server : bump httplib to 0.19.0 (#11908)

This patch fixes a typo in command help.
prefx -> prefix

Signed-off-by: Masanari Iida <standby24x7@gmail.com>

.devops/cpu.Dockerfile

@@ -2,6 +2,10 @@ ARG UBUNTU_VERSION=22.04
 
 FROM ubuntu:$UBUNTU_VERSION AS build
 
+ARG TARGETARCH
+
+ARG GGML_CPU_ARM_ARCH=armv8-a
+
 RUN apt-get update && \
     apt-get install -y build-essential git cmake libcurl4-openssl-dev
 
@@ -9,7 +13,14 @@ WORKDIR /app
 
 COPY . .
 
-RUN cmake -S . -B build -DGGML_BACKEND_DL=ON -DGGML_NATIVE=OFF -DGGML_CPU_ALL_VARIANTS=ON -DLLAMA_CURL=ON -DCMAKE_BUILD_TYPE=Release && \
+RUN if [ "$TARGETARCH" = "amd64" ]; then \
+        cmake -S . -B build -DCMAKE_BUILD_TYPE=Release -DLLAMA_CURL=ON -DGGML_NATIVE=OFF -DGGML_BACKEND_DL=ON -DGGML_CPU_ALL_VARIANTS=ON; \
+    elif [ "$TARGETARCH" = "arm64" ]; then \
+        cmake -S . -B build -DCMAKE_BUILD_TYPE=Release -DLLAMA_CURL=ON -DGGML_NATIVE=OFF -DGGML_CPU_ARM_ARCH=${GGML_CPU_ARM_ARCH}; \
+    else \
+        echo "Unsupported architecture"; \
+        exit 1; \
+    fi && \
     cmake --build build -j $(nproc)
 
 RUN mkdir -p /app/lib && \

.devops/cuda.Dockerfile

@@ -1,6 +1,6 @@
 ARG UBUNTU_VERSION=22.04
 # This needs to generally match the container host's environment.
-ARG CUDA_VERSION=12.6.0
+ARG CUDA_VERSION=12.4.0
 # Target the CUDA build image
 ARG BASE_CUDA_DEV_CONTAINER=nvidia/cuda:${CUDA_VERSION}-devel-ubuntu${UBUNTU_VERSION}
 

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

@@ -17,10 +17,10 @@ Version:        %( date "+%%Y%%m%%d" )
 Release:        1%{?dist}
 Summary:        CPU Inference of LLaMA model in pure C/C++ (no CUDA/OpenCL)
 License:        MIT
-Source0:        https://github.com/ggerganov/llama.cpp/archive/refs/heads/master.tar.gz
+Source0:        https://github.com/ggml-org/llama.cpp/archive/refs/heads/master.tar.gz
 BuildRequires:  coreutils make gcc-c++ git cuda-toolkit
 Requires:       cuda-toolkit
-URL:            https://github.com/ggerganov/llama.cpp
+URL:            https://github.com/ggml-org/llama.cpp
 
 %define debug_package %{nil}
 %define source_date_epoch_from_changelog 0

.devops/llama-cpp.srpm.spec

@@ -18,10 +18,10 @@ Version:        %( date "+%%Y%%m%%d" )
 Release:        1%{?dist}
 Summary:        CPU Inference of LLaMA model in pure C/C++ (no CUDA/OpenCL)
 License:        MIT
-Source0:        https://github.com/ggerganov/llama.cpp/archive/refs/heads/master.tar.gz
+Source0:        https://github.com/ggml-org/llama.cpp/archive/refs/heads/master.tar.gz
 BuildRequires:  coreutils make gcc-c++ git libstdc++-devel
 Requires:       libstdc++
-URL:            https://github.com/ggerganov/llama.cpp
+URL:            https://github.com/ggml-org/llama.cpp
 
 %define debug_package %{nil}
 %define source_date_epoch_from_changelog 0

.devops/musa.Dockerfile

@@ -1,6 +1,6 @@
 ARG UBUNTU_VERSION=22.04
 # This needs to generally match the container host's environment.
-ARG MUSA_VERSION=rc3.1.0
+ARG MUSA_VERSION=rc3.1.1
 # Target the MUSA build image
 ARG BASE_MUSA_DEV_CONTAINER=mthreads/musa:${MUSA_VERSION}-devel-ubuntu${UBUNTU_VERSION}
 

.devops/nix/package.nix

@@ -133,12 +133,12 @@ effectiveStdenv.mkDerivation (finalAttrs: {
       --replace '[bundle pathForResource:@"default" ofType:@"metallib"];' "@\"$out/bin/default.metallib\";"
   '';
 
-  # With PR#6015 https://github.com/ggerganov/llama.cpp/pull/6015,
+  # With PR#6015 https://github.com/ggml-org/llama.cpp/pull/6015,
   # `default.metallib` may be compiled with Metal compiler from XCode
   # and we need to escape sandbox on MacOS to access Metal compiler.
   # `xcrun` is used find the path of the Metal compiler, which is varible
   # and not on $PATH
-  # see https://github.com/ggerganov/llama.cpp/pull/6118 for discussion
+  # see https://github.com/ggml-org/llama.cpp/pull/6118 for discussion
   __noChroot = effectiveStdenv.isDarwin && useMetalKit && precompileMetalShaders;
 
   nativeBuildInputs =
@@ -220,7 +220,7 @@ effectiveStdenv.mkDerivation (finalAttrs: {
     broken = (useMetalKit && !effectiveStdenv.isDarwin);
 
     description = "Inference of LLaMA model in pure C/C++${descriptionSuffix}";
-    homepage = "https://github.com/ggerganov/llama.cpp/";
+    homepage = "https://github.com/ggml-org/llama.cpp/";
     license = lib.licenses.mit;
 
     # Accommodates `nix run` and `lib.getExe`

.devops/rocm.Dockerfile

@@ -11,7 +11,7 @@ ARG BASE_ROCM_DEV_CONTAINER=rocm/dev-ubuntu-${UBUNTU_VERSION}:${ROCM_VERSION}-co
 FROM ${BASE_ROCM_DEV_CONTAINER} AS build
 
 # Unless otherwise specified, we make a fat build.
-# List from https://github.com/ggerganov/llama.cpp/pull/1087#issuecomment-1682807878
+# List from https://github.com/ggml-org/llama.cpp/pull/1087#issuecomment-1682807878
 # This is mostly tied to rocBLAS supported archs.
 # gfx803, gfx900, gfx1032, gfx1101, gfx1102,not officialy supported
 # gfx906 is deprecated

.devops/tools.sh

@@ -13,9 +13,13 @@ elif [[ "$arg1" == '--quantize' || "$arg1" == '-q' ]]; then
     exec ./llama-quantize "$@"
 elif [[ "$arg1" == '--run' || "$arg1" == '-r' ]]; then
     exec ./llama-cli "$@"
+elif [[ "$arg1" == '--bench' || "$arg1" == '-b' ]]; then
+    exec ./llama-bench "$@"
+elif [[ "$arg1" == '--perplexity' || "$arg1" == '-p' ]]; then
+    exec ./llama-perplexity "$@"
 elif [[ "$arg1" == '--all-in-one' || "$arg1" == '-a' ]]; then
     echo "Converting PTH to GGML..."
-    for i in `ls $1/$2/ggml-model-f16.bin*`; do
+    for i in $(ls $1/$2/ggml-model-f16.bin*); do
         if [ -f "${i/f16/q4_0}" ]; then
             echo "Skip model quantization, it already exists: ${i/f16/q4_0}"
         else
@@ -30,6 +34,10 @@ else
     echo "Available commands: "
     echo "  --run (-r): Run a model previously converted into ggml"
     echo "              ex: -m /models/7B/ggml-model-q4_0.bin -p \"Building a website can be done in 10 simple steps:\" -n 512"
+    echo "  --bench (-b): Benchmark the performance of the inference for various parameters."
+    echo "              ex: -m model.gguf"
+    echo "  --perplexity (-p): Measure the perplexity of a model over a given text."
+    echo "              ex: -m model.gguf -f file.txt"
     echo "  --convert (-c): Convert a llama model into ggml"
     echo "              ex: --outtype f16 \"/models/7B/\" "
     echo "  --quantize (-q): Optimize with quantization process ggml"

.devops/vulkan.Dockerfile

@@ -1,4 +1,4 @@
-ARG UBUNTU_VERSION=jammy
+ARG UBUNTU_VERSION=24.04
 
 FROM ubuntu:$UBUNTU_VERSION AS build
 
@@ -7,7 +7,7 @@ RUN apt update && apt install -y git build-essential cmake wget
 
 # Install Vulkan SDK and cURL
 RUN wget -qO - https://packages.lunarg.com/lunarg-signing-key-pub.asc | apt-key add - && \
-    wget -qO /etc/apt/sources.list.d/lunarg-vulkan-jammy.list https://packages.lunarg.com/vulkan/lunarg-vulkan-jammy.list && \
+    wget -qO /etc/apt/sources.list.d/lunarg-vulkan-noble.list https://packages.lunarg.com/vulkan/lunarg-vulkan-noble.list && \
     apt update -y && \
     apt-get install -y vulkan-sdk libcurl4-openssl-dev curl
 
@@ -34,7 +34,7 @@ RUN mkdir -p /app/full \
 FROM ubuntu:$UBUNTU_VERSION AS base
 
 RUN apt-get update \
-    && apt-get install -y libgomp1 curl\
+    && apt-get install -y libgomp1 curl libvulkan-dev \
     && apt autoremove -y \
     && apt clean -y \
     && rm -rf /tmp/* /var/tmp/* \
@@ -55,8 +55,9 @@ RUN apt-get update \
     git \
     python3 \
     python3-pip \
-    && pip install --upgrade pip setuptools wheel \
-    && pip install -r requirements.txt \
+    python3-wheel \
+    && pip install --break-system-packages --upgrade setuptools \
+    && pip install --break-system-packages -r requirements.txt \
     && apt autoremove -y \
     && apt clean -y \
     && rm -rf /tmp/* /var/tmp/* \

.editorconfig

@@ -40,3 +40,11 @@ indent_style = tab
 [examples/cvector-generator/*.txt]
 trim_trailing_whitespace = unset
 insert_final_newline = unset
+
+[models/templates/*.jinja]
+indent_style = unset
+indent_size = unset
+end_of_line = unset
+charset = unset
+trim_trailing_whitespace = unset
+insert_final_newline = unset

.github/ISSUE_TEMPLATE/020-enhancement.yml

@@ -6,7 +6,7 @@ body:
   - type: markdown
     attributes:
       value: |
-        [Please post your idea first in Discussion if there is not yet a consensus for this enhancement request. This will help to keep this issue tracker focused on enhancements that the community has agreed needs to be implemented.](https://github.com/ggerganov/llama.cpp/discussions/categories/ideas)
+        [Please post your idea first in Discussion if there is not yet a consensus for this enhancement request. This will help to keep this issue tracker focused on enhancements that the community has agreed needs to be implemented.](https://github.com/ggml-org/llama.cpp/discussions/categories/ideas)
 
   - type: checkboxes
     id: prerequisites
@@ -16,11 +16,11 @@ body:
       options:
         - label: I am running the latest code. Mention the version if possible as well.
           required: true
-        - label: I carefully followed the [README.md](https://github.com/ggerganov/llama.cpp/blob/master/README.md).
+        - label: I carefully followed the [README.md](https://github.com/ggml-org/llama.cpp/blob/master/README.md).
           required: true
         - label: I searched using keywords relevant to my issue to make sure that I am creating a new issue that is not already open (or closed).
           required: true
-        - label: I reviewed the [Discussions](https://github.com/ggerganov/llama.cpp/discussions), and have a new and useful enhancement to share.
+        - label: I reviewed the [Discussions](https://github.com/ggml-org/llama.cpp/discussions), and have a new and useful enhancement to share.
           required: true
 
   - type: textarea

.github/ISSUE_TEMPLATE/030-research.yml

@@ -6,7 +6,7 @@ body:
   - type: markdown
     attributes:
       value: |
-        Don't forget to check for any [duplicate research issue tickets](https://github.com/ggerganov/llama.cpp/issues?q=is%3Aopen+is%3Aissue+label%3A%22research+%F0%9F%94%AC%22)
+        Don't forget to check for any [duplicate research issue tickets](https://github.com/ggml-org/llama.cpp/issues?q=is%3Aopen+is%3Aissue+label%3A%22research+%F0%9F%94%AC%22)
 
   - type: checkboxes
     id: research-stage

.github/ISSUE_TEMPLATE/040-refactor.yml

@@ -6,8 +6,8 @@ body:
   - type: markdown
     attributes:
       value: |
-        Don't forget to [check for existing refactor issue tickets](https://github.com/ggerganov/llama.cpp/issues?q=is%3Aopen+is%3Aissue+label%3Arefactoring) in case it's already covered.
-        Also you may want to check [Pull request refactor label as well](https://github.com/ggerganov/llama.cpp/pulls?q=is%3Aopen+is%3Apr+label%3Arefactoring) for duplicates too.
+        Don't forget to [check for existing refactor issue tickets](https://github.com/ggml-org/llama.cpp/issues?q=is%3Aopen+is%3Aissue+label%3Arefactoring) in case it's already covered.
+        Also you may want to check [Pull request refactor label as well](https://github.com/ggml-org/llama.cpp/pulls?q=is%3Aopen+is%3Apr+label%3Arefactoring) for duplicates too.
 
