test zero max buffer size

2026-04-23 16:37:33 +03:00 · 2024-06-26 17:11:09 +02:00
427 changed files with 3800 additions and 5510 deletions
--- a/.devops/full-cuda.Dockerfile
+++ b/.devops/full-cuda.Dockerfile
@@ -27,7 +27,7 @@ COPY . .
 # Set nvcc architecture
 ENV CUDA_DOCKER_ARCH=${CUDA_DOCKER_ARCH}
 # Enable CUDA
-ENV GGML_CUDA=1
+ENV LLAMA_CUDA=1
 # Enable cURL
 ENV LLAMA_CURL=1

--- a/.devops/full-rocm.Dockerfile
+++ b/.devops/full-rocm.Dockerfile
@@ -36,7 +36,7 @@ COPY . .
 # Set nvcc architecture
 ENV GPU_TARGETS=${ROCM_DOCKER_ARCH}
 # Enable ROCm
-ENV GGML_HIPBLAS=1
+ENV LLAMA_HIPBLAS=1
 ENV CC=/opt/rocm/llvm/bin/clang
 ENV CXX=/opt/rocm/llvm/bin/clang++

--- a/.devops/llama-cli-cuda.Dockerfile
+++ b/.devops/llama-cli-cuda.Dockerfile
@@ -21,7 +21,7 @@ COPY . .
 # Set nvcc architecture
 ENV CUDA_DOCKER_ARCH=${CUDA_DOCKER_ARCH}
 # Enable CUDA
-ENV GGML_CUDA=1
+ENV LLAMA_CUDA=1

 RUN make -j$(nproc) llama-cli

--- a/.devops/llama-cli-intel.Dockerfile
+++ b/.devops/llama-cli-intel.Dockerfile
@@ -2,7 +2,7 @@ ARG ONEAPI_VERSION=2024.1.1-devel-ubuntu22.04

 FROM intel/oneapi-basekit:$ONEAPI_VERSION as build

-ARG GGML_SYCL_F16=OFF
+ARG LLAMA_SYCL_F16=OFF
 RUN apt-get update && \
    apt-get install -y git

@@ -10,11 +10,11 @@ WORKDIR /app

 COPY . .

-RUN if [ "${GGML_SYCL_F16}" = "ON" ]; then \
-        echo "GGML_SYCL_F16 is set" && \
-        export OPT_SYCL_F16="-DGGML_SYCL_F16=ON"; \
+RUN if [ "${LLAMA_SYCL_F16}" = "ON" ]; then \
+        echo "LLAMA_SYCL_F16 is set" && \
+        export OPT_SYCL_F16="-DLLAMA_SYCL_F16=ON"; \
    fi && \
-    cmake -B build -DGGML_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx ${OPT_SYCL_F16} && \
+    cmake -B build -DLLAMA_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx ${OPT_SYCL_F16} && \
    cmake --build build --config Release --target llama-cli

 FROM intel/oneapi-basekit:$ONEAPI_VERSION as runtime
--- a/.devops/llama-cli-rocm.Dockerfile
+++ b/.devops/llama-cli-rocm.Dockerfile
@@ -36,7 +36,7 @@ COPY . .
 # Set nvcc architecture
 ENV GPU_TARGETS=${ROCM_DOCKER_ARCH}
 # Enable ROCm
-ENV GGML_HIPBLAS=1
+ENV LLAMA_HIPBLAS=1
 ENV CC=/opt/rocm/llvm/bin/clang
 ENV CXX=/opt/rocm/llvm/bin/clang++

--- a/.devops/llama-cli-vulkan.Dockerfile
+++ b/.devops/llama-cli-vulkan.Dockerfile
@@ -14,7 +14,7 @@ RUN wget -qO - https://packages.lunarg.com/lunarg-signing-key-pub.asc | apt-key
 # Build it
 WORKDIR /app
 COPY . .
-RUN cmake -B build -DGGML_VULKAN=1 && \
+RUN cmake -B build -DLLAMA_VULKAN=1 && \
    cmake --build build --config Release --target llama-cli

 # Clean up
--- a/.devops/llama-cpp-clblast.srpm.spec
+++ b/.devops/llama-cpp-clblast.srpm.spec
@@ -0,0 +1,84 @@
+# SRPM for building from source and packaging an RPM for RPM-based distros.
+# https://docs.fedoraproject.org/en-US/quick-docs/creating-rpm-packages
+# Built and maintained by John Boero - boeroboy@gmail.com
+# In honor of Seth Vidal https://www.redhat.com/it/blog/thank-you-seth-vidal
+
+# Notes for llama.cpp:
+# 1. Tags are currently based on hash - which will not sort asciibetically.
+#    We need to declare standard versioning if people want to sort latest releases.
+# 2. Builds for CUDA/OpenCL support are separate, with different depenedencies.
+# 3. NVidia's developer repo must be enabled with nvcc, cublas, clblas, etc installed.
+#    Example: https://developer.download.nvidia.com/compute/cuda/repos/fedora37/x86_64/cuda-fedora37.repo
+# 4. OpenCL/CLBLAST support simply requires the ICD loader and basic opencl libraries.
+#    It is up to the user to install the correct vendor-specific support.
+
+Name:           llama.cpp-clblast
+Version:        %( date "+%%Y%%m%%d" )
+Release:        1%{?dist}
+Summary:        OpenCL Inference of LLaMA model in C/C++
+License:        MIT
+Source0:        https://github.com/ggerganov/llama.cpp/archive/refs/heads/master.tar.gz
+BuildRequires:  coreutils make gcc-c++ git mesa-libOpenCL-devel clblast-devel
+Requires:       clblast
+URL:            https://github.com/ggerganov/llama.cpp
+
+%define debug_package %{nil}
+%define source_date_epoch_from_changelog 0
+
+%description
+CPU inference for Meta's Lllama2 models using default options.
+
+%prep
+%setup -n llama.cpp-master
+
+%build
+make -j LLAMA_CLBLAST=1
+
+%install
+mkdir -p %{buildroot}%{_bindir}/
+cp -p llama-cli %{buildroot}%{_bindir}/llama-clblast-cli
+cp -p llama-server %{buildroot}%{_bindir}/llama-clblast-server
+cp -p llama-simple %{buildroot}%{_bindir}/llama-clblast-simple
+
+mkdir -p %{buildroot}/usr/lib/systemd/system
+%{__cat} <<EOF  > %{buildroot}/usr/lib/systemd/system/llamaclblast.service
+[Unit]
+Description=Llama.cpp server, CPU only (no GPU support in this build).
+After=syslog.target network.target local-fs.target remote-fs.target nss-lookup.target
+
+[Service]
+Type=simple
+EnvironmentFile=/etc/sysconfig/llama
+ExecStart=/usr/bin/llama-clblast-server $LLAMA_ARGS
+ExecReload=/bin/kill -s HUP $MAINPID
+Restart=never
+
+[Install]
+WantedBy=default.target
+EOF
+
+mkdir -p %{buildroot}/etc/sysconfig
+%{__cat} <<EOF  > %{buildroot}/etc/sysconfig/llama
+LLAMA_ARGS="-m /opt/llama2/ggml-model-f32.bin"
+EOF
+
+%clean
+rm -rf %{buildroot}
+rm -rf %{_builddir}/*
+
+%files
+%{_bindir}/llama-clblast-cli
+%{_bindir}/llama-clblast-server
+%{_bindir}/llama-clblast-simple
+/usr/lib/systemd/system/llamaclblast.service
+%config /etc/sysconfig/llama
+
+
+%pre
+
+%post
+
+%preun
+%postun
+
+%changelog
--- a/.devops/llama-cpp-cuda.srpm.spec
+++ b/.devops/llama-cpp-cuda.srpm.spec
@@ -32,7 +32,7 @@ CPU inference for Meta's Lllama2 models using default options.
 %setup -n llama.cpp-master

 %build
-make -j GGML_CUDA=1
+make -j LLAMA_CUDA=1

 %install
 mkdir -p %{buildroot}%{_bindir}/
--- a/.devops/llama-server-cuda.Dockerfile
+++ b/.devops/llama-server-cuda.Dockerfile
@@ -21,7 +21,7 @@ COPY . .
 # Set nvcc architecture
 ENV CUDA_DOCKER_ARCH=${CUDA_DOCKER_ARCH}
 # Enable CUDA
-ENV GGML_CUDA=1
+ENV LLAMA_CUDA=1
 # Enable cURL
 ENV LLAMA_CURL=1

--- a/.devops/llama-server-intel.Dockerfile
+++ b/.devops/llama-server-intel.Dockerfile
@@ -2,7 +2,7 @@ ARG ONEAPI_VERSION=2024.1.1-devel-ubuntu22.04

 FROM intel/oneapi-basekit:$ONEAPI_VERSION as build

-ARG GGML_SYCL_F16=OFF
+ARG LLAMA_SYCL_F16=OFF
 RUN apt-get update && \
    apt-get install -y git libcurl4-openssl-dev

@@ -10,11 +10,11 @@ WORKDIR /app

 COPY . .

-RUN if [ "${GGML_SYCL_F16}" = "ON" ]; then \
-        echo "GGML_SYCL_F16 is set" && \
-        export OPT_SYCL_F16="-DGGML_SYCL_F16=ON"; \
+RUN if [ "${LLAMA_SYCL_F16}" = "ON" ]; then \
+        echo "LLAMA_SYCL_F16 is set" && \
+        export OPT_SYCL_F16="-DLLAMA_SYCL_F16=ON"; \
    fi && \
-    cmake -B build -DGGML_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx -DLLAMA_CURL=ON ${OPT_SYCL_F16} && \
+    cmake -B build -DLLAMA_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx -DLLAMA_CURL=ON ${OPT_SYCL_F16} && \
    cmake --build build --config Release --target llama-server

 FROM intel/oneapi-basekit:$ONEAPI_VERSION as runtime
--- a/.devops/llama-server-rocm.Dockerfile
+++ b/.devops/llama-server-rocm.Dockerfile
@@ -36,7 +36,7 @@ COPY . .
 # Set nvcc architecture
 ENV GPU_TARGETS=${ROCM_DOCKER_ARCH}
 # Enable ROCm
-ENV GGML_HIPBLAS=1
+ENV LLAMA_HIPBLAS=1
 ENV CC=/opt/rocm/llvm/bin/clang
 ENV CXX=/opt/rocm/llvm/bin/clang++

--- a/.devops/llama-server-vulkan.Dockerfile
+++ b/.devops/llama-server-vulkan.Dockerfile
@@ -14,7 +14,7 @@ RUN wget -qO - https://packages.lunarg.com/lunarg-signing-key-pub.asc | apt-key
 # Build it
 WORKDIR /app
 COPY . .
-RUN cmake -B build -DGGML_VULKAN=1 -DLLAMA_CURL=1 && \
+RUN cmake -B build -DLLAMA_VULKAN=1 -DLLAMA_CURL=1 && \
    cmake --build build --config Release --target llama-server

 # Clean up
--- a/.devops/nix/package.nix
+++ b/.devops/nix/package.nix
@@ -17,18 +17,19 @@
  rocmPackages,
  vulkan-headers,
  vulkan-loader,
-  curl,
+  clblast,
  useBlas ? builtins.all (x: !x) [
    useCuda
    useMetalKit
+    useOpenCL
    useRocm
    useVulkan
  ] && blas.meta.available,
  useCuda ? config.cudaSupport,
-  useMetalKit ? stdenv.isAarch64 && stdenv.isDarwin,
+  useMetalKit ? stdenv.isAarch64 && stdenv.isDarwin && !useOpenCL,
  useMpi ? false, # Increases the runtime closure size by ~700M
+  useOpenCL ? false,
  useRocm ? config.rocmSupport,
-  enableCurl ? true,
  useVulkan ? false,
  llamaVersion ? "0.0.0", # Arbitrary version, substituted by the flake

@@ -55,6 +56,7 @@ let
    ++ lib.optionals useCuda [ "CUDA" ]
    ++ lib.optionals useMetalKit [ "MetalKit" ]
    ++ lib.optionals useMpi [ "MPI" ]
+    ++ lib.optionals useOpenCL [ "OpenCL" ]
    ++ lib.optionals useRocm [ "ROCm" ]
    ++ lib.optionals useVulkan [ "Vulkan" ];

@@ -158,9 +160,9 @@ effectiveStdenv.mkDerivation (
    };

    postPatch = ''
-      substituteInPlace ./ggml/src/ggml-metal.m \
+      substituteInPlace ./ggml-metal.m \
        --replace '[bundle pathForResource:@"ggml-metal" ofType:@"metal"];' "@\"$out/bin/ggml-metal.metal\";"
-      substituteInPlace ./ggml/src/ggml-metal.m \
+      substituteInPlace ./ggml-metal.m \
        --replace '[bundle pathForResource:@"default" ofType:@"metallib"];' "@\"$out/bin/default.metallib\";"
    '';

@@ -196,24 +198,24 @@ effectiveStdenv.mkDerivation (
      optionals effectiveStdenv.isDarwin darwinBuildInputs
      ++ optionals useCuda cudaBuildInputs
      ++ optionals useMpi [ mpi ]
+      ++ optionals useOpenCL [ clblast ]
      ++ optionals useRocm rocmBuildInputs
      ++ optionals useBlas [ blas ]
-      ++ optionals useVulkan vulkanBuildInputs
-      ++ optionals enableCurl [ curl ];
+      ++ optionals useVulkan vulkanBuildInputs;

    cmakeFlags =
      [
+        (cmakeBool "LLAMA_NATIVE" false)
        (cmakeBool "LLAMA_BUILD_SERVER" true)
        (cmakeBool "BUILD_SHARED_LIBS" (!enableStatic))
        (cmakeBool "CMAKE_SKIP_BUILD_RPATH" true)
-        (cmakeBool "LLAMA_CURL" enableCurl)
-        (cmakeBool "GGML_NATIVE" false)
-        (cmakeBool "GGML_BLAS" useBlas)
-        (cmakeBool "GGML_CUDA" useCuda)
-        (cmakeBool "GGML_HIPBLAS" useRocm)
-        (cmakeBool "GGML_METAL" useMetalKit)
-        (cmakeBool "GGML_VULKAN" useVulkan)
-        (cmakeBool "GGML_STATIC" enableStatic)
+        (cmakeBool "LLAMA_BLAS" useBlas)
+        (cmakeBool "LLAMA_CLBLAST" useOpenCL)
+        (cmakeBool "LLAMA_CUDA" useCuda)
+        (cmakeBool "LLAMA_HIPBLAS" useRocm)
+        (cmakeBool "LLAMA_METAL" useMetalKit)
+        (cmakeBool "LLAMA_VULKAN" useVulkan)
+        (cmakeBool "LLAMA_STATIC" enableStatic)
      ]
      ++ optionals useCuda [
        (
@@ -229,7 +231,7 @@ effectiveStdenv.mkDerivation (
      ]
      ++ optionals useMetalKit [
        (lib.cmakeFeature "CMAKE_C_FLAGS" "-D__ARM_FEATURE_DOTPROD=1")
-        (cmakeBool "GGML_METAL_EMBED_LIBRARY" (!precompileMetalShaders))
+        (cmakeBool "LLAMA_METAL_EMBED_LIBRARY" (!precompileMetalShaders))
      ];

    # Environment variables needed for ROCm
@@ -242,7 +244,7 @@ effectiveStdenv.mkDerivation (
    # if they haven't been added yet.
    postInstall = ''
      mkdir -p $out/include
-      cp $src/include/llama.h $out/include/
+      cp $src/llama.h $out/include/
    '';

    # Define the shells here, but don't add in the inputsFrom to avoid recursion.
@@ -252,6 +254,7 @@ effectiveStdenv.mkDerivation (
        useCuda
        useMetalKit
        useMpi
+        useOpenCL
        useRocm
        useVulkan
        ;
@@ -278,7 +281,7 @@ effectiveStdenv.mkDerivation (
      # Configurations we don't want even the CI to evaluate. Results in the
      # "unsupported platform" messages. This is mostly a no-op, because
      # cudaPackages would've refused to evaluate anyway.
-      badPlatforms = optionals useCuda lib.platforms.darwin;
+      badPlatforms = optionals (useCuda || useOpenCL) lib.platforms.darwin;

      # Configurations that are known to result in build failures. Can be
      # overridden by importing Nixpkgs with `allowBroken = true`.
--- a/.github/ISSUE_TEMPLATE/config.yml
+++ b/.github/ISSUE_TEMPLATE/config.yml
@@ -9,3 +9,5 @@ contact_links:
  - name: Want to contribute?
    url: https://github.com/ggerganov/llama.cpp/wiki/contribute
    about: Head to the contribution guide page of the wiki for areas you can help with
+
+
--- a/.github/labeler.yml
+++ b/.github/labeler.yml
@@ -2,31 +2,31 @@
 Kompute:
    - changed-files:
        - any-glob-to-any-file:
-            - ggml/include/ggml-kompute.h
-            - ggml/src/ggml-kompute.cpp
+            - ggml-kompute.h
+            - ggml-kompute.cpp
            - README-kompute.md
 Apple Metal:
    - changed-files:
        - any-glob-to-any-file:
-            - ggml/include/ggml-metal.h
-            - ggml/src/ggml-metal.cpp
+            - ggml-metal.h
+            - ggml-metal.cpp
            - README-metal.md
 SYCL:
    - changed-files:
        - any-glob-to-any-file:
-            - ggml/include/ggml-sycl.h
-            - ggml/src/ggml-sycl.cpp
+            - ggml-sycl.h
+            - ggml-sycl.cpp
            - README-sycl.md
 Nvidia GPU:
    - changed-files:
        - any-glob-to-any-file:
-            - ggml/include/ggml-cuda.h
-            - ggml/src/ggml-cuda/**
+            - ggml-cuda.h
+            - ggml-cuda/**
 Vulkan:
    - changed-files:
        - any-glob-to-any-file:
-            - ggml/ggml_vk_generate_shaders.py
-            - ggml/src/ggml-vulkan*
+            - ggml_vk_generate_shaders.py
+            - ggml-vulkan*
 documentation:
    - changed-files:
        - any-glob-to-any-file:
@@ -73,10 +73,10 @@ server:
 ggml:
    - changed-files:
        - any-glob-to-any-file:
-            - ggml/include/ggml*.h
-            - ggml/src/ggml*.c
-            - ggml/src/ggml*.cpp
-            - ggml/src/ggml*.h
+            - ggml.c
+            - ggml.h
+            - ggml-*.c
+            - ggml-*.h
            - ggml-cuda/**
 nix:
    - changed-files:
--- a/.github/workflows/bench.yml
+++ b/.github/workflows/bench.yml
@@ -109,7 +109,7 @@ jobs:
        run: |
          set -eux
          cmake -B build \
-              -DGGML_NATIVE=OFF \
+              -DLLAMA_NATIVE=OFF \
              -DLLAMA_BUILD_SERVER=ON \
              -DLLAMA_CURL=ON \
              -DLLAMA_CUBLAS=ON \
--- a/.github/workflows/build.yml
+++ b/.github/workflows/build.yml
@@ -10,10 +10,10 @@ on:
  push:
    branches:
      - master
-    paths: ['.github/workflows/build.yml', '**/CMakeLists.txt', '**/Makefile', '**/*.h', '**/*.hpp', '**/*.c', '**/*.cpp', '**/*.cu', '**/*.cuh', '**/*.swift', '**/*.m', '**/*.metal']
+    paths: ['.github/workflows/**', '**/CMakeLists.txt', '**/Makefile', '**/*.h', '**/*.hpp', '**/*.c', '**/*.cpp', '**/*.cu', '**/*.swift', '**/*.m']
  pull_request:
    types: [opened, synchronize, reopened]
-    paths: ['.github/workflows/build.yml', '**/CMakeLists.txt', '**/Makefile', '**/*.h', '**/*.hpp', '**/*.c', '**/*.cpp', '**/*.cu', '**/*.cuh', '**/*.swift', '**/*.m', '**/*.metal']
+    paths: ['.github/workflows/build.yml', '**/CMakeLists.txt', '**/Makefile', '**/*.h', '**/*.hpp', '**/*.c', '**/*.cpp', '**/*.cu', '**/*.cuh', '**/*.swift', '**/*.m']

 concurrency:
  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
@@ -47,7 +47,7 @@ jobs:
          sysctl -a
          mkdir build
          cd build
-          cmake -DLLAMA_FATAL_WARNINGS=ON -DGGML_METAL_EMBED_LIBRARY=ON -DLLAMA_CURL=ON -DBUILD_SHARED_LIBS=OFF ..
+          cmake -DLLAMA_FATAL_WARNINGS=ON -DLLAMA_METAL_EMBED_LIBRARY=ON -DLLAMA_CURL=ON ..
          cmake --build . --config Release -j $(sysctl -n hw.logicalcpu)

      - name: Test
@@ -105,7 +105,7 @@ jobs:
          sysctl -a
          # Metal is disabled due to intermittent failures with Github runners not having a GPU:
          # https://github.com/ggerganov/llama.cpp/actions/runs/8635935781/job/23674807267#step:5:2313
-          cmake -B build -DLLAMA_FATAL_WARNINGS=ON -DGGML_METAL=OFF -DLLAMA_CURL=ON -DBUILD_SHARED_LIBS=OFF
+          cmake -B build -DLLAMA_FATAL_WARNINGS=ON -DLLAMA_METAL=OFF -DLLAMA_CURL=ON
          cmake --build build --config Release -j $(sysctl -n hw.logicalcpu)

      - name: Test
@@ -222,7 +222,7 @@ jobs:
        run: |
          mkdir build
          cd build
-          cmake .. -DLLAMA_FATAL_WARNINGS=ON -DLLAMA_CURL=ON -DBUILD_SHARED_LIBS=OFF
+          cmake .. -DLLAMA_FATAL_WARNINGS=ON -DLLAMA_CURL=ON
          cmake --build . --config Release -j $(nproc)

      - name: Test
@@ -305,7 +305,7 @@ jobs:
        run: |
          mkdir build
          cd build
-          cmake .. -DLLAMA_FATAL_WARNINGS=ON -DLLAMA_SANITIZE_${{ matrix.sanitizer }}=ON -DCMAKE_BUILD_TYPE=${{ matrix.build_type }} -DGGML_OPENMP=OFF
+          cmake .. -DLLAMA_FATAL_WARNINGS=ON -DLLAMA_SANITIZE_${{ matrix.sanitizer }}=ON -DCMAKE_BUILD_TYPE=${{ matrix.build_type }} -DLLAMA_OPENMP=OFF
          cmake --build . --config ${{ matrix.build_type }} -j $(nproc)

      - name: Test
@@ -335,7 +335,7 @@ jobs:
        run: |
          mkdir build
          cd build
-          cmake -DGGML_RPC=ON ..
+          cmake -DLLAMA_RPC=ON ..
          cmake --build . --config Release -j $(nproc)

      - name: Test
@@ -363,7 +363,7 @@ jobs:
        run: |
          mkdir build
          cd build
-          cmake -DGGML_VULKAN=ON ..
+          cmake -DLLAMA_VULKAN=ON ..
          cmake --build . --config Release -j $(nproc)

  ubuntu-22-cmake-hip:
@@ -384,13 +384,13 @@ jobs:
      - name: Build with native CMake HIP support
        id: cmake_build
        run: |
-          cmake -B build -S . -DCMAKE_HIP_COMPILER="$(hipconfig -l)/clang" -DGGML_HIPBLAS=ON
+          cmake -B build -S . -DCMAKE_HIP_COMPILER="$(hipconfig -l)/clang" -DLLAMA_HIPBLAS=ON
          cmake --build build --config Release -j $(nproc)

