/Mellum-4b-base", },
-    {"name": "modern-bert",      "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/answerdotai/ModernBERT-base", },
-    {"name": "afmoe",            "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/arcee-ai/Trinity-Tokenizer", },
-    {"name": "bailingmoe2",      "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/inclusionAI/Ling-mini-base-2.0", },
-    {"name": "granite-docling",  "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/ibm-granite/granite-docling-258M", },
-    {"name": "minimax-m2",       "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/MiniMaxAI/MiniMax-M2", },
-    {"name": "kormo",            "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/KORMo-Team/KORMo-tokenizer", },
 ]
 
 # some models are known to be broken upstream, so we will skip them as exceptions
@@ -153,18 +138,14 @@ pre_computed_hashes = [
     {"name": "chatglm-bpe", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/THUDM/glm-4-9b-chat", "chkhsh": "b6e8e1518dc4305be2fe39c313ed643381c4da5db34a98f6a04c093f8afbe99b"},
     {"name": "chatglm-bpe", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/THUDM/glm-4-9b-chat", "chkhsh": "81d72c7348a9f0ebe86f23298d37debe0a5e71149e29bd283904c02262b27516"},
     {"name": "glm4", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/THUDM/glm-4-9b-hf", "chkhsh": "a1336059768a55c99a734006ffb02203cd450fed003e9a71886c88acf24fdbc2"},
-    {"name": "glm4", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/zai-org/GLM-4.5-Air", "chkhsh": "9ca2dd618e8afaf09731a7cf6e2105b373ba6a1821559f258b272fe83e6eb902"},
     {"name": "minerva-7b", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/sapienzanlp/Minerva-7B-base-v1.0", "chkhsh": "1431a23e583c97432bc230bff598d103ddb5a1f89960c8f1d1051aaa944d0b35"},
     {"name": "hunyuan", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/tencent/Hunyuan-A13B-Instruct", "chkhsh": "7e57df22b1fe23a7b1e1c7f3dc4e3f96d43a4eb0836d0c6bdc3436d7b2f1c664"},
-    {"name": "hunyuan-dense", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/tencent/Hunyuan-4B-Instruct", "chkhsh": "bba3b3366b646dbdded5dbc42d59598b849371afc42f7beafa914afaa5b70aa6"},
     # falcon-h1 series uses 4 different tokenizers across model sizes (0.5b - 34b), hence we need to define 4 different hashes
     {"name": "falcon-h1", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/tiiuae/Falcon-H1-0.5B-Base", "chkhsh": "a6b57017d60e6edb4d88ecc2845188e0eb333a70357e45dcc9b53964a73bbae6"},
     {"name": "falcon-h1", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/tiiuae/Falcon-H1-1B-Base", "chkhsh": "60476e1243776c4fb1b993dbd7a5f15ac22f83c80afdf425fa5ae01c8d44ef86"},
     {"name": "falcon-h1", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/tiiuae/Falcon-H1-7B-Base", "chkhsh": "3eda48b4c4dc7de733d1a8b3e3b4a85243dbbf704da2ee9d42c6beced8897896"},
     {"name": "falcon-h1", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/tiiuae/Falcon-H1-34B-Base", "chkhsh": "48f8e02c0359c0bbdd82f26909171fac1c18a457bb47573ed1fe3bbb2c1cfd4b"},
     {"name": "kimi-k2",   "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/moonshotai/Kimi-K2-Base",   "chkhsh": "81212dc7cdb7e0c1074ca62c5aeab0d43c9f52b8a737be7b12a777c953027890"},
-    {"name": "qwen2",     "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/Qwen/Qwen3-Embedding-0.6B", "chkhsh": "d4540891389ea895b53b399da6ac824becc30f2fba0e9ddbb98f92e55ca0e97c"},
-    {"name": "grok-2",    "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/alvarobartt/grok-2-tokenizer", "chkhsh": "66b8d4e19ab16c3bfd89bce5d785fb7e0155e8648708a1f42077cb9fe002c273"},
 ]
 