   - type: textarea
     id: background-description

.github/ISSUE_TEMPLATE/config.yml

@@ -1,11 +1,11 @@
 blank_issues_enabled: true
 contact_links:
   - name: Got an idea?
-    url: https://github.com/ggerganov/llama.cpp/discussions/categories/ideas
+    url: https://github.com/ggml-org/llama.cpp/discussions/categories/ideas
     about: Pop it there. It may then become an enhancement ticket.
   - name: Got a question?
-    url: https://github.com/ggerganov/llama.cpp/discussions/categories/q-a
+    url: https://github.com/ggml-org/llama.cpp/discussions/categories/q-a
     about: Ask a question there!
   - name: Want to contribute?
-    url: https://github.com/ggerganov/llama.cpp/wiki/contribute
+    url: https://github.com/ggml-org/llama.cpp/wiki/contribute
     about: Head to the contribution guide page of the wiki for areas you can help with

.github/pull_request_template.md

@@ -1 +1 @@
-*Make sure to read the [contributing guidelines](https://github.com/ggerganov/llama.cpp/blob/master/CONTRIBUTING.md) before submitting a PR*
+*Make sure to read the [contributing guidelines](https://github.com/ggml-org/llama.cpp/blob/master/CONTRIBUTING.md) before submitting a PR*

.github/workflows/bench.yml.disabled

@@ -1,5 +1,5 @@
 # TODO: there have been some issues with the workflow, so disabling for now
-#       https://github.com/ggerganov/llama.cpp/issues/7893
+#       https://github.com/ggml-org/llama.cpp/issues/7893
 #
 # Benchmark
 name: Benchmark
@@ -57,17 +57,7 @@ jobs:
 
     if: |
       inputs.gpu-series == 'Standard_NC4as_T4_v3'
-      || (
-        github.event_name == 'schedule'
-        && github.ref_name == 'master'
-        && github.repository_owner == 'ggerganov'
-      )
       || github.event_name == 'pull_request_target'
-      || (
-        github.event_name == 'push'
-        && github.event.ref == 'refs/heads/master'
-        && github.repository_owner == 'ggerganov'
-      )
     steps:
       - name: Clone
         id: checkout

.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/build.yml', '**/CMakeLists.txt', '**/Makefile', '**/*.h', '**/*.hpp', '**/*.c', '**/*.cpp', '**/*.cu', '**/*.cuh', '**/*.swift', '**/*.m', '**/*.metal', '**/*.comp']
   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', '**/*.metal', '**/*.comp']
 
 concurrency:
   group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
@@ -43,6 +43,12 @@ jobs:
         with:
           fetch-depth: 0
 
+      - name: ccache
+        uses: hendrikmuhs/ccache-action@v1.2.16
+        with:
+          key: macOS-latest-cmake-arm64
+          evict-old-files: 1d
+
       - name: Dependencies
         id: depends
         continue-on-error: true
@@ -53,15 +59,14 @@ jobs:
         id: cmake_build
         run: |
           sysctl -a
-          mkdir build
-          cd build
-          cmake .. \
+          cmake -B build \
+            -DCMAKE_BUILD_RPATH="@loader_path" \
             -DLLAMA_FATAL_WARNINGS=ON \
             -DLLAMA_CURL=ON \
             -DGGML_METAL_USE_BF16=ON \
             -DGGML_METAL_EMBED_LIBRARY=ON \
             -DGGML_RPC=ON
-          cmake --build . --config Release -j $(sysctl -n hw.logicalcpu)
+          cmake --build build --config Release -j $(sysctl -n hw.logicalcpu)
 
       - name: Test
         id: cmake_test
@@ -87,6 +92,7 @@ jobs:
         if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }}
         run: |
           cp LICENSE ./build/bin/
+          cp examples/run/linenoise.cpp/LICENSE ./build/bin/LICENSE.linenoise.cpp
           zip -r llama-${{ steps.tag.outputs.name }}-bin-macos-arm64.zip ./build/bin/*
 
       - name: Upload artifacts
@@ -106,6 +112,12 @@ jobs:
         with:
           fetch-depth: 0
 
+      - name: ccache
+        uses: hendrikmuhs/ccache-action@v1.2.16
+        with:
+          key: macOS-latest-cmake-x64
+          evict-old-files: 1d
+
       - name: Dependencies
         id: depends
         continue-on-error: true
@@ -117,8 +129,9 @@ jobs:
         run: |
           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
+          # https://github.com/ggml-org/llama.cpp/actions/runs/8635935781/job/23674807267#step:5:2313
           cmake -B build \
+            -DCMAKE_BUILD_RPATH="@loader_path" \
             -DLLAMA_FATAL_WARNINGS=ON \
             -DLLAMA_CURL=ON \
             -DGGML_METAL=OFF \
@@ -149,6 +162,7 @@ jobs:
         if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }}
         run: |
           cp LICENSE ./build/bin/
+          cp examples/run/linenoise.cpp/LICENSE ./build/bin/LICENSE.linenoise.cpp
           zip -r llama-${{ steps.tag.outputs.name }}-bin-macos-x64.zip ./build/bin/*
 
       - name: Upload artifacts
@@ -158,8 +172,8 @@ jobs:
           path: llama-${{ steps.tag.outputs.name }}-bin-macos-x64.zip
           name: llama-bin-macos-x64.zip
 
-  ubuntu-latest-cmake:
-    runs-on: ubuntu-latest
+  ubuntu-cpu-cmake:
+    runs-on: ubuntu-22.04
 
     steps:
       - name: Clone
@@ -168,6 +182,12 @@ jobs:
         with:
           fetch-depth: 0
 
+      - name: ccache
+        uses: hendrikmuhs/ccache-action@v1.2.16
+        with:
+          key: ubuntu-cpu-cmake
+          evict-old-files: 1d
+
       - name: Dependencies
         id: depends
         run: |
@@ -177,10 +197,11 @@ jobs:
       - name: Build
         id: cmake_build
         run: |
-          mkdir build
-          cd build
-          cmake .. -DLLAMA_FATAL_WARNINGS=ON -DLLAMA_CURL=ON -DGGML_RPC=ON
-          cmake --build . --config Release -j $(nproc)
+          cmake -B build \
+            -DLLAMA_FATAL_WARNINGS=ON \
+            -DLLAMA_CURL=ON \
+            -DGGML_RPC=ON
+          cmake --build build --config Release -j $(nproc)
 
       - name: Test
         id: cmake_test
@@ -217,6 +238,7 @@ jobs:
         if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }}
         run: |
           cp LICENSE ./build/bin/
+          cp examples/run/linenoise.cpp/LICENSE ./build/bin/LICENSE.linenoise.cpp
           zip -r llama-${{ steps.tag.outputs.name }}-bin-ubuntu-x64.zip ./build/bin/*
 
       - name: Upload artifacts
@@ -234,7 +256,54 @@ jobs:
     strategy:
       matrix:
         sanitizer: [ADDRESS, THREAD, UNDEFINED]
-        build_type: [Debug, Release]
+        build_type: [Debug]
+
+    steps:
+      - name: Clone
+        id: checkout
+        uses: actions/checkout@v4
+
+      - name: ccache
+        uses: hendrikmuhs/ccache-action@v1.2.16
+        with:
+          key: ubuntu-latest-cmake-sanitizer-${{ matrix.sanitizer }}
+          evict-old-files: 1d
+
+      - name: Dependencies
+        id: depends
+        run: |
+          sudo apt-get update
+          sudo apt-get install build-essential
+
+      - name: Build
+        id: cmake_build
+        if: ${{ matrix.sanitizer != 'THREAD' }}
+        run: |
+          cmake -B build \
+            -DLLAMA_FATAL_WARNINGS=ON \
+            -DLLAMA_SANITIZE_${{ matrix.sanitizer }}=ON \
+            -DCMAKE_BUILD_TYPE=${{ matrix.build_type }}
+          cmake --build build --config ${{ matrix.build_type }} -j $(nproc)
+
+      - name: Build (no OpenMP)
+        id: cmake_build_no_openmp
+        if: ${{ matrix.sanitizer == 'THREAD' }}
+        run: |
+          cmake -B build \
+            -DLLAMA_FATAL_WARNINGS=ON \
+            -DLLAMA_SANITIZE_${{ matrix.sanitizer }}=ON \
+            -DCMAKE_BUILD_TYPE=${{ matrix.build_type }} \
+            -DGGML_OPENMP=OFF
+          cmake --build build --config ${{ matrix.build_type }} -j $(nproc)
+
+      - name: Test
+        id: cmake_test
+        run: |
+          cd build
+          ctest -L main --verbose --timeout 900
+
+  ubuntu-latest-llguidance:
+    runs-on: ubuntu-latest
 
     steps:
       - name: Clone
@@ -249,21 +318,13 @@ jobs:
 
       - name: Build
         id: cmake_build
-        if: ${{ matrix.sanitizer != 'THREAD' }}
         run: |
           mkdir build
           cd build
-          cmake .. -DLLAMA_FATAL_WARNINGS=ON -DLLAMA_SANITIZE_${{ matrix.sanitizer }}=ON -DCMAKE_BUILD_TYPE=${{ matrix.build_type }}
-          cmake --build . --config ${{ matrix.build_type }} -j $(nproc)
-
-      - name: Build (no OpenMP)
-        id: cmake_build_no_openmp
-        if: ${{ matrix.sanitizer == 'THREAD' }}
-        run: |
-          mkdir build
-          cd build
-          cmake .. -DLLAMA_FATAL_WARNINGS=ON -DLLAMA_SANITIZE_${{ matrix.sanitizer }}=ON -DCMAKE_BUILD_TYPE=${{ matrix.build_type }} -DGGML_OPENMP=OFF
-          cmake --build . --config ${{ matrix.build_type }} -j $(nproc)
+          cmake .. \
+            -DLLAMA_FATAL_WARNINGS=ON \
+            -DLLAMA_LLGUIDANCE=ON
+          cmake --build . --config Release -j $(nproc)
 
       - name: Test
         id: cmake_test
@@ -281,6 +342,12 @@ jobs:
         id: checkout
         uses: actions/checkout@v4
 
+      - name: ccache
+        uses: hendrikmuhs/ccache-action@v1.2.16
+        with:
+          key: ubuntu-latest-cmake-rpc
+          evict-old-files: 1d
+
       - name: Dependencies
         id: depends
         run: |
@@ -290,10 +357,9 @@ jobs:
       - name: Build
         id: cmake_build
         run: |
-          mkdir build
-          cd build
-          cmake -DGGML_RPC=ON ..
-          cmake --build . --config Release -j $(nproc)
+          cmake -B build \
+            -DGGML_RPC=ON
+          cmake --build build --config Release -j $(nproc)
 