      - name: Build with legacy HIP support
        id: cmake_build_legacy_hip
        run: |
-          cmake -B build2 -S . -DCMAKE_C_COMPILER=hipcc -DCMAKE_CXX_COMPILER=hipcc -DGGML_HIPBLAS=ON
+          cmake -B build2 -S . -DCMAKE_C_COMPILER=hipcc -DCMAKE_CXX_COMPILER=hipcc -DLLAMA_HIPBLAS=ON
          cmake --build build2 --config Release -j $(nproc)

  ubuntu-22-cmake-sycl:
@@ -431,7 +431,7 @@ jobs:
          source /opt/intel/oneapi/setvars.sh
          mkdir build
          cd build
-          cmake -DGGML_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx ..
+          cmake -DLLAMA_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx ..
          cmake --build . --config Release -j $(nproc)

  ubuntu-22-cmake-sycl-fp16:
@@ -472,10 +472,10 @@ jobs:
          source /opt/intel/oneapi/setvars.sh
          mkdir build
          cd build
-          cmake -DGGML_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx -DGGML_SYCL_F16=ON ..
+          cmake -DLLAMA_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx -DLLAMA_SYCL_F16=ON ..
          cmake --build . --config Release -j $(nproc)

-  # TODO: build with GGML_NO_METAL because test-backend-ops fail on "Apple Paravirtual device" and I don't know
+  # TODO: build with LLAMA_NO_METAL because test-backend-ops fail on "Apple Paravirtual device" and I don't know
  #       how to debug it.
  #       ref: https://github.com/ggerganov/llama.cpp/actions/runs/7131777249/job/19420981052#step:5:1124
  macOS-latest-make:
@@ -497,15 +497,15 @@ jobs:
        env:
            LLAMA_FATAL_WARNINGS: 1
        run: |
-          GGML_NO_METAL=1 make -j $(sysctl -n hw.logicalcpu)
+          LLAMA_NO_METAL=1 make -j $(sysctl -n hw.logicalcpu)

      - name: Test
        id: make_test
        run: |
-          GGML_NO_METAL=1 make tests -j $(sysctl -n hw.logicalcpu)
-          GGML_NO_METAL=1 make test  -j $(sysctl -n hw.logicalcpu)
+          LLAMA_NO_METAL=1 make tests -j $(sysctl -n hw.logicalcpu)
+          LLAMA_NO_METAL=1 make test  -j $(sysctl -n hw.logicalcpu)

-  # TODO: build with GGML_METAL=OFF because test-backend-ops fail on "Apple Paravirtual device" and I don't know
+  # TODO: build with LLAMA_METAL=OFF because test-backend-ops fail on "Apple Paravirtual device" and I don't know
  #       how to debug it.
  #       ref: https://github.com/ggerganov/llama.cpp/actions/runs/7132125951/job/19422043567?pr=4359#step:5:6584
  #       would be great if we fix these
@@ -529,7 +529,7 @@ jobs:
          sysctl -a
          mkdir build
          cd build
-          cmake -DLLAMA_FATAL_WARNINGS=ON -DGGML_METAL=OFF ..
+          cmake -DLLAMA_FATAL_WARNINGS=ON -DLLAMA_METAL=OFF ..
          cmake --build . --config Release -j $(sysctl -n hw.logicalcpu)

      - name: Test
@@ -559,14 +559,13 @@ jobs:
          mkdir build
          cd build
          cmake -G Xcode .. \
-            -DGGML_METAL_EMBED_LIBRARY=ON \
+            -DLLAMA_METAL_EMBED_LIBRARY=ON \
            -DLLAMA_BUILD_EXAMPLES=OFF \
            -DLLAMA_BUILD_TESTS=OFF \
            -DLLAMA_BUILD_SERVER=OFF \
            -DCMAKE_SYSTEM_NAME=iOS \
-            -DCMAKE_OSX_DEPLOYMENT_TARGET=14.0 \
-            -DCMAKE_XCODE_ATTRIBUTE_DEVELOPMENT_TEAM=ggml
-          cmake --build . --config Release -j $(sysctl -n hw.logicalcpu) -- CODE_SIGNING_ALLOWED=NO
+            -DCMAKE_OSX_DEPLOYMENT_TARGET=14.0
+          cmake --build . --config Release -j $(sysctl -n hw.logicalcpu)

  macOS-latest-cmake-tvos:
    runs-on: macos-latest
@@ -589,14 +588,13 @@ jobs:
          mkdir build
          cd build
          cmake -G Xcode .. \
-            -DGGML_METAL_EMBED_LIBRARY=ON \
+            -DLLAMA_METAL_EMBED_LIBRARY=ON \
            -DLLAMA_BUILD_EXAMPLES=OFF \
            -DLLAMA_BUILD_TESTS=OFF \
            -DLLAMA_BUILD_SERVER=OFF \
            -DCMAKE_SYSTEM_NAME=tvOS \
-            -DCMAKE_OSX_DEPLOYMENT_TARGET=14.0 \
-            -DCMAKE_XCODE_ATTRIBUTE_DEVELOPMENT_TEAM=ggml
-          cmake --build . --config Release -j $(sysctl -n hw.logicalcpu) -- CODE_SIGNING_ALLOWED=NO
+            -DCMAKE_OSX_DEPLOYMENT_TARGET=14.0
+          cmake --build . --config Release -j $(sysctl -n hw.logicalcpu)

  macOS-latest-swift:
    runs-on: macos-latest
@@ -664,7 +662,7 @@ jobs:
      - name: Build using make w/ OpenBLAS
        shell: msys2 {0}
        run: |
-            make GGML_OPENBLAS=1 -j $(nproc)
+            make LLAMA_OPENBLAS=1 -j $(nproc)

      - name: Build using CMake
        shell: msys2 {0}
@@ -680,7 +678,7 @@ jobs:
      - name: Build using CMake w/ OpenBLAS
        shell: msys2 {0}
        run: |
-            cmake -B build -DGGML_BLAS=ON -DGGML_BLAS_VENDOR=OpenBLAS
+            cmake -B build -DLLAMA_BLAS=ON -DLLAMA_BLAS_VENDOR=OpenBLAS
            cmake --build build --config ${{ matrix.build }} -j $(nproc)

  windows-latest-cmake:
@@ -695,25 +693,25 @@ jobs:
      matrix:
        include:
          - build: 'rpc-x64'
-            defines: '-DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_RPC=ON -DBUILD_SHARED_LIBS=ON'
+            defines: '-DLLAMA_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DLLAMA_RPC=ON -DBUILD_SHARED_LIBS=ON'
          - build: 'noavx-x64'
-            defines: '-DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_AVX=OFF -DGGML_AVX2=OFF -DGGML_FMA=OFF -DBUILD_SHARED_LIBS=ON'
+            defines: '-DLLAMA_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DLLAMA_AVX=OFF -DLLAMA_AVX2=OFF -DLLAMA_FMA=OFF -DBUILD_SHARED_LIBS=ON'
          - build: 'avx2-x64'
-            defines: '-DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DBUILD_SHARED_LIBS=ON'
+            defines: '-DLLAMA_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DBUILD_SHARED_LIBS=ON'
          - build: 'avx-x64'
-            defines: '-DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_AVX2=OFF -DBUILD_SHARED_LIBS=ON'
+            defines: '-DLLAMA_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DLLAMA_AVX2=OFF -DBUILD_SHARED_LIBS=ON'
          - build: 'avx512-x64'
-            defines: '-DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_AVX512=ON -DBUILD_SHARED_LIBS=ON'
+            defines: '-DLLAMA_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DLLAMA_AVX512=ON -DBUILD_SHARED_LIBS=ON'
          - build: 'openblas-x64'
-            defines: '-DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_BLAS=ON -DBUILD_SHARED_LIBS=ON -DGGML_BLAS_VENDOR=OpenBLAS -DBLAS_INCLUDE_DIRS="$env:RUNNER_TEMP/openblas/include" -DBLAS_LIBRARIES="$env:RUNNER_TEMP/openblas/lib/openblas.lib"'
+            defines: '-DLLAMA_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DLLAMA_BLAS=ON -DBUILD_SHARED_LIBS=ON -DLLAMA_BLAS_VENDOR=OpenBLAS -DBLAS_INCLUDE_DIRS="$env:RUNNER_TEMP/openblas/include" -DBLAS_LIBRARIES="$env:RUNNER_TEMP/openblas/lib/openblas.lib"'
          - build: 'kompute-x64'
-            defines: '-DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_KOMPUTE=ON -DKOMPUTE_OPT_DISABLE_VULKAN_VERSION_CHECK=ON -DBUILD_SHARED_LIBS=ON'
+            defines: '-DLLAMA_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DLLAMA_KOMPUTE=ON -DKOMPUTE_OPT_DISABLE_VULKAN_VERSION_CHECK=ON -DBUILD_SHARED_LIBS=ON'
          - build: 'vulkan-x64'
-            defines: '-DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_VULKAN=ON -DBUILD_SHARED_LIBS=ON'
+            defines: '-DLLAMA_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DLLAMA_VULKAN=ON -DBUILD_SHARED_LIBS=ON'
          - build: 'llvm-arm64'
-            defines: '-G "Ninja Multi-Config" -D CMAKE_TOOLCHAIN_FILE=cmake/arm64-windows-llvm.cmake -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DBUILD_SHARED_LIBS=ON'
+            defines: '-G "Ninja Multi-Config" -D CMAKE_TOOLCHAIN_FILE=cmake/arm64-windows-llvm.cmake -DLLAMA_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DBUILD_SHARED_LIBS=ON'
          - build: 'msvc-arm64'
-            defines: '-G "Ninja Multi-Config" -D CMAKE_TOOLCHAIN_FILE=cmake/arm64-windows-msvc.cmake -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DBUILD_SHARED_LIBS=ON'
+            defines: '-G "Ninja Multi-Config" -D CMAKE_TOOLCHAIN_FILE=cmake/arm64-windows-msvc.cmake -DLLAMA_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DBUILD_SHARED_LIBS=ON'

    steps:
      - name: Clone
@@ -726,7 +724,7 @@ jobs:
        id: clone_kompute
        if: ${{ matrix.build == 'kompute-x64' }}
        run: |
-          git submodule update --init ggml/src/kompute
+          git submodule update --init kompute

      - name: Download OpenBLAS
        id: get_openblas
@@ -799,7 +797,6 @@ jobs:
          7z x "-o${env:RUNNER_TEMP}" $env:RUNNER_TEMP/sde.tar
          $sde = $(join-path $env:RUNNER_TEMP sde-external-${env:SDE_VERSION}-win/sde.exe)
          cd build
-          $env:LLAMA_SKIP_TESTS_SLOW_ON_EMULATOR = 1
          & $sde -future -- ctest -L main -C Release --verbose --timeout 900

      - name: Determine tag name
@@ -857,7 +854,7 @@ jobs:
        run: |
          mkdir build
          cd build
-          cmake .. -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=ON
+          cmake .. -DLLAMA_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DLLAMA_CUDA=ON -DBUILD_SHARED_LIBS=ON
          cmake --build . --config Release -j ${env:NUMBER_OF_PROCESSORS}

      - name: Determine tag name
@@ -990,7 +987,7 @@ jobs:
        run: |
          $env:HIP_PATH=$(Resolve-Path 'C:\Program Files\AMD\ROCm\*\bin\clang.exe' | split-path | split-path)
          $env:CMAKE_PREFIX_PATH="${env:HIP_PATH}"
-          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" -DGGML_HIPBLAS=ON
+          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" -DLLAMA_HIPBLAS=ON
          cmake --build build --config Release

  ios-xcode-build:
--- a/.github/workflows/docker.yml
+++ b/.github/workflows/docker.yml
@@ -14,7 +14,6 @@ on:
  push:
    branches:
      - master
-    paths: ['.github/workflows/docker.yml', '.devops/*.Dockerfile', '**/CMakeLists.txt', '**/Makefile', '**/*.h', '**/*.hpp', '**/*.c', '**/*.cpp', '**/*.cu', '**/*.cuh', '**/*.swift', '**/*.m', '**/*.metal']

 concurrency:
  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
--- a/.github/workflows/server.yml
+++ b/.github/workflows/server.yml
@@ -92,12 +92,12 @@ jobs:
        if: ${{ matrix.sanitizer == 'THREAD' }}
        run: |
          cmake -B build \
-              -DGGML_NATIVE=OFF \
+              -DLLAMA_NATIVE=OFF \
              -DLLAMA_BUILD_SERVER=ON \
              -DLLAMA_CURL=ON \
              -DCMAKE_BUILD_TYPE=${{ matrix.build_type }} \
              -DLLAMA_SANITIZE_${{ matrix.sanitizer }}=ON \
-              -DGGML_OPENMP=OFF ;
+              -DLLAMA_OPENMP=OFF ;
          cmake --build build --config ${{ matrix.build_type }} -j $(nproc) --target llama-server

      - name: Build
@@ -105,7 +105,7 @@ jobs:
        if: ${{ matrix.sanitizer != 'THREAD' }}
        run: |
          cmake -B build \
-              -DGGML_NATIVE=OFF \
+              -DLLAMA_NATIVE=OFF \
              -DLLAMA_BUILD_SERVER=ON \
              -DLLAMA_CURL=ON \
              -DCMAKE_BUILD_TYPE=${{ matrix.build_type }} \
--- a/.gitignore
+++ b/.gitignore
@@ -56,7 +56,6 @@ CMakeSettings.json
 compile_commands.json
 ggml-metal-embed.metal
 llama-batched-swift
-/rpc-server
 out/
 tmp/

--- a/.gitmodules
+++ b/.gitmodules
@@ -1,3 +1,3 @@
 [submodule "kompute"]
-	path = ggml/src/kompute
+	path = kompute
 	url = https://github.com/nomic-ai/kompute.git
--- a/129
+++ b/129
@@ -1,9 +1,8 @@
-# date: Wed Jun 26 19:36:34 EEST 2024
+# date: Tue Apr  9 09:17:14 EEST 2024
 # this file is auto-generated by scripts/gen-authors.sh