 
@@ -239,13 +220,12 @@ if not args.full:
     all_models = models.copy()
     models = [model for model in all_models if model["name"] not in existing_models]
 
-if not args.check_missing:
-    logging.info(f"Downloading {len(models)} models...")
-    for model in models:
-        try:
-            download_model(model)
-        except Exception as e:
-            logger.error(f"Failed to download model {model['name']}. Error: {e}")
+logging.info(f"Downloading {len(models)} models...")
+for model in models:
+    try:
+        download_model(model)
+    except Exception as e:
+        logger.error(f"Failed to download model {model['name']}. Error: {e}")
 
 
 # generate the source code for the convert_hf_to_gguf.py:get_vocab_base_pre() function:
@@ -439,7 +419,7 @@ for model in models:
             tokenizer = AutoTokenizer.from_pretrained(f"models/tokenizers/{name}", use_fast=False)
         else:
             tokenizer = AutoTokenizer.from_pretrained(f"models/tokenizers/{name}")
-    except (OSError, TypeError) as e:
+    except OSError as e:
         logger.error(f"Failed to load tokenizer for model {name}. Error: {e}")
         continue  # Skip this model and continue with the next one in the loop
 

convert_lora_to_gguf.py

@@ -12,7 +12,7 @@ import json
 from math import prod
 from pathlib import Path
 from typing import TYPE_CHECKING, Any, Callable, Iterable, Iterator, Sequence, SupportsIndex, cast
-from transformers import AutoConfig, AutoTokenizer
+from transformers import AutoConfig
 
 import torch
 
@@ -26,8 +26,6 @@ import gguf
 # reuse model definitions from convert_hf_to_gguf.py
 from convert_hf_to_gguf import LazyTorchTensor, ModelBase
 
-from gguf.constants import GGUFValueType
-
 logger = logging.getLogger("lora-to-gguf")
 
 
@@ -242,7 +240,7 @@ def parse_args() -> argparse.Namespace:
         help="path to write to; default: based on input. {ftype} will be replaced by the outtype.",
     )
     parser.add_argument(
-        "--outtype", type=str, choices=["f32", "f16", "bf16", "q8_0", "auto"], default="f32",
+        "--outtype", type=str, choices=["f32", "f16", "bf16", "q8_0", "auto"], default="f16",
         help="output format - use f32 for float32, f16 for float16, bf16 for bfloat16, q8_0 for Q8_0, auto for the highest-fidelity 16-bit float type depending on the first loaded tensor type",
     )
     parser.add_argument(
@@ -277,15 +275,10 @@ def parse_args() -> argparse.Namespace:
     return parser.parse_args()
 
 
-def load_hparams_from_hf(hf_model_id: str) -> tuple[dict[str, Any], Path | None]:
-    from huggingface_hub import try_to_load_from_cache
-
+def load_hparams_from_hf(hf_model_id: str) -> dict[str, Any]:
     # normally, adapter does not come with base model config, we need to load it from AutoConfig
     config = AutoConfig.from_pretrained(hf_model_id)
-    cache_dir = try_to_load_from_cache(hf_model_id, "config.json")
-    cache_dir = Path(cache_dir).parent if isinstance(cache_dir, str) else None
-
-    return config.to_dict(), cache_dir
+    return config.to_dict()
 
 
 if __name__ == '__main__':
@@ -330,13 +323,13 @@ if __name__ == '__main__':
     # load base model
     if base_model_id is not None:
         logger.info(f"Loading base model from Hugging Face: {base_model_id}")
-        hparams, dir_base_model = load_hparams_from_hf(base_model_id)
+        hparams = load_hparams_from_hf(base_model_id)
     elif dir_base_model is None:
         if "base_model_name_or_path" in lparams:
             model_id = lparams["base_model_name_or_path"]
             logger.info(f"Loading base model from Hugging Face: {model_id}")
             try:
-                hparams, dir_base_model = load_hparams_from_hf(model_id)
+                hparams = load_hparams_from_hf(model_id)
             except OSError as e:
                 logger.error(f"Failed to load base model config: {e}")
                 logger.error("Please try downloading the base model and add its path to --base")
@@ -347,7 +340,7 @@ if __name__ == '__main__':
             sys.exit(1)
     else:
         logger.info(f"Loading base model: {dir_base_model.name}")
-        hparams = ModelBase.load_hparams(dir_base_model, False)
+        hparams = ModelBase.load_hparams(dir_base_model)
 