       - name: Test
         id: cmake_test
@@ -309,6 +375,12 @@ jobs:
         id: checkout
         uses: actions/checkout@v4
 
+      - name: ccache
+        uses: hendrikmuhs/ccache-action@v1.2.16
+        with:
+          key: ubuntu-22-cmake-vulkan
+          evict-old-files: 1d
+
       - name: Dependencies
         id: depends
         run: |
@@ -320,16 +392,44 @@ jobs:
       - name: Build
         id: cmake_build
         run: |
-          mkdir build
-          cd build
-          cmake -DGGML_VULKAN=ON ..
-          cmake --build . --config Release -j $(nproc)
+          cmake -B build \
+            -DGGML_VULKAN=ON
+          cmake --build build --config Release -j $(nproc)
 
       - name: Test
         id: cmake_test
         run: |
           cd build
-          ctest -L main --verbose --timeout 900
+          # This is using llvmpipe and runs slower than other backends
+          ctest -L main --verbose --timeout 2700
+
+      - name: Determine tag name
+        id: tag
+        shell: bash
+        run: |
+          BUILD_NUMBER="$(git rev-list --count HEAD)"
+          SHORT_HASH="$(git rev-parse --short=7 HEAD)"
+          if [[ "${{ env.BRANCH_NAME }}" == "master" ]]; then
+            echo "name=b${BUILD_NUMBER}" >> $GITHUB_OUTPUT
+          else
+            SAFE_NAME=$(echo "${{ env.BRANCH_NAME }}" | tr '/' '-')
+            echo "name=${SAFE_NAME}-b${BUILD_NUMBER}-${SHORT_HASH}" >> $GITHUB_OUTPUT
+          fi
+
+      - name: Pack artifacts
+        id: pack_artifacts
+        if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }}
+        run: |
+          cp LICENSE ./build/bin/
+          cp examples/run/linenoise.cpp/LICENSE ./build/bin/LICENSE.linenoise.cpp
+          zip -r llama-${{ steps.tag.outputs.name }}-bin-ubuntu-vulkan-x64.zip ./build/bin/*
+
+      - name: Upload artifacts
+        if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }}
+        uses: actions/upload-artifact@v4
+        with:
+          path: llama-${{ steps.tag.outputs.name }}-bin-ubuntu-vulkan-x64.zip
+          name: llama-bin-ubuntu-vulkan-x64.zip
 
   ubuntu-22-cmake-hip:
     runs-on: ubuntu-22.04
@@ -346,21 +446,32 @@ jobs:
           sudo apt-get update
           sudo apt-get install -y build-essential git cmake rocblas-dev hipblas-dev
 
+      - name: ccache
+        uses: hendrikmuhs/ccache-action@v1.2.16
+        with:
+          key: ubuntu-22-cmake-hip
+          evict-old-files: 1d
+
       - name: Build with native CMake HIP support
         id: cmake_build
         run: |
-          cmake -B build -S . -DCMAKE_HIP_COMPILER="$(hipconfig -l)/clang" -DGGML_HIP=ON
+          cmake -B build -S . \
+            -DCMAKE_HIP_COMPILER="$(hipconfig -l)/clang" \
+            -DGGML_HIP=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_HIP=ON
+          cmake -B build2 -S . \
+            -DCMAKE_C_COMPILER=hipcc \
+            -DCMAKE_CXX_COMPILER=hipcc \
+            -DGGML_HIP=ON
           cmake --build build2 --config Release -j $(nproc)
 
   ubuntu-22-cmake-musa:
     runs-on: ubuntu-22.04
-    container: mthreads/musa:rc3.1.0-devel-ubuntu22.04
+    container: mthreads/musa:rc3.1.1-devel-ubuntu22.04
 
     steps:
       - name: Clone
@@ -373,10 +484,17 @@ jobs:
           apt-get update
           apt-get install -y build-essential git cmake libcurl4-openssl-dev
 
+      - name: ccache
+        uses: hendrikmuhs/ccache-action@v1.2.16
+        with:
+          key: ubuntu-22-cmake-musa
+          evict-old-files: 1d
+
       - name: Build with native CMake MUSA support
         id: cmake_build
         run: |
-          cmake -B build -S . -DGGML_MUSA=ON
+          cmake -B build -S . \
+            -DGGML_MUSA=ON
           cmake --build build --config Release -j $(nproc)
 
   ubuntu-22-cmake-sycl:
@@ -411,14 +529,21 @@ jobs:
         id: checkout
         uses: actions/checkout@v4
 
+      - name: ccache
+        uses: hendrikmuhs/ccache-action@v1.2.16
+        with:
+          key: ubuntu-22-cmake-sycl
+          evict-old-files: 1d
+
       - name: Build
         id: cmake_build
         run: |
           source /opt/intel/oneapi/setvars.sh
-          mkdir build
-          cd build
-          cmake -DGGML_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx ..
-          cmake --build . --config Release -j $(nproc)
+          cmake -B build \
+            -DGGML_SYCL=ON \
+            -DCMAKE_C_COMPILER=icx \
+            -DCMAKE_CXX_COMPILER=icpx
+          cmake --build build --config Release -j $(nproc)
 
   ubuntu-22-cmake-sycl-fp16:
     runs-on: ubuntu-22.04
@@ -452,47 +577,22 @@ jobs:
         id: checkout
         uses: actions/checkout@v4
 
+      - name: ccache
+        uses: hendrikmuhs/ccache-action@v1.2.16
+        with:
+          key: ubuntu-22-cmake-sycl-fp16
+          evict-old-files: 1d
+
       - name: Build
         id: cmake_build
         run: |
           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 --build . --config Release -j $(nproc)
-
-  # TODO: build with GGML_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
-  macOS-latest-cmake:
-    runs-on: macos-latest
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v4
-
-      - name: Dependencies
-        id: depends
-        continue-on-error: true
-        run: |
-          brew update
-
-      - name: Build
-        id: cmake_build
-        run: |
-          sysctl -a
-          mkdir build
-          cd build
-          cmake -DLLAMA_FATAL_WARNINGS=ON -DGGML_METAL=OFF ..
-          cmake --build . --config Release -j $(sysctl -n hw.logicalcpu)
-
-      - name: Test
-        id: cmake_test
-        run: |
-          cd build
-          ctest -L main --verbose --timeout 900
+          cmake -B build \
+            -DGGML_SYCL=ON \
+            -DCMAKE_C_COMPILER=icx \
+            -DCMAKE_CXX_COMPILER=icpx \
+            -DGGML_SYCL_F16=ON
+          cmake --build build --config Release -j $(nproc)
 
   macOS-latest-cmake-ios:
     runs-on: macos-latest
@@ -502,6 +602,12 @@ jobs:
         id: checkout
         uses: actions/checkout@v4
 
+      - name: ccache
+        uses: hendrikmuhs/ccache-action@v1.2.16
+        with:
+          key: macOS-latest-cmake-ios
+          evict-old-files: 1d
+
       - name: Dependencies
         id: depends
         continue-on-error: true
@@ -512,9 +618,7 @@ jobs:
         id: cmake_build
         run: |
           sysctl -a
-          mkdir build
-          cd build
-          cmake -G Xcode .. \
+          cmake -B build -G Xcode \
             -DGGML_METAL_USE_BF16=ON \
             -DGGML_METAL_EMBED_LIBRARY=ON \
             -DLLAMA_BUILD_EXAMPLES=OFF \
@@ -523,7 +627,7 @@ jobs:
             -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
+          cmake --build build --config Release -j $(sysctl -n hw.logicalcpu) -- CODE_SIGNING_ALLOWED=NO
 
   macOS-latest-cmake-tvos:
     runs-on: macos-latest
@@ -533,6 +637,12 @@ jobs:
         id: checkout
         uses: actions/checkout@v4
 
+      - name: ccache
+        uses: hendrikmuhs/ccache-action@v1.2.16
+        with:
+          key: macOS-latest-cmake-tvos
+          evict-old-files: 1d
+
       - name: Dependencies
         id: depends
         continue-on-error: true
@@ -543,9 +653,7 @@ jobs:
         id: cmake_build
         run: |
           sysctl -a
-          mkdir build
-          cd build
-          cmake -G Xcode .. \
+          cmake -B build -G Xcode \
             -DGGML_METAL_USE_BF16=ON \
             -DGGML_METAL_EMBED_LIBRARY=ON \
             -DLLAMA_BUILD_EXAMPLES=OFF \
@@ -554,7 +662,7 @@ jobs:
             -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
+          cmake --build build --config Release -j $(sysctl -n hw.logicalcpu) -- CODE_SIGNING_ALLOWED=NO
 
   macOS-latest-swift:
     runs-on: macos-latest
@@ -568,6 +676,12 @@ jobs:
         id: checkout
         uses: actions/checkout@v4
 
+      - name: ccache
+        uses: hendrikmuhs/ccache-action@v1.2.16
+        with:
+          key: macOS-latest-swift
+          evict-old-files: 1d
+
       - name: Dependencies
         id: depends
         continue-on-error: true
@@ -578,17 +692,15 @@ jobs:
         id: cmake_build
         run: |
           sysctl -a
-          mkdir build
-          cd build
-          cmake -G Xcode .. \
+          cmake -B build -G Xcode \
             -DGGML_METAL_USE_BF16=ON \
             -DGGML_METAL_EMBED_LIBRARY=ON \
             -DLLAMA_BUILD_EXAMPLES=OFF \
             -DLLAMA_BUILD_TESTS=OFF \
             -DLLAMA_BUILD_SERVER=OFF \
             -DCMAKE_OSX_ARCHITECTURES="arm64;x86_64"
-          cmake --build . --config Release -j $(sysctl -n hw.logicalcpu)
-          sudo cmake --install . --config Release
+          cmake --build build --config Release -j $(sysctl -n hw.logicalcpu)
+          sudo cmake --install build --config Release
 
       - name: xcodebuild for swift package
         id: xcodebuild
@@ -609,6 +721,13 @@ jobs:
       - name: Clone
         uses: actions/checkout@v4
 
+      - name: ccache
+        uses: hendrikmuhs/ccache-action@v1.2.16
+        with:
+          key: windows-msys2
+          variant: sccache
+          evict-old-files: 1d
+
       - name: Setup ${{ matrix.sys }}
         uses: msys2/setup-msys2@v2
         with:
@@ -616,6 +735,7 @@ jobs:
           msystem: ${{matrix.sys}}
           install: >-
             base-devel
+            git
             mingw-w64-${{matrix.env}}-toolchain
             mingw-w64-${{matrix.env}}-cmake
             mingw-w64-${{matrix.env}}-openblas
@@ -676,6 +796,13 @@ jobs:
         with:
           fetch-depth: 0
 
+      - name: ccache
+        uses: hendrikmuhs/ccache-action@v1.2.16
+        with:
+          key: windows-latest-cmake-${{ matrix.build }}
+          variant: sccache
+          evict-old-files: 1d
+
       - name: Clone Kompute submodule
         id: clone_kompute
         if: ${{ matrix.build == 'kompute-x64' }}
@@ -715,21 +842,19 @@ jobs:
         run: |
           git clone https://github.com/KhronosGroup/OpenCL-Headers
           cd OpenCL-Headers
-          mkdir build && cd build
-          cmake .. `
+          cmake -B build `
             -DBUILD_TESTING=OFF `
             -DOPENCL_HEADERS_BUILD_TESTING=OFF `
             -DOPENCL_HEADERS_BUILD_CXX_TESTS=OFF `
             -DCMAKE_INSTALL_PREFIX="$env:RUNNER_TEMP/opencl-arm64-release"
-          cmake --build . --target install
+          cmake --build build --target install
           git clone https://github.com/KhronosGroup/OpenCL-ICD-Loader
           cd OpenCL-ICD-Loader
-          mkdir build-arm64-release && cd build-arm64-release
-          cmake .. `
+          cmake -B build-arm64-release `
             -A arm64 `
             -DCMAKE_PREFIX_PATH="$env:RUNNER_TEMP/opencl-arm64-release" `
             -DCMAKE_INSTALL_PREFIX="$env:RUNNER_TEMP/opencl-arm64-release"
-          cmake --build . --target install --config release
+          cmake --build build-arm64-release --target install --config release
 