 0cc4m <picard12@live.de>
 0xspringtime <110655352+0xspringtime@users.noreply.github.com>
-20kdc <asdd2808@gmail.com>
 2f38b454 <dxf@protonmail.com>
 3ooabkhxtn <31479382+3ooabkhxtn@users.noreply.github.com>
 44670 <44670@users.noreply.github.com>
@@ -12,18 +11,14 @@ AT <manyoso@users.noreply.github.com>
 Aarni Koskela <akx@iki.fi>
 Aaron Miller <apage43@ninjawhale.com>
 Aaryaman Vasishta <aaryaman.vasishta@amd.com>
-Abheek Gulati <abheekg@hotmail.com>
 Abhilash Majumder <30946547+abhilash1910@users.noreply.github.com>
 Abhishek Gopinath K <31348521+overtunned@users.noreply.github.com>
 Adithya Balaji <adithya.b94@gmail.com>
 AdithyanI <adithyan.i4internet@gmail.com>
 Adrian <smith.adriane@gmail.com>
 Adrian Hesketh <a-h@users.noreply.github.com>
-Ahmet Zeer <ahmed.zeer@std.yildiz.edu.tr>
 AidanBeltonS <87009434+AidanBeltonS@users.noreply.github.com>
 Aisuko <urakiny@gmail.com>
-Akarshan Biswas <akarshanbiswas@fedoraproject.org>
-Albert Jin <albert.jin@gmail.com>
 Alberto <57916483+albbus-stack@users.noreply.github.com>
 Alex <awhill19@icloud.com>
 Alex Azarov <alex@azarov.by>
@@ -40,24 +35,19 @@ Ali Nehzat <ali.nehzat@thanks.dev>
 Ali Tariq <ali.tariq@10xengineers.ai>
 Alon <alonfaraj@gmail.com>
 AlpinDale <52078762+AlpinDale@users.noreply.github.com>
-Amir <amir_zia@outlook.com>
 AmirAli Mirian <37371367+amiralimi@users.noreply.github.com>
 Ananta Bastola <anantarajbastola@gmail.com>
 Anas Ahouzi <112881240+aahouzi@users.noreply.github.com>
 András Salamon <ott2@users.noreply.github.com>
 Andrei <abetlen@gmail.com>
 Andrew Canis <andrew.canis@gmail.com>
-Andrew Downing <andrew2085@gmail.com>
 Andrew Duffy <a10y@users.noreply.github.com>
 Andrew Godfrey <AndrewGodfrey@users.noreply.github.com>
-Andy Tai <andy-tai@users.noreply.github.com>
 Arik Poznanski <arikpoz@users.noreply.github.com>
 Artem <guinmoon@gmail.com>
-Artem Zinnatullin <ceo@abstractny.gay>
 Artyom Lebedev <vagran.ast@gmail.com>
 Asbjørn Olling <asbjornolling@gmail.com>
 Ásgeir Bjarni Ingvarsson <asgeir@fundinn.org>
-Ashish <1856117+ashishdatta@users.noreply.github.com>
 Ashok Gelal <401055+ashokgelal@users.noreply.github.com>
 Ashraful Islam <ashraful.meche@gmail.com>
 Atsushi Tatsuma <yoshoku@outlook.com>
@@ -67,46 +57,35 @@ BADR <contact@pythops.com>
 Bach Le <bach@bullno1.com>
 Bailey Chittle <39804642+bachittle@users.noreply.github.com>
 BarfingLemurs <128182951+BarfingLemurs@users.noreply.github.com>
-Bartowski <ckealty1182@gmail.com>
 Behnam M <58621210+ibehnam@users.noreply.github.com>
-Ben Ashbaugh <ben.ashbaugh@intel.com>
 Ben Garney <bengarney@users.noreply.github.com>
 Ben Siraphob <bensiraphob@gmail.com>
 Ben Williams <ben@719ben.com>
-Benjamin Findley <39356821+Kartoffelsaft@users.noreply.github.com>
 Benjamin Lecaillon <84293038+blecaillon@users.noreply.github.com>
 Bernat Vadell <hounter.caza@gmail.com>
-Bingan <70050083+binganao@users.noreply.github.com>
 Bodo Graumann <mail@bodograumann.de>
 Bono Lv <lvscar@users.noreply.github.com>
 Borislav Stanimirov <b.stanimirov@abv.bg>
 Branden Butler <bwtbutler@hotmail.com>
 Brian <mofosyne@gmail.com>
 Bruce MacDonald <brucewmacdonald@gmail.com>
-Bryan Honof <bryanhonof@gmail.com>
 CJ Pais <cj@cjpais.com>
 CRD716 <crd716@gmail.com>
-Calvin Laurenson <calvin@laurenson.dev>
 Cameron <csteele@steelecameron.com>
 Cameron Kaiser <classilla@users.noreply.github.com>
-Carolinabanana <140120812+Carolinabanana@users.noreply.github.com>
 Casey Primozic <casey@cprimozic.net>
 Casey Primozic <me@ameo.link>
 CausalLM <148736309+CausalLM@users.noreply.github.com>
 Cebtenzzre <cebtenzzre@gmail.com>
 Chad Brewbaker <crb002@gmail.com>
-Chao Jiang <jc19chaoj@zoho.com>
 Cheng Shao <terrorjack@type.dance>
-Chris Elrod <elrodc@gmail.com>
 Chris Kuehl <ckuehl@ckuehl.me>
 Christian Demsar <christian@github.email.demsar.us>
 Christian Demsar <crasm@git.vczf.us>
 Christian Falch <875252+chrfalch@users.noreply.github.com>
 Christian Kögler <ck3d@gmx.de>
-Christian Zhou-Zheng <59622928+christianazinn@users.noreply.github.com>
 Clark Saben <76020733+csaben@users.noreply.github.com>
 Clint Herron <hanclinto@gmail.com>
-CrispStrobe <154636388+CrispStrobe@users.noreply.github.com>
 Cuong Trinh Manh <nguoithichkhampha@gmail.com>
 DAN™ <dranger003@gmail.com>
 Damian Stewart <d@damianstewart.com>
@@ -116,12 +95,8 @@ Daniel Bevenius <daniel.bevenius@gmail.com>
 Daniel Drake <drake@endlessos.org>
 Daniel Hiltgen <dhiltgen@users.noreply.github.com>
 Daniel Illescas Romero <illescas.daniel@protonmail.com>
-Daniele <57776841+daniandtheweb@users.noreply.github.com>
 DannyDaemonic <DannyDaemonic@gmail.com>
 Dat Quoc Nguyen <2412555+datquocnguyen@users.noreply.github.com>
-Dave <dave-fl@users.noreply.github.com>
-Dave Airlie <airlied@gmail.com>
-Dave Airlie <airlied@redhat.com>
 Dave Della Costa <ddellacosta+github@gmail.com>
 David Friehs <david@friehs.info>
 David Kennedy <dakennedyd@gmail.com>
@@ -129,13 +104,10 @@ David Pflug <david@pflug.email>
 David Renshaw <dwrenshaw@gmail.com>
 David Sommers <12738+databyte@users.noreply.github.com>
 David Yang <davidyang6us@gmail.com>
-Dawid Potocki <github@dawidpotocki.com>
 Dawid Wysocki <62249621+TortillaZHawaii@users.noreply.github.com>
 Dean <Dean.Sinaean@gmail.com>
 Deins <deinsegle@gmail.com>
-Deven Mistry <31466137+deven367@users.noreply.github.com>
 Didzis Gosko <didzis@users.noreply.github.com>
-Djip007 <djip.perois@free.fr>
 Don Mahurin <dmahurin@users.noreply.github.com>
 DooWoong Lee (David) <manics99@naver.com>
 Doomsdayrs <38189170+Doomsdayrs@users.noreply.github.com>
@@ -144,11 +116,8 @@ Dr. Tom Murphy VII Ph.D <499244+tom7@users.noreply.github.com>
 Ebey Abraham <ebey97@gmail.com>
 Ed Lee <edilee@mozilla.com>
 Ed Lepedus <ed.lepedus@googlemail.com>
-Eddie-Wang <wangjinheng1120@163.com>
 Edward Taylor <edeetee@gmail.com>
-Elaine <elaine.zosa@gmail.com>
 Elbios <141279586+Elbios@users.noreply.github.com>
-Elton Kola <eltonkola@gmail.com>
 Engininja2 <139037756+Engininja2@users.noreply.github.com>
 Equim <sayaka@ekyu.moe>
 Eric Sommerlade <es0m@users.noreply.github.com>
@@ -174,47 +143,37 @@ Firat <firatkiral@gmail.com>
 Folko-Ven <71110216+Folko-Ven@users.noreply.github.com>
 Foul-Tarnished <107711110+Foul-Tarnished@users.noreply.github.com>
 Francisco Melo <43780565+francis2tm@users.noreply.github.com>
-Frank Mai <thxcode0824@gmail.com>
 FrankHB <frankhb1989@gmail.com>
-Fred Douglas <43351173+fredlas@users.noreply.github.com>
 Frederik Vogel <Schaltfehler@users.noreply.github.com>
 Gabe Goodhart <gabe.l.hart@gmail.com>
 GainLee <perfecter.gen@gmail.com>
 Galunid <karolek1231456@gmail.com>
 Gary Linscott <glinscott@gmail.com>
 Gary Mulder <gjmulder@gmail.com>
-Gavin Zhao <gavinzhaojw@protonmail.com>
 Genkagaku.GPT <hlhr202@163.com>
 Georgi Gerganov <ggerganov@gmail.com>
 Gilad S <giladgd@users.noreply.github.com>
-Giuseppe Scrivano <giuseppe@scrivano.org>
 GiviMAD <GiviMAD@users.noreply.github.com>
 Govlzkoy <gotope@users.noreply.github.com>
 Guillaume "Vermeille" Sanchez <Guillaume.V.Sanchez@gmail.com>
 Guillaume Wenzek <gwenzek@users.noreply.github.com>
 Guoteng <32697156+SolenoidWGT@users.noreply.github.com>
 Gustavo Rocha Dias <91472747+gustrd@users.noreply.github.com>
-Haggai Nuchi <h.nuchi@gmail.com>
 Halalaluyafail3 <55773281+Halalaluyafail3@users.noreply.github.com>
-Hamdoud Hakem <90524568+hamdoudhakem@users.noreply.github.com>
-HanishKVC <hanishkvc@gmail.com>
 Haohui Mai <ricetons@gmail.com>
 Haoxiang Fei <tonyfettes@tonyfettes.com>
 Harald Fernengel <harald.fernengel@here.com>
 Hatsune Miku <129688334+at8u@users.noreply.github.com>
-HatsuneMikuUwU33 <173229399+HatsuneMikuUwU33@users.noreply.github.com>
 Henk Poley <HenkPoley@gmail.com>
 Henri Vasserman <henv@hot.ee>
 Henrik Forstén <henrik.forsten@gmail.com>
 Herman Semenov <GermanAizek@yandex.ru>
 Hesen Peng <hesen.peng@gmail.com>
 Hoang Nguyen <hugo53@users.noreply.github.com>
-Hong Bo PENG <penghb@cn.ibm.com>
 Hongyu Ouyang <96765450+casavaca@users.noreply.github.com>
 Howard Su <howard0su@gmail.com>
 Hua Jiang <allenhjiang@outlook.com>
 Huawei Lin <huaweilin.cs@gmail.com>
-Hugo Roussel <hugo.rous@gmail.com>
 Ian Bull <irbull@eclipsesource.com>
 Ian Bull <irbull@gmail.com>
 Ian Scrivener <github@zilogy.asia>
@@ -231,10 +190,8 @@ Ivan Stepanov <ivanstepanovftw@gmail.com>
 JH23X <165871467+JH23X@users.noreply.github.com>
 Jack Mousseau <jmousseau@users.noreply.github.com>
 JackJollimore <130917767+JackJollimore@users.noreply.github.com>
-Jaemin Son <woalsdnd@gmail.com>
 Jag Chadha <jagtesh@gmail.com>
 Jakub N <jakubniemczyk97@gmail.com>
-James A Capozzoli <157492257+jac-jim@users.noreply.github.com>
 James Reynolds <magnusviri@users.noreply.github.com>
 Jan Boon <jan.boon@kaetemi.be>
 Jan Boon <kaetemi@gmail.com>
@@ -248,17 +205,12 @@ Jean-Michaël Celerier <jeanmichael.celerier+github@gmail.com>
 Jed Fox <git@jedfox.com>
 Jeffrey Quesnelle <emozilla@nousresearch.com>
 Jesse Jojo Johnson <williamsaintgeorge@gmail.com>
-Jeximo <jeximo@gmail.com>
 Jhen-Jie Hong <iainst0409@gmail.com>
 Jiahao Li <liplus17@163.com>
 Jian Liao <jianliao@users.noreply.github.com>
 JidongZhang-THU <1119708529@qq.com>
 Jinwoo Jeong <33892306+williamjeong2@users.noreply.github.com>
 Jiří Podivín <66251151+jpodivin@users.noreply.github.com>
-Jiří Sejkora <Sejseloid@gmail.com>
-Joan Fontanals <jfontanalsmartinez@gmail.com>
-Joan Fontanals <joan.fontanals.martinez@jina.ai>
-Johan <JohanAR@users.noreply.github.com>
 Johannes Gäßler <johannesg@5d6.de>
 Johannes Rudolph <johannes.rudolph@gmail.com>
 John <78893154+cmp-nct@users.noreply.github.com>
@@ -269,19 +221,15 @@ Jonas Wunderlich <32615971+jonas-w@users.noreply.github.com>
 Jorge A <161275481+jorgealias@users.noreply.github.com>
 Jose Maldonado <63384398+yukiteruamano@users.noreply.github.com>
 Joseph Stahl <1269177+josephst@users.noreply.github.com>
-Josh Ramer <josh.ramer@icloud.com>
 Joyce <joycebrum@google.com>
 Juan Calderon-Perez <835733+gaby@users.noreply.github.com>
 Judd <foldl@users.noreply.github.com>
 Julius Arkenberg <arki05@users.noreply.github.com>
 Jun Jie <71215065+junnjiee16@users.noreply.github.com>
-Junyang Lin <justinlin930319@hotmail.com>
 Juraj Bednar <juraj@bednar.io>
 Justin Parker <jparkerweb@gmail.com>
 Justin Suess <justin.suess@westpoint.edu>
-Justina Cho <justcho5@gmail.com>
 Justine Tunney <jtunney@gmail.com>
-Justine Tunney <jtunney@mozilla.com>
 Juuso Alasuutari <juuso.alasuutari@gmail.com>
 KASR <karim.asrih@gmail.com>
 Kamil Tomšík <info@tomsik.cz>
@@ -294,7 +242,6 @@ Kawrakow <48489457+ikawrakow@users.noreply.github.com>
 Keiichi Tabata <keiichi.tabata@outlook.com>
 Kenvix ⭐ <kenvixzure@live.com>
 Kerfuffle <44031344+KerfuffleV2@users.noreply.github.com>
-Kevin Gibbons <bakkot@gmail.com>
 Kevin Ji <1146876+kevinji@users.noreply.github.com>
 Kevin Kwok <antimatter15@gmail.com>
 Kevin Lo <kevlo@kevlo.org>
@@ -310,7 +257,6 @@ Laura <Tijntje_7@msn.com>
 Lee <44310445+lx200916@users.noreply.github.com>
 Lee Drake <b.lee.drake@gmail.com>
 Leng Yue <lengyue@lengyue.me>
-Leon Knauer <git@leonknauer.com>
 LeonEricsson <70749762+LeonEricsson@users.noreply.github.com>
 Leonardo Neumann <leonardo@neumann.dev.br>
 Li Tan <tanliboy@gmail.com>
@@ -319,26 +265,20 @@ LoganDark <github@logandark.mozmail.com>
 LostRuins <39025047+LostRuins@users.noreply.github.com>
 Luciano <lucianostrika44@gmail.com>
 Luo Tian <lt@basecity.com>
-Lyle Dean <dean@lyle.dev>
 M. Yusuf Sarıgöz <yusufsarigoz@gmail.com>
 Maarten ter Huurne <maarten@treewalker.org>
 Mack Straight <eiz@users.noreply.github.com>
 Maël Kerbiriou <m431.kerbiriou@gmail.com>
 MaggotHATE <clay1326@gmail.com>
-Manuel <44313466+makuche@users.noreply.github.com>
 Marc Köhlbrugge <subscriptions@marckohlbrugge.com>
 Marco Matthies <71844+marcom@users.noreply.github.com>
 Marcus Dunn <51931484+MarcusDunn@users.noreply.github.com>
 Marian Cepok <marian.cepok@gmail.com>
 Mark Fairbairn <thebaron88@gmail.com>
 Marko Tasic <mtasic85@gmail.com>
-Markus Tavenrath <mtavenrath@users.noreply.github.com>
-Martin Delille <martin@delille.org>
 Martin Krasser <krasserm@googlemail.com>
 Martin Schwaighofer <mschwaig@users.noreply.github.com>
 Marvin Gießing <marvin.giessing@gmail.com>
-Masaya, Kato <62578291+msy-kato@users.noreply.github.com>
-MasterYi1024 <39848311+MasterYi1024@users.noreply.github.com>
 Mateusz Charytoniuk <mateusz.charytoniuk@protonmail.com>
 Matheus C. França <matheus-catarino@hotmail.com>
 Matheus Gabriel Alves Silva <matheusgasource@gmail.com>
@@ -347,11 +287,8 @@ Mathijs de Bruin <mathijs@mathijsfietst.nl>
 Matt Clayton <156335168+mattjcly@users.noreply.github.com>
 Matt Pulver <matt.pulver@heavy.ai>
 Matteo Boschini <12133566+mbosc@users.noreply.github.com>
-Mattheus Chediak <shammcity00@gmail.com>
 Matthew Tejo <matthew.tejo@gmail.com>
 Matvey Soloviev <blackhole89@gmail.com>
-Max Krasnyansky <max.krasnyansky@gmail.com>
-Max Krasnyansky <quic_maxk@quicinc.com>
 Maxime <672982+maximegmd@users.noreply.github.com>
 Maximilian Winter <maximilian.winter.91@gmail.com>
 Meng Zhang <meng@tabbyml.com>
@@ -363,41 +300,32 @@ Michael Kesper <mkesper@schokokeks.org>
 Michael Klimenko <mklimenko29@gmail.com>
 Michael Podvitskiy <podvitskiymichael@gmail.com>
 Michael Potter <NanoTekGuy@Gmail.com>
-Michael de Gans <michael.john.degans@gmail.com>
 Michaël de Vries <vriesdemichael@gmail.com>
 Mihai <mihai.chirculescu@yahoo.com>
 Mike <ytianhui2004@gmail.com>
-Mikko Juola <mikjuo@gmail.com>
 Minsoo Cheong <54794500+mscheong01@users.noreply.github.com>
 Mirko185 <mirkosig@gmail.com>
 Mirror Azure <54669636+MirrorAzure@users.noreply.github.com>
 Miwa / Ensan <63481257+ensan-hcl@users.noreply.github.com>
 Mohammadreza Hendiani <hendiani.mohammadreza@gmail.com>
-Mohammadreza Hendiani <mohammad.r.hendiani@gmail.com>
 Murilo Santana <mvrilo@gmail.com>
 Musab Gultekin <musabgultekin@users.noreply.github.com>
 Nam D. Tran <42194884+namtranase@users.noreply.github.com>
-Nathan Epstein <nate2@umbc.edu>
 NawafAlansari <72708095+NawafAlansari@users.noreply.github.com>
 Nebula <infinitewormhole@gmail.com>
-Neo Zhang <14088817+arthw@users.noreply.github.com>
-Neo Zhang <zhang.jianyu@outlook.com>
 Neo Zhang Jianyu <jianyu.zhang@intel.com>
 Neuman Vong <neuman.vong@gmail.com>
 Nexesenex <124105151+Nexesenex@users.noreply.github.com>
 Niall Coates <1349685+Niall-@users.noreply.github.com>
 Nicolai Weitkemper <kontakt@nicolaiweitkemper.de>
-Nicolás Pérez <nicolas_perez@brown.edu>
 Nigel Bosch <pnigelb@gmail.com>
 Niklas Korz <niklas@niklaskorz.de>
-Nikolas <127742645+nneubacher@users.noreply.github.com>
 Nindaleth <Nindaleth@users.noreply.github.com>
 Oleksandr Nikitin <oleksandr@tvori.info>
 Oleksii Maryshchenko <oleksii.maryshchenko@gmail.com>
 Olivier Chafik <ochafik@users.noreply.github.com>
 Ondřej Čertík <ondrej@certik.us>
 Ouadie EL FAROUKI <ouadie.elfarouki@codeplay.com>
-Patrice Ferlet <metal3d@gmail.com>
 Paul Tsochantaris <ptsochantaris@icloud.com>
 Pavol Rusnak <pavol@rusnak.io>
 Pedro Cuenca <pedro@huggingface.co>
@@ -415,14 +343,9 @@ RJ Adriaansen <adriaansen@eshcc.eur.nl>
 Radoslav Gerganov <rgerganov@gmail.com>
 Radosław Gryta <radek.gryta@gmail.com>
 Rahul Vivek Nair <68507071+RahulVivekNair@users.noreply.github.com>
-Raj Hammeer Singh Hada <hammeerraj@gmail.com>
-Ralph Soika <ralph.soika@imixs.com>
 Rand Xie <randxiexyy29@gmail.com>
 Randall Fitzgerald <randall@dasaku.net>
 Reinforce-II <fate@eastal.com>
-Ren Xuancheng <jklj077@users.noreply.github.com>
-Rene Leonhardt <65483435+reneleonhardt@users.noreply.github.com>
-RhinoDevel <RhinoDevel@users.noreply.github.com>
 Riceball LEE <snowyu.lee@gmail.com>
 Richard Kiss <him@richardkiss.com>
 Richard Roberson <richardr1126@gmail.com>
@@ -450,7 +373,6 @@ Rowan Hart <rowanbhart@gmail.com>
 Rune <43761327+Rune-AI@users.noreply.github.com>
 Ryan Landay <rlanday@gmail.com>
 Ryder Wishart <ryderwishart@gmail.com>
-Ryuei <louixs@users.noreply.github.com>
 Rőczey Barnabás <31726601+An0nie@users.noreply.github.com>
 SakuraUmi <yukinon244@gmail.com>
 Salvador E. Tropea <stropea@inti.gob.ar>
@@ -464,7 +386,6 @@ SebastianApel <13675545+SebastianApel@users.noreply.github.com>
 Senemu <10880819+Senemu@users.noreply.github.com>
 Sergey Alirzaev <zl29ah@gmail.com>
 Sergio López <slp@sinrega.org>
-Sertaç Özercan <852750+sozercan@users.noreply.github.com>
 SeungWon Jeong <65549245+redlion0929@users.noreply.github.com>
 ShadovvBeast <ShadovvBeast@gmail.com>
 Shakhar Dasgupta <shakhardasgupta@gmail.com>
@@ -473,7 +394,6 @@ Shijie <821898965@qq.com>
 Shintarou Okada <kokuzen@gmail.com>
 Shouzheng Liu <61452103+lshzh-ww@users.noreply.github.com>
 Shouzheng Liu <lshzh.hi@gmail.com>
-Shuichi Tsutsumi <shuichi0526@gmail.com>
 Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
 Simon Willison <swillison@gmail.com>
 Siwen Yu <yusiwen@gmail.com>
@@ -485,14 +405,11 @@ Someone <sergei.kozlukov@aalto.fi>
 Someone Serge <sergei.kozlukov@aalto.fi>
 Sourab Mangrulkar <13534540+pacman100@users.noreply.github.com>
 Spencer Sutton <spencersutton@users.noreply.github.com>
-Srihari-mcw <96763064+Srihari-mcw@users.noreply.github.com>
 Srinivas Billa <nivibilla@gmail.com>
 Stefan Sydow <stefan@sydow.email>
-Steffen Röcker <sroecker@gmail.com>
 Stephan Walter <stephan@walter.name>
 Stephen Nichols <snichols@users.noreply.github.com>
 Steve Grubb <ausearch.1@gmail.com>
-Steven Prichard <spprichard20@gmail.com>
 Steven Roussey <sroussey@gmail.com>
 Steward Garcia <57494570+FSSRepo@users.noreply.github.com>
 Suaj Carrot <72162667+SuajCarrot@users.noreply.github.com>
@@ -517,19 +434,16 @@ Tom C <tom.corelis@gmail.com>
 Tom Jobbins <784313+TheBloke@users.noreply.github.com>
 Tomas <tom.tomas.36478119@gmail.com>
 Tomáš Pazdiora <tomas.pazdiora@gmail.com>
-Tristan Druyen <tristan@vault81.mozmail.com>
 Tristan Ross <rosscomputerguy@protonmail.com>
 Tungsten842 <886724vf@anonaddy.me>
 Tungsten842 <quantmint@protonmail.com>
 Tushar <ditsuke@protonmail.com>
 UEXTM.com <84163508+uextm@users.noreply.github.com>
-Ulrich Drepper <drepper@gmail.com>
 Uzo Nweke <uzoechi@gmail.com>
 Vaibhav Srivastav <vaibhavs10@gmail.com>
 Val Kharitonov <mail@kharvd.com>
 Valentin Konovalov <valle.ketsujin@gmail.com>
 Valentyn Bezshapkin <61702053+valentynbez@users.noreply.github.com>
-Victor Nogueira <felladrin@gmail.com>
 Victor Z. Peng <ziliangdotme@gmail.com>
 Vlad <spitfireage@gmail.com>
 Vladimir <bogdad@gmail.com>
@@ -541,9 +455,7 @@ Weird Constructor <weirdconstructor@gmail.com>
 Welby Seely <welbyseely@gmail.com>
 Wentai Zhang <rchardx@gmail.com>
 WillCorticesAI <150854901+WillCorticesAI@users.noreply.github.com>
-William Tambellini <william.tambellini@gmail.com>
 Willy Tarreau <w@1wt.eu>
-Wouter <9594229+DifferentialityDevelopment@users.noreply.github.com>
 Wu Jian Ping <wujjpp@hotmail.com>
 Wu Jian Ping <wujp@greatld.com>
 Xiake Sun <xiake.sun@intel.com>
@@ -554,8 +466,6 @@ Xiaoyi Chen <cxychina@gmail.com>
 Xingchen Song(宋星辰) <xingchensong1996@163.com>
 Xuan Son Nguyen <thichthat@gmail.com>
 Yann Follet <131855179+YannFollet@users.noreply.github.com>
-Yaroslav <yaroslav.yashin@me.com>
-Yazan Agha-Schrader <mountaiin@icloud.com>
 Yiming Cui <conandiy@vip.qq.com>
 Yishuo Wang <MeouSker77@outlook.com>
 Yueh-Po Peng <94939112+y10ab1@users.noreply.github.com>
@@ -567,7 +477,6 @@ Zane Shannon <z@zcs.me>
 Zay <95888118+isaiahbjork@users.noreply.github.com>
 Zenix <zenixls2@gmail.com>
 Zhang Peiyuan <a1286225768@gmail.com>
-Zheng.Deng <32841220+dengzheng-cloud@users.noreply.github.com>
 ZhouYuChen <zhouyuchen@naver.com>
 Ziad Ben Hadj-Alouane <zied.benhadjalouane@gmail.com>
 Ziang Wu <97337387+ZiangWu-77@users.noreply.github.com>
@@ -575,18 +484,14 @@ Zsapi <martin1.zsapka@gmail.com>
 a-n-n-a-l-e-e <150648636+a-n-n-a-l-e-e@users.noreply.github.com>
 adel boussaken <netdur@gmail.com>
 afrideva <95653597+afrideva@users.noreply.github.com>
-agray3 <agray3@users.noreply.github.com>
 akawrykow <142945436+akawrykow@users.noreply.github.com>
 alexpinel <93524949+alexpinel@users.noreply.github.com>
 alonfaraj <alonfaraj@gmail.com>
-alwqx <kenan3015@gmail.com>
-amd-lalithnc <lalithnc@amd.com>
 andrijdavid <david@geek.mg>
 anon998 <131767832+anon998@users.noreply.github.com>
 anzz1 <anzz1@live.com>
 apaz <aarpazdera@gmail.com>
 apcameron <37645737+apcameron@users.noreply.github.com>
-arch-btw <57669023+arch-btw@users.noreply.github.com>
 arcrank <arcrank@gmail.com>
 arlo-phoenix <140345165+arlo-phoenix@users.noreply.github.com>
 at8u <129688334+at8u@users.noreply.github.com>
@@ -609,17 +514,13 @@ cocktailpeanut <121128867+cocktailpeanut@users.noreply.github.com>
 coezbek <c.oezbek@gmail.com>
 comex <comexk@gmail.com>
 compilade <113953597+compilade@users.noreply.github.com>
-compilade <git@compilade.net>
-cpumaxx <163466046+cpumaxx@users.noreply.github.com>
 crasm <crasm@git.vczf.net>
 crasm <crasm@git.vczf.us>
 daboe01 <daboe01@googlemail.com>
 david raistrick <keen99@users.noreply.github.com>
-ddh0 <dylanhalladay02@icloud.com>
 ddpasa <112642920+ddpasa@users.noreply.github.com>
 deepdiffuser <112834445+deepdiffuser@users.noreply.github.com>
 divinity76 <divinity76@gmail.com>
-dm4 <sunrisedm4@gmail.com>
 dotpy314 <33351922+dotpy314@users.noreply.github.com>
 drbh <david.richard.holtz@gmail.com>
 ds5t5 <145942675+ds5t5@users.noreply.github.com>
@@ -628,7 +529,6 @@ eastriver <lee@eastriver.dev>
 ebraminio <ebraminio@gmail.com>
 eiery <19350831+eiery@users.noreply.github.com>
 eric8607242 <e0928021388@gmail.com>
-fairydreaming <166155368+fairydreaming@users.noreply.github.com>
 fraxy-v <65565042+fraxy-v@users.noreply.github.com>
 github-actions[bot] <github-actions[bot]@users.noreply.github.com>
 gliptic <gliptic@users.noreply.github.com>
@@ -639,7 +539,6 @@ h-h-h-h <13482553+h-h-h-h@users.noreply.github.com>
 hankcs <cnhankmc@gmail.com>
 hoangmit <hoangmit@users.noreply.github.com>
 hongbo.mo <352280764@qq.com>
-hopkins385 <98618192+hopkins385@users.noreply.github.com>
 howlger <eclipse@voormann.de>
 howlger <github@voormann.de>
 hutli <6594598+hutli@users.noreply.github.com>
@@ -650,22 +549,14 @@ hydai <z54981220@gmail.com>
 iSma <ismail.senhaji@gmail.com>
 iacore <74560659+iacore@users.noreply.github.com>
 igarnier <igarnier@protonmail.com>
-intelmatt <61025942+intelmatt@users.noreply.github.com>
 iohub <rickyang.pro@gmail.com>
 jacobi petrucciani <8117202+jpetrucciani@users.noreply.github.com>
-jaime-m-p <167997752+jaime-m-p@users.noreply.github.com>
 jameswu2014 <545426914@qq.com>
-jiez <373447296@qq.com>
 jneem <joeneeman@gmail.com>
-joecryptotoo <80373433+joecryptotoo@users.noreply.github.com>
 johnson442 <56517414+johnson442@users.noreply.github.com>
-jojorne <jojorne@users.noreply.github.com>
 jon-chuang <9093549+jon-chuang@users.noreply.github.com>
 jp-x-g <jpxg-dev@protonmail.com>
-jukofyork <69222624+jukofyork@users.noreply.github.com>
-junchao-loongson <68935141+junchao-loongson@users.noreply.github.com>
 jwj7140 <32943891+jwj7140@users.noreply.github.com>
-k.h.lai <adrian.k.h.lai@outlook.com>
 kaizau <kaizau@users.noreply.github.com>
 kalomaze <66376113+kalomaze@users.noreply.github.com>
 kang <tpdns9032100@gmail.com>
@@ -684,15 +575,11 @@ ldwang <ftgreat@163.com>
 le.chang <cljs118@126.com>
 leejet <leejet714@gmail.com>
 limitedAtonement <limitedAtonement@users.noreply.github.com>
-liuwei-git <14815172+liuwei-git@users.noreply.github.com>
 lon <114724657+longregen@users.noreply.github.com>
-loonerin <132926317+loonerin@users.noreply.github.com>
-luoyu-intel <yu.luo@intel.com>
 m3ndax <adrian.goessl@outlook.com>
 maddes8cht <55592906+maddes8cht@users.noreply.github.com>
 makomk <makosoft@googlemail.com>
 manikbhandari <mbbhandarimanik2@gmail.com>
-maor-ps <154728172+maor-ps@users.noreply.github.com>
 mdrokz <mohammadmunshi@gmail.com>
 mgroeber9110 <45620825+mgroeber9110@users.noreply.github.com>
 minarchist <minarchist@users.noreply.github.com>
@@ -706,19 +593,15 @@ ngc92 <7938269+ngc92@users.noreply.github.com>
 nhamanasu <45545786+nhamanasu@users.noreply.github.com>
 niansa/tuxifan <anton-sa@web.de>
 niansa/tuxifan <tuxifan@posteo.de>
-nickp27 <nb.porter@gmail.com>
 ningshanwutuobang <ningshanwutuobang@gmail.com>
 nold <Nold360@users.noreply.github.com>
 nopperl <54780682+nopperl@users.noreply.github.com>
 nusu-github <29514220+nusu-github@users.noreply.github.com>
 olexiyb <olexiyb@gmail.com>
-omahs <73983677+omahs@users.noreply.github.com>
 oobabooga <112222186+oobabooga@users.noreply.github.com>
 opparco <parco.opaai@gmail.com>
 ostix360 <55257054+ostix360@users.noreply.github.com>
-pengxin99 <pengxin.yuan@intel.com>
 perserk <perserk@gmail.com>
-pmysl <piotr.myslinski@outlook.com>
 postmasters <namnguyen@google.com>
 pudepiedj <pudepiedj@gmail.com>
 qingfengfenga <41416092+qingfengfenga@users.noreply.github.com>
@@ -731,19 +614,16 @@ rhuddleston <ryan.huddleston@percona.com>
 rimoliga <53384203+rimoliga@users.noreply.github.com>
 runfuture <runfuture@users.noreply.github.com>
 sandyiscool <sandyiscool@gmail.com>
-sasha0552 <admin@sasha0552.org>
 semidark <me@semidark.net>
 sharpHL <132747147+sharpHL@users.noreply.github.com>
 shibe2 <shibe@tuta.io>
 singularity <12184989+singularity-s0@users.noreply.github.com>
 sjinzh <sjinzh@gmail.com>
-sjxx <63994076+ylsdamxssjxxdd@users.noreply.github.com>
 slaren <2141330+slaren@users.noreply.github.com>
 slaren <slarengh@gmail.com>
 snadampal <87143774+snadampal@users.noreply.github.com>
 staviq <staviq@gmail.com>
 stduhpf <stephduh@live.fr>
-strawberrymelonpanda <152940198+strawberrymelonpanda@users.noreply.github.com>
 swittk <switt1995@gmail.com>
 takov751 <40316768+takov751@users.noreply.github.com>
 tarcey <cey.tarik@gmail.com>
@@ -756,16 +636,12 @@ uint256_t <konndennsa@gmail.com>
 uint256_t <maekawatoshiki1017@gmail.com>
 unbounded <haakon@likedan.net>
 valiray <133289098+valiray@users.noreply.github.com>
-vik <vikhyatk@gmail.com>
-viric <viric@viric.name>
 vodkaslime <646329483@qq.com>
 vvhg1 <94630311+vvhg1@users.noreply.github.com>
 vxiiduu <73044267+vxiiduu@users.noreply.github.com>
 wbpxre150 <100937007+wbpxre150@users.noreply.github.com>
 whoreson <139810751+whoreson@users.noreply.github.com>
-woachk <24752637+woachk@users.noreply.github.com>
 wonjun Jang <strutive07@gmail.com>
-woodx <124784234+woodx9@users.noreply.github.com>
 wzy <32936898+Freed-Wu@users.noreply.github.com>
 xaedes <xaedes@gmail.com>
 xaedes <xaedes@googlemail.com>
@@ -773,10 +649,7 @@ xloem <0xloem@gmail.com>
 yangli2 <yangli2@gmail.com>
 yuiseki <yuiseki@gmail.com>
 zakkor <edward.partenie@gmail.com>
-zhangkaihuo <zhangkaihuo@gmail.com>
 zhouwg <6889919+zhouwg@users.noreply.github.com>
-zhouwg <zhouwg2000@gmail.com>
 zrm <trustiosity.zrm@gmail.com>
-Ștefan-Gabriel Muscalu <legraphista@users.noreply.github.com>
 源文雨 <41315874+fumiama@users.noreply.github.com>
 Нияз Гарифзянов <112617865+garrnizon@users.noreply.github.com>
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
--- a/CMakePresets.json
+++ b/CMakePresets.json
@@ -19,15 +19,14 @@
        "cacheVariables": {
            "CMAKE_EXPORT_COMPILE_COMMANDS": "ON",
            "CMAKE_CXX_COMPILER": "icx",
-            "CMAKE_C_COMPILER": "cl",
-            "GGML_SYCL": "ON",
+            "LLAMA_SYCL": "ON",
            "CMAKE_INSTALL_RPATH": "$ORIGIN;$ORIGIN/.."
        }
    },
    { "name": "debug",   "hidden": true, "cacheVariables": { "CMAKE_BUILD_TYPE": "Debug" } },
    { "name": "release", "hidden": true, "cacheVariables": { "CMAKE_BUILD_TYPE": "Release" } },
-    { "name": "reldbg",  "hidden": true, "cacheVariables": { "CMAKE_BUILD_TYPE": "RelWithDebInfo" } },
-    { "name": "static",  "hidden": true, "cacheVariables": { "GGML_STATIC": "ON" } },
+    { "name": "reldbg", "hidden": true, "cacheVariables": { "CMAKE_BUILD_TYPE": "RelWithDebInfo" } },
+    { "name": "static",  "hidden": true, "cacheVariables": { "LLAMA_STATIC": "ON" } },