     with torch.inference_mode():
         try:
@@ -376,31 +369,7 @@ if __name__ == '__main__':
                 self.gguf_writer.add_string(gguf.Keys.Adapter.TYPE, "lora")
 
             def set_gguf_parameters(self):
-                logger.debug("GGUF KV: %s = %d", gguf.Keys.Adapter.LORA_ALPHA, self.lora_alpha)
                 self.gguf_writer.add_float32(gguf.Keys.Adapter.LORA_ALPHA, self.lora_alpha)
-                alora_invocation_tokens = lparams.get("alora_invocation_tokens")
-                invocation_string = lparams.get("invocation_string")
-                if invocation_string and not alora_invocation_tokens:
-                    logger.debug("Tokenizing invocation_string -> alora_invocation_tokens")
-                    base_model_path_or_id = hparams.get("_name_or_path")
-                    try:
-                        tokenizer = AutoTokenizer.from_pretrained(base_model_path_or_id)
-                    except ValueError:
-                        logger.error("Unable to load tokenizer from %s", base_model_path_or_id)
-                        raise
-                    # NOTE: There's an off-by-one with the older aLoRAs where
-                    # the invocation string includes the "<|start_of_turn|>"
-                    # token, but the adapters themselves were trained to
-                    # activate _after_ that first token, so we drop it here.
-                    alora_invocation_tokens = tokenizer(invocation_string)["input_ids"][1:]
-                if alora_invocation_tokens:
-                    logger.debug("GGUF KV: %s = %s", gguf.Keys.Adapter.ALORA_INVOCATION_TOKENS, alora_invocation_tokens)
-                    self.gguf_writer.add_key_value(
-                        gguf.Keys.Adapter.ALORA_INVOCATION_TOKENS,
-                        alora_invocation_tokens,
-                        GGUFValueType.ARRAY,
-                        GGUFValueType.UINT32,
-                    )
 
             def generate_extra_tensors(self) -> Iterable[tuple[str, Tensor]]:
                 # Never add extra tensors (e.g. rope_freqs) for LoRA adapters
@@ -485,7 +454,6 @@ if __name__ == '__main__':
             dir_lora_model=dir_lora,
             lora_alpha=alpha,
             hparams=hparams,
-            remote_hf_model_id=base_model_id,
         )
 
         logger.info("Exporting model...")

docs/android.md

@@ -1,27 +1,7 @@
 
 # Android
 
-## Build GUI binding using Android Studio
-
-Import the `examples/llama.android` directory into Android Studio, then perform a Gradle sync and build the project.
-![Project imported into Android Studio](./android/imported-into-android-studio.jpg)
-
-This Android binding supports hardware acceleration up to `SME2` for **Arm** and `AMX` for **x86-64** CPUs on Android and ChromeOS devices.
-It automatically detects the host's hardware to load compatible kernels. As a result, it runs seamlessly on both the latest premium devices and older devices that may lack modern CPU features or have limited RAM, without requiring any manual configuration.
-
-A minimal Android app frontend is included to showcase the binding’s core functionalities:
-1.	**Parse GGUF metadata** via `GgufMetadataReader` from either a `ContentResolver` provided `Uri` from shared storage, or a local `File` from your app's private storage.
-2.	**Obtain a `InferenceEngine`** instance through the `AiChat` facade and load your selected model via its app-private file path.
-3.	**Send a raw user prompt** for automatic template formatting, prefill, and batch decoding. Then collect the generated tokens in a Kotlin `Flow`.
-
-For a production-ready experience that leverages advanced features such as system prompts and benchmarks, plus friendly UI features such as model management and Arm feature visualizer, check out [Arm AI Chat](https://play.google.com/store/apps/details?id=com.arm.aichat) on Google Play.
-This project is made possible through a collaborative effort by Arm's **CT-ML**, **CE-ML** and **STE** groups:
-
-| ![Home screen](https://naco-siren.github.io/ai-chat/policy/index/1-llm-starter-pack.png)  | ![System prompt](https://naco-siren.github.io/ai-chat/policy/index/5-system-prompt.png)  | !["Haiku"](https://naco-siren.github.io/ai-chat/policy/index/4-metrics.png)  |
-|:------------------------------------------------------:|:----------------------------------------------------:|:--------------------------------------------------------:|
-|                      Home screen                       |                    System prompt                     |                         "Haiku"                          |
-
-## Build CLI on Android using Termux
+## Build on Android using Termux
 