       - name: Build
         id: cmake_build
@@ -796,6 +921,7 @@ jobs:
         if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }}
         run: |
           Copy-Item LICENSE .\build\bin\Release\llama.cpp.txt
+          Copy-Item .\examples\run\linenoise.cpp\LICENSE .\build\bin\Release\linenoise.cpp.txt
           7z a llama-${{ steps.tag.outputs.name }}-bin-win-${{ matrix.build }}.zip .\build\bin\Release\*
 
       - name: Upload artifacts
@@ -813,6 +939,8 @@ jobs:
         - name: Clone
           id: checkout
           uses: actions/checkout@v4
+          with:
+            fetch-depth: 0
 
         - name: Install dependencies
           env:
@@ -821,9 +949,21 @@ jobs:
               apt update
               apt install -y cmake build-essential ninja-build libgomp1 git
 
+        - name: ccache
+          uses: hendrikmuhs/ccache-action@v1.2.16
+          with:
+            key: ubuntu-latest-cmake-cuda
+            evict-old-files: 1d
+
         - name: Build with CMake
           run: |
-            cmake -S . -B build -G Ninja -DCMAKE_BUILD_TYPE=Release -DGGML_NATIVE=OFF -DGGML_CUDA=ON -DCMAKE_CUDA_ARCHITECTURES=89-real -DCMAKE_EXE_LINKER_FLAGS=-Wl,--allow-shlib-undefined -DLLAMA_FATAL_WARNINGS=ON
+            cmake -S . -B build -G Ninja \
+              -DCMAKE_BUILD_TYPE=Release \
+              -DCMAKE_CUDA_ARCHITECTURES=89-real \
+              -DCMAKE_EXE_LINKER_FLAGS=-Wl,--allow-shlib-undefined \
+              -DLLAMA_FATAL_WARNINGS=ON \
+              -DGGML_NATIVE=OFF \
+              -DGGML_CUDA=ON
             cmake --build build
 
   windows-2019-cmake-cuda:
@@ -841,6 +981,13 @@ jobs:
         with:
             fetch-depth: 0
 
+      - name: Install ccache
+        uses: hendrikmuhs/ccache-action@v1.2.16
+        with:
+          key: ${{ github.job }}-${{ matrix.cuda }}-${{ matrix.build }}
+          variant: sccache
+          evict-old-files: 1d
+
       - name: Install Cuda Toolkit 11.7
         if: ${{ matrix.cuda == '11.7' }}
         run: |
@@ -897,11 +1044,6 @@ jobs:
           echo "CUDA_PATH=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.4" | Out-File -FilePath $env:GITHUB_ENV -Append -Encoding utf8
           echo "CUDA_PATH_V12_4=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.4" | Out-File -FilePath $env:GITHUB_ENV -Append -Encoding utf8
 
-      - name: Install ccache
-        uses: hendrikmuhs/ccache-action@v1.2
-        with:
-          key: ${{ github.job }}-${{ matrix.cuda }}-${{ matrix.build }}
-
       - name: Install Ninja
         id: install_ninja
         run: |
@@ -912,7 +1054,11 @@ jobs:
         shell: cmd
         run: |
           call "C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\VC\Auxiliary\Build\vcvars64.bat"
-          cmake -S . -B build -G "Ninja Multi-Config" -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_CUDA=ON -DGGML_RPC=ON
+          cmake -S . -B build -G "Ninja Multi-Config" ^
+            -DLLAMA_BUILD_SERVER=ON ^
+            -DGGML_NATIVE=OFF ^
+            -DGGML_CUDA=ON ^
+            -DGGML_RPC=ON
           set /A NINJA_JOBS=%NUMBER_OF_PROCESSORS%-1
           cmake --build build --config Release -j %NINJA_JOBS% -t ggml
           cmake --build build --config Release
@@ -977,6 +1123,13 @@ jobs:
         with:
           fetch-depth: 0
 
+      - name: ccache
+        uses: hendrikmuhs/ccache-action@v1.2.16
+        with:
+          key: windows-latest-cmake-sycl
+          variant: sccache
+          evict-old-files: 1d
+
       - name: Install
         run:  |
           scripts/install-oneapi.bat $WINDOWS_BASEKIT_URL $WINDOWS_DPCPP_MKL
@@ -1056,16 +1209,22 @@ jobs:
           & 'C:\Program Files\AMD\ROCm\*\bin\clang.exe' --version
 
       - name: Install ccache
-        uses: hendrikmuhs/ccache-action@v1.2
+        uses: hendrikmuhs/ccache-action@v1.2.16
         with:
           key: ${{ github.job }}
+          evict-old-files: 1d
 
       - name: Build
         id: cmake_build
         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_HIP=ON -DCMAKE_BUILD_TYPE=Release -DGGML_RPC=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" `
+            -DCMAKE_BUILD_TYPE=Release `
+            -DGGML_HIP=ON `
+            -DGGML_RPC=ON
           cmake --build build -j ${env:NUMBER_OF_PROCESSORS}
 
   windows-latest-cmake-hip-release:
@@ -1083,6 +1242,12 @@ jobs:
         with:
             fetch-depth: 0
 
+      - name: ccache
+        uses: hendrikmuhs/ccache-action@v1.2.16
+        with:
+          key: windows-latest-cmake-hip-release
+          evict-old-files: 1d
+
       - name: Install
         id: depends
         run: |
@@ -1103,7 +1268,13 @@ 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_HIP=ON -DCMAKE_BUILD_TYPE=Release -DAMDGPU_TARGETS=${{ matrix.gpu_target }} -DGGML_RPC=ON
+          cmake -G "Unix Makefiles" -B build -S . `
+            -DCMAKE_C_COMPILER="${env:HIP_PATH}\bin\clang.exe" `
+            -DCMAKE_CXX_COMPILER="${env:HIP_PATH}\bin\clang++.exe" `
+            -DCMAKE_BUILD_TYPE=Release `
+            -DAMDGPU_TARGETS=${{ matrix.gpu_target }} `
+            -DGGML_HIP=ON `
+            -DGGML_RPC=ON
           cmake --build build -j ${env:NUMBER_OF_PROCESSORS}
           md "build\bin\rocblas\library\"
           cp "${env:HIP_PATH}\bin\hipblas.dll" "build\bin\"
@@ -1145,9 +1316,7 @@ jobs:
         id: cmake_build
         run: |
           sysctl -a
-          mkdir build
-          cd build
-          cmake -G Xcode .. \
+          cmake -B build -G Xcode \
             -DGGML_METAL_USE_BF16=ON \
             -DGGML_METAL_EMBED_LIBRARY=ON \
             -DLLAMA_BUILD_EXAMPLES=OFF \
@@ -1156,8 +1325,8 @@ jobs:
             -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
-          sudo cmake --install . --config Release
+          cmake --build build --config Release -j $(sysctl -n hw.logicalcpu) -- CODE_SIGNING_ALLOWED=NO
+          sudo cmake --install build --config Release
 
       - name: xcodebuild for swift package
         id: xcodebuild
@@ -1174,6 +1343,12 @@ jobs:
       - name: Clone
         uses: actions/checkout@v4
 
+      - name: ccache
+        uses: hendrikmuhs/ccache-action@v1.2.16
+        with:
+          key: android-build
+          evict-old-files: 1d
+
       - name: Set up JDK
         uses: actions/setup-java@v3
         with:
@@ -1197,8 +1372,7 @@ jobs:
     runs-on: ubuntu-latest
 
     needs:
-      - ubuntu-latest-cmake
-      - macOS-latest-cmake
+      - ubuntu-cpu-cmake
       - windows-latest-cmake
       - windows-2019-cmake-cuda
       - windows-latest-cmake-hip-release
@@ -1212,6 +1386,12 @@ jobs:
         with:
           fetch-depth: 0
 
+      - name: ccache
+        uses: hendrikmuhs/ccache-action@v1.2.16
+        with:
+          key: release
+          evict-old-files: 1d
+
       - name: Determine tag name
         id: tag
         shell: bash
@@ -1457,3 +1637,37 @@ jobs:
 #          popd
 #          emcmake cmake . -DCMAKE_BUILD_TYPE=${{ matrix.build }}
 #          make
+
+  openEuler-latest-cmake-cann:
+    if: ${{ github.event_name != 'pull_request' || contains(github.event.pull_request.labels.*.name, 'Ascend NPU') }}
+    defaults:
+      run:
+       shell: bash -el {0}
+    runs-on: ubuntu-24.04-arm
+    strategy:
+      matrix:
+        cann:
+          - '8.0.rc3.beta1-910b-openeuler22.03-py3.10'
+        device:
+          - 'ascend910b3'
+        build:
+          - 'Release'
+    container: ascendai/cann:${{ matrix.cann }}
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+
+      - name: Dependencies
+        run: |
+          yum update -y
+          yum install -y git gcc gcc-c++ make cmake
+
+      - name: Build
+        run: |
+          export LD_LIBRARY_PATH=${ASCEND_TOOLKIT_HOME}/lib64:${ASCEND_TOOLKIT_HOME}/$(uname -m)-linux/devlib/:${LD_LIBRARY_PATH}
+
+          cmake -S . -B build \
+              -DCMAKE_BUILD_TYPE=${{ matrix.build }} \
+              -DGGML_CANN=on \
+              -DSOC_TYPE=${{ matrix.device }}
+          cmake --build build -j $(nproc)

.github/workflows/close-issue.yml

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

.github/workflows/docker.yml

@@ -28,10 +28,11 @@ jobs:
   push_to_registry:
     name: Push Docker image to Docker Hub
 
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-22.04
     env:
       COMMIT_SHA: ${{ github.sha }}
     strategy:
+      fail-fast: false
       matrix:
         config:
           # Multi-stage build
@@ -50,6 +51,8 @@ jobs:
 
       - name: Set up QEMU
         uses: docker/setup-qemu-action@v3
+        with:
+          image: tonistiigi/binfmt:qemu-v7.0.0-28
 
       - name: Set up Docker Buildx
         uses: docker/setup-buildx-action@v3

.github/workflows/labeler.yml

@@ -11,7 +11,7 @@ jobs:
     steps:
     - uses: actions/checkout@v4
       with:
-        repository: "ggerganov/llama.cpp"
+        repository: "ggml-org/llama.cpp"
     - uses: actions/labeler@v5
       with:
         configuration-path: '.github/labeler.yml'

.github/workflows/server.yml

@@ -81,13 +81,36 @@ jobs:
         with:
           node-version: '22.11.0'
 
+      - name: WebUI - Install dependencies
+        id: webui_lint
+        run: |
+          cd examples/server/webui
+          npm ci
+
+      - name: WebUI - Check code format
+        id: webui_format
+        run: |
+          git config --global --add safe.directory $(realpath .)
+          cd examples/server/webui
+          git status
+
+          npm run format
+          git status
+          modified_files="$(git status -s)"
+          echo "Modified files: ${modified_files}"
+          if [ -n "${modified_files}" ]; then
+            echo "Files do not follow coding style. To fix: npm run format"
+            echo "${modified_files}"
+            exit 1
+          fi
+
       - name: Verify bundled index.html
         id: verify_server_index_html
         run: |
           git config --global --add safe.directory $(realpath .)
           cd examples/server/webui
           git status
-          npm ci
+
           npm run build
           git status
           modified_files="$(git status -s)"
@@ -112,9 +135,9 @@ jobs:
               -DGGML_OPENMP=OFF ;
           cmake --build build --config ${{ matrix.build_type }} -j $(nproc) --target llama-server
 
-      - name: Build
-        id: cmake_build
-        if: ${{ matrix.sanitizer != 'THREAD' }}
+      - name: Build (sanitizers)
+        id: cmake_build_sanitizers
+        if: ${{ matrix.sanitizer != '' && matrix.sanitizer != 'THREAD' }}
         run: |
           cmake -B build \
               -DGGML_NATIVE=OFF \
@@ -124,12 +147,31 @@ jobs:
               -DLLAMA_SANITIZE_${{ matrix.sanitizer }}=ON ;
           cmake --build build --config ${{ matrix.build_type }} -j $(nproc) --target llama-server
 