    {
        "name": "arm64-windows-msvc", "hidden": true,
--- a/1075
+++ b/1075
--- a/Package.swift
+++ b/Package.swift
@@ -3,13 +3,14 @@
 import PackageDescription

 var sources = [
-    "src/llama.cpp",
-    "src/unicode.cpp",
-    "src/unicode-data.cpp",
-    "ggml/src/ggml.c",
-    "ggml/src/ggml-alloc.c",
-    "ggml/src/ggml-backend.c",
-    "ggml/src/ggml-quants.c",
+    "ggml.c",
+    "sgemm.cpp",
+    "llama.cpp",
+    "unicode.cpp",
+    "unicode-data.cpp",
+    "ggml-alloc.c",
+    "ggml-backend.c",
+    "ggml-quants.c",
 ]

 var resources: [Resource] = []
@@ -25,8 +26,8 @@ var cSettings: [CSetting] =  [
 ]

 #if canImport(Darwin)
-sources.append("ggml/src/ggml-metal.m")
-resources.append(.process("ggml/src/ggml-metal.metal"))
+sources.append("ggml-metal.m")
+resources.append(.process("ggml-metal.metal"))
 linkerSettings.append(.linkedFramework("Accelerate"))
 cSettings.append(
    contentsOf: [
@@ -62,6 +63,8 @@ let package = Package(
               "models",
               "tests",
               "CMakeLists.txt",
+               "ggml-cuda.cu",
+               "ggml-cuda.h",
               "Makefile"
            ],
            sources: sources,
--- a/README-sycl.md
+++ b/README-sycl.md
@@ -115,12 +115,12 @@ The docker build option is currently limited to *intel GPU* targets.
 ### Build image
 ```sh
 # Using FP16
-docker build -t llama-cpp-sycl --build-arg="GGML_SYCL_F16=ON" -f .devops/llama-cli-intel.Dockerfile .
+docker build -t llama-cpp-sycl --build-arg="LLAMA_SYCL_F16=ON" -f .devops/llama-cli-intel.Dockerfile .
 ```

 *Notes*:

-To build in default FP32 *(Slower than FP16 alternative)*, you can remove the `--build-arg="GGML_SYCL_F16=ON"` argument from the previous command.
+To build in default FP32 *(Slower than FP16 alternative)*, you can remove the `--build-arg="LLAMA_SYCL_F16=ON"` argument from the previous command.

 You can also use the `.devops/llama-server-intel.Dockerfile`, which builds the *"server"* alternative.

@@ -244,10 +244,10 @@ source /opt/intel/oneapi/setvars.sh
 # Build LLAMA with MKL BLAS acceleration for intel GPU

 # Option 1: Use FP32 (recommended for better performance in most cases)
-cmake -B build -DGGML_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx
+cmake -B build -DLLAMA_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx

 # Option 2: Use FP16
-cmake -B build -DGGML_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx -DGGML_SYCL_F16=ON
+cmake -B build -DLLAMA_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx -DLLAMA_SYCL_F16=ON

 # build all binary
 cmake --build build --config Release -j -v
@@ -264,10 +264,10 @@ export CPLUS_INCLUDE_DIR=/path/to/oneMKL/include:$CPLUS_INCLUDE_DIR
 # Build LLAMA with Nvidia BLAS acceleration through SYCL

 # Option 1: Use FP32 (recommended for better performance in most cases)
-cmake -B build -DGGML_SYCL=ON -DGGML_SYCL_TARGET=NVIDIA -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx
+cmake -B build -DLLAMA_SYCL=ON -DLLAMA_SYCL_TARGET=NVIDIA -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx

 # Option 2: Use FP16
-cmake -B build -DGGML_SYCL=ON -DGGML_SYCL_TARGET=NVIDIA -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx -DGGML_SYCL_F16=ON
+cmake -B build -DLLAMA_SYCL=ON -DLLAMA_SYCL_TARGET=NVIDIA -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx -DLLAMA_SYCL_F16=ON

 # build all binary
 cmake --build build --config Release -j -v
@@ -422,10 +422,10 @@ On the oneAPI command line window, step into the llama.cpp main directory and ru
@call "C:\Program Files (x86)\Intel\oneAPI\setvars.bat" intel64 --force

 # Option 1: Use FP32 (recommended for better performance in most cases)
-cmake -B build -G "Ninja" -DGGML_SYCL=ON -DCMAKE_C_COMPILER=cl -DCMAKE_CXX_COMPILER=icx  -DCMAKE_BUILD_TYPE=Release
+cmake -B build -G "Ninja" -DLLAMA_SYCL=ON -DCMAKE_C_COMPILER=cl -DCMAKE_CXX_COMPILER=icx  -DCMAKE_BUILD_TYPE=Release

 # Option 2: Or FP16
-cmake -B build -G "Ninja" -DGGML_SYCL=ON -DCMAKE_C_COMPILER=cl -DCMAKE_CXX_COMPILER=icx  -DCMAKE_BUILD_TYPE=Release -DGGML_SYCL_F16=ON
+cmake -B build -G "Ninja" -DLLAMA_SYCL=ON -DCMAKE_C_COMPILER=cl -DCMAKE_CXX_COMPILER=icx  -DCMAKE_BUILD_TYPE=Release -DLLAMA_SYCL_F16=ON

 cmake --build build --config Release -j
 ```
@@ -440,7 +440,7 @@ Or, use CMake presets to build:
 cmake --preset x64-windows-sycl-release
 cmake --build build-x64-windows-sycl-release -j --target llama-cli

-cmake -DGGML_SYCL_F16=ON --preset x64-windows-sycl-release
+cmake -DLLAMA_SYCL_F16=ON --preset x64-windows-sycl-release
 cmake --build build-x64-windows-sycl-release -j --target llama-cli

 cmake --preset x64-windows-sycl-debug
@@ -544,9 +544,9 @@ use 1 SYCL GPUs: [0] with Max compute units:512

 | Name               | Value                             | Function                                    |
 |--------------------|-----------------------------------|---------------------------------------------|
-| GGML_SYCL          | ON (mandatory)                    | Enable build with SYCL code path.           |
-| GGML_SYCL_TARGET   | INTEL *(default)* \| NVIDIA       | Set the SYCL target device type.            |
-| GGML_SYCL_F16      | OFF *(default)* \|ON *(optional)* | Enable FP16 build with SYCL code path.      |
+| LLAMA_SYCL         | ON (mandatory)                    | Enable build with SYCL code path.           |
+| LLAMA_SYCL_TARGET  | INTEL *(default)* \| NVIDIA       | Set the SYCL target device type.            |
+| LLAMA_SYCL_F16     | OFF *(default)* \|ON *(optional)* | Enable FP16 build with SYCL code path.      |
 | CMAKE_C_COMPILER   | icx                               | Set *icx* compiler for SYCL code path.      |
 | CMAKE_CXX_COMPILER | icpx *(Linux)*, icx *(Windows)*   | Set `icpx/icx` compiler for SYCL code path. |

--- a/README.md
+++ b/README.md
@@ -15,7 +15,6 @@ Inference of Meta's [LLaMA](https://arxiv.org/abs/2302.13971) model (and others)

 ### Recent API changes

- [2024 Jun 26] The source code and CMake build scripts have been restructured https://github.com/ggerganov/llama.cpp/pull/8006
 - [2024 Apr 21] `llama_token_to_piece` can now optionally render special tokens https://github.com/ggerganov/llama.cpp/pull/6807
 - [2024 Apr 4] State and session file functions reorganized under `llama_state_*` https://github.com/ggerganov/llama.cpp/pull/6341
 - [2024 Mar 26] Logits and embeddings API updated for compactness https://github.com/ggerganov/llama.cpp/pull/6122
@@ -108,7 +107,6 @@ Typically finetunes of the base models below are supported as well.
 - [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)
- [X] [BERT](https://github.com/ggerganov/llama.cpp/pull/5423)
 - [X] [Koala](https://bair.berkeley.edu/blog/2023/04/03/koala/)
 - [X] [Baichuan 1 & 2](https://huggingface.co/models?search=baichuan-inc/Baichuan) + [derivations](https://huggingface.co/hiyouga/baichuan-7b-sft)
 - [X] [Aquila 1 & 2](https://huggingface.co/models?search=BAAI/Aquila)
@@ -218,11 +216,6 @@ Unless otherwise noted these projects are open-source with permissive licensing:
 **Tools:**

 - [akx/ggify](https://github.com/akx/ggify) – download PyTorch models from HuggingFace Hub and convert them to GGML
- [crashr/gppm](https://github.com/crashr/gppm) – launch llama.cpp instances utilizing NVIDIA Tesla P40 or P100 GPUs with reduced idle power consumption
-
-**Infrastructure:**
-
- [Paddler](https://github.com/distantmagic/paddler) - Stateful load balancer custom-tailored for llama.cpp

 ---

@@ -422,7 +415,7 @@ Flox follows the nixpkgs build of llama.cpp.
 ### Metal Build

 On MacOS, Metal is enabled by default. Using Metal makes the computation run on the GPU.
-To disable the Metal build at compile time use the `GGML_NO_METAL=1` flag or the `GGML_METAL=OFF` cmake option.
+To disable the Metal build at compile time use the `LLAMA_NO_METAL=1` flag or the `LLAMA_METAL=OFF` cmake option.

 When built with Metal support, you can explicitly disable GPU inference with the `--n-gpu-layers|-ngl 0` command-line
 argument.
@@ -442,7 +435,7 @@ Building the program with BLAS support may lead to some performance improvements
  - Using `make`:
    - On Linux:
      ```bash
-      make GGML_OPENBLAS=1
+      make LLAMA_OPENBLAS=1
      ```

    - On Windows:
@@ -457,13 +450,13 @@ Building the program with BLAS support may lead to some performance improvements
      8. From here you can run:

          ```bash
-          make GGML_OPENBLAS=1
+          make LLAMA_OPENBLAS=1
          ```

  - Using `CMake` on Linux:

      ```bash
-      cmake -B build -DGGML_BLAS=ON -DGGML_BLAS_VENDOR=OpenBLAS
+      cmake -B build -DLLAMA_BLAS=ON -DLLAMA_BLAS_VENDOR=OpenBLAS
      cmake --build build --config Release
      ```

@@ -482,10 +475,10 @@ Building the program with BLAS support may lead to some performance improvements
  Building through oneAPI compilers will make avx_vnni instruction set available for intel processors that do not support avx512 and avx512_vnni. Please note that this build config **does not support Intel GPU**. For Intel GPU support, please refer to [llama.cpp for SYCL](./README-sycl.md).

  - Using manual oneAPI installation:
-    By default, `GGML_BLAS_VENDOR` is set to `Generic`, so if you already sourced intel environment script and assign `-DGGML_BLAS=ON` in cmake, the mkl version of Blas will automatically been selected. Otherwise please install oneAPI and follow the below steps:
+    By default, `LLAMA_BLAS_VENDOR` is set to `Generic`, so if you already sourced intel environment script and assign `-DLLAMA_BLAS=ON` in cmake, the mkl version of Blas will automatically been selected. Otherwise please install oneAPI and follow the below steps:
      ```bash
      source /opt/intel/oneapi/setvars.sh # You can skip this step if  in oneapi-basekit docker image, only required for manual installation
-      cmake -B build -DGGML_BLAS=ON -DGGML_BLAS_VENDOR=Intel10_64lp -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx -DGGML_NATIVE=ON
+      cmake -B build -DLLAMA_BLAS=ON -DLLAMA_BLAS_VENDOR=Intel10_64lp -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx -DLLAMA_NATIVE=ON
      cmake --build build --config Release
      ```

@@ -502,28 +495,28 @@ Building the program with BLAS support may lead to some performance improvements

  - Using `make`:
    ```bash
-    make GGML_CUDA=1
+    make LLAMA_CUDA=1
    ```
  - Using `CMake`:

    ```bash
-    cmake -B build -DGGML_CUDA=ON
+    cmake -B build -DLLAMA_CUDA=ON
    cmake --build build --config Release
    ```

  The environment variable [`CUDA_VISIBLE_DEVICES`](https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#env-vars) can be used to specify which GPU(s) will be used. The following compilation options are also available to tweak performance:

-  | Option                        | Legal values           | Default | Description                                                                                                                                                                                                                                                                             |
-  |-------------------------------|------------------------|---------|-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
-  | GGML_CUDA_FORCE_DMMV          | Boolean                | false   | Force the use of dequantization + matrix vector multiplication kernels instead of using kernels that do matrix vector multiplication on quantized data. By default the decision is made based on compute capability (MMVQ for 6.1/Pascal/GTX 1000 or higher). Does not affect k-quants. |
-  | GGML_CUDA_DMMV_X              | Positive integer >= 32 | 32      | Number of values in x direction processed by the CUDA dequantization + matrix vector multiplication kernel per iteration. Increasing this value can improve performance on fast GPUs. Power of 2 heavily recommended. Does not affect k-quants.                                         |
-  | GGML_CUDA_MMV_Y               | Positive integer       | 1       | Block size in y direction for the CUDA mul mat vec kernels. Increasing this value can improve performance on fast GPUs. Power of 2 recommended.                                                                                                                                         |
-  | GGML_CUDA_FORCE_MMQ           | Boolean                | false   | Force the use of custom matrix multiplication kernels for quantized models instead of FP16 cuBLAS even if there is no int8 tensor core implementation available (affects V100, RDNA3). MMQ kernels are enabled by default on GPUs with int8 tensor core support. With MMQ force enabled, speed for large batch sizes will be worse but VRAM consumption will be lower.                       |
-  | GGML_CUDA_FORCE_CUBLAS        | Boolean                | false   | Force the use of FP16 cuBLAS instead of custom matrix multiplication kernels for quantized models                                                                                                                                                                                       |
-  | GGML_CUDA_F16                 | Boolean                | false   | If enabled, use half-precision floating point arithmetic for the CUDA dequantization + mul mat vec kernels and for the q4_1 and q5_1 matrix matrix multiplication kernels. Can improve performance on relatively recent GPUs.                                                           |
-  | GGML_CUDA_KQUANTS_ITER        | 1 or 2                 | 2       | Number of values processed per iteration and per CUDA thread for Q2_K and Q6_K quantization formats. Setting this value to 1 can improve performance for slow GPUs.                                                                                                                     |
-  | GGML_CUDA_PEER_MAX_BATCH_SIZE | Positive integer       | 128     | Maximum batch size for which to enable peer access between multiple GPUs. Peer access requires either Linux or NVLink. When using NVLink enabling peer access for larger batch sizes is potentially beneficial.                                                                         |
-  | GGML_CUDA_FA_ALL_QUANTS       | Boolean                | false   | Compile support for all KV cache quantization type (combinations) for the FlashAttention CUDA kernels. More fine-grained control over KV cache size but compilation takes much longer.                                                                                                  |
+  | Option                         | Legal values           | Default | Description                                                                                                                                                                                                                                                                             |
+  |--------------------------------|------------------------|---------|-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
+  | LLAMA_CUDA_FORCE_DMMV          | Boolean                | false   | Force the use of dequantization + matrix vector multiplication kernels instead of using kernels that do matrix vector multiplication on quantized data. By default the decision is made based on compute capability (MMVQ for 6.1/Pascal/GTX 1000 or higher). Does not affect k-quants. |
+  | LLAMA_CUDA_DMMV_X              | Positive integer >= 32 | 32      | Number of values in x direction processed by the CUDA dequantization + matrix vector multiplication kernel per iteration. Increasing this value can improve performance on fast GPUs. Power of 2 heavily recommended. Does not affect k-quants.                                         |
+  | LLAMA_CUDA_MMV_Y               | Positive integer       | 1       | Block size in y direction for the CUDA mul mat vec kernels. Increasing this value can improve performance on fast GPUs. Power of 2 recommended.                                                                                                                                         |
+  | LLAMA_CUDA_FORCE_MMQ           | Boolean                | false   | Force the use of custom matrix multiplication kernels for quantized models instead of FP16 cuBLAS even if there is no int8 tensor core implementation available (affects V100, RDNA3). MMQ kernels are enabled by default on GPUs with int8 tensor core support. With MMQ force enabled, speed for large batch sizes will be worse but VRAM consumption will be lower.                       |
+  | LLAMA_CUDA_FORCE_CUBLAS        | Boolean                | false   | Force the use of FP16 cuBLAS instead of custom matrix multiplication kernels for quantized models                                                                                                                                                                                       |
+  | LLAMA_CUDA_F16                 | Boolean                | false   | If enabled, use half-precision floating point arithmetic for the CUDA dequantization + mul mat vec kernels and for the q4_1 and q5_1 matrix matrix multiplication kernels. Can improve performance on relatively recent GPUs.                                                           |
+  | LLAMA_CUDA_KQUANTS_ITER        | 1 or 2                 | 2       | Number of values processed per iteration and per CUDA thread for Q2_K and Q6_K quantization formats. Setting this value to 1 can improve performance for slow GPUs.                                                                                                                     |
+  | LLAMA_CUDA_PEER_MAX_BATCH_SIZE | Positive integer       | 128     | Maximum batch size for which to enable peer access between multiple GPUs. Peer access requires either Linux or NVLink. When using NVLink enabling peer access for larger batch sizes is potentially beneficial.                                                                         |
+  | LLAMA_CUDA_FA_ALL_QUANTS       | Boolean                | false   | Compile support for all KV cache quantization type (combinations) for the FlashAttention CUDA kernels. More fine-grained control over KV cache size but compilation takes much longer.                                                                                                  |

 - #### hipBLAS

@@ -533,15 +526,15 @@ Building the program with BLAS support may lead to some performance improvements

  - Using `make`:
    ```bash
-    make GGML_HIPBLAS=1
+    make LLAMA_HIPBLAS=1
    ```
  - Using `CMake` for Linux (assuming a gfx1030-compatible AMD GPU):
    ```bash
    HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" \
-        cmake -S . -B build -DGGML_HIPBLAS=ON -DAMDGPU_TARGETS=gfx1030 -DCMAKE_BUILD_TYPE=Release \
+        cmake -S . -B build -DLLAMA_HIPBLAS=ON -DAMDGPU_TARGETS=gfx1030 -DCMAKE_BUILD_TYPE=Release \
        && cmake --build build --config Release -- -j 16
    ```
-    On Linux it is also possible to use unified memory architecture (UMA) to share main memory between the CPU and integrated GPU by setting `-DGGML_HIP_UMA=ON`.
+    On Linux it is also possible to use unified memory architecture (UMA) to share main memory between the CPU and integrated GPU by setting `-DLLAMA_HIP_UMA=ON`.
    However, this hurts performance for non-integrated GPUs (but enables working with integrated GPUs).

    Note that if you get the following error:
@@ -555,19 +548,19 @@ Building the program with BLAS support may lead to some performance improvements
    ```bash
    HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -p)" \
    HIP_DEVICE_LIB_PATH=<directory-you-just-found> \
-        cmake -S . -B build -DGGML_HIPBLAS=ON -DAMDGPU_TARGETS=gfx1030 -DCMAKE_BUILD_TYPE=Release \
+        cmake -S . -B build -DLLAMA_HIPBLAS=ON -DAMDGPU_TARGETS=gfx1030 -DCMAKE_BUILD_TYPE=Release \
        && cmake --build build -- -j 16
    ```

  - Using `make` (example for target gfx1030, build with 16 CPU threads):
    ```bash
-    make -j16 GGML_HIPBLAS=1 GGML_HIP_UMA=1 AMDGPU_TARGETS=gfx1030
+    make -j16 LLAMA_HIPBLAS=1 LLAMA_HIP_UMA=1 AMDGPU_TARGETS=gfx1030
    ```

  - Using `CMake` for Windows (using x64 Native Tools Command Prompt for VS, and assuming a gfx1100-compatible AMD GPU):
    ```bash
    set PATH=%HIP_PATH%\bin;%PATH%
-    cmake -S . -B build -G Ninja -DAMDGPU_TARGETS=gfx1100 -DGGML_HIPBLAS=ON -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ -DCMAKE_BUILD_TYPE=Release
+    cmake -S . -B build -G Ninja -DAMDGPU_TARGETS=gfx1100 -DLLAMA_HIPBLAS=ON -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ -DCMAKE_BUILD_TYPE=Release
    cmake --build build
    ```
    Make sure that `AMDGPU_TARGETS` is set to the GPU arch you want to compile for. The above example uses `gfx1100` that corresponds to Radeon RX 7900XTX/XT/GRE. You can find a list of targets [here](https://llvm.org/docs/AMDGPUUsage.html#processors)
@@ -578,11 +571,11 @@ Building the program with BLAS support may lead to some performance improvements
  If your GPU is not officially supported you can use the environment variable [`HSA_OVERRIDE_GFX_VERSION`] set to a similar GPU, for example 10.3.0 on RDNA2 (e.g. gfx1030, gfx1031, or gfx1035) or 11.0.0 on RDNA3.
  The following compilation options are also available to tweak performance (yes, they refer to CUDA, not HIP, because it uses the same code as the cuBLAS version above):

-  | Option                 | Legal values           | Default | Description                                                                                                                                                                                                                                    |
-  |------------------------|------------------------|---------|------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
-  | GGML_CUDA_DMMV_X       | Positive integer >= 32 | 32      | Number of values in x direction processed by the HIP dequantization + matrix vector multiplication kernel per iteration. Increasing this value can improve performance on fast GPUs. Power of 2 heavily recommended. Does not affect k-quants. |
-  | GGML_CUDA_MMV_Y        | Positive integer       | 1       | Block size in y direction for the HIP mul mat vec kernels. Increasing this value can improve performance on fast GPUs. Power of 2 recommended. Does not affect k-quants.                                                                       |
-  | GGML_CUDA_KQUANTS_ITER | 1 or 2                 | 2       | Number of values processed per iteration and per HIP thread for Q2_K and Q6_K quantization formats. Setting this value to 1 can improve performance for slow GPUs.                                                                             |
+  | Option                  | Legal values           | Default | Description                                                                                                                                                                                                                                    |
+  |-------------------------|------------------------|---------|------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
+  | LLAMA_CUDA_DMMV_X       | Positive integer >= 32 | 32      | Number of values in x direction processed by the HIP dequantization + matrix vector multiplication kernel per iteration. Increasing this value can improve performance on fast GPUs. Power of 2 heavily recommended. Does not affect k-quants. |
+  | LLAMA_CUDA_MMV_Y        | Positive integer       | 1       | Block size in y direction for the HIP mul mat vec kernels. Increasing this value can improve performance on fast GPUs. Power of 2 recommended. Does not affect k-quants.                                                                       |
+  | LLAMA_CUDA_KQUANTS_ITER | 1 or 2                 | 2       | Number of values processed per iteration and per HIP thread for Q2_K and Q6_K quantization formats. Setting this value to 1 can improve performance for slow GPUs.                                                                             |

 - #### Vulkan

@@ -620,7 +613,7 @@ Building the program with BLAS support may lead to some performance improvements
  Then, build llama.cpp using the cmake command below:

  ```bash
-  cmake -B build -DGGML_VULKAN=1
+  cmake -B build -DLLAMA_VULKAN=1
  cmake --build build --config Release
  # Test the output binary (with "-ngl 33" to offload all layers to GPU)
  ./bin/llama-cli -m "PATH_TO_MODEL" -p "Hi you how are you" -n 50 -e -ngl 33 -t 4
--- a/ci/run.sh
+++ b/ci/run.sh
@@ -36,11 +36,11 @@ SRC=`pwd`
 CMAKE_EXTRA="-DLLAMA_FATAL_WARNINGS=ON"

 if [ ! -z ${GG_BUILD_METAL} ]; then
-    CMAKE_EXTRA="${CMAKE_EXTRA} -DGGML_METAL=ON"
+    CMAKE_EXTRA="${CMAKE_EXTRA} -DLLAMA_METAL_SHADER_DEBUG=ON"
 fi

 if [ ! -z ${GG_BUILD_CUDA} ]; then
-    CMAKE_EXTRA="${CMAKE_EXTRA} -DGGML_CUDA=1"
+    CMAKE_EXTRA="${CMAKE_EXTRA} -DLLAMA_CUDA=1"
 fi

 if [ ! -z ${GG_BUILD_SYCL} ]; then
@@ -50,7 +50,7 @@ if [ ! -z ${GG_BUILD_SYCL} ]; then
        exit 1
    fi

-    CMAKE_EXTRA="${CMAKE_EXTRA} -DGGML_SYCL=1 DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx -DGGML_SYCL_F16=ON"
+    CMAKE_EXTRA="${CMAKE_EXTRA} -DLLAMA_SYCL=1 DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx -DLLAMA_SYCL_F16=ON"
 fi
 ## helpers

@@ -284,7 +284,7 @@ function gg_run_open_llama_7b_v2 {

    set -e

-    (time cmake -DCMAKE_BUILD_TYPE=Release ${CMAKE_EXTRA} -DGGML_CUDA=1 .. ) 2>&1 | tee -a $OUT/${ci}-cmake.log
+    (time cmake -DCMAKE_BUILD_TYPE=Release ${CMAKE_EXTRA} -DLLAMA_CUDA=1 .. ) 2>&1 | tee -a $OUT/${ci}-cmake.log
    (time make -j                                                           ) 2>&1 | tee -a $OUT/${ci}-make.log

    python3 ../examples/convert-legacy-llama.py ${path_models} --outfile ${path_models}/ggml-model-f16.gguf
@@ -550,7 +550,7 @@ function gg_run_pythia_2_8b {

    set -e

-    (time cmake -DCMAKE_BUILD_TYPE=Release ${CMAKE_EXTRA} -DGGML_CUDA=1 .. ) 2>&1 | tee -a $OUT/${ci}-cmake.log
+    (time cmake -DCMAKE_BUILD_TYPE=Release ${CMAKE_EXTRA} -DLLAMA_CUDA=1 .. ) 2>&1 | tee -a $OUT/${ci}-cmake.log
    (time make -j                                                           ) 2>&1 | tee -a $OUT/${ci}-make.log

    python3 ../convert-hf-to-gguf.py ${path_models} --outfile ${path_models}/ggml-model-f16.gguf
--- a/ggml/cmake/FindSIMD.cmake
+++ b/ggml/cmake/FindSIMD.cmake
@@ -79,22 +79,22 @@ endmacro()
 # flags are for MSVC only!
 check_sse("AVX" " ;/arch:AVX")
 if (NOT ${AVX_FOUND})
-    set(GGML_AVX OFF)
+    set(LLAMA_AVX OFF)
 else()
-    set(GGML_AVX ON)
+    set(LLAMA_AVX ON)
 endif()

 check_sse("AVX2" " ;/arch:AVX2")
 check_sse("FMA" " ;/arch:AVX2")
 if ((NOT ${AVX2_FOUND}) OR (NOT ${FMA_FOUND}))
-    set(GGML_AVX2 OFF)
+    set(LLAMA_AVX2 OFF)
 else()
-    set(GGML_AVX2 ON)
+    set(LLAMA_AVX2 ON)
 endif()

 check_sse("AVX512" " ;/arch:AVX512")
 if (NOT ${AVX512_FOUND})
-    set(GGML_AVX512 OFF)
+    set(LLAMA_AVX512 OFF)
 else()
-    set(GGML_AVX512 ON)
+    set(LLAMA_AVX512 ON)
 endif()
--- a/cmake/git-vars.cmake
+++ b/cmake/git-vars.cmake
@@ -1,22 +0,0 @@
-find_package(Git)
-
-# the commit's SHA1
-execute_process(COMMAND
-    "${GIT_EXECUTABLE}" describe --match=NeVeRmAtCh --always --abbrev=8
-    WORKING_DIRECTORY "${CMAKE_SOURCE_DIR}"
-    OUTPUT_VARIABLE GIT_SHA1
-    ERROR_QUIET OUTPUT_STRIP_TRAILING_WHITESPACE)
-
-# the date of the commit
-execute_process(COMMAND
-    "${GIT_EXECUTABLE}" log -1 --format=%ad --date=local
-    WORKING_DIRECTORY "${CMAKE_SOURCE_DIR}"
-    OUTPUT_VARIABLE GIT_DATE
-    ERROR_QUIET OUTPUT_STRIP_TRAILING_WHITESPACE)
-
-# the subject of the commit
-execute_process(COMMAND
-    "${GIT_EXECUTABLE}" log -1 --format=%s
-    WORKING_DIRECTORY "${CMAKE_SOURCE_DIR}"
-    OUTPUT_VARIABLE GIT_COMMIT_SUBJECT
-    ERROR_QUIET OUTPUT_STRIP_TRAILING_WHITESPACE)
--- a/common/CMakeLists.txt
+++ b/common/CMakeLists.txt
@@ -1,6 +1,5 @@
 # common

-find_package(Threads REQUIRED)

 # Build info header
 #
@@ -37,7 +36,7 @@ add_custom_command(
    COMMENT "Generating build details from Git"
    COMMAND ${CMAKE_COMMAND} -DMSVC=${MSVC} -DCMAKE_C_COMPILER_VERSION=${CMAKE_C_COMPILER_VERSION}
            -DCMAKE_C_COMPILER_ID=${CMAKE_C_COMPILER_ID} -DCMAKE_VS_PLATFORM_NAME=${CMAKE_VS_PLATFORM_NAME}
-            -DCMAKE_C_COMPILER=${CMAKE_C_COMPILER} -P "${CMAKE_CURRENT_SOURCE_DIR}/cmake/build-info-gen-cpp.cmake"
+            -DCMAKE_C_COMPILER=${CMAKE_C_COMPILER} -P "${CMAKE_CURRENT_SOURCE_DIR}/../scripts/gen-build-info-cpp.cmake"
    WORKING_DIRECTORY "${CMAKE_CURRENT_SOURCE_DIR}/.."
    DEPENDS "${CMAKE_CURRENT_SOURCE_DIR}/build-info.cpp.in" ${GIT_INDEX}
    VERBATIM
@@ -84,5 +83,5 @@ if (LLAMA_CURL)
 endif ()

 target_include_directories(${TARGET} PUBLIC .)
-target_compile_features   (${TARGET} PUBLIC cxx_std_11)
-target_link_libraries     (${TARGET} PRIVATE ${LLAMA_COMMON_EXTRA_LIBS} PUBLIC llama Threads::Threads)
+target_compile_features(${TARGET} PUBLIC cxx_std_11)
+target_link_libraries(${TARGET} PRIVATE ${LLAMA_COMMON_EXTRA_LIBS} PUBLIC llama Threads::Threads)
--- a/common/common.cpp
+++ b/common/common.cpp
@@ -757,7 +757,7 @@ bool gpt_params_find_arg(int argc, char ** argv, const std::string & arg, gpt_pa
        params.cache_type_v = argv[++i];
        return true;
    }
-    if (arg == "-mli" || arg == "--multiline-input") {
+    if (arg == "--multiline-input") {
        params.multiline_input = true;
        return true;
    }
@@ -1014,23 +1014,16 @@ bool gpt_params_find_arg(int argc, char ** argv, const std::string & arg, gpt_pa
    }
    if (arg == "--in-prefix-bos") {
        params.input_prefix_bos = true;
-        params.enable_chat_template = false;
        return true;
    }
    if (arg == "--in-prefix") {
        CHECK_ARG
        params.input_prefix = argv[i];
-        params.enable_chat_template = false;
        return true;
    }
    if (arg == "--in-suffix") {
        CHECK_ARG
        params.input_suffix = argv[i];
-        params.enable_chat_template = false;
-        return true;
-    }
-    if (arg == "--spm-infill") {
-        params.spm_infill = true;
        return true;
    }
    if (arg == "--grammar") {
@@ -1409,15 +1402,13 @@ void gpt_params_print_usage(int /*argc*/, char ** argv, const gpt_params & param
                                                                        "halt generation at PROMPT, return control in interactive mode\n"
                                                                        "can be specified more than once for multiple prompts" });
    options.push_back({ "main",        "-sp,   --special",              "special tokens output enabled (default: %s)", params.special ? "true" : "false" });
-    options.push_back({ "main",        "-cnv,  --conversation",         "run in conversation mode (does not print special tokens and suffix/prefix, use default chat template) (default: %s)", params.conversation ? "true" : "false" });
+    options.push_back({ "main",        "-cnv,  --conversation",         "run in conversation mode (does not print special tokens and suffix/prefix) (default: %s)", params.conversation ? "true" : "false" });
    options.push_back({ "main infill", "-i,    --interactive",          "run in interactive mode (default: %s)", params.interactive ? "true" : "false" });
    options.push_back({ "main infill", "-if,   --interactive-first",    "run in interactive mode and wait for input right away (default: %s)", params.interactive_first ? "true" : "false" });
    options.push_back({ "main infill", "-mli,  --multiline-input",      "allows you to write or paste multiple lines without ending each in '\\'" });
    options.push_back({ "main infill", "       --in-prefix-bos",        "prefix BOS to user inputs, preceding the `--in-prefix` string" });
    options.push_back({ "main infill", "       --in-prefix STRING",     "string to prefix user inputs with (default: empty)" });
    options.push_back({ "main infill", "       --in-suffix STRING",     "string to suffix after user inputs with (default: empty)" });
-    options.push_back({ "server infill",
-                                       "       --spm-infill",           "use Suffix/Prefix/Middle pattern for infill (instead of Prefix/Suffix/Middle) as some models prefer this. (default: %s)", params.spm_infill ? "enabled" : "disabled" });

    options.push_back({ "sampling" });
    options.push_back({ "*",           "       --samplers SAMPLERS",    "samplers that will be used for generation in the order, separated by \';\'\n"
@@ -2627,7 +2618,6 @@ std::string llama_chat_apply_template(const struct llama_model * model,
        const std::vector<llama_chat_msg> & msgs,
        bool add_ass) {
    int alloc_size = 0;
-    bool fallback = false; // indicate if we must fallback to default chatml
    std::vector<llama_chat_message> chat;
    for (auto & msg : msgs) {
        chat.push_back({msg.role.c_str(), msg.content.c_str()});
@@ -2640,26 +2630,10 @@ std::string llama_chat_apply_template(const struct llama_model * model,
    // run the first time to get the total output length
    int32_t res = llama_chat_apply_template(model, ptr_tmpl, chat.data(), chat.size(), add_ass, buf.data(), buf.size());

-    // error: chat template is not supported
-    if (res < 0) {
-        if (ptr_tmpl != nullptr) {
-            // if the custom "tmpl" is not supported, we throw an error
-            // this is a bit redundant (for good), since we're not sure if user validated the custom template with llama_chat_verify_template()
-            throw std::runtime_error("this custom template is not supported");
-        } else {
-            // If the built-in template is not supported, we default to chatml
-            res = llama_chat_apply_template(nullptr, "chatml", chat.data(), chat.size(), add_ass, buf.data(), buf.size());
-            fallback = true;
-        }
-    }
-
    // if it turns out that our buffer is too small, we resize it
    if ((size_t) res > buf.size()) {
        buf.resize(res);
-        res = llama_chat_apply_template(
-            fallback ? nullptr : model,
-            fallback ? "chatml" : ptr_tmpl,
-            chat.data(), chat.size(), add_ass, buf.data(), buf.size());
+        res = llama_chat_apply_template(model, ptr_tmpl, chat.data(), chat.size(), add_ass, buf.data(), buf.size());
    }

    std::string formatted_chat(buf.data(), res);
@@ -2671,19 +2645,12 @@ std::string llama_chat_format_single(const struct llama_model * model,
        const std::vector<llama_chat_msg> & past_msg,
        const llama_chat_msg & new_msg,
        bool add_ass) {
-    std::ostringstream ss;
    auto fmt_past_msg = llama_chat_apply_template(model, tmpl, past_msg, false);
    std::vector<llama_chat_msg> chat_new(past_msg);
-    // if the past_msg ends with a newline, we must preserve it in the formatted version
-    if (add_ass && !fmt_past_msg.empty() && fmt_past_msg.back() == '\n') {
-        ss << "\n";
-    };
-    // format chat with new_msg
    chat_new.push_back(new_msg);
    auto fmt_new_msg = llama_chat_apply_template(model, tmpl, chat_new, add_ass);
-    // get the diff part
-    ss << fmt_new_msg.substr(fmt_past_msg.size(), fmt_new_msg.size() - fmt_past_msg.size());
-    return ss.str();
+    auto formatted = fmt_new_msg.substr(fmt_past_msg.size(), fmt_new_msg.size() - fmt_past_msg.size());
+    return formatted;
 }

 std::string llama_chat_format_example(const struct llama_model * model,
@@ -2837,87 +2804,125 @@ float llama_embd_similarity_cos(const float * embd1, const float * embd2, int n)
 //

 static llama_control_vector_data llama_control_vector_load_one(const llama_control_vector_load_info & load_info) {
+    int32_t n_tensors;
+
+    size_t n_bytes = 0;
+
+    uint32_t max_direction_layer = 0;
+
    llama_control_vector_data result = { -1, {} };

-    ggml_context * ctx = nullptr;
-    struct gguf_init_params meta_gguf_params = {
-        /* .no_alloc = */ false,
-        /* .ctx      = */ &ctx,
-    };
-    struct gguf_context * ctx_gguf = gguf_init_from_file(load_info.fname.c_str(), meta_gguf_params);
-    if (!ctx_gguf) {
-        fprintf(stderr, "%s: failed to load control vector file from %s\n", __func__, load_info.fname.c_str());
+    // calculate size of ctx needed for tensors, ensure tensors are f32, and find max layer
+    {
+        struct ggml_init_params meta_params = {
+            /* .mem_size   = */ ggml_tensor_overhead() * 128 + ggml_graph_overhead(),
+            /* .mem_buffer = */ nullptr,
+            /* .no_alloc   = */ true,
+        };
+        ggml_context * meta_ctx = ggml_init(meta_params);
+        struct gguf_init_params meta_gguf_params = {
+            /* .no_alloc = */ true,
+            /* .ctx      = */ &meta_ctx,
+        };
+        struct gguf_context * meta_ctx_gguf = gguf_init_from_file(load_info.fname.c_str(), meta_gguf_params);
+        if (!meta_ctx_gguf) {
+            fprintf(stderr, "%s: failed to load control vector from %s\n", __func__, load_info.fname.c_str());
+            ggml_free(meta_ctx);
+            return result;
+        }
+
+        n_tensors = gguf_get_n_tensors(meta_ctx_gguf);
+        for (int i = 0; i < n_tensors; i++) {
+            std::string name = gguf_get_tensor_name(meta_ctx_gguf, i);
+
+            // split on '.'
+            size_t dotpos = name.find('.');
+            if (dotpos != std::string::npos && name.substr(0, dotpos) == "direction") {
+                try {
+                    uint32_t layer = std::stoi(name.substr(dotpos + 1));
+                    if (layer == 0) {
+                        fprintf(stderr, "%s: direction tensor invalid in %s\n", __func__, load_info.fname.c_str());
+                        ggml_free(meta_ctx);
+                        gguf_free(meta_ctx_gguf);
+                        return result;
+                    }
+                    if (layer > max_direction_layer) {
+                        max_direction_layer = layer;
+                    }
+                } catch (...) {
+                    fprintf(stderr, "%s: direction tensor invalid in %s\n", __func__, load_info.fname.c_str());
+                    ggml_free(meta_ctx);
+                    gguf_free(meta_ctx_gguf);
+                    return result;
+                }
+            }
+
+            struct ggml_tensor * tensor_meta = ggml_get_tensor(meta_ctx, name.c_str());
+            if (tensor_meta->type != GGML_TYPE_F32 || ggml_n_dims(tensor_meta) != 1) {
+                fprintf(stderr, "%s: direction tensor invalid in %s\n", __func__, load_info.fname.c_str());
+                ggml_free(meta_ctx);
+                gguf_free(meta_ctx_gguf);
+                return result;
+            }
+            if (result.n_embd == -1) {
+                result.n_embd = ggml_nelements(tensor_meta);
+            } else if (ggml_nelements(tensor_meta) != result.n_embd) {
+                fprintf(stderr, "%s: direction tensor sizes mismatched in %s\n", __func__, load_info.fname.c_str());
+                ggml_free(meta_ctx);
+                gguf_free(meta_ctx_gguf);
+                return result;
+            }
+            n_bytes += ggml_nbytes(tensor_meta);
+        }
+        ggml_free(meta_ctx);
+        gguf_free(meta_ctx_gguf);
+    }
+
+    if (n_tensors == 0) {
+        fprintf(stderr, "%s: no direction tensors found in %s\n", __func__, load_info.fname.c_str());
        return result;
    }

-    int32_t n_tensors = gguf_get_n_tensors(ctx_gguf);
-    if (n_tensors == 0) {
-        fprintf(stderr, "%s: no direction tensors found in %s\n", __func__, load_info.fname.c_str());
+    // load and scale tensors into final control vector context
+    struct ggml_init_params ggml_params = {
+        /* .mem_size   = */ ggml_tensor_overhead() * n_tensors + n_bytes,
+        /* .mem_buffer = */ nullptr,
+        /* .no_alloc   = */ false,
+    };
+    struct ggml_context * ctx = ggml_init(ggml_params);
+
+    struct gguf_init_params params = {
+        /*.no_alloc = */ false,
+        /*.ctx      = */ &ctx,
+    };
+    struct gguf_context * ctx_gguf = gguf_init_from_file(load_info.fname.c_str(), params);
+    if (!ctx_gguf) {
+        fprintf(stderr, "%s: failed to load control vector from %s\n", __func__, load_info.fname.c_str());
+        ggml_free(ctx);
+        return result;
    }

-    for (int i = 0; i < n_tensors; i++) {
-        std::string name = gguf_get_tensor_name(ctx_gguf, i);
+    // do not store data for layer 0 (it's not used)
+    result.data.resize(result.n_embd * max_direction_layer);

-        int layer_idx = -1;
+    for (uint32_t il = 1; il <= max_direction_layer; il++) {
+        const std::string name = "direction." + std::to_string(il);
+        const ggml_tensor * tensor = ggml_get_tensor(ctx, name.c_str());

-        // split on '.'
-        size_t dotpos = name.find('.');
-        if (dotpos != std::string::npos && name.substr(0, dotpos) == "direction") {
-            try {
-                layer_idx = std::stoi(name.substr(dotpos + 1));
-            } catch (...) {
-                layer_idx = -1;
+        float * dst = result.data.data() + result.n_embd * (il - 1);
+
+        if (tensor) {
+            const float * src = (const float *) tensor->data;
+            for (int j = 0; j < result.n_embd; j++) {
+                dst[j] = src[j] * load_info.strength;
+            }
+        } else {
+            for (int j = 0; j < result.n_embd; j++) {
+                dst[j] = 0.0f;
            }
        }
-        if (layer_idx < 0) {
-            fprintf(stderr, "%s: invalid/unparsable direction tensor layer index in %s\n", __func__, load_info.fname.c_str());
-            result.n_embd = -1;
-            break;
-        } else if (layer_idx == 0) {
-            fprintf(stderr, "%s: invalid (zero) direction tensor layer index in %s\n", __func__, load_info.fname.c_str());
-            result.n_embd = -1;
-            break;
-        }
-
-        struct ggml_tensor * tensor = ggml_get_tensor(ctx, name.c_str());
-        if (tensor->type != GGML_TYPE_F32) {
-            fprintf(stderr, "%s: invalid (non-F32) direction tensor type in %s\n", __func__, load_info.fname.c_str());
-            result.n_embd = -1;
-            break;
-        }
-        if (ggml_n_dims(tensor) != 1) {
-            fprintf(stderr, "%s: invalid (non-1D) direction tensor shape in %s\n", __func__, load_info.fname.c_str());
-            result.n_embd = -1;
-            break;
-        }
-
-        if (result.n_embd == -1) {
-            result.n_embd = ggml_nelements(tensor);
-        } else if (ggml_nelements(tensor) != result.n_embd) {
-            fprintf(stderr, "%s: direction tensor in %s does not match previous dimensions\n", __func__, load_info.fname.c_str());
-            result.n_embd = -1;
-            break;
-        }
-
-        // extend if necessary - do not store data for layer 0 (it's not used)
-        result.data.resize(std::max(result.data.size(), static_cast<size_t>(result.n_embd * layer_idx)), 0.0f);
-
-        const float * src = (const float *) tensor->data;
-        float * dst = result.data.data() + result.n_embd * (layer_idx - 1);  // layer 1 at [0]
-        for (int j = 0; j < result.n_embd; j++) {
-            dst[j] += src[j] * load_info.strength;  // allows multiple directions for same layer in same file
-        }
-
    }