 [Termux](https://termux.dev/en/) is an Android terminal emulator and Linux environment app (no root required). As of writing, Termux is available experimentally in the Google Play Store; otherwise, it may be obtained directly from the project repo or on F-Droid.
 
@@ -52,7 +32,7 @@ To see what it might look like visually, here's an old demo of an interactive se
 
 https://user-images.githubusercontent.com/271616/225014776-1d567049-ad71-4ef2-b050-55b0b3b9274c.mp4
 
-## Cross-compile CLI using Android NDK
+## Cross-compile using Android NDK
 It's possible to build `llama.cpp` for Android on your host system via CMake and the Android NDK. If you are interested in this path, ensure you already have an environment prepared to cross-compile programs for Android (i.e., install the Android SDK). Note that, unlike desktop environments, the Android environment ships with a limited set of native libraries, and so only those libraries are available to CMake when building with the Android NDK (see: https://developer.android.com/ndk/guides/stable_apis.)
 
 Once you're ready and have cloned `llama.cpp`, invoke the following in the project directory:

docs/backend/CANN.md

@@ -293,37 +293,22 @@ We would like to thank Tuo Dai, Shanni Li, and all of the project maintainers fr
 
 ## Environment variable setup
 
+### GGML_CANN_ASYNC_MODE
+
+Enables asynchronous operator submission. Disabled by default.
+
 ### GGML_CANN_MEM_POOL
 
-Specifies the memory pool management strategy, Default is vmm.
+Specifies the memory pool management strategy:
 
 - vmm: Utilizes a virtual memory manager pool. If hardware support for VMM is unavailable, falls back to the legacy (leg) memory pool.
 
 - prio: Employs a priority queue-based memory pool management.
-
 - leg: Uses a fixed-size buffer pool.
 
 ### GGML_CANN_DISABLE_BUF_POOL_CLEAN
 
 Controls automatic cleanup of the memory pool. This option is only effective when using the prio or leg memory pool strategies.
 
-### GGML_CANN_WEIGHT_NZ
-
-Converting the matmul weight format from ND to NZ to improve performance. Enabled by default.
-
-### GGML_CANN_ACL_GRAPH
-
-Operators are executed using ACL graph execution, rather than in op-by-op (eager) mode. Enabled by default. This option is only effective if `USE_ACL_GRAPH` was enabled at compilation time. To enable it, recompile using:
-
-```sh
-cmake -B build -DGGML_CANN=on -DCMAKE_BUILD_TYPE=release -DUSE_ACL_GRAPH=ON
-cmake --build build --config release
-```
-
-### GGML_CANN_GRAPH_CACHE_CAPACITY
-
-Maximum number of compiled CANN graphs kept in the LRU cache, default is 12. When the number of cached graphs exceeds this capacity, the least recently used graph will be evicted.
-
-### GGML_CANN_PREFILL_USE_GRAPH
-
-Enable ACL graph execution during the prefill stage, default is false. This option is only effective when FA is enabled.
+## TODO
+- Support more models and data types.