+      - name: Build (sanitizers)
+        id: cmake_build
+        if: ${{ matrix.sanitizer == '' }}
+        run: |
+          cmake -B build \
+              -DGGML_NATIVE=OFF \
+              -DLLAMA_BUILD_SERVER=ON \
+              -DLLAMA_CURL=ON \
+              -DCMAKE_BUILD_TYPE=${{ matrix.build_type }} ;
+          cmake --build build --config ${{ matrix.build_type }} -j $(nproc) --target llama-server
+
       - name: Tests
         id: server_integration_tests
+        if: ${{ matrix.sanitizer == '' }}
         run: |
           cd examples/server/tests
           ./tests.sh
 
+      - name: Tests (sanitizers)
+        id: server_integration_tests_sanitizers
+        if: ${{ matrix.sanitizer != '' }}
+        run: |
+          cd examples/server/tests
+          LLAMA_SANITIZE=1 ./tests.sh
+
       - name: Slow tests
         id: server_integration_tests_slow
         if: ${{ (github.event.schedule || github.event.inputs.slow_tests == 'true') && matrix.build_type == 'Release' }}
@@ -186,7 +228,7 @@ jobs:
         run: |
           cd examples/server/tests
           $env:PYTHONIOENCODING = ":replace"
-          pytest -v -x
+          pytest -v -x -m "not slow"
 
       - name: Slow tests
         id: server_integration_tests_slow

.gitignore

@@ -18,6 +18,7 @@
 *.metallib
 *.o
 *.so
+*.swp
 *.tmp
 
 # IDE / OS

AUTHORS

@@ -1,4 +1,4 @@
-# date: Thu Nov 28 20:46:15 EET 2024
+# date: Tue Feb  4 13:04:05 EET 2025
 # this file is auto-generated by scripts/gen-authors.sh
 