-    if (result.n_embd == -1) {
-        fprintf(stderr, "%s: skipping %s due to invalid direction tensors\n", __func__, load_info.fname.c_str());
-        result.data.clear();
-    }
-
-    gguf_free(ctx_gguf);
-    ggml_free(ctx);
-
    return result;
 }

@@ -2928,19 +2933,16 @@ llama_control_vector_data llama_control_vector_load(const std::vector<llama_cont
        auto cur = llama_control_vector_load_one(info);

        if (cur.n_embd == -1) {
-            result.n_embd = -1;
-            break;
+            return result;
        }
-        if (result.n_embd != -1 && result.n_embd != cur.n_embd) {
-            fprintf(stderr, "%s: control vectors in %s does not match previous dimensions\n", __func__, info.fname.c_str());
-            result.n_embd = -1;
-            break;
+        if (result.n_embd != -1 && (result.n_embd != cur.n_embd || result.data.size() != cur.data.size())) {
+            fprintf(stderr, "%s: control vector in %s does not match previous vector dimensions\n", __func__, info.fname.c_str());
+            return result;
        }

        if (result.n_embd == -1) {
            result = std::move(cur);
        } else {
-            result.data.resize(std::max(result.data.size(), cur.data.size()), 0.0f);  // extend if necessary
            for (size_t i = 0; i < cur.data.size(); i++) {
                result.data[i] += cur.data[i];
            }
@@ -2948,8 +2950,7 @@ llama_control_vector_data llama_control_vector_load(const std::vector<llama_cont
    }

    if (result.n_embd == -1) {
-        fprintf(stderr, "%s: no valid control vector files passed\n", __func__);
-        result.data.clear();
+        fprintf(stderr, "%s: no vectors passed\n", __func__);
    }

    return result;
--- a/common/common.h
+++ b/common/common.h
@@ -200,7 +200,6 @@ struct gpt_params {
    std::string public_path   = "";
    std::string chat_template = "";
    std::string system_prompt = "";
-    bool enable_chat_template = true;

    std::vector<std::string> api_keys;

@@ -251,8 +250,6 @@ struct gpt_params {
    std::string cvector_outfile       = "control_vector.gguf";
    std::string cvector_positive_file = "examples/cvector-generator/positive.txt";
    std::string cvector_negative_file = "examples/cvector-generator/negative.txt";
-
-    bool spm_infill = false; // suffix/prefix/middle pattern for infill
 };

 void gpt_params_handle_model_default(gpt_params & params);
@@ -383,8 +380,6 @@ struct llama_chat_msg {
 bool llama_chat_verify_template(const std::string & tmpl);

 // CPP wrapper for llama_chat_apply_template
-// If the built-in template is not supported, we default to chatml
-// If the custom "tmpl" is not supported, we throw an error
 std::string llama_chat_apply_template(const struct llama_model * model,
        const std::string & tmpl,
        const std::vector<llama_chat_msg> & chat,
@@ -459,3 +454,4 @@ void yaml_dump_string_multiline(FILE * stream, const char * prop_name, const cha
 void yaml_dump_non_result_info(
    FILE * stream, const gpt_params & params, const llama_context * lctx,
    const std::string & timestamp, const std::vector<int> & prompt_tokens, const char * model_desc);
+
--- a/common/json-schema-to-grammar.cpp
+++ b/common/json-schema-to-grammar.cpp
@@ -316,7 +316,7 @@ std::unordered_map<char, std::string> GRAMMAR_LITERAL_ESCAPES = {
 };

 std::unordered_set<char> NON_LITERAL_SET = {'|', '.', '(', ')', '[', ']', '{', '}', '*', '+', '?'};
-std::unordered_set<char> ESCAPED_IN_REGEXPS_BUT_NOT_IN_LITERALS = {'^', '$', '.', '[', ']', '(', ')', '|', '{', '}', '*', '+', '?'};
+std::unordered_set<char> ESCAPED_IN_REGEXPS_BUT_NOT_IN_LITERALS = {'[', ']', '(', ')', '|', '{', '}', '*', '+', '?'};

 template <typename Iterator>
 std::string join(Iterator begin, Iterator end, const std::string & separator) {
@@ -720,7 +720,7 @@ private:
            }
            prop_names.push_back(prop_name);
        }
-        if ((additional_properties.is_boolean() && additional_properties.get<bool>()) || additional_properties.is_object()) {
+        if (!(additional_properties.is_boolean() && !additional_properties.get<bool>())) {
            std::string sub_name = name + (name.empty() ? "" : "-") + "additional";
            std::string value_rule =
                additional_properties.is_object() ? visit(additional_properties, sub_name + "-value")
--- a/convert-hf-to-gguf-update.py
+++ b/convert-hf-to-gguf-update.py
@@ -85,10 +85,6 @@ models = [
    {"name": "smaug-bpe",      "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/abacusai/Smaug-Llama-3-70B-Instruct", },
    {"name": "poro-chat",      "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/LumiOpen/Poro-34B-chat", },
    {"name": "jina-v2-code",   "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/jinaai/jina-embeddings-v2-base-code", },
-    {"name": "viking",         "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/LumiOpen/Viking-7B", }, # Also used for Viking 13B and 33B
-    {"name": "gemma",          "tokt": TOKENIZER_TYPE.SPM, "repo": "https://huggingface.co/google/gemma-2b", },
-    {"name": "gemma-2",        "tokt": TOKENIZER_TYPE.SPM, "repo": "https://huggingface.co/google/gemma-2-9b", },
-    {"name": "jais",           "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/core42/jais-13b", },
 ]


@@ -275,8 +271,7 @@ tests = [
    "3333333",
    "33333333",
    "333333333",
-    "Cửa Việt", # llama-bpe fails on this
-    " discards",
+    # "Cửa Việt", # llama-bpe fails on this
    chktxt,
 ]

--- a/convert-hf-to-gguf.py
+++ b/convert-hf-to-gguf.py
@@ -487,12 +487,6 @@ class Model:
        if chkhsh == "7967bfa498ade6b757b064f31e964dddbb80f8f9a4d68d4ba7998fcf281c531a":
            # ref: https://huggingface.co/jinaai/jina-embeddings-v2-base-code
            res = "jina-v2-code"
-        if chkhsh == "7fc505bd3104ca1083b150b17d088b59534ede9bde81f0dd2090967d7fe52cee":
-            # ref: https://huggingface.co/LumiOpen/Viking-7B
-            res = "viking"
-        if chkhsh == "b53802fb28e26d645c3a310b34bfe07da813026ec7c7716883404d5e0f8b1901":
-            # ref: https://huggingface.co/core42/jais-13b
-            res = "jais"

        if res is None:
            logger.warning("\n")
@@ -579,19 +573,7 @@ class Model:
        special_vocab._set_special_token("unk", tokenizer.special_tokens["<|endoftext|>"])
        special_vocab.add_to_gguf(self.gguf_writer)

-    def _set_vocab_sentencepiece(self, add_to_gguf=True):
-        tokens, scores, toktypes = self._create_vocab_sentencepiece()
-
-        self.gguf_writer.add_tokenizer_model("llama")
-        self.gguf_writer.add_tokenizer_pre("default")
-        self.gguf_writer.add_token_list(tokens)
-        self.gguf_writer.add_token_scores(scores)
-        self.gguf_writer.add_token_types(toktypes)
-
-        special_vocab = gguf.SpecialVocab(self.dir_model, n_vocab=len(tokens))
-        special_vocab.add_to_gguf(self.gguf_writer)
-
-    def _create_vocab_sentencepiece(self):
+    def _set_vocab_sentencepiece(self):
        from sentencepiece import SentencePieceProcessor

        tokenizer_path = self.dir_model / 'tokenizer.model'
@@ -653,7 +635,14 @@ class Model:
                scores.append(-1000.0)
                toktypes.append(SentencePieceTokenTypes.UNUSED)

-        return tokens, scores, toktypes
+        self.gguf_writer.add_tokenizer_model("llama")
+        self.gguf_writer.add_tokenizer_pre("default")
+        self.gguf_writer.add_token_list(tokens)
+        self.gguf_writer.add_token_scores(scores)
+        self.gguf_writer.add_token_types(toktypes)
+
+        special_vocab = gguf.SpecialVocab(self.dir_model, n_vocab=len(tokens))
+        special_vocab.add_to_gguf(self.gguf_writer)

    def _set_vocab_llama_hf(self):
        vocab = gguf.LlamaHfVocab(self.dir_model)
@@ -1942,7 +1931,7 @@ class Phi3MiniModel(Model):
        if len(rope_scaling_type) == 0:
            raise KeyError('Missing the required key rope_scaling.type')

-        if rope_scaling_type == 'su' or rope_scaling_type == 'longrope':
+        if rope_scaling_type == 'su':
            attn_factor = math.sqrt(1 + math.log(scale) / math.log(orig_max_pos_embds)) if scale > 1.0 else 1.0
        elif rope_scaling_type == 'yarn':
            attn_factor = 0.1 * math.log(scale) + 1.0 if scale > 1.0 else 1.0
@@ -2316,8 +2305,6 @@ class GemmaModel(Model):
        special_vocab._set_special_token("eot",    107)
        special_vocab.add_to_gguf(self.gguf_writer)

-        self.gguf_writer.add_add_space_prefix(False)
-
    def set_gguf_parameters(self):
        hparams = self.hparams
        block_count = hparams["num_hidden_layers"]
@@ -2350,71 +2337,6 @@ class GemmaModel(Model):
        return [(self.map_tensor_name(name), data_torch)]


-@Model.register("Gemma2ForCausalLM")
-class Gemma2Model(Model):
-    model_arch = gguf.MODEL_ARCH.GEMMA2
-
-    def set_vocab(self):
-        tokens, scores, toktypes = self._create_vocab_sentencepiece()
-        # hack: This is required so that we can properly use start/end-of-turn for chat template
-        for i in range(108):
-            # including <unusedX>, <start_of_turn>, <end_of_turn>
-            toktypes[i] = SentencePieceTokenTypes.CONTROL
-        self.gguf_writer.add_tokenizer_model("llama")
-        self.gguf_writer.add_tokenizer_pre("default")
-        self.gguf_writer.add_token_list(tokens)
-        self.gguf_writer.add_token_scores(scores)
-        self.gguf_writer.add_token_types(toktypes)
-
-        special_vocab = gguf.SpecialVocab(self.dir_model, n_vocab=len(tokens))
-        special_vocab.add_to_gguf(self.gguf_writer)
-
-        self.gguf_writer.add_add_space_prefix(False)
-
-    def set_gguf_parameters(self):
-        hparams = self.hparams
-        block_count = hparams["num_hidden_layers"]
-
-        self.gguf_writer.add_name(self.dir_model.name if self.model_name is None else self.model_name)
-        self.gguf_writer.add_context_length(hparams["max_position_embeddings"])
-        self.gguf_writer.add_embedding_length(hparams["hidden_size"])
-        self.gguf_writer.add_block_count(block_count)
-        self.gguf_writer.add_feed_forward_length(hparams["intermediate_size"])
-        self.gguf_writer.add_head_count(hparams["num_attention_heads"])
-        self.gguf_writer.add_head_count_kv(self.hparams["num_key_value_heads"] if "num_key_value_heads" in hparams else hparams["num_attention_heads"])
-        self.gguf_writer.add_layer_norm_rms_eps(self.hparams["rms_norm_eps"])
-        self.gguf_writer.add_key_length(hparams["head_dim"])
-        self.gguf_writer.add_value_length(hparams["head_dim"])
-        self.gguf_writer.add_file_type(self.ftype)
-        self.gguf_writer.add_attn_logit_softcapping(
-            self.hparams["attn_logit_softcapping"]
-        )
-        self.gguf_writer.add_final_logit_softcapping(
-            self.hparams["final_logit_softcapping"]
-        )
-        self.gguf_writer.add_sliding_window(self.hparams["sliding_window"])
-
-        # sanity check
-        attn_scalar = self.hparams["query_pre_attn_scalar"]
-        if attn_scalar != hparams["hidden_size"] / hparams["num_attention_heads"]:
-            raise ValueError("query_pre_attn_scalar must be equal to n_embd / n_head")
-
-    def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
-        del bid  # unusem
-
-        # lm_head is not used in llama.cpp, while autoawq will include this tensor in model
-        # To prevent errors, skip loading lm_head.weight.
-        if name == "lm_head.weight":
-            logger.debug(f"Skipping get tensor {name!r} in safetensors so that convert can end normally.")
-            return []
-
-        # ref: https://github.com/huggingface/transformers/blob/fc37f38915372c15992b540dfcbbe00a916d4fc6/src/transformers/models/gemma/modeling_gemma.py#L89
-        if name.endswith("norm.weight"):
-            data_torch = data_torch + 1
-
-        return [(self.map_tensor_name(name), data_torch)]
-
-
@Model.register("Starcoder2ForCausalLM")
 class StarCoder2Model(Model):
    model_arch = gguf.MODEL_ARCH.STARCODER2
@@ -2971,96 +2893,6 @@ class T5Model(Model):
        return [(self.map_tensor_name(name), data_torch)]


-@Model.register("JAISLMHeadModel")
-class JaisModel(Model):
-    model_arch = gguf.MODEL_ARCH.JAIS
-
-    def __init__(self, *args, **kwargs):
-        super().__init__(*args, **kwargs)
-
-        # SwigLU activation
-        assert self.hparams["activation_function"] == "swiglu"
-        # ALiBi position embedding
-        assert self.hparams["position_embedding_type"] == "alibi"
-
-        # Embeddings scale
-        self.embeddings_scale = 1.0
-        # note: For some JAIS flavors, output is tied to (same as) wte in original model
-        self.output_is_wte = False
-        if 'mup_embeddings_scale' in self.hparams:
-            self.output_is_wte = True   # Hack (?)
-            self.embeddings_scale = self.hparams['mup_embeddings_scale']
-        elif 'embeddings_scale' in self.hparams:
-            self.embeddings_scale = self.hparams['embeddings_scale']
-        else:
-            assert False
-
-        self.width_scale = 1.0
-        if 'mup_output_alpha' in self.hparams:
-            assert 'mup_width_scale' in self.hparams
-            self.width_scale = self.hparams['mup_output_alpha'] * self.hparams['mup_width_scale']
-        elif 'width_scale' in self.hparams:
-            self.width_scale = self.hparams['width_scale']
-        else:
-            assert False
-
-        self.max_alibi_bias = 8.0
-
-    def set_vocab(self):
-        self._set_vocab_gpt2()
-
-    def set_gguf_parameters(self):
-        self.gguf_writer.add_name(self.dir_model.name)
-        self.gguf_writer.add_block_count(self.hparams["n_layer"])
-        self.gguf_writer.add_context_length(self.hparams["n_positions"])
-        self.gguf_writer.add_embedding_length(self.hparams["n_embd"])
-        self.gguf_writer.add_feed_forward_length(self.hparams["n_inner"])
-        self.gguf_writer.add_head_count(self.hparams["n_head"])
-        self.gguf_writer.add_layer_norm_eps(self.hparams["layer_norm_epsilon"])
-        self.gguf_writer.add_file_type(self.ftype)
-
-    def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
-        del bid  # unused
-
-        tensors: list[tuple[str, Tensor]] = []
-
-        # we don't need these
-        if name.endswith((".attn.bias")):
-            return tensors
-
-        if name.endswith(("relative_pe.slopes")):
-            # Calculate max ALiBi bias (this is the inverse of the ALiBi calculation)
-            # Some other models has max_alibi_bias spelled out explicitly in the hyperparams,
-            # but Jais's PyTorch model simply precalculates the slope values and places them
-            # in relative_pes.slopes
-            n_head_closest_log2 = 2 ** math.floor(math.log2(self.hparams["n_head"]))
-            first_val = float(data_torch._data[0])
-            self.max_alibi_bias = -round(math.log2(first_val) * n_head_closest_log2)
-
-            return tensors
-
-        if name.endswith((".c_attn.weight", ".c_proj.weight", ".c_fc.weight", ".c_fc2.weight")):
-            data_torch = data_torch.transpose(1, 0)
-
-        new_name = self.map_tensor_name(name)
-
-        if new_name == self.format_tensor_name(gguf.MODEL_TENSOR.TOKEN_EMBD):
-            tensors.append((new_name, data_torch * self.embeddings_scale))
-            if self.output_is_wte:
-                tensors.append((self.format_tensor_name(gguf.MODEL_TENSOR.OUTPUT), data_torch * self.width_scale))
-        elif new_name == self.format_tensor_name(gguf.MODEL_TENSOR.OUTPUT):
-            assert not self.output_is_wte
-            tensors.append((new_name, data_torch * self.width_scale))
-        else:
-            tensors.append((new_name, data_torch))
-
-        return tensors
-
-    def write_tensors(self):
-        super().write_tensors()
-        self.gguf_writer.add_max_alibi_bias(self.max_alibi_bias)
-
-
 ###### CONVERSION LOGIC ######


@@ -3216,8 +3048,7 @@ def main() -> None:
        "auto": gguf.LlamaFileType.GUESSED,
    }

-    is_split = args.split_max_tensors > 0 or args.split_max_size != "0"
-    if args.use_temp_file and is_split:
+    if args.use_temp_file and (args.split_max_tensors > 0 or args.split_max_size != "0"):
        logger.error("Error: Cannot use temp file when splitting")
        sys.exit(1)

@@ -3254,12 +3085,11 @@ def main() -> None:
        if args.vocab_only:
            logger.info("Exporting model vocab...")
            model_instance.write_vocab()
-            logger.info(f"Model vocab successfully exported to {model_instance.fname_out}")
+            logger.info("Model vocab successfully exported.")
        else:
            logger.info("Exporting model...")
            model_instance.write()
-            out_path = f"{model_instance.fname_out.parent}{os.sep}" if is_split else model_instance.fname_out
-            logger.info(f"Model successfully exported to {out_path}")
+            logger.info("Model successfully exported.")


 if __name__ == '__main__':
--- a/docs/BLIS.md
+++ b/docs/BLIS.md
@@ -30,8 +30,8 @@ We recommend using openmp since it's easier to modify the cores being used.
 Makefile:

 ```bash
-make GGML_BLIS=1 -j
-# make GGML_BLIS=1 llama-benchmark-matmult
+make LLAMA_BLIS=1 -j
+# make LLAMA_BLIS=1 benchmark-matmult
 ```

 CMake:
@@ -39,7 +39,7 @@ CMake:
 ```bash
 mkdir build
 cd build
-cmake -DGGML_BLAS=ON -DGGML_BLAS_VENDOR=FLAME ..
+cmake -DLLAMA_BLAS=ON -DLLAMA_BLAS_VENDOR=FLAME ..
 make -j
 ```