docs/backend/OPENCL.md

@@ -39,23 +39,18 @@ The llama.cpp OpenCL backend is designed to enable llama.cpp on **Qualcomm Adren
 | Adreno 830 (Snapdragon 8 Elite)      | Support |
 | Adreno X85 (Snapdragon X Elite)      | Support |
 
-> A6x GPUs with a recent driver and compiler are supported; they are usually found in IoT platforms.
-However, A6x GPUs in phones are likely not supported due to the outdated driver and compiler.
-
 ## DataType Supports
 
 | DataType               | Status                     |
 |:----------------------:|:--------------------------:|
 | Q4_0                   | Support                    |
 | Q6_K                   | Support, but not optimized |
-| Q8_0                   | Support                    |
-| MXFP4                  | Support                    |
 
 ## Model Preparation
 
-You can refer to the general [llama-quantize tool](/tools/quantize/README.md) for steps to convert a model in Hugging Face safetensor format to GGUF with quantization.
+You can refer to the general [*Prepare and Quantize*](README.md#prepare-and-quantize) guide for model prepration.
 
-Currently we support `Q4_0` quantization and have optimized for it. To achieve best performance on Adreno GPU, add `--pure` to `llama-quantize` (i.e., make all weights in `Q4_0`). For example,
+Currently we support `Q4_0` quantization and have optimize for it. To achieve best performance on Adreno GPU, add `--pure` to `llama-quantize`. For example,
 
 ```sh
 ./llama-quantize --pure ggml-model-qwen2.5-3b-f16.gguf ggml-model-qwen-3b-Q4_0.gguf Q4_0
@@ -63,17 +58,6 @@ Currently we support `Q4_0` quantization and have optimized for it. To achieve b
 
 Since `Q6_K` is also supported, `Q4_0` quantization without `--pure` will also work. However, the performance will be worse compared to pure `Q4_0` quantization.
 
-### `MXFP4` MoE Models
-
-OpenAI gpt-oss models are MoE models in `MXFP4`. The quantized model will be in `MXFP4_MOE`, a mixture of `MXFP4` and `Q8_0`.
-For this quantization, there is no need to specify `--pure`.
-For gpt-oss-20b model, you can directly [download](https://huggingface.co/ggml-org/gpt-oss-20b-GGUF) the quantized GGUF file in `MXFP4_MOE` from Hugging Face.
-
-Although it is possible to quantize gpt-oss-20b model in pure `Q4_0` (all weights in `Q4_0`), it is not recommended since `MXFP4` has been optimized for MoE while `Q4_0` is not. In addition, accuracy should degrade with such pure `Q4_0` quantization.
-Hence, using the default `MXFP4_MOE` quantization (see the link above) is recommended for this model.
-
-> Note that the `Q4_0` model found [here](https://huggingface.co/unsloth/gpt-oss-20b-GGUF/blob/main/gpt-oss-20b-Q4_0.gguf) is a mixture of `Q4_0`, `Q8_0` and `MXFP4` and gives better performance than `MXFP4_MOE` quantization.
-
 ## CMake Options
 
 The OpenCL backend has the following CMake options that control the behavior of the backend.
@@ -162,13 +146,10 @@ A Snapdragon X Elite device with Windows 11 Arm64 is used. Make sure the followi
 * Ninja
 * Visual Studio 2022
 * Powershell 7
-* Python
 
 Visual Studio provides necessary headers and libraries although it is not directly used for building.
 Alternatively, Visual Studio Build Tools can be installed instead of the full Visual Studio.
 
-> Note that building using Visual Studio's cl compiler is not supported. Clang must be used. Clang depends on libraries provided by Visual Studio to work. Therefore, Visual Studio must be installed. Alternatively, Visual Studio Build Tools can be installed instead of the full Visual Studio.
-
 Powershell 7 is used for the following commands.
 If an older version of Powershell is used, these commands may not work as they are.
 
@@ -220,12 +201,9 @@ ninja
 
 ## Known Issues
 
-- Flash attention does not always improve performance.