 0cc4m <picard12@live.de>
@@ -20,6 +20,8 @@ Adithya Balaji <adithya.b94@gmail.com>
 AdithyanI <adithyan.i4internet@gmail.com>
 Adrian <smith.adriane@gmail.com>
 Adrian Hesketh <a-h@users.noreply.github.com>
+Adrien Gallouët <adrien@gallouet.fr>
+Adrien Gallouët <angt@huggingface.co>
 Ahmad Tameem <113388789+Tameem-10xE@users.noreply.github.com>
 Ahmet Zeer <ahmed.zeer@std.yildiz.edu.tr>
 AidanBeltonS <87009434+AidanBeltonS@users.noreply.github.com>
@@ -55,6 +57,7 @@ Ananta Bastola <anantarajbastola@gmail.com>
 Anas Ahouzi <112881240+aahouzi@users.noreply.github.com>
 András Salamon <ott2@users.noreply.github.com>
 Andreas (Andi) Kunar <andreask@msn.com>
+Andreas Kieslinger <47689530+aendk@users.noreply.github.com>
 Andrei <abetlen@gmail.com>
 Andrew Canis <andrew.canis@gmail.com>
 Andrew Downing <andrew2085@gmail.com>
@@ -91,13 +94,17 @@ 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>
+Benson Wong <mostlygeek@gmail.com>
 Bernat Vadell <hounter.caza@gmail.com>
+Bernhard M. Wiedemann <githubbmwprimary@lsmod.de>
 Bert Wagner <github@bertwagner.com>
+Billel Mokeddem <billel.mokeddem.ml@gmail.com>
 Bingan <70050083+binganao@users.noreply.github.com>
 Bjarke Viksøe <164612031+bviksoe@users.noreply.github.com>
 Bodo Graumann <mail@bodograumann.de>
 Bono Lv <lvscar@users.noreply.github.com>
 Borislav Stanimirov <b.stanimirov@abv.bg>
+Borislav Stanimirov <b@ibob.bg>
 Branden Butler <bwtbutler@hotmail.com>
 Brandon Squizzato <35474886+bsquizz@users.noreply.github.com>
 Brian <mofosyne@gmail.com>
@@ -117,6 +124,7 @@ Casey Primozic <casey@cprimozic.net>
 Casey Primozic <me@ameo.link>
 CausalLM <148736309+CausalLM@users.noreply.github.com>
 Cebtenzzre <cebtenzzre@gmail.com>
+CentricStorm <CentricStorm@users.noreply.github.com>
 Chad Brewbaker <crb002@gmail.com>
 Changyeon Kim <cyzero.kim@samsung.com>
 Chao Jiang <jc19chaoj@zoho.com>
@@ -131,12 +139,15 @@ 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 Kastner <ckk@kvr.at>
 Christian Kögler <ck3d@gmx.de>
 Christian Köhnenkamp <cvk5@me.com>
 Christian Zhou-Zheng <59622928+christianazinn@users.noreply.github.com>
+Christopher Nielsen <62156882+mascguy@users.noreply.github.com>
 Clark Saben <76020733+csaben@users.noreply.github.com>
 Clint Herron <hanclinto@gmail.com>
 Conrad Kramer <conrad@conradkramer.com>
+Corentin REGAL <corentin.regal@gmail.com>
 CrispStrobe <154636388+CrispStrobe@users.noreply.github.com>
 Csaba Kecskemeti <csaba.kecskemeti@gmail.com>
 Cuong Trinh Manh <nguoithichkhampha@gmail.com>
@@ -176,6 +187,7 @@ Dibakar Gope <dibakar.gope@arm.com>
 Didzis Gosko <didzis@users.noreply.github.com>
 Diego Devesa <slarengh@gmail.com>
 Diogo Teles Sant'Anna <diogoteles@google.com>
+Djip007 <3705339+Djip007@users.noreply.github.com>
 Djip007 <djip.perois@free.fr>
 Don Mahurin <dmahurin@users.noreply.github.com>
 DooWoong Lee (David) <manics99@naver.com>
@@ -193,6 +205,7 @@ Edward Taylor <edeetee@gmail.com>
 Elaine <elaine.zosa@gmail.com>
 Elbios <141279586+Elbios@users.noreply.github.com>
 Elton Kola <eltonkola@gmail.com>
+Emreerdog <34742675+Emreerdog@users.noreply.github.com>
 Engininja2 <139037756+Engininja2@users.noreply.github.com>
 Equim <sayaka@ekyu.moe>
 Eric Curtin <ecurtin@redhat.com>
@@ -233,6 +246,7 @@ Fred Douglas <43351173+fredlas@users.noreply.github.com>
 Frederik Vogel <Schaltfehler@users.noreply.github.com>
 Gabe Goodhart <gabe.l.hart@gmail.com>
 Gabe Goodhart <ghart@us.ibm.com>
+Gaetan Bisson <gaetan@fenua.org>
 GainLee <perfecter.gen@gmail.com>
 Galunid <karolek1231456@gmail.com>
 Gary Linscott <glinscott@gmail.com>
@@ -249,6 +263,7 @@ Guillaume "Vermeille" Sanchez <Guillaume.V.Sanchez@gmail.com>
 Guillaume Wenzek <gwenzek@users.noreply.github.com>
 Guoliang Hua <32868157+nbcsm@users.noreply.github.com>
 Guoteng <32697156+SolenoidWGT@users.noreply.github.com>
+Guspan Tanadi <36249910+guspan-tanadi@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>
@@ -259,11 +274,13 @@ 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>
+Haus1 <haus.xda@gmail.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>
+HimariO <dsfhe49854@gmail.com>
 Hoang Nguyen <hugo53@users.noreply.github.com>
 Hong Bo PENG <penghb@cn.ibm.com>
 Hongyu Ouyang <96765450+casavaca@users.noreply.github.com>
@@ -280,6 +297,7 @@ Icecream95 <the.real.icecream95@gmail.com>
 Ido S <ido.pluto@gmail.com>
 IgnacioFDM <ignaciofdm@gmail.com>
 Igor Okulist <okigan@gmail.com>
+Ihar Hrachyshka <ihrachys@redhat.com>
 Ikko Eltociear Ashimine <eltociear@gmail.com>
 Ilya Kurdyukov <59548320+ilyakurdyukov@users.noreply.github.com>
 Ionoclast Laboratories <brigham@ionoclast.com>
@@ -289,12 +307,14 @@ Ivan <nekotekina@gmail.com>
 Ivan Filipov <159561759+vanaka11@users.noreply.github.com>
 Ivan Komarov <Ivan.Komarov@dfyz.info>
 Ivan Stepanov <ivanstepanovftw@gmail.com>
+JFLFY2255 <JFLFY2255@163.com>
 JH23X <165871467+JH23X@users.noreply.github.com>
 Jack Mousseau <jack@software.inc>
 Jack Mousseau <jmousseau@users.noreply.github.com>
 JackJollimore <130917767+JackJollimore@users.noreply.github.com>
 Jaeden Amero <jaeden@patater.com>
 Jaemin Son <woalsdnd@gmail.com>
+Jafar Uruç <jafar.uruc@gmail.com>
 Jag Chadha <jagtesh@gmail.com>
 Jakub N <jakubniemczyk97@gmail.com>
 James A Capozzoli <157492257+jac-jim@users.noreply.github.com>
@@ -315,6 +335,7 @@ Jeffrey Morgan <jmorganca@gmail.com>
 Jeffrey Quesnelle <emozilla@nousresearch.com>
 Jeroen Mostert <jeroen.mostert@cm.com>
 Jesse Jojo Johnson <williamsaintgeorge@gmail.com>
+Jett Janiak <jettjaniak@gmail.com>
 Jeximo <jeximo@gmail.com>
 Jhen-Jie Hong <iainst0409@gmail.com>
 Jiahao Li <liplus17@163.com>
@@ -343,6 +364,7 @@ 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>
+Juk Armstrong <69222624+jukofyork@users.noreply.github.com>
 Julius Arkenberg <arki05@users.noreply.github.com>
 Jun Hee Yoo <contact.jhyoo@gmail.com>
 Jun Jie <71215065+junnjiee16@users.noreply.github.com>
@@ -357,6 +379,7 @@ Justine Tunney <jtunney@mozilla.com>
 Juuso Alasuutari <juuso.alasuutari@gmail.com>
 KASR <karim.asrih@gmail.com>
 Kamil Tomšík <info@tomsik.cz>
+Karol Kontny <82021046+kkontny@users.noreply.github.com>
 Karsten Weiss <knweiss@gmail.com>
 Karthick <j.karthic2004@gmail.com>
 Karthik Kumar Viswanathan <195178+guilt@users.noreply.github.com>
@@ -376,6 +399,7 @@ Kolen Cheung <ickc@users.noreply.github.com>
 Konstantin Herud <konstantin.herud@denkbares.com>
 Konstantin Zhuravlyov <konstantin.zhuravlyov@amd.com>
 Kunshang Ji <kunshang.ji@intel.com>
+Kyle Bruene <KyleBruene@users.noreply.github.com>
 Kyle Liang <liangmanlai@gmail.com>
 Kyle Mistele <kyle@mistele.com>
 Kylin <56434533+KyL0N@users.noreply.github.com>
@@ -394,6 +418,7 @@ Liu Jia <jia3.liu@intel.com>
 LoganDark <github@logandark.mozmail.com>
 Loïc Carrère <loic.carrere@gmail.com>
 LostRuins <39025047+LostRuins@users.noreply.github.com>
+LostRuins Concedo <39025047+LostRuins@users.noreply.github.com>
 Luciano <lucianostrika44@gmail.com>
 Luo Tian <lt@basecity.com>
 Lyle Dean <dean@lyle.dev>
@@ -423,6 +448,7 @@ 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>
+Mathieu Baudier <mbaudier@argeo.org>
 Mathieu Geli <mathieu.geli@gmail.com>
 Mathieu Nayrolles <MathieuNls@users.noreply.github.com>
 Mathijs Henquet <mathijs.henquet@gmail.com>
@@ -444,6 +470,7 @@ Meng, Hengyu <hengyu.meng@intel.com>
 Mengqing Cao <cmq0113@163.com>
 Merrick Christensen <merrick.christensen@gmail.com>
 Michael Coppola <m18coppola@gmail.com>
+Michael Engel <mengel@redhat.com>
 Michael Francis <edude03@gmail.com>
 Michael Hueschen <m@mhueschen.dev>
 Michael Kesper <mkesper@schokokeks.org>
@@ -452,7 +479,9 @@ Michael Podvitskiy <podvitskiymichael@gmail.com>
 Michael Potter <NanoTekGuy@Gmail.com>
 Michael de Gans <michael.john.degans@gmail.com>
 Michaël de Vries <vriesdemichael@gmail.com>
+Michał Moskal <michal@moskal.me>
 Michał Tuszyński <srgtuszy@gmail.com>
+Michelle Tan <41475767+MichelleTanPY@users.noreply.github.com>
 Mihai <mihai.chirculescu@yahoo.com>
 Mike <ytianhui2004@gmail.com>
 Mikko Juola <mikjuo@gmail.com>
@@ -477,6 +506,7 @@ 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>
+NeverLucky <92274250+nvrxq@users.noreply.github.com>
 Nexes the Old <124105151+Nexesenex@users.noreply.github.com>
 Nexesenex <124105151+Nexesenex@users.noreply.github.com>
 Niall Coates <1349685+Niall-@users.noreply.github.com>
@@ -484,11 +514,15 @@ Nicholai Tukanov <nicholaitukanov@gmail.com>
 Nico Bosshard <nico@bosshome.ch>
 Nicolai Weitkemper <kontakt@nicolaiweitkemper.de>
 Nicolás Pérez <nicolas_perez@brown.edu>
+Nicolò Scipione <nicolo.scipione@codeplay.com>
 Nigel Bosch <pnigelb@gmail.com>
+Nikita Sarychev <42014488+sARY77@users.noreply.github.com>
 Niklas Korz <niklas@niklaskorz.de>
 NikolaiLyssogor <59844691+NikolaiLyssogor@users.noreply.github.com>
+Nikolaos Pothitos <pothitos@di.uoa.gr>
 Nikolas <127742645+nneubacher@users.noreply.github.com>
 Nindaleth <Nindaleth@users.noreply.github.com>
+Nuno <rare-magma@posteo.eu>
 OSecret <135510162+OLSecret@users.noreply.github.com>
 Oleksandr Nikitin <oleksandr@tvori.info>
 Oleksii Maryshchenko <oleksii.maryshchenko@gmail.com>
@@ -504,6 +538,7 @@ Pavel Zloi <github.com@drteam.rocks>
 Pavol Rusnak <pavol@rusnak.io>
 Paweł Wodnicki <151604+32bitmicro@users.noreply.github.com>
 Pedro Cuenca <pedro@huggingface.co>
+Peter <peter277@users.noreply.github.com>
 Peter Sugihara <peter@campsh.com>
 Phil H <5756783+phiharri@users.noreply.github.com>
 Philip Taron <philip.taron@gmail.com>
@@ -529,9 +564,12 @@ Rand Xie <randxiexyy29@gmail.com>
 Randall Fitzgerald <randall@dasaku.net>
 Random Fly <renfei8@live.cn>
 Reinforce-II <fate@eastal.com>
+Rémy Oudompheng <oudomphe@phare.normalesup.org>
 Ren Xuancheng <jklj077@users.noreply.github.com>
 Rene Leonhardt <65483435+reneleonhardt@users.noreply.github.com>
+Reza Kakhki <rezakakhki.de@gmail.com>
 RhinoDevel <RhinoDevel@users.noreply.github.com>
+Riccardo Orlando <Riccorl@users.noreply.github.com>
 Riceball LEE <snowyu.lee@gmail.com>
 Rich Dougherty <rich@rd.nz>
 Richard Kiss <him@richardkiss.com>
@@ -544,6 +582,8 @@ Riley Stewart <ristew@users.noreply.github.com>
 Rinne <AsakusaRinne@gmail.com>
 Rinne <liu_yaohui1998@126.com>
 Robert Brisita <986796+rbrisita@users.noreply.github.com>
+Robert Collins <roberto.tomas.cuentas@gmail.com>
+Robert Ormandi <52251610+ormandi@users.noreply.github.com>
 Robert Sung-wook Shin <edp1096@users.noreply.github.com>
 Robey Holderith <robey@flaminglunchbox.net>
 Robyn <robyngraf@users.noreply.github.com>
@@ -559,7 +599,9 @@ Roni <sulpher@gmx.net>
 Ronny Brendel <ronnybrendel@gmail.com>
 Ronsor <ronsor@ronsor.pw>
 Rowan Hart <rowanbhart@gmail.com>
+Ruan <47767371+ruanych@users.noreply.github.com>
 Ruchira Hasaranga <ruchira66@gmail.com>
+Rudi Servo <rudiservo@gmail.com>
 Ruixin Huang <18860020911@163.com>
 Rune <43761327+Rune-AI@users.noreply.github.com>
 RunningLeon <maningsheng@sensetime.com>
@@ -623,12 +665,14 @@ Steven Roussey <sroussey@gmail.com>
 Steward Garcia <57494570+FSSRepo@users.noreply.github.com>
 StrangeBytesDev <141275258+StrangeBytesDev@users.noreply.github.com>
 Suaj Carrot <72162667+SuajCarrot@users.noreply.github.com>
+Sukriti Sharma <Ssukriti@users.noreply.github.com>
 SuperUserNameMan <yoann@terminajones.com>
 Sutou Kouhei <kou@cozmixng.org>
 Tai Duc Nguyen <taiducnguyen.drexel@gmail.com>
 Taikono-Himazin <kazu@po.harenet.ne.jp>
 Tameem <113388789+AhmadTameem@users.noreply.github.com>
 Tamotsu Takahashi <ttakah+github@gmail.com>
+Tei Home <taiteitonghome@proton.me>
 Thái Hoàng Tâm <75922889+RoyalHeart@users.noreply.github.com>
 Thatcher Chamberlin <j.thatcher.c@gmail.com>
 Theia Vogel <theia@vgel.me>
@@ -640,6 +684,7 @@ Tim Miller <drasticactions@users.noreply.github.com>
 Tim Wang <overocean@gmail.com>
 Timmy Knight <r2d2fish@gmail.com>
 Timothy Cronin <40186632+4imothy@users.noreply.github.com>
+Ting Lou <louting@189.cn>
 Ting Lou <ting.lou@gmail.com>
 Ting Sun <suntcrick@gmail.com>
 Tobias Lütke <tobi@shopify.com>
@@ -661,6 +706,7 @@ Uzo Nweke <uzoechi@gmail.com>
 Vaibhav Srivastav <vaibhavs10@gmail.com>
 Val Kharitonov <mail@kharvd.com>
 Valentin Konovalov <valle.ketsujin@gmail.com>
+Valentin Mamedov <45292985+Inf1delis@users.noreply.github.com>
 Valentyn Bezshapkin <61702053+valentynbez@users.noreply.github.com>
 Vali Malinoiu <0x4139@gmail.com>
 Victor Nogueira <felladrin@gmail.com>
@@ -673,13 +719,17 @@ Vladimir Malyutin <first-leon@yandex.ru>
 Vladimir Zorin <vladimir@deviant.guru>
 VoidIsVoid <343750470@qq.com>
 Volodymyr Vitvitskyi <72226+signalpillar@users.noreply.github.com>
+Wang Qin <37098874+wangqin0@users.noreply.github.com>
+Wang Ran (汪然) <wangr@smail.nju.edu.cn>
 WangHaoranRobin <56047610+WangHaoranRobin@users.noreply.github.com>
 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>
+William Tambellini <wtambellini@sdl.com>
 Willy Tarreau <w@1wt.eu>
+Woof Dog <197125663+woof-dog@users.noreply.github.com>
 Wouter <9594229+DifferentialityDevelopment@users.noreply.github.com>
 Wu Jian Ping <wujjpp@hotmail.com>
 Wu Jian Ping <wujp@greatld.com>
@@ -692,6 +742,7 @@ Xie Yanbo <xieyanbo@gmail.com>
 Xingchen Song(宋星辰) <xingchensong1996@163.com>
 Xinpeng Dou <81913537+Dou-Git@users.noreply.github.com>
 Xuan Son Nguyen <thichthat@gmail.com>
+Xuan-Son Nguyen <thichthat@gmail.com>
 Yaiko <elyaiko@hotmail.com>
 Yann Follet <131855179+YannFollet@users.noreply.github.com>
 Yaroslav <yaroslav.yashin@me.com>
@@ -702,7 +753,9 @@ Yoshi Suhara <y.suhara@gmail.com>
 Yoshi Suhara <ysuhara@nvidia.com>
 Younes Belkada <49240599+younesbelkada@users.noreply.github.com>
 Yueh-Po Peng <94939112+y10ab1@users.noreply.github.com>
+Yüg <eugeniosegalaweb@gmail.com>
 Yui <dev@sleepyyui.com>
+Yun Dou <dixyes@gmail.com>
 Yuri Khrustalev <ykhrustalev@users.noreply.github.com>
 Yusuf Kağan Hanoğlu <hanoglu@yahoo.com>
 Yuval Peled <31162840+Yuval-Peled@users.noreply.github.com>
@@ -714,18 +767,23 @@ Zhang Peiyuan <a1286225768@gmail.com>
 Zheng.Deng <32841220+dengzheng-cloud@users.noreply.github.com>
 Zhenwei Jin <109658203+kylo5aby@users.noreply.github.com>
 Zhiyuan Li <lizhiyuan@uniartisan.com>
+Zhiyuan Li <uniartisan2017@gmail.com>
 ZhouYuChen <zhouyuchen@naver.com>
 Ziad Ben Hadj-Alouane <zied.benhadjalouane@gmail.com>
 Ziang Wu <97337387+ZiangWu-77@users.noreply.github.com>
 Zsapi <martin1.zsapka@gmail.com>
 a-n-n-a-l-e-e <150648636+a-n-n-a-l-e-e@users.noreply.github.com>
+a3sh <38979186+A3shTnT@users.noreply.github.com>
 adel boussaken <netdur@gmail.com>
 afrideva <95653597+afrideva@users.noreply.github.com>
+ag2s20150909 <19373730+ag2s20150909@users.noreply.github.com>
 agray3 <agray3@users.noreply.github.com>
 akawrykow <142945436+akawrykow@users.noreply.github.com>
+alek3y <44779186+alek3y@users.noreply.github.com>
 alexpinel <93524949+alexpinel@users.noreply.github.com>
 alonfaraj <alonfaraj@gmail.com>
 alwqx <kenan3015@gmail.com>
+amd-dwang <dong.wang@amd.com>
 amd-lalithnc <lalithnc@amd.com>
 amritahs-ibm <amritahs@linux.vnet.ibm.com>
 andrijdavid <david@geek.mg>
@@ -737,6 +795,7 @@ arch-btw <57669023+arch-btw@users.noreply.github.com>
 arcrank <arcrank@gmail.com>
 ardfork <134447697+ardfork@users.noreply.github.com>
 arlo-phoenix <140345165+arlo-phoenix@users.noreply.github.com>
+aryantandon01 <80969509+aryantandon01@users.noreply.github.com>
 at8u <129688334+at8u@users.noreply.github.com>
 automaticcat <daogiatuank54@gmail.com>
 awatuna <23447591+awatuna@users.noreply.github.com>
@@ -751,12 +810,14 @@ bryanSwk <93190252+bryanSwk@users.noreply.github.com>
 bsilvereagle <bsilvereagle@users.noreply.github.com>
 bssrdf <merlintiger@hotmail.com>
 byte-6174 <88070277+byte-6174@users.noreply.github.com>
+cduk <19917266+cduk@users.noreply.github.com>
 cebtenzzre <cebtenzzre@gmail.com>
 chaihahaha <chai836275709@gmail.com>
 chiranko <96988916+chiranko@users.noreply.github.com>
 clibdev <52199778+clibdev@users.noreply.github.com>
 clyang <clyang@clyang.net>
 cocktailpeanut <121128867+cocktailpeanut@users.noreply.github.com>
+codezjx <code.zjx@gmail.com>
 coezbek <c.oezbek@gmail.com>
 comex <comexk@gmail.com>
 compilade <113953597+compilade@users.noreply.github.com>
@@ -780,14 +841,17 @@ drbh <david.richard.holtz@gmail.com>
 ds5t5 <145942675+ds5t5@users.noreply.github.com>
 dylan <canardleteer@users.noreply.github.com>
 eastriver <lee@eastriver.dev>
+ebraminio <ebrahim@gnu.org>
 ebraminio <ebraminio@gmail.com>
 eiery <19350831+eiery@users.noreply.github.com>
 eric8607242 <e0928021388@gmail.com>
 fairydreaming <166155368+fairydreaming@users.noreply.github.com>
 fengerhu1 <2748250768@qq.com>
+fj-y-saito <85871716+fj-y-saito@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>
+gn64 <yukikaze.jp@gmail.com>
 goerch <jhr.walter@t-online.de>
 grahameth <96447521+grahameth@users.noreply.github.com>
 gtygo <gtydoit@gmail.com>
@@ -812,10 +876,12 @@ icppWorld <124377669+icppWorld@users.noreply.github.com>
 igarnier <igarnier@protonmail.com>
 intelmatt <61025942+intelmatt@users.noreply.github.com>
 iohub <rickyang.pro@gmail.com>
+issixx <46835150+issixx@users.noreply.github.com>
 jacobi petrucciani <8117202+jpetrucciani@users.noreply.github.com>
 jaime-m-p <167997752+jaime-m-p@users.noreply.github.com>
 jameswu2014 <545426914@qq.com>
 jdomke <28772296+jdomke@users.noreply.github.com>
+jiahao su <damow890@gmail.com>
 jiez <373447296@qq.com>
 jneem <joeneeman@gmail.com>
 joecryptotoo <80373433+joecryptotoo@users.noreply.github.com>
@@ -828,6 +894,7 @@ 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>
+kallewoof <kalle.alm@gmail.com>
 kalomaze <66376113+kalomaze@users.noreply.github.com>
 kang <tpdns9032100@gmail.com>
 katsu560 <118887472+katsu560@users.noreply.github.com>
@@ -835,6 +902,7 @@ kchro3 <62481661+kchro3@users.noreply.github.com>
 khimaros <me@khimaros.com>
 kiltyj <kiltyj@gmail.com>
 klosax <131523366+klosax@users.noreply.github.com>
+krystiancha <krystian@krystianch.com>
 kunal-vaishnavi <115581922+kunal-vaishnavi@users.noreply.github.com>
 kunnis <kunnis@users.noreply.github.com>
 kuronekosaiko <EvanChanJ@163.com>
@@ -847,6 +915,8 @@ ldwang <ftgreat@163.com>
 le.chang <cljs118@126.com>
 leejet <leejet714@gmail.com>
 leo-pony <nengjunma@outlook.com>
+lexasub <lexakopp2212@gmail.com>
+lhez <quic_lih@quicinc.com>
 limitedAtonement <limitedAtonement@users.noreply.github.com>
 liuwei-git <14815172+liuwei-git@users.noreply.github.com>
 lon <114724657+longregen@users.noreply.github.com>
@@ -855,10 +925,13 @@ ltoniazzi <61414566+ltoniazzi@users.noreply.github.com>
 luoyu-intel <yu.luo@intel.com>
 m3ndax <adrian.goessl@outlook.com>
 maddes8cht <55592906+maddes8cht@users.noreply.github.com>
+mahorozte <41834471+mahorozte@users.noreply.github.com>
 makomk <makosoft@googlemail.com>
 manikbhandari <mbbhandarimanik2@gmail.com>
 maor-ps <154728172+maor-ps@users.noreply.github.com>
+mashdragon <122402293+mashdragon@users.noreply.github.com>
 matiaslin <45382001+matiaslin@users.noreply.github.com>
+matt23654 <matthew.webber@protonmail.com>
 matteo <matteogeniaccio@yahoo.it>
 mdrokz <mohammadmunshi@gmail.com>
 mgroeber9110 <45620825+mgroeber9110@users.noreply.github.com>
@@ -868,6 +941,7 @@ mmyjona <jonathan.gonse@gmail.com>
 momonga <115213907+mmnga@users.noreply.github.com>
 momonga <146910567+mmngays@users.noreply.github.com>
 moritzbrantner <31051084+moritzbrantner@users.noreply.github.com>
+musoles <135031143+musoles@users.noreply.github.com>
 mzcu <milos.cubrilo@gmail.com>
 nanahi <130121847+na-na-hi@users.noreply.github.com>
 ngc92 <7938269+ngc92@users.noreply.github.com>
@@ -885,6 +959,7 @@ oobabooga <112222186+oobabooga@users.noreply.github.com>
 opparco <parco.opaai@gmail.com>
 ostix360 <55257054+ostix360@users.noreply.github.com>
 pculliton <phillipculliton@gmail.com>
+peidaqi <peidaqi@gmail.com>
 pengxin99 <pengxin.yuan@intel.com>
 perserk <perserk@gmail.com>
 piDack <104877312+piDack@users.noreply.github.com>
@@ -892,10 +967,12 @@ pmysl <piotr.myslinski@outlook.com>
 postmasters <namnguyen@google.com>
 pudepiedj <pudepiedj@gmail.com>
 qingfengfenga <41416092+qingfengfenga@users.noreply.github.com>
+qingy1337 <qxli2@students.everettcc.edu>
 qouoq <qouoq@fastmail.com>
 qunash <anzoria@gmail.com>
 rabidcopy <rabidcopy@yahoo.com>
 rankaiyx <rankaiyx@rankaiyx.com>
+redbeard <bharrington@alticon.net>
 rhjdvsgsgks <26178113+rhjdvsgsgks@users.noreply.github.com>
 rhuddleston <ryan.huddleston@percona.com>
 rimoliga <53384203+rimoliga@users.noreply.github.com>
@@ -912,6 +989,7 @@ 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>
+someone13574 <81528246+someone13574@users.noreply.github.com>
 standby24x7 <standby24x7@gmail.com>
 staviq <staviq@gmail.com>
 stduhpf <stephduh@live.fr>
@@ -931,6 +1009,7 @@ uint256_t <konndennsa@gmail.com>
 uint256_t <maekawatoshiki1017@gmail.com>
 unbounded <haakon@likedan.net>
 uvos <devnull@uvos.xyz>
+uvos <philipp@uvos.xyz>
 valiray <133289098+valiray@users.noreply.github.com>
 vb <vaibhavs10@gmail.com>
 vik <vikhyatk@gmail.com>
@@ -951,6 +1030,7 @@ xaedes <xaedes@googlemail.com>
 xctan <axunlei@gmail.com>
 xloem <0xloem@gmail.com>
 yangli2 <yangli2@gmail.com>
+ymcki <84055651+ymcki@users.noreply.github.com>
 yuiseki <yuiseki@gmail.com>
 yuri@FreeBSD <yurivict@users.noreply.github.com>
 zakkor <edward.partenie@gmail.com>
@@ -963,4 +1043,5 @@ zrm <trustiosity.zrm@gmail.com>
 杨朱 · Kiki <baofa.fan@daocloud.io>
 源文雨 <41315874+fumiama@users.noreply.github.com>
 蕭澧邦 <45505768+shou692199@users.noreply.github.com>
+谢乃闻 <sienaiwun@users.noreply.github.com>
 Нияз Гарифзянов <112617865+garrnizon@users.noreply.github.com>