--- a/examples/CMakeLists.txt
+++ b/examples/CMakeLists.txt
@@ -39,13 +39,13 @@ else()
    add_subdirectory(quantize-stats)
    add_subdirectory(quantize)
    add_subdirectory(retrieval)
-    if (GGML_RPC)
+    if (LLAMA_RPC)
        add_subdirectory(rpc)
    endif()
    if (LLAMA_BUILD_SERVER)
    add_subdirectory(server)
    endif()
-    if (GGML_SYCL)
+    if (LLAMA_SYCL)
        add_subdirectory(sycl)
    endif()
    add_subdirectory(save-load-state)
--- a/examples/embedding/README.md
+++ b/examples/embedding/README.md
@@ -58,3 +58,4 @@ The above command will output space-separated float values.
 ```powershell
 embedding.exe -p 'Castle<#sep#>Stronghold<#sep#>Dog<#sep#>Cat' --embd-separator '<#sep#>' --embd-normalize 2  --embd-output-format '' -m './path/to/model.gguf' --n-gpu-layers 99 --log-disable 2>/dev/null
 ```
+
--- a/examples/imatrix/README.md
+++ b/examples/imatrix/README.md
@@ -25,7 +25,7 @@ For faster computation, make sure to use GPU offloading via the `-ngl` argument
 ## Example

 ```bash
-GGML_CUDA=1 make -j
+LLAMA_CUDA=1 make -j

 # generate importance matrix (imatrix.dat)
 ./llama-imatrix -m ggml-model-f16.gguf -f train-data.txt -ngl 99
--- a/examples/infill/README.md
+++ b/examples/infill/README.md
@@ -15,7 +15,6 @@ In this section, we cover the most commonly used options for running the `infill
 -   `-i, --interactive`: Run the program in interactive mode, allowing you to provide input directly and receive real-time responses.
 -   `-n N, --n-predict N`: Set the number of tokens to predict when generating text. Adjusting this value can influence the length of the generated text.
 -   `-c N, --ctx-size N`: Set the size of the prompt context. The default is 512, but LLaMA models were built with a context of 2048, which will provide better results for longer input/inference.
-   `--spm-infill`: Use Suffix/Prefix/Middle pattern for infill (instead of Prefix/Suffix/Middle) as some models prefer this.

 ## Input Prompts

--- a/examples/infill/infill.cpp
+++ b/examples/infill/infill.cpp
@@ -210,7 +210,6 @@ int main(int argc, char ** argv) {
        suff_rm_leading_spc = false;
    }
    std::vector<llama_token> embd_inp;
-    std::vector<llama_token> embd_end;
    std::vector<llama_token> inp_pfx = ::llama_tokenize(ctx, params.input_prefix, false);
    std::vector<llama_token> inp_sfx = ::llama_tokenize(ctx, params.input_suffix, false);
    const int space_token = 29871;
@@ -218,13 +217,12 @@ int main(int argc, char ** argv) {
        inp_sfx.erase(inp_sfx.begin());
    }
    inp_pfx.insert(inp_pfx.begin(), llama_token_prefix(model));
-    inp_sfx.insert(inp_sfx.begin(), llama_token_suffix(model));
-    embd_inp = params.spm_infill ? inp_sfx : inp_pfx;
-    embd_end = params.spm_infill ? inp_pfx : inp_sfx;
    if (add_bos) {
-        embd_inp.insert(embd_inp.begin(), llama_token_bos(model));
+        inp_pfx.insert(inp_pfx.begin(), llama_token_bos(model));
    }
-    embd_inp.insert(embd_inp.end(), embd_end.begin(), embd_end.end());
+    inp_sfx.insert(inp_sfx.begin(), llama_token_suffix(model));
+    embd_inp = inp_pfx;
+    embd_inp.insert(embd_inp.end(), inp_sfx.begin(), inp_sfx.end());

    const llama_token middle_token = llama_token_middle(model);
    if (middle_token >= 0) {
@@ -528,14 +526,14 @@ int main(int argc, char ** argv) {
                    inp_sfx.erase(inp_sfx.begin());
                }
                inp_pfx.insert(inp_pfx.begin(), llama_token_prefix(model));
-                inp_sfx.insert(inp_sfx.begin(), llama_token_suffix(model));
-                embd_inp = params.spm_infill ? inp_sfx : inp_pfx;
-                embd_end = params.spm_infill ? inp_pfx : inp_sfx;
                if (add_bos) {
-                    embd_inp.insert(embd_inp.begin(), llama_token_bos(model));
+                    inp_pfx.insert(inp_pfx.begin(), llama_token_bos(model));
                }
-                embd_inp.insert(embd_inp.end(), embd_end.begin(), embd_end.end());
+                inp_sfx.insert(inp_sfx.begin(), llama_token_suffix(model));
+                embd_inp = inp_pfx;
+                embd_inp.insert(embd_inp.end(), inp_sfx.begin(), inp_sfx.end());

+                const llama_token middle_token = llama_token_middle(model);
                if (middle_token >= 0) {
                    embd_inp.push_back(middle_token);
                }
@@ -659,3 +657,4 @@ int main(int argc, char ** argv) {

    return 0;
 }
+
--- a/examples/json_schema_to_grammar.py
+++ b/examples/json_schema_to_grammar.py
@@ -231,7 +231,7 @@ GRAMMAR_RANGE_LITERAL_ESCAPE_RE = re.compile(r'[\r\n"\]\-\\]')
 GRAMMAR_LITERAL_ESCAPES = {'\r': '\\r', '\n': '\\n', '"': '\\"', '-': '\\-', ']': '\\]'}

 NON_LITERAL_SET = set('|.()[]{}*+?')
-ESCAPED_IN_REGEXPS_BUT_NOT_IN_LITERALS = set('^$.[]()|{}*+?')
+ESCAPED_IN_REGEXPS_BUT_NOT_IN_LITERALS = set('[]()|{}*+?')


 class SchemaConverter:
@@ -602,7 +602,7 @@ class SchemaConverter:
                else:
                    add_component(t, is_required=True)

-            return self._add_rule(rule_name, self._build_object_rule(properties, required, hybrid_name, additional_properties=None))
+            return self._add_rule(rule_name, self._build_object_rule(properties, required, hybrid_name, additional_properties=[]))

        elif schema_type in (None, 'array') and ('items' in schema or 'prefixItems' in schema):
            items = schema.get('items') or schema['prefixItems']
@@ -691,7 +691,7 @@ class SchemaConverter:
        required_props = [k for k in sorted_props if k in required]
        optional_props = [k for k in sorted_props if k not in required]

-        if additional_properties is not None and additional_properties != False:
+        if additional_properties != False:
            sub_name = f'{name}{"-" if name else ""}additional'
            value_rule = self.visit(additional_properties, f'{sub_name}-value') if isinstance(additional_properties, dict) else \
                self._add_primitive('value', PRIMITIVE_RULES['value'])
--- a/examples/llama.android/llama/CMakeLists.txt
+++ b/examples/llama.android/llama/CMakeLists.txt
@@ -0,0 +1,55 @@
+
+# For more information about using CMake with Android Studio, read the
+# documentation: https://d.android.com/studio/projects/add-native-code.html.
+# For more examples on how to use CMake, see https://github.com/android/ndk-samples.
+
+# Sets the minimum CMake version required for this project.
+cmake_minimum_required(VERSION 3.22.1)
+
+# Declares the project name. The project name can be accessed via ${ PROJECT_NAME},
+# Since this is the top level CMakeLists.txt, the project name is also accessible
+# with ${CMAKE_PROJECT_NAME} (both CMake variables are in-sync within the top level
+# build script scope).
+project("llama-android")
+
+## Fetch latest llama.cpp from GitHub
+#include(FetchContent)
+#FetchContent_Declare(
+#        llama
+#        GIT_REPOSITORY https://github.com/ggerganov/llama.cpp
+#        GIT_TAG        master
+#)
+#
+## Also provides "common"
+#FetchContent_MakeAvailable(llama)
+
+# llama.cpp CI uses the code from the current branch
+# ref: https://github.com/ggerganov/llama.cpp/pull/7341#issuecomment-2117617700
+add_subdirectory(../../../../../../ build-llama)
+
+# Creates and names a library, sets it as either STATIC
+# or SHARED, and provides the relative paths to its source code.
+# You can define multiple libraries, and CMake builds them for you.
+# Gradle automatically packages shared libraries with your APK.
+#
+# In this top level CMakeLists.txt, ${CMAKE_PROJECT_NAME} is used to define
+# the target library name; in the sub-module's CMakeLists.txt, ${PROJECT_NAME}
+# is preferred for the same purpose.
+#
+# In order to load a library into your app from Java/Kotlin, you must call
+# System.loadLibrary() and pass the name of the library defined here;
+# for GameActivity/NativeActivity derived applications, the same library name must be
+# used in the AndroidManifest.xml file.
+add_library(${CMAKE_PROJECT_NAME} SHARED
+    # List C/C++ source files with relative paths to this CMakeLists.txt.
+        llama-android.cpp)
+
+# Specifies libraries CMake should link to your target library. You
+# can link libraries from various origins, such as libraries defined in this
+# build script, prebuilt third-party libraries, or Android system libraries.
+target_link_libraries(${CMAKE_PROJECT_NAME}
+    # List libraries link to the target library
+    llama
+    common
+    android
+    log)
--- a/examples/llama.android/llama/src/main/cpp/CMakeLists.txt
+++ b/examples/llama.android/llama/src/main/cpp/CMakeLists.txt
@@ -11,15 +11,15 @@ cmake_minimum_required(VERSION 3.22.1)
 # build script scope).
 project("llama-android")

-#include(FetchContent)
-#FetchContent_Declare(
-#        llama
-#        GIT_REPOSITORY https://github.com/ggerganov/llama.cpp
-#        GIT_TAG        master
-#)
+include(FetchContent)
+FetchContent_Declare(
+        llama
+        GIT_REPOSITORY https://github.com/ggerganov/llama.cpp
+        GIT_TAG        master
+)

 # Also provides "common"
-#FetchContent_MakeAvailable(llama)
+FetchContent_MakeAvailable(llama)

 # Creates and names a library, sets it as either STATIC
 # or SHARED, and provides the relative paths to its source code.
@@ -30,10 +30,6 @@ project("llama-android")
 # the target library name; in the sub-module's CMakeLists.txt, ${PROJECT_NAME}
 # is preferred for the same purpose.
 #
-
-#load local llama.cpp
-add_subdirectory(../../../../../../ build-llama)
-
 # In order to load a library into your app from Java/Kotlin, you must call
 # System.loadLibrary() and pass the name of the library defined here;
 # for GameActivity/NativeActivity derived applications, the same library name must be
--- a/examples/llama.android/llama/src/main/cpp/llama-android.cpp
+++ b/examples/llama.android/llama/src/main/cpp/llama-android.cpp
@@ -5,7 +5,7 @@
 #include <string>
 #include <unistd.h>
 #include "llama.h"
-#include "common.h"
+#include "common/common.h"

 // Write C++ code here.
 //
--- a/examples/llava/MobileVLM-README.md
+++ b/examples/llava/MobileVLM-README.md
@@ -194,7 +194,7 @@ llama_print_timings:       total time =   44411.01 ms /   377 tokens
 ## Orin compile and run
 ### compile
 ```sh
-make GGML_CUDA=1 CUDA_DOCKER_ARCH=sm_87 GGML_CUDA_F16=1 -j 32
+make LLAMA_CUDA=1 CUDA_DOCKER_ARCH=sm_87 LLAMA_CUDA_F16=1 -j 32
 ```
 ### run on Orin
 ### case 1
--- a/examples/llava/clip.cpp
+++ b/examples/llava/clip.cpp
@@ -1121,20 +1121,20 @@ struct clip_ctx * clip_model_load(const char * fname, const int verbosity = 1) {
            }
            if (n < 32)
                hparams.image_grid_pinpoints[n] = 0;
-        } catch (std::runtime_error & /*e*/) {
+        } catch (std::runtime_error & e) {
            hparams.image_grid_pinpoints[0]=0;
        }

        try {
            int idx = get_key_idx(ctx, KEY_MM_PATCH_MERGE_TYPE);
            strcpy(hparams.mm_patch_merge_type, gguf_get_val_str(ctx, idx));
-        } catch (std::runtime_error & /*e*/) {
+        } catch (std::runtime_error & e) {
            strcpy(hparams.mm_patch_merge_type, "flat");
        }

        try {
            hparams.image_crop_resolution = get_u32(ctx, KEY_IMAGE_CROP_RESOLUTION); // llava-1.6
-        } catch(const std::exception& /*e*/) {
+        } catch(const std::exception& e) {
            hparams.image_crop_resolution = hparams.image_size;
        }

@@ -1173,7 +1173,7 @@ struct clip_ctx * clip_model_load(const char * fname, const int verbosity = 1) {
        try {
            vision_model.class_embedding  = get_tensor(new_clip->ctx_data, TN_CLASS_EMBD);
            new_clip->has_class_embedding = true;
-        } catch (const std::exception& /*e*/) {
+        } catch (const std::exception& e) {
            new_clip->has_class_embedding = false;
        }

@@ -1181,7 +1181,7 @@ struct clip_ctx * clip_model_load(const char * fname, const int verbosity = 1) {
            vision_model.pre_ln_w  = get_tensor(new_clip->ctx_data, format(TN_LN_PRE, "v", "weight"));
            vision_model.pre_ln_b  = get_tensor(new_clip->ctx_data, format(TN_LN_PRE, "v", "bias"));
            new_clip->has_pre_norm = true;
-        } catch (std::exception & /*e*/) {
+        } catch (std::exception & e) {
            new_clip->has_pre_norm = false;
        }

@@ -1189,21 +1189,21 @@ struct clip_ctx * clip_model_load(const char * fname, const int verbosity = 1) {
            vision_model.post_ln_w  = get_tensor(new_clip->ctx_data, format(TN_LN_POST, "v", "weight"));
            vision_model.post_ln_b  = get_tensor(new_clip->ctx_data, format(TN_LN_POST, "v", "bias"));
            new_clip->has_post_norm = true;
-        } catch (std::exception & /*e*/) {
+        } catch (std::exception & e) {
            new_clip->has_post_norm = false;
        }

        try {
            vision_model.patch_bias = get_tensor(new_clip->ctx_data, TN_PATCH_BIAS);
            new_clip->has_patch_bias = true;
-        } catch (std::exception & /*e*/) {
+        } catch (std::exception & e) {
            new_clip->has_patch_bias = false;
        }

        try {
            vision_model.patch_embeddings    = get_tensor(new_clip->ctx_data, TN_PATCH_EMBD);
            vision_model.position_embeddings = get_tensor(new_clip->ctx_data, format(TN_POS_EMBD, "v"));
-        } catch(const std::exception& /*e*/) {
+        } catch(const std::exception& e) {
            LOG_TEE("%s: failed to load vision model tensors\n", __func__);
        }

@@ -1215,26 +1215,26 @@ struct clip_ctx * clip_model_load(const char * fname, const int verbosity = 1) {
                // Yi-type llava
                vision_model.mm_1_w = get_tensor(new_clip->ctx_data, format(TN_LLAVA_PROJ, 1, "weight"));
                vision_model.mm_1_b = get_tensor(new_clip->ctx_data, format(TN_LLAVA_PROJ, 1, "bias"));
-            } catch (std::runtime_error & /*e*/) { }
+            } catch (std::runtime_error & e) {  }
            try {
                // missing in Yi-type llava
                vision_model.mm_2_w              = get_tensor(new_clip->ctx_data, format(TN_LLAVA_PROJ, 2, "weight"));
                vision_model.mm_2_b              = get_tensor(new_clip->ctx_data, format(TN_LLAVA_PROJ, 2, "bias"));
-            } catch (std::runtime_error & /*e*/) { }
+            } catch (std::runtime_error & e) {  }
            try {
                // Yi-type llava
                vision_model.mm_3_w = get_tensor(new_clip->ctx_data, format(TN_LLAVA_PROJ, 3, "weight"));
                vision_model.mm_3_b = get_tensor(new_clip->ctx_data, format(TN_LLAVA_PROJ, 3, "bias"));
-            } catch (std::runtime_error & /*e*/) { }
+            } catch (std::runtime_error & e) {  }
            try {
                // Yi-type llava
                vision_model.mm_4_w = get_tensor(new_clip->ctx_data, format(TN_LLAVA_PROJ, 4, "weight"));
                vision_model.mm_4_b = get_tensor(new_clip->ctx_data, format(TN_LLAVA_PROJ, 4, "bias"));
-            } catch (std::runtime_error & /*e*/) { }
+            } catch (std::runtime_error & e) {  }
            try {
                vision_model.image_newline = get_tensor(new_clip->ctx_data, TN_IMAGE_NEWLINE);
                // LOG_TEE("%s: image_newline tensor (llava-1.6) found\n", __func__);
-            } catch (std::runtime_error & /*e*/) { }
+            } catch (std::runtime_error & e) {  }
        } else if (new_clip->proj_type == PROJECTOR_TYPE_LDP) {
            // MobileVLM projection
            vision_model.mm_model_mlp_1_w               = get_tensor(new_clip->ctx_data, format(TN_MVLM_PROJ_MLP, 1, "weight"));
--- a/examples/lookup/README.md
+++ b/examples/lookup/README.md
@@ -10,3 +10,4 @@ More info:

 https://github.com/ggerganov/llama.cpp/pull/4484
 https://github.com/ggerganov/llama.cpp/issues/4226
+
--- a/examples/main-cmake-pkg/.gitignore
+++ b/examples/main-cmake-pkg/.gitignore
@@ -48,3 +48,4 @@
 build*/
 out/
 tmp/
+
--- a/examples/main-cmake-pkg/CMakeLists.txt
+++ b/examples/main-cmake-pkg/CMakeLists.txt
@@ -30,3 +30,4 @@ target_include_directories(${TARGET} PRIVATE ${_common_path})
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
 target_compile_features(${TARGET} PRIVATE cxx_std_11)
+
--- a/examples/main/main.cpp
+++ b/examples/main/main.cpp
@@ -261,7 +261,7 @@ int main(int argc, char ** argv) {
    std::vector<llama_token> embd_inp;

    {
-        auto prompt = (params.conversation && params.enable_chat_template)
+        auto prompt = params.conversation
            ? chat_add_and_format(model, chat_msgs, "system", params.prompt) // format the system prompt in conversation mode
            : params.prompt;
        if (params.interactive_first || !params.prompt.empty() || session_tokens.empty()) {
@@ -810,9 +810,7 @@ int main(int argc, char ** argv) {
                        is_antiprompt = true;
                    }

-                    if (params.enable_chat_template) {
-                        chat_add_and_format(model, chat_msgs, "assistant", assistant_ss.str());
-                    }
+                    chat_add_and_format(model, chat_msgs, "system", assistant_ss.str());
                    is_interacting = true;
                    printf("\n");
                }
@@ -874,13 +872,12 @@ int main(int argc, char ** argv) {
                        string_process_escapes(buffer);
                    }

-                    bool format_chat = params.conversation && params.enable_chat_template;
-                    std::string user_inp = format_chat
+                    std::string user_inp = params.conversation
                        ? chat_add_and_format(model, chat_msgs, "user", std::move(buffer))
                        : std::move(buffer);
                    // TODO: one inconvenient of current chat template implementation is that we can't distinguish between user input and special tokens (prefix/postfix)
                    const auto line_pfx = ::llama_tokenize(ctx, params.input_prefix, false, true);
-                    const auto line_inp = ::llama_tokenize(ctx, user_inp,            false, format_chat);
+                    const auto line_inp = ::llama_tokenize(ctx, user_inp,            false, params.conversation);
                    const auto line_sfx = ::llama_tokenize(ctx, params.input_suffix, false, true);

                    LOG("input tokens: %s\n", LOG_TOKENS_TOSTR_PRETTY(ctx, line_inp).c_str());
--- a/examples/perplexity/perplexity.cpp
+++ b/examples/perplexity/perplexity.cpp
@@ -1991,12 +1991,6 @@ int main(int argc, char ** argv) {
        params.n_batch = std::min(params.n_batch, n_kv);
    } else {
        params.n_batch = std::min(params.n_batch, params.n_ctx);
-        if (params.kl_divergence) {
-            params.n_parallel = 1;
-        } else {
-            // ensure there's at least enough seq_ids for HellaSwag
-            params.n_parallel = std::max(4, params.n_parallel);
-        }
    }

    if (params.ppl_stride > 0) {
@@ -2021,6 +2015,9 @@ int main(int argc, char ** argv) {
    llama_model * model;
    llama_context * ctx;

+    // ensure there's at least enough seq_ids for HellaSwag
+    params.n_parallel = std::max(4, params.n_parallel);
+
    // load the model and apply lora adapter, if any
    std::tie(model, ctx) = llama_init_from_gpt_params(params);
    if (model == NULL) {
--- a/examples/rpc/README.md
+++ b/examples/rpc/README.md
@@ -29,13 +29,13 @@ You can also run multiple `rpc-server` instances on the same host, each with a d

 ## Usage

-On each host, build the corresponding backend with `cmake` and add `-DGGML_RPC=ON` to the build options.
+On each host, build the corresponding backend with `cmake` and add `-DLLAMA_RPC=ON` to the build options.
 For example, to build the CUDA backend with RPC support:

 ```bash
 mkdir build-rpc-cuda
 cd build-rpc-cuda
-cmake .. -DGGML_CUDA=ON -DGGML_RPC=ON
+cmake .. -DLLAMA_CUDA=ON -DLLAMA_RPC=ON
 cmake --build . --config Release
 ```

@@ -58,12 +58,12 @@ $ CUDA_VISIBLE_DEVICES=0 bin/rpc-server -p 50052
 This way you can run multiple `rpc-server` instances on the same host, each with a different CUDA device.


-On the main host build `llama.cpp` only with `-DGGML_RPC=ON`:
+On the main host build `llama.cpp` only with `-DLLAMA_RPC=ON`:

 ```bash
 mkdir build-rpc
 cd build-rpc
-cmake .. -DGGML_RPC=ON
+cmake .. -DLLAMA_RPC=ON
 cmake --build . --config Release
 ```

--- a/examples/server-embd.py
+++ b/examples/server-embd.py
@@ -31,3 +31,4 @@ for i in range(n-1):
        embedding2 = np.array(result[j])
        similarity = np.dot(embedding1, embedding2) / (np.linalg.norm(embedding1) * np.linalg.norm(embedding2))
        print(f"Similarity between {i} and {j}: {similarity:.2f}")
+
--- a/examples/server/CMakeLists.txt
+++ b/examples/server/CMakeLists.txt
@@ -1,14 +1,7 @@
 set(TARGET llama-server)
 option(LLAMA_SERVER_VERBOSE "Build verbose logging option for Server" ON)
-option(LLAMA_SERVER_SSL     "Build SSL support for the server"        OFF)
-
+option(LLAMA_SERVER_SSL "Build SSL support for the server" OFF)
 include_directories(${CMAKE_CURRENT_SOURCE_DIR} ${CMAKE_CURRENT_BINARY_DIR})
-
-if (MINGW)
-    # fix: https://github.com/ggerganov/llama.cpp/actions/runs/9651004652/job/26617901362?pr=8006
-    add_compile_definitions(_WIN32_WINNT=${GGML_WIN_VER})
-endif()
-
 set(TARGET_SRCS
    server.cpp
    utils.hpp
@@ -31,7 +24,6 @@ set(PUBLIC_ASSETS
    prompt-formats.js
    json-schema-to-grammar.mjs
 )
-
 foreach(asset ${PUBLIC_ASSETS})
    set(input "${CMAKE_CURRENT_SOURCE_DIR}/public/${asset}")
    set(output "${CMAKE_CURRENT_BINARY_DIR}/${asset}.hpp")
@@ -42,23 +34,18 @@ foreach(asset ${PUBLIC_ASSETS})
        COMMAND "${CMAKE_COMMAND}" "-DINPUT=${input}" "-DOUTPUT=${output}" -P "${PROJECT_SOURCE_DIR}/scripts/xxd.cmake"
    )
 endforeach()
-
 add_executable(${TARGET} ${TARGET_SRCS})
 install(TARGETS ${TARGET} RUNTIME)
 target_compile_definitions(${TARGET} PRIVATE
    SERVER_VERBOSE=$<BOOL:${LLAMA_SERVER_VERBOSE}>
 )
-
 target_link_libraries(${TARGET} PRIVATE common ${CMAKE_THREAD_LIBS_INIT})
-
 if (LLAMA_SERVER_SSL)
    find_package(OpenSSL REQUIRED)
    target_link_libraries(${TARGET} PRIVATE OpenSSL::SSL OpenSSL::Crypto)
    target_compile_definitions(${TARGET} PRIVATE CPPHTTPLIB_OPENSSL_SUPPORT)
 endif()
-
 if (WIN32)
    TARGET_LINK_LIBRARIES(${TARGET} PRIVATE ws2_32)
 endif()
-
 target_compile_features(${TARGET} PRIVATE cxx_std_11)
--- a/examples/server/README.md
+++ b/examples/server/README.md
@@ -73,7 +73,6 @@ The project is under active development, and we are [looking for feedback and co
 - `-fa`, `--flash-attn` : enable flash attention (default: disabled).
 - `-ctk TYPE`, `--cache-type-k TYPE` : KV cache data type for K (default: `f16`, options `f32`, `f16`, `q8_0`, `q4_0`, `q4_1`, `iq4_nl`, `q5_0`, or `q5_1`)
 - `-ctv TYPE`, `--cache-type-v TYPE` : KV cache type for V (default `f16`, see `-ctk` for options)
- `--spm-infill` : Use Suffix/Prefix/Middle pattern for infill (instead of Prefix/Suffix/Middle) as some models prefer this.

 **If compiled with `LLAMA_SERVER_SSL=ON`**
 - `--ssl-key-file FNAME`: path to file a PEM-encoded SSL private key
--- a/examples/server/public/json-schema-to-grammar.mjs
+++ b/examples/server/public/json-schema-to-grammar.mjs
@@ -259,7 +259,7 @@ const GRAMMAR_RANGE_LITERAL_ESCAPE_RE = /[\n\r"\]\-\\]/g;
 const GRAMMAR_LITERAL_ESCAPES = { '\r': '\\r', '\n': '\\n', '"': '\\"', '-': '\\-', ']': '\\]' };

 const NON_LITERAL_SET = new Set('|.()[]{}*+?');
-const ESCAPED_IN_REGEXPS_BUT_NOT_IN_LITERALS = new Set('^$.[]()|{}*+?');
+const ESCAPED_IN_REGEXPS_BUT_NOT_IN_LITERALS = new Set('[]()|{}*+?');

 export class SchemaConverter {
  constructor(options) {
@@ -751,7 +751,7 @@ export class SchemaConverter {
    const requiredProps = sortedProps.filter(k => required.has(k));
    const optionalProps = sortedProps.filter(k => !required.has(k));

-    if (additionalProperties) {
+    if (additionalProperties !== false) {
      const subName = `${name ?? ''}${name ? '-' : ''}additional`;
      const valueRule =
        additionalProperties != null && typeof additionalProperties === 'object' ? this.visit(additionalProperties, `${subName}-value`)
--- a/examples/server/server.cpp
+++ b/examples/server/server.cpp
@@ -2020,7 +2020,6 @@ struct server_context {
                        slot.t_start_generation = 0;

                        if (slot.infill) {
-                            const bool add_bos = llama_should_add_bos_token(model);
                            bool suff_rm_leading_spc = true;
                            if (params.input_suffix.find_first_of(' ') == 0 && params.input_suffix.size() > 1) {
                                params.input_suffix.erase(0, 1);
@@ -2036,21 +2035,16 @@ struct server_context {
                            }

                            prefix_tokens.insert(prefix_tokens.begin(), llama_token_prefix(model));
-                            suffix_tokens.insert(suffix_tokens.begin(), llama_token_suffix(model));
-
-                            auto embd_inp = params.spm_infill ? suffix_tokens : prefix_tokens;
-                            auto embd_end = params.spm_infill ? prefix_tokens : suffix_tokens;
-                            if (add_bos) {
-                                embd_inp.insert(embd_inp.begin(), llama_token_bos(model));
-                            }
-                            embd_inp.insert(embd_inp.end(), embd_end.begin(), embd_end.end());
+                            prefix_tokens.insert(prefix_tokens.begin(), llama_token_bos(model)); // always add BOS
+                            prefix_tokens.insert(prefix_tokens.end(),   llama_token_suffix(model));
+                            prefix_tokens.insert(prefix_tokens.end(),   suffix_tokens.begin(), suffix_tokens.end());

                            const llama_token middle_token = llama_token_middle(model);
                            if (middle_token >= 0) {
-                                embd_inp.push_back(middle_token);
+                                prefix_tokens.push_back(middle_token);
                            }

-                            prompt_tokens = embd_inp;
+                            prompt_tokens = prefix_tokens;
                        } else {
                            prompt_tokens = tokenize(slot.prompt, system_prompt.empty()); // add BOS if there isn't system prompt
                        }
--- a/examples/server/tests/features/passkey.feature
+++ b/examples/server/tests/features/passkey.feature
@@ -52,3 +52,4 @@ Feature: Passkey / Self-extend with context shift
      #| TheBloke/Llama-2-7B-GGUF        | llama-2-7b.Q2_K.gguf        | 4096        | 3   | 16384 | 512     | 4    | 512    | 500    | 300   | 1234    | 5           | 1234           |
      #| TheBloke/Mixtral-8x7B-v0.1-GGUF | mixtral-8x7b-v0.1.Q2_K.gguf | 32768       | 2   | 16384 | 512     | 4    | 512    | 500    | 100   | 0987    | 5           | 0
      # 987           |
+
--- a/examples/server/themes/buttons-top/index.html
+++ b/examples/server/themes/buttons-top/index.html
@@ -1054,3 +1054,4 @@
 </body>

 </html>
+
--- a/examples/server/themes/wild/index.html
+++ b/examples/server/themes/wild/index.html
@@ -1058,3 +1058,4 @@
 </body>

 </html>
+
--- a/examples/sycl/build.sh
+++ b/examples/sycl/build.sh
@@ -8,10 +8,10 @@ cd build
 source /opt/intel/oneapi/setvars.sh

 #for FP16
-#cmake .. -DGGML_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx -DGGML_SYCL_F16=ON # faster for long-prompt inference
+#cmake .. -DLLAMA_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx -DLLAMA_SYCL_F16=ON # faster for long-prompt inference

 #for FP32
-cmake .. -DGGML_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx
+cmake .. -DLLAMA_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx

 #build example/main
 #cmake --build . --config Release --target main
--- a/examples/sycl/run-llama2.sh
+++ b/examples/sycl/run-llama2.sh
@@ -34,3 +34,4 @@ fi

 #use multiple GPUs with same max compute units
 #ZES_ENABLE_SYSMAN=1 ./build/bin/llama-cli -m models/llama-2-7b.Q4_0.gguf -p "${INPUT2}" -n 400 -e -ngl 33 -s 0
+
--- a/examples/sycl/win-build-sycl.bat
+++ b/examples/sycl/win-build-sycl.bat
@@ -13,10 +13,10 @@ if %errorlevel% neq 0 goto ERROR

 ::  for FP16
 ::  faster for long-prompt inference
-::  cmake -G "MinGW Makefiles" ..  -DGGML_SYCL=ON -DCMAKE_CXX_COMPILER=icx -DBUILD_SHARED_LIBS=ON -DCMAKE_BUILD_TYPE=Release -DGGML_SYCL_F16=ON
+::  cmake -G "MinGW Makefiles" ..  -DLLAMA_SYCL=ON -DCMAKE_CXX_COMPILER=icx -DBUILD_SHARED_LIBS=ON -DCMAKE_BUILD_TYPE=Release -DLLAMA_SYCL_F16=ON

 ::  for FP32
-cmake -G "Ninja" ..  -DGGML_SYCL=ON -DCMAKE_C_COMPILER=cl -DCMAKE_CXX_COMPILER=icx -DBUILD_SHARED_LIBS=ON -DCMAKE_BUILD_TYPE=Release
+cmake -G "Ninja" ..  -DLLAMA_SYCL=ON -DCMAKE_C_COMPILER=cl -DCMAKE_CXX_COMPILER=icx -DBUILD_SHARED_LIBS=ON -DCMAKE_BUILD_TYPE=Release
 if %errorlevel% neq 0 goto ERROR
 ::  build example/main only
 ::  make main
@@ -31,3 +31,4 @@ exit /B 0
 :ERROR
 echo comomand error: %errorlevel%
 exit /B %errorlevel%
+
--- a/examples/sycl/win-run-llama2.bat
+++ b/examples/sycl/win-run-llama2.bat
@@ -7,3 +7,5 @@ set INPUT2="Building a website can be done in 10 simple steps:\nStep 1:"


 .\build\bin\main.exe -m models\llama-2-7b.Q4_0.gguf -p %INPUT2% -n 400 -e -ngl 33 -s 0
+
+
--- a/flake.lock
+++ b/flake.lock
@@ -20,11 +20,11 @@
    },
    "nixpkgs": {
      "locked": {
-        "lastModified": 1719506693,
-        "narHash": "sha256-C8e9S7RzshSdHB7L+v9I51af1gDM5unhJ2xO1ywxNH8=",
+        "lastModified": 1718318537,
+        "narHash": "sha256-4Zu0RYRcAY/VWuu6awwq4opuiD//ahpc2aFHg2CWqFY=",
        "owner": "NixOS",
        "repo": "nixpkgs",
-        "rev": "b2852eb9365c6de48ffb0dc2c9562591f652242a",
+        "rev": "e9ee548d90ff586a6471b4ae80ae9cfcbceb3420",
        "type": "github"
      },
      "original": {
--- a/ggml/src/ggml-alloc.c
+++ b/ggml/src/ggml-alloc.c
@@ -988,19 +988,19 @@ ggml_backend_buffer_t ggml_backend_alloc_ctx_tensors_from_buft(struct ggml_conte
            this_size = GGML_PAD(ggml_backend_buft_get_alloc_size(buft, t), alignment);
        }

-        if (this_size > max_size) {
-            fprintf(stderr, "%s: tensor %s is too large to fit in a %s buffer (tensor size: %zu, max buffer size: %zu)\n",
-                    __func__, t->name,
-                    ggml_backend_buft_name(buft),
-                    this_size, max_size);
-            for (size_t i = 0; i < n_buffers; i++) {
-                ggml_backend_buffer_free(buffers[i]);
-            }
-            free(buffers);
-            return NULL;
-        }
+        //if (this_size > max_size) {
+        //    fprintf(stderr, "%s: tensor %s is too large to fit in a %s buffer (tensor size: %zu, max buffer size: %zu)\n",
+        //            __func__, t->name,
+        //            ggml_backend_buft_name(buft),
+        //            this_size, max_size);
+        //    for (size_t i = 0; i < n_buffers; i++) {
+        //        ggml_backend_buffer_free(buffers[i]);
+        //    }
+        //    free(buffers);
+        //    return NULL;
+        //}

-        if ((cur_buf_size + this_size) > max_size) {
+        if (cur_buf_size != 0 && (cur_buf_size + this_size) > max_size) {
            // allocate tensors in the current buffer
            if (!alloc_tensor_range(ctx, first, t, buft, cur_buf_size, &buffers, &n_buffers)) {
                return NULL;
--- a/ggml/include/ggml-alloc.h
+++ b/ggml/include/ggml-alloc.h
--- a/ggml/src/ggml-backend-impl.h
+++ b/ggml/src/ggml-backend-impl.h
--- a/ggml/src/ggml-backend.c
+++ b/ggml/src/ggml-backend.c
--- a/ggml/include/ggml-backend.h
+++ b/ggml/include/ggml-backend.h
--- a/ggml/src/ggml-blas.cpp
+++ b/ggml/src/ggml-blas.cpp
--- a/ggml/include/ggml-blas.h
+++ b/ggml/include/ggml-blas.h
--- a/ggml/src/ggml-common.h
+++ b/ggml/src/ggml-common.h
@@ -106,19 +106,19 @@ typedef sycl::half2 ggml_half2;
 #define QR6_K 2

 #define QI2_XXS (QK_K / (4*QR2_XXS))
-#define QR2_XXS 4
+#define QR2_XXS 8

 #define QI2_XS (QK_K / (4*QR2_XS))
-#define QR2_XS 4
+#define QR2_XS 8

 #define QI2_S (QK_K / (4*QR2_S))
-#define QR2_S 4
+#define QR2_S 8

 #define QI3_XXS (QK_K / (4*QR3_XXS))
-#define QR3_XXS 4
+#define QR3_XXS 8

 #define QI3_XS (QK_K / (4*QR3_XS))
-#define QR3_XS 4
+#define QR3_XS 8

 #define QI1_S (QK_K / (4*QR1_S))
 #define QR1_S 8
@@ -130,10 +130,10 @@ typedef sycl::half2 ggml_half2;
 #define QR4_NL 2

 #define QI4_XS (QK_K / (4*QR4_XS))
-#define QR4_XS 2
+#define QR4_XS 8

 #define QI3_S (QK_K / (4*QR3_S))
-#define QR3_S 4
+#define QR3_S 8

 #endif // GGML_COMMON_DECL_CUDA || GGML_COMMON_DECL_HIP

--- a/ggml/src/ggml-cuda.cu
+++ b/ggml/src/ggml-cuda.cu
@@ -576,6 +576,11 @@ GGML_CALL static size_t ggml_backend_cuda_buffer_type_get_alignment(ggml_backend
    GGML_UNUSED(buft);
 }

+GGML_CALL static size_t ggml_backend_cuda_buffer_type_get_max_size(ggml_backend_buffer_type_t buft) {
+    ggml_backend_cuda_buffer_type_context * buft_ctx = (ggml_backend_cuda_buffer_type_context *)buft->context;
+    return ggml_cuda_info().devices[buft_ctx->device].vmm_granularity;
+}
+
 GGML_CALL static size_t ggml_backend_cuda_buffer_type_get_alloc_size(ggml_backend_buffer_type_t buft, const ggml_tensor * tensor) {
    size_t size = ggml_nbytes(tensor);
    int64_t ne0 = tensor->ne[0];
@@ -595,7 +600,7 @@ static ggml_backend_buffer_type_i ggml_backend_cuda_buffer_type_interface = {
    /* .get_name         = */ ggml_backend_cuda_buffer_type_name,
    /* .alloc_buffer     = */ ggml_backend_cuda_buffer_type_alloc_buffer,
    /* .get_alignment    = */ ggml_backend_cuda_buffer_type_get_alignment,
-    /* .get_max_size     = */ NULL, // defaults to SIZE_MAX
+    /* .get_max_size     = */ ggml_backend_cuda_buffer_type_get_max_size,
    /* .get_alloc_size   = */ ggml_backend_cuda_buffer_type_get_alloc_size,
    /* .is_host          = */ NULL,
 };
@@ -1882,11 +1887,6 @@ static void ggml_cuda_mul_mat(ggml_backend_cuda_context & ctx, const ggml_tensor
    bool              use_mul_mat_q =  ggml_is_quantized(src0->type)
        && src1->type == GGML_TYPE_F32 && dst->type == GGML_TYPE_F32;

-    // if mmvq is available it's a better choice than dmmv:
-#ifndef GGML_CUDA_FORCE_DMMV
-    use_dequantize_mul_mat_vec = use_dequantize_mul_mat_vec && !use_mul_mat_vec_q;
-#endif // GGML_CUDA_FORCE_DMMV
-
    bool any_gpus_with_slow_fp16 = false;

    if (split) {
@@ -1899,15 +1899,22 @@ static void ggml_cuda_mul_mat(ggml_backend_cuda_context & ctx, const ggml_tensor
            }

            const int cc            = ggml_cuda_info().devices[id].cc;
+            use_mul_mat_vec_q       = use_mul_mat_vec_q       && cc >= MIN_CC_DP4A;
            use_mul_mat_q           = use_mul_mat_q           && ggml_cuda_should_use_mmq(src0->type, cc, src1->ne[1]);
            any_gpus_with_slow_fp16 = any_gpus_with_slow_fp16 || !fast_fp16_available(cc);
        }
    } else {
        const int cc            = ggml_cuda_info().devices[ctx.device].cc;
+        use_mul_mat_vec_q       = use_mul_mat_vec_q       && cc >= MIN_CC_DP4A;
        use_mul_mat_q           = use_mul_mat_q           && ggml_cuda_should_use_mmq(src0->type, cc, src1->ne[1]);
        any_gpus_with_slow_fp16 = any_gpus_with_slow_fp16 || !fast_fp16_available(cc);
    }

+    // if mmvq is available it's a better choice than dmmv:
+#ifndef GGML_CUDA_FORCE_DMMV
+    use_dequantize_mul_mat_vec = use_dequantize_mul_mat_vec && !use_mul_mat_vec_q;
+#endif // GGML_CUDA_FORCE_DMMV
+
    // debug helpers
    //printf("src0: %8d %8d %8d %8d\n", src0->ne[0], src0->ne[1], src0->ne[2], src0->ne[3]);
    //printf("      %8d %8d %8d %8d\n", src0->nb[0], src0->nb[1], src0->nb[2], src0->nb[3]);
@@ -2711,40 +2718,27 @@ GGML_CALL static bool ggml_backend_cuda_supports_op(ggml_backend_t backend, cons
        case GGML_OP_MUL_MAT:
        case GGML_OP_MUL_MAT_ID:
            {
-                struct ggml_tensor * a = op->src[0];
+                struct ggml_tensor * a;
+                struct ggml_tensor * b;
                if (op->op == GGML_OP_MUL_MAT) {
-                    struct ggml_tensor * b = op->src[1];
-                    if (a->ne[3] != b->ne[3]) {
+                    a = op->src[0];
+                    b = op->src[1];
+                } else {
+                    a = op->src[2];
+                    b = op->src[1];
+                }
+                if (a->ne[3] != b->ne[3]) {
+                    return false;
+                }
+                ggml_type a_type = a->type;
+                if (a_type == GGML_TYPE_IQ2_XXS || a_type == GGML_TYPE_IQ2_XS || a_type == GGML_TYPE_IQ3_XXS ||
+                    a_type == GGML_TYPE_IQ1_S   || a_type == GGML_TYPE_IQ4_NL || a_type == GGML_TYPE_IQ3_S   ||
+                    a_type == GGML_TYPE_IQ1_M   || a_type == GGML_TYPE_IQ2_S  || a_type == GGML_TYPE_IQ4_XS) {
+                    if (b->ne[1] == 1 && ggml_nrows(b) > 1) {
                        return false;
                    }
                }
-                switch (a->type) {
-                    case GGML_TYPE_F32:
-                    case GGML_TYPE_F16:
-                    case GGML_TYPE_Q4_0:
-                    case GGML_TYPE_Q4_1:
-                    case GGML_TYPE_Q5_0:
-                    case GGML_TYPE_Q5_1:
-                    case GGML_TYPE_Q8_0:
-                    case GGML_TYPE_Q2_K:
-                    case GGML_TYPE_Q3_K:
-                    case GGML_TYPE_Q4_K:
-                    case GGML_TYPE_Q5_K:
-                    case GGML_TYPE_Q6_K:
-                    case GGML_TYPE_Q8_K:
-                    case GGML_TYPE_IQ1_M:
-                    case GGML_TYPE_IQ1_S:
-                    case GGML_TYPE_IQ2_S:
-                    case GGML_TYPE_IQ2_XS:
-                    case GGML_TYPE_IQ2_XXS:
-                    case GGML_TYPE_IQ3_S:
-                    case GGML_TYPE_IQ3_XXS:
-                    case GGML_TYPE_IQ4_NL:
-                    case GGML_TYPE_IQ4_XS:
-                        return true;
-                    default:
-                        return false;
-                }
+                return true;
            } break;
        case GGML_OP_GET_ROWS:
            {
--- a/ggml/include/ggml-cuda.h
+++ b/ggml/include/ggml-cuda.h
--- a/ggml/src/ggml-cuda/acc.cu
+++ b/ggml/src/ggml-cuda/acc.cu
--- a/ggml/src/ggml-cuda/acc.cuh
+++ b/ggml/src/ggml-cuda/acc.cuh
--- a/ggml/src/ggml-cuda/arange.cu
+++ b/ggml/src/ggml-cuda/arange.cu
--- a/ggml/src/ggml-cuda/arange.cuh
+++ b/ggml/src/ggml-cuda/arange.cuh
--- a/ggml/src/ggml-cuda/argsort.cu
+++ b/ggml/src/ggml-cuda/argsort.cu
--- a/ggml/src/ggml-cuda/argsort.cuh
+++ b/ggml/src/ggml-cuda/argsort.cuh
--- a/ggml/src/ggml-cuda/binbcast.cu
+++ b/ggml/src/ggml-cuda/binbcast.cu
--- a/ggml/src/ggml-cuda/binbcast.cuh
+++ b/ggml/src/ggml-cuda/binbcast.cuh
--- a/ggml/src/ggml-cuda/clamp.cu
+++ b/ggml/src/ggml-cuda/clamp.cu
--- a/ggml/src/ggml-cuda/clamp.cuh
+++ b/ggml/src/ggml-cuda/clamp.cuh
--- a/ggml/src/ggml-cuda/common.cuh
+++ b/ggml/src/ggml-cuda/common.cuh
@@ -3,7 +3,6 @@
 #include "ggml.h"
 #include "ggml-cuda.h"

-#include <cstdint>
 #include <memory>

 #if defined(GGML_USE_HIPBLAS)
@@ -227,10 +226,6 @@ typedef float2 dfloat2;
 #define RDNA2
 #endif

-#if defined(__gfx1010__) || defined(__gfx1012__)
-#define RDNA1
-#endif
-
 #ifndef __has_builtin
    #define __has_builtin(x) 0
 #endif
@@ -273,15 +268,30 @@ static __device__ __forceinline__ unsigned int __vcmpeq4(unsigned int a, unsigne
    return c;
 }

-static __device__ __forceinline__ unsigned int __vcmpne4(unsigned int a, unsigned int b) {
-    const uint8x4_t& va = reinterpret_cast<const uint8x4_t&>(a);
-    const uint8x4_t& vb = reinterpret_cast<const uint8x4_t&>(b);
-    unsigned int c;
-    uint8x4_t& vc = reinterpret_cast<uint8x4_t&>(c);
-#pragma unroll
-    for (int i = 0; i < 4; ++i) {
-        vc[i] = va[i] == vb[i] ? 0x00 : 0xff;
-    }
+static __device__ __forceinline__ int __dp4a(const int a, const int b, int c) {
+#if defined(__gfx906__) || defined(__gfx908__) || defined(__gfx90a__) || defined(__gfx1030__)
+    c = __builtin_amdgcn_sdot4(a, b, c, false);
+#elif defined(RDNA3)
+    c = __builtin_amdgcn_sudot4( true, a, true, b, c, false);
+#elif defined(__gfx1010__) || defined(__gfx900__)
+    int tmp1;
+    int tmp2;
+    asm("\n \
+        v_mul_i32_i24 %1, sext(%3), sext(%4) dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_0 \n \
+        v_mul_i32_i24 %2, sext(%3), sext(%4) dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:BYTE_1 \n \
+        v_add3_u32 %0, %1, %2, %0 \n \
+        v_mul_i32_i24 %1, sext(%3), sext(%4) dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_2 src1_sel:BYTE_2 \n \
+        v_mul_i32_i24 %2, sext(%3), sext(%4) dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:BYTE_3 \n \
+        v_add3_u32 %0, %1, %2, %0 \n \
+        "
+        : "+v"(c), "=&v"(tmp1), "=&v"(tmp2)
+        : "v"(a), "v"(b)
+    );
+#else
+    const int8x4_t va = reinterpret_cast<const int8x4_t&>(a);
+    const int8x4_t vb = reinterpret_cast<const int8x4_t&>(b);
+    c += va[0] * vb[0] + va[1] * vb[1] + va[2] * vb[2] + va[3] * vb[3];
+#endif
    return c;
 }

@@ -457,48 +467,8 @@ static __device__ __forceinline__ uint32_t __hgt2_mask(const half2 a, const half
 }
 #endif // CUDART_VERSION < 12000

-static __device__ __forceinline__ int ggml_cuda_dp4a(const int a, const int b, int c) {
-#if defined(GGML_USE_HIPBLAS) && defined(__HIP_PLATFORM_AMD__)
-#if defined(__gfx906__) || defined(__gfx908__) || defined(__gfx90a__) || defined(__gfx1030__)
-    c = __builtin_amdgcn_sdot4(a, b, c, false);
-#elif defined(RDNA3)
-    c = __builtin_amdgcn_sudot4( true, a, true, b, c, false);
-#elif defined(__gfx1010__) || defined(__gfx900__)
-    int tmp1;
-    int tmp2;
-    asm("\n \
-        v_mul_i32_i24 %1, sext(%3), sext(%4) dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:BYTE_0 \n \
-        v_mul_i32_i24 %2, sext(%3), sext(%4) dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:BYTE_1 \n \
-        v_add3_u32 %0, %1, %2, %0 \n \
-        v_mul_i32_i24 %1, sext(%3), sext(%4) dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_2 src1_sel:BYTE_2 \n \
-        v_mul_i32_i24 %2, sext(%3), sext(%4) dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_3 src1_sel:BYTE_3 \n \
-        v_add3_u32 %0, %1, %2, %0 \n \
-        "
-        : "+v"(c), "=&v"(tmp1), "=&v"(tmp2)
-        : "v"(a), "v"(b)
-    );
-#else
-    const int8x4_t va = reinterpret_cast<const int8x4_t&>(a);
-    const int8x4_t vb = reinterpret_cast<const int8x4_t&>(b);
-    c += va[0] * vb[0] + va[1] * vb[1] + va[2] * vb[2] + va[3] * vb[3];
-#endif
-    return c;
-
-#else // defined(GGML_USE_HIPBLAS) && defined(__HIP_PLATFORM_AMD__)
-
-#if __CUDA_ARCH__ >= MIN_CC_DP4A
-    return __dp4a(a, b, c);
-#else // __CUDA_ARCH__ >= MIN_CC_DP4A
-    const int8_t * a8 = (const int8_t *) &a;
-    const int8_t * b8 = (const int8_t *) &b;
-    return c + a8[0]*b8[0] + a8[1]*b8[1] + a8[2]*b8[2] + a8[3]*b8[3];
-#endif // __CUDA_ARCH__ >= MIN_CC_DP4A
-
-#endif // defined(GGML_USE_HIPBLAS) && defined(__HIP_PLATFORM_AMD__)
-}
-
 // TODO: move to ggml-common.h
-static constexpr __device__ int8_t kvalues_iq4nl[16] = {-127, -104, -83, -65, -49, -35, -22, -10, 1, 13, 25, 38, 53, 69, 89, 113};
+static const __device__ int8_t kvalues_iq4nl[16] = {-127, -104, -83, -65, -49, -35, -22, -10, 1, 13, 25, 38, 53, 69, 89, 113};

 typedef void (*dequantize_kernel_t)(const void * vx, const int64_t ib, const int iqs, dfloat2 & v);

--- a/ggml/src/ggml-cuda/concat.cu
+++ b/ggml/src/ggml-cuda/concat.cu
--- a/ggml/src/ggml-cuda/concat.cuh
+++ b/ggml/src/ggml-cuda/concat.cuh
--- a/ggml/src/ggml-cuda/convert.cu
+++ b/ggml/src/ggml-cuda/convert.cu
--- a/ggml/src/ggml-cuda/convert.cuh
+++ b/ggml/src/ggml-cuda/convert.cuh
--- a/ggml/src/ggml-cuda/cpy.cu
+++ b/ggml/src/ggml-cuda/cpy.cu
@@ -487,3 +487,4 @@ void* ggml_cuda_cpy_fn(const ggml_tensor * src0, ggml_tensor * src1) {
        GGML_ASSERT(false);
    }
 }
+
--- a/ggml/src/ggml-cuda/cpy.cuh
+++ b/ggml/src/ggml-cuda/cpy.cuh
--- a/ggml/src/ggml-cuda/dequantize.cuh
+++ b/ggml/src/ggml-cuda/dequantize.cuh
--- a/ggml/src/ggml-cuda/diagmask.cu
+++ b/ggml/src/ggml-cuda/diagmask.cu
--- a/ggml/src/ggml-cuda/diagmask.cuh
+++ b/ggml/src/ggml-cuda/diagmask.cuh
--- a/ggml/src/ggml-cuda/dmmv.cu
+++ b/ggml/src/ggml-cuda/dmmv.cu
--- a/ggml/src/ggml-cuda/dmmv.cuh
+++ b/ggml/src/ggml-cuda/dmmv.cuh
--- a/Show More
+++ b/Show More
				`@@ -7,3 +7,5 @@ set INPUT2="Building a website can be done in 10 simple steps:\nStep 1:"`


				`.\build\bin\main.exe -m models\llama-2-7b.Q4_0.gguf -p %INPUT2% -n 400 -e -ngl 33 -s 0`