-- Currently OpenCL backend works on A6xx GPUs with recent drivers and compilers (usually found in IoT platforms).
-  However, it does not work on A6xx GPUs found in phones with old drivers and compilers.
+- Currently OpenCL backend does not work on Adreno 6xx GPUs.
 
 ## TODO
 
 - Optimization for Q6_K
 - Support and optimization for Q4_K
-- Improve flash attention

docs/backend/SYCL.md

@@ -42,9 +42,6 @@ The following releases are verified and recommended:
 
 ## News
 
-- 2025.11
-  - Support malloc memory on device more than 4GB.
-
 - 2025.2
   - Optimize MUL_MAT Q4_0 on Intel GPU for all dGPUs and built-in GPUs since MTL. Increase the performance of LLM (llama-2-7b.Q4_0.gguf) 21%-87% on Intel GPUs (MTL, ARL-H, Arc, Flex, PVC).
     |GPU|Base tokens/s|Increased tokens/s|Percent|
@@ -103,8 +100,6 @@ SYCL backend supports Intel GPU Family:
 - Intel Built-in Arc GPU
 - Intel iGPU in Core CPU (11th Generation Core CPU and newer, refer to [oneAPI supported GPU](https://www.intel.com/content/www/us/en/developer/articles/system-requirements/intel-oneapi-base-toolkit-system-requirements.html#inpage-nav-1-1)).
 
-On older Intel GPUs, you may try [OpenCL](/docs/backend/OPENCL.md) although the performance is not optimal, and some GPUs may not support OpenCL nor have any GPGPU capabilities.
-
 #### Verified devices
 
 | Intel GPU                     | Status  | Verified Model                        |
@@ -150,13 +145,12 @@ The docker build option is currently limited to *Intel GPU* targets.
 ```sh
 # Using FP16
 docker build -t llama-cpp-sycl --build-arg="GGML_SYCL_F16=ON" --target light -f .devops/intel.Dockerfile .
-
-# Using FP32
-docker build -t llama-cpp-sycl --build-arg="GGML_SYCL_F16=OFF" --target light -f .devops/intel.Dockerfile .
 ```
 
 *Notes*:
 
+To build in default FP32 *(Slower than FP16 alternative)*, set `--build-arg="GGML_SYCL_F16=OFF"` in the previous command.
+
 You can also use the `.devops/llama-server-intel.Dockerfile`, which builds the *"server"* alternative.
 Check the [documentation for Docker](../docker.md) to see the available images.
 
@@ -166,7 +160,7 @@ Check the [documentation for Docker](../docker.md) to see the available images.
 # First, find all the DRI cards
 ls -la /dev/dri
 # Then, pick the card that you want to use (here for e.g. /dev/dri/card1).
-docker run -it --rm -v "/path/to/models:/models" --device /dev/dri/renderD128:/dev/dri/renderD128 --device /dev/dri/card0:/dev/dri/card0 llama-cpp-sycl -m /models/7B/ggml-model-q4_0.gguf -p "Building a website can be done in 10 simple steps:" -n 400 -e -ngl 33 -c 4096 -s 0
+docker run -it --rm -v "$(pwd):/app:Z" --device /dev/dri/renderD128:/dev/dri/renderD128 --device /dev/dri/card1:/dev/dri/card1 llama-cpp-sycl -m "/app/models/YOUR_MODEL_FILE" -p "Building a website can be done in 10 simple steps:" -n 400 -e -ngl 33
 ```
 
 *Notes:*
@@ -221,19 +215,9 @@ To target AMD GPUs with SYCL, the ROCm stack must be installed first.
 
 2. **Install Intel® oneAPI Base toolkit**
 
-SYCL backend depends on:
-  - Intel® oneAPI DPC++/C++ compiler/running-time.
-  - Intel® oneAPI DPC++/C++ library (oneDPL).
-  - Intel® oneAPI Deep Neural Network Library (oneDNN).
-  - Intel® oneAPI Math Kernel Library (oneMKL).
-
 - **For Intel GPU**
 
-All above are included in both **Intel® oneAPI Base toolkit** and **Intel® Deep Learning Essentials** packages.
-
-It's recommended to install **Intel® Deep Learning Essentials** which only provides the necessary libraries with less size.
-
-The **Intel® oneAPI Base toolkit** and **Intel® Deep Learning Essentials** can be obtained from the official [Intel® oneAPI Base Toolkit](https://www.intel.com/content/www/us/en/developer/tools/oneapi/base-toolkit.html) page.
+The base toolkit can be obtained from the official [Intel® oneAPI Base Toolkit](https://www.intel.com/content/www/us/en/developer/tools/oneapi/base-toolkit.html) page.
 