CMakeLists.txt

@@ -16,6 +16,7 @@ endif()
 list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/cmake/")
 
 set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/bin)
+set(CMAKE_LIBRARY_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/bin)
 
 if (CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR)
     set(LLAMA_STANDALONE ON)
@@ -49,6 +50,8 @@ endif()
 if (MSVC)
     add_compile_options("$<$<COMPILE_LANGUAGE:C>:/utf-8>")
     add_compile_options("$<$<COMPILE_LANGUAGE:CXX>:/utf-8>")
+    add_compile_options("$<$<COMPILE_LANGUAGE:C>:/bigobj>")
+    add_compile_options("$<$<COMPILE_LANGUAGE:CXX>:/bigobj>")
 endif()
 
 #
@@ -77,17 +80,15 @@ option(LLAMA_BUILD_SERVER   "llama: build server example" ${LLAMA_STANDALONE})
 
 # 3rd party libs
 option(LLAMA_CURL "llama: use libcurl to download model from an URL" OFF)
+option(LLAMA_LLGUIDANCE "llama-common: include LLGuidance library for structured output in common utils" OFF)
 
 # Required for relocatable CMake package
 include(${CMAKE_CURRENT_SOURCE_DIR}/cmake/build-info.cmake)
 include(${CMAKE_CURRENT_SOURCE_DIR}/cmake/common.cmake)
 
 # override ggml options
-set(GGML_SANITIZE_THREAD    ${LLAMA_SANITIZE_THREAD})
-set(GGML_SANITIZE_ADDRESS   ${LLAMA_SANITIZE_ADDRESS})
-set(GGML_SANITIZE_UNDEFINED ${LLAMA_SANITIZE_UNDEFINED})
-set(GGML_ALL_WARNINGS       ${LLAMA_ALL_WARNINGS})
-set(GGML_FATAL_WARNINGS     ${LLAMA_FATAL_WARNINGS})
+set(GGML_ALL_WARNINGS   ${LLAMA_ALL_WARNINGS})
+set(GGML_FATAL_WARNINGS ${LLAMA_FATAL_WARNINGS})
 
 # change the default for these ggml options
 if (NOT DEFINED GGML_LLAMAFILE)
@@ -117,16 +118,62 @@ llama_option_depr(WARNING     LLAMA_SYCL                GGML_SYCL)
 llama_option_depr(WARNING     LLAMA_SYCL_F16            GGML_SYCL_F16)
 llama_option_depr(WARNING     LLAMA_CANN                GGML_CANN)
 
+if (NOT MSVC)
+    if (LLAMA_SANITIZE_THREAD)
+        message(STATUS "Using -fsanitize=thread")
+
+        add_compile_options(-fsanitize=thread)
+        link_libraries     (-fsanitize=thread)
+    endif()
+
+    if (LLAMA_SANITIZE_ADDRESS)
+        message(STATUS "Using -fsanitize=address")
+
+        add_compile_options(-fsanitize=address -fno-omit-frame-pointer)
+        link_libraries     (-fsanitize=address)
+    endif()
+
+    if (LLAMA_SANITIZE_UNDEFINED)
+        message(STATUS "Using -fsanitize=undefined")
+
+        add_compile_options(-fsanitize=undefined)
+        link_libraries     (-fsanitize=undefined)
+    endif()
+endif()
+
 #
-# build the library
+# 3rd-party
 #
 
 if (NOT TARGET ggml)
     add_subdirectory(ggml)
     # ... otherwise assume ggml is added by a parent CMakeLists.txt
 endif()
+
+#
+# build the library
+#
+
 add_subdirectory(src)
 
+#
+# utils, programs, examples and tests
+#
+
+if (LLAMA_BUILD_COMMON)
+    add_subdirectory(common)
+endif()
+
+if (LLAMA_BUILD_COMMON AND LLAMA_BUILD_TESTS AND NOT CMAKE_JS_VERSION)
+    include(CTest)
+    add_subdirectory(tests)
+endif()
+
+if (LLAMA_BUILD_COMMON AND LLAMA_BUILD_EXAMPLES)
+    add_subdirectory(examples)
+    add_subdirectory(pocs)
+endif()
+
 #
 # install
 #
@@ -142,27 +189,14 @@ set(LLAMA_INCLUDE_INSTALL_DIR ${CMAKE_INSTALL_INCLUDEDIR} CACHE PATH "Location o
 set(LLAMA_LIB_INSTALL_DIR     ${CMAKE_INSTALL_LIBDIR}     CACHE PATH "Location of library files")
 set(LLAMA_BIN_INSTALL_DIR     ${CMAKE_INSTALL_BINDIR}     CACHE PATH "Location of binary  files")
 