 Please follow the instructions for downloading and installing the Toolkit for Linux, and preferably keep the default installation values unchanged, notably the installation path *(`/opt/intel/oneapi` by default)*.
 
@@ -241,12 +225,6 @@ Following guidelines/code snippets assume the default installation values. Other
 
 Upon a successful installation, SYCL is enabled for the available intel devices, along with relevant libraries such as oneAPI oneDNN for Intel GPUs.
 
-|Verified release|
-|-|
-|2025.2.1|
-|2025.1|
-|2024.1|
-
 - **Adding support to Nvidia GPUs**
 
 **oneAPI Plugin**: In order to enable SYCL support on Nvidia GPUs, please install the [Codeplay oneAPI Plugin for Nvidia GPUs](https://developer.codeplay.com/products/oneapi/nvidia/download). User should also make sure the plugin version matches the installed base toolkit one *(previous step)* for a seamless "oneAPI on Nvidia GPU" setup.
@@ -277,11 +255,10 @@ sycl-ls
 When targeting an intel GPU, the user should expect one or more devices among the available SYCL devices. Please make sure that at least one GPU is present via `sycl-ls`, for instance `[level_zero:gpu]` in the sample output below:
 
 ```
-[level_zero:gpu][level_zero:0] Intel(R) oneAPI Unified Runtime over Level-Zero, Intel(R) Arc(TM) A770 Graphics 12.55.8 [1.3.29735+27]
-[level_zero:gpu][level_zero:1] Intel(R) oneAPI Unified Runtime over Level-Zero, Intel(R) UHD Graphics 730 12.2.0 [1.3.29735+27]
-[opencl:cpu][opencl:0] Intel(R) OpenCL, 13th Gen Intel(R) Core(TM) i5-13400 OpenCL 3.0 (Build 0) [2025.20.8.0.06_160000]
-[opencl:gpu][opencl:1] Intel(R) OpenCL Graphics, Intel(R) Arc(TM) A770 Graphics OpenCL 3.0 NEO  [24.39.31294]
-[opencl:gpu][opencl:2] Intel(R) OpenCL Graphics, Intel(R) UHD Graphics 730 OpenCL 3.0 NEO  [24.39.31294]
+[opencl:acc][opencl:0] Intel(R) FPGA Emulation Platform for OpenCL(TM), Intel(R) FPGA Emulation Device OpenCL 1.2  [2023.16.10.0.17_160000]
+[opencl:cpu][opencl:1] Intel(R) OpenCL, 13th Gen Intel(R) Core(TM) i7-13700K OpenCL 3.0 (Build 0) [2023.16.10.0.17_160000]
+[opencl:gpu][opencl:2] Intel(R) OpenCL Graphics, Intel(R) Arc(TM) A770 Graphics OpenCL 3.0 NEO  [23.30.26918.50]
+[level_zero:gpu][level_zero:0] Intel(R) Level-Zero, Intel(R) Arc(TM) A770 Graphics 1.3 [1.3.26918]
 ```
 
 - **Nvidia GPU**
@@ -376,7 +353,7 @@ cmake --build build --config Release -j -v
 
 #### Retrieve and prepare model
 
-You can refer to the general [*Prepare and Quantize*](README.md#prepare-and-quantize) guide for model preparation, or download an already quantized model like [llama-2-7b.Q4_0.gguf](https://huggingface.co/TheBloke/Llama-2-7B-GGUF/resolve/main/llama-2-7b.Q4_0.gguf?download=true) or [Meta-Llama-3-8B-Instruct-Q4_0.gguf](https://huggingface.co/aptha/Meta-Llama-3-8B-Instruct-Q4_0-GGUF/resolve/main/Meta-Llama-3-8B-Instruct-Q4_0.gguf).
+You can refer to the general [*Prepare and Quantize*](README.md#prepare-and-quantize) guide for model preparation, or download an already quantized model like [llama-2-7b.Q4_0.gguf](https://huggingface.co/TheBloke/Llama-2-7B-GGUF/blob/main/llama-2-7b.Q4_0.gguf) or [Meta-Llama-3-8B-Instruct-Q4_0.gguf](https://huggingface.co/aptha/Meta-Llama-3-8B-Instruct-Q4_0-GGUF/resolve/main/Meta-Llama-3-8B-Instruct-Q4_0.gguf).
 
 ##### Check device
 
@@ -489,17 +466,7 @@ If you already have a recent version of Microsoft Visual Studio, you can skip th
 
 3. Install Intel® oneAPI Base toolkit
 
-SYCL backend depends on:
-  - Intel® oneAPI DPC++/C++ compiler/running-time.
-  - Intel® oneAPI DPC++/C++ library (oneDPL).
-  - Intel® oneAPI Deep Neural Network Library (oneDNN).
-  - Intel® oneAPI Math Kernel Library (oneMKL).
-
-All above are included in both **Intel® oneAPI Base toolkit** and **Intel® Deep Learning Essentials** packages.
-
-It's recommended to install **Intel® Deep Learning Essentials** which only provides the necessary libraries with less size.
-
-The **Intel® oneAPI Base toolkit** and **Intel® Deep Learning Essentials** can be obtained from the official [Intel® oneAPI Base Toolkit](https://www.intel.com/content/www/us/en/developer/tools/oneapi/base-toolkit.html) page.
+The base toolkit can be obtained from the official [Intel® oneAPI Base Toolkit](https://www.intel.com/content/www/us/en/developer/tools/oneapi/base-toolkit.html) page.
 
 Please follow the instructions for downloading and installing the Toolkit for Windows, and preferably keep the default installation values unchanged, notably the installation path *(`C:\Program Files (x86)\Intel\oneAPI` by default)*.
 
@@ -794,8 +761,6 @@ use 1 SYCL GPUs: [0] with Max compute units:512
 | GGML_SYCL_DISABLE_GRAPH | 0 or 1 (default) | Disable running computations through SYCL Graphs feature. Disabled by default because graph performance isn't yet better than non-graph performance. |
 | GGML_SYCL_DISABLE_DNN | 0 (default) or 1 | Disable running computations through oneDNN and always use oneMKL. |
 | ZES_ENABLE_SYSMAN | 0 (default) or 1 | Support to get free memory of GPU by sycl::aspect::ext_intel_free_memory.<br>Recommended to use when --split-mode = layer |
-| UR_L0_ENABLE_RELAXED_ALLOCATION_LIMITS | 0 (default) or 1 | Support malloc device memory more than 4GB.|
-
 
 
 ## Known Issues
@@ -842,14 +807,6 @@ use 1 SYCL GPUs: [0] with Max compute units:512
   | The default context is too big. It leads to excessive memory usage.|Set `-c 8192` or a smaller value.|
   | The model is too big and requires more memory than what is available.|Choose a smaller model or change to a smaller quantization, like Q5 -> Q4;<br>Alternatively, use more than one device to load model.|
 
-- `ggml_backend_sycl_buffer_type_alloc_buffer: can't allocate 5000000000 Bytes of memory on device`
-
-  You need to enable to support 4GB memory malloc by:
-  ```
-    export UR_L0_ENABLE_RELAXED_ALLOCATION_LIMITS=1
-    set UR_L0_ENABLE_RELAXED_ALLOCATION_LIMITS=1
-  ```
-
 ### **GitHub contribution**:
 Please add the `SYCL :` prefix/tag in issues/PRs titles to help the SYCL contributors to check/address them without delay.