-# At the moment some compile definitions are placed within the ggml/src
-# directory but not exported on the `ggml` target. This could be improved by
-# determining _precisely_ which defines are necessary for the llama-config
-# package.
-#
-set(GGML_TRANSIENT_DEFINES)
-get_target_property(GGML_DIRECTORY ggml SOURCE_DIR)
-get_directory_property(GGML_DIR_DEFINES DIRECTORY ${GGML_DIRECTORY} COMPILE_DEFINITIONS)
-if (GGML_DIR_DEFINES)
-    list(APPEND GGML_TRANSIENT_DEFINES ${GGML_DIR_DEFINES})
-endif()
-get_target_property(GGML_TARGET_DEFINES ggml COMPILE_DEFINITIONS)
-if (GGML_TARGET_DEFINES)
-    list(APPEND GGML_TRANSIENT_DEFINES ${GGML_TARGET_DEFINES})
-endif()
-get_target_property(GGML_LINK_LIBRARIES ggml LINK_LIBRARIES)
-# all public headers
 set(LLAMA_PUBLIC_HEADERS
     ${CMAKE_CURRENT_SOURCE_DIR}/include/llama.h
     ${CMAKE_CURRENT_SOURCE_DIR}/include/llama-cpp.h)
-set_target_properties(llama PROPERTIES PUBLIC_HEADER "${LLAMA_PUBLIC_HEADERS}")
+
+set_target_properties(llama
+    PROPERTIES
+        PUBLIC_HEADER "${LLAMA_PUBLIC_HEADERS}")
+
 install(TARGETS llama LIBRARY PUBLIC_HEADER)
 
 configure_package_config_file(
@@ -199,22 +233,4 @@ configure_file(cmake/llama.pc.in
         @ONLY)
 
 install(FILES "${CMAKE_CURRENT_BINARY_DIR}/llama.pc"
-        DESTINATION lib/pkgconfig)
-
-#
-# utils, programs, examples and tests
-#
-
-if (LLAMA_BUILD_COMMON)
-    add_subdirectory(common)
-endif()
-
-if (LLAMA_BUILD_COMMON AND LLAMA_BUILD_TESTS AND NOT CMAKE_JS_VERSION)
-    include(CTest)
-    add_subdirectory(tests)
-endif()
-
-if (LLAMA_BUILD_COMMON AND LLAMA_BUILD_EXAMPLES)
-    add_subdirectory(examples)
-    add_subdirectory(pocs)
-endif()
+        DESTINATION ${CMAKE_INSTALL_LIBDIR}/pkgconfig)

CONTRIBUTING.md

@@ -12,7 +12,7 @@
 
 - Squash-merge PRs
 - Use the following format for the squashed commit title: `<module> : <commit title> (#<issue_number>)`. For example: `utils : fix typo in utils.py (#1234)`
-- Optionally pick a `<module>` from here: https://github.com/ggerganov/llama.cpp/wiki/Modules
+- Optionally pick a `<module>` from here: https://github.com/ggml-org/llama.cpp/wiki/Modules
 - Consider adding yourself to [CODEOWNERS](CODEOWNERS)
 
 # Coding guidelines
@@ -40,14 +40,14 @@
 - Try to follow the existing patterns in the code (indentation, spaces, etc.). In case of doubt use `clang-format` to format the added code
 - For anything not covered in the current guidelines, refer to the [C++ Core Guidelines](https://isocpp.github.io/CppCoreGuidelines/CppCoreGuidelines)
 - Tensors store data in row-major order. We refer to dimension 0 as columns, 1 as rows, 2 as matrices
-- Matrix multiplication is unconventional: [`C = ggml_mul_mat(ctx, A, B)`](https://github.com/ggerganov/llama.cpp/blob/880e352277fc017df4d5794f0c21c44e1eae2b84/ggml.h#L1058-L1064) means $C^T = A B^T \Leftrightarrow C = B A^T.$
+- Matrix multiplication is unconventional: [`C = ggml_mul_mat(ctx, A, B)`](https://github.com/ggml-org/llama.cpp/blob/880e352277fc017df4d5794f0c21c44e1eae2b84/ggml.h#L1058-L1064) means $C^T = A B^T \Leftrightarrow C = B A^T.$
 
 ![matmul](media/matmul.png)
 
 # Naming guidelines
 
 - Use `snake_case` for function, variable and type names
-- Naming usually optimizes for longest common prefix (see https://github.com/ggerganov/ggml/pull/302#discussion_r1243240963)
+- Naming usually optimizes for longest common prefix (see https://github.com/ggml-org/ggml/pull/302#discussion_r1243240963)
 
     ```cpp
     // not OK
@@ -122,4 +122,4 @@
 
 The Github issues, PRs and discussions contain a lot of information that can be useful to get familiar with the codebase. For convenience, some of the more important information is referenced from Github projects:
 
-https://github.com/ggerganov/llama.cpp/projects
+https://github.com/ggml-org/llama.cpp/projects

Makefile

@@ -1,5 +1,5 @@
 ifndef LLAMA_MAKEFILE
-$(error The Makefile build is deprecated. Use the CMake build instead. For more details, see https://github.com/ggerganov/llama.cpp/blob/master/docs/build.md)
+$(error The Makefile build is deprecated. Use the CMake build instead. For more details, see https://github.com/ggml-org/llama.cpp/blob/master/docs/build.md)
 endif
 
 # Define the default target now so that it is always the first target
@@ -52,6 +52,7 @@ TEST_TARGETS = \
 	tests/test-arg-parser \
 	tests/test-autorelease \
 	tests/test-backend-ops \
+	tests/test-chat \
 	tests/test-chat-template \
 	tests/test-double-float \
 	tests/test-grammar-integration \
@@ -462,7 +463,7 @@ endif
 ifneq '' '$(findstring mingw,$(shell $(CC) -dumpmachine))'
 	# The stack is only 16-byte aligned on Windows, so don't let gcc emit aligned moves.
 	# https://gcc.gnu.org/bugzilla/show_bug.cgi?id=54412
-	# https://github.com/ggerganov/llama.cpp/issues/2922
+	# https://github.com/ggml-org/llama.cpp/issues/2922
 	MK_CFLAGS   += -Xassembler -muse-unaligned-vector-move
 	MK_CXXFLAGS += -Xassembler -muse-unaligned-vector-move
 
@@ -595,7 +596,7 @@ ifdef GGML_RPC
 	OBJ_GGML_EXT += ggml/src/ggml-rpc.o
 endif # GGML_RPC
 
-OBJ_CUDA_TMPL      = $(patsubst %.cu,%.o,$(wildcard ggml/src/ggml-cuda/template-instances/fattn-wmma*.cu))
+OBJ_CUDA_TMPL      = $(patsubst %.cu,%.o,$(wildcard ggml/src/ggml-cuda/template-instances/fattn-mma*.cu))
 OBJ_CUDA_TMPL     += $(patsubst %.cu,%.o,$(wildcard ggml/src/ggml-cuda/template-instances/mmq*.cu))
 
 ifdef GGML_CUDA_FA_ALL_QUANTS
@@ -983,6 +984,7 @@ OBJ_COMMON = \
 	$(DIR_COMMON)/ngram-cache.o \
 	$(DIR_COMMON)/sampling.o \
 	$(DIR_COMMON)/speculative.o \
+	$(DIR_COMMON)/chat.o \
 	$(DIR_COMMON)/build-info.o \
 	$(DIR_COMMON)/json-schema-to-grammar.o
 
@@ -1076,8 +1078,8 @@ endif
 ifdef REMOVE_WARNING
 $(info !!! REMOVAL WARNING !!!)
 $(info The following LLAMA_ options have been removed and are no longer supported)
-$(info   - LLAMA_DISABLE_LOGS   (https://github.com/ggerganov/llama.cpp/pull/9418))
-$(info   - LLAMA_SERVER_VERBOSE (https://github.com/ggerganov/llama.cpp/pull/9418))
+$(info   - LLAMA_DISABLE_LOGS   (https://github.com/ggml-org/llama.cpp/pull/9418))
+$(info   - LLAMA_SERVER_VERBOSE (https://github.com/ggml-org/llama.cpp/pull/9418))
 $(info )
 endif
 
@@ -1361,7 +1363,11 @@ llama-server: \
 	examples/server/httplib.h \
 	examples/server/index.html.hpp \
 	examples/server/loading.html.hpp \
+	common/chat.cpp \
+	common/chat.hpp \
+	common/chat-template.hpp \
 	common/json.hpp \
+	common/minja.hpp \
 	$(OBJ_ALL)
 	$(CXX) $(CXXFLAGS) -c $< -o $(call GET_OBJ_FILE, $<)
 	$(CXX) $(CXXFLAGS) $(filter-out %.h %.hpp $<,$^) -Iexamples/server $(call GET_OBJ_FILE, $<) -o $@ $(LDFLAGS) $(LWINSOCK2)
@@ -1469,6 +1475,11 @@ tests/test-json-schema-to-grammar: tests/test-json-schema-to-grammar.cpp \
 	$(CXX) $(CXXFLAGS) -Iexamples/server -c $< -o $(call GET_OBJ_FILE, $<)
 	$(CXX) $(CXXFLAGS) $(filter-out %.h $<,$^) $(call GET_OBJ_FILE, $<) -o $@ $(LDFLAGS)
 
+tests/test-chat: tests/test-chat.cpp \
+	$(OBJ_ALL)
+	$(CXX) $(CXXFLAGS) -Iexamples/server -c $< -o $(call GET_OBJ_FILE, $<)
+	$(CXX) $(CXXFLAGS) $(filter-out %.h $<,$^) $(call GET_OBJ_FILE, $<) -o $@ $(LDFLAGS)
+
 tests/test-opt: tests/test-opt.cpp \
 	$(OBJ_GGML)
 	$(CXX) $(CXXFLAGS) -c $< -o $(call GET_OBJ_FILE, $<)

README.md

@@ -3,22 +3,33 @@
 ![llama](https://user-images.githubusercontent.com/1991296/230134379-7181e485-c521-4d23-a0d6-f7b3b61ba524.png)
 
 [![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](https://opensource.org/licenses/MIT)
-[![Server](https://github.com/ggerganov/llama.cpp/actions/workflows/server.yml/badge.svg)](https://github.com/ggerganov/llama.cpp/actions/workflows/server.yml)
+[![Server](https://github.com/ggml-org/llama.cpp/actions/workflows/server.yml/badge.svg)](https://github.com/ggml-org/llama.cpp/actions/workflows/server.yml)
 
-[Roadmap](https://github.com/users/ggerganov/projects/7) / [Project status](https://github.com/ggerganov/llama.cpp/discussions/3471) / [Manifesto](https://github.com/ggerganov/llama.cpp/discussions/205) / [ggml](https://github.com/ggerganov/ggml)
+[Roadmap](https://github.com/users/ggml-org/projects/7) / [Project status](https://github.com/ggml-org/llama.cpp/discussions/3471) / [Manifesto](https://github.com/ggml-org/llama.cpp/discussions/205) / [ggml](https://github.com/ggml-org/ggml)
 
 Inference of Meta's [LLaMA](https://arxiv.org/abs/2302.13971) model (and others) in pure C/C++
 
+> [!IMPORTANT]
+> New `llama.cpp` package location: [ggml-org/llama.cpp](https://github.com/ggml-org/llama.cpp/pkgs/container/llama.cpp)
+>
+> Update your container URLs to: `ghcr.io/ggml-org/llama.cpp`
+>
+> More info: https://github.com/ggml-org/llama.cpp/discussions/11801
+
 ## Recent API changes
 
-- [Changelog for `libllama` API](https://github.com/ggerganov/llama.cpp/issues/9289)
-- [Changelog for `llama-server` REST API](https://github.com/ggerganov/llama.cpp/issues/9291)
+- [Changelog for `libllama` API](https://github.com/ggml-org/llama.cpp/issues/9289)
+- [Changelog for `llama-server` REST API](https://github.com/ggml-org/llama.cpp/issues/9291)
 
 ## Hot topics
 
-- **Introducing GGUF-my-LoRA** https://github.com/ggerganov/llama.cpp/discussions/10123
-- Hugging Face Inference Endpoints now support GGUF out of the box! https://github.com/ggerganov/llama.cpp/discussions/9669
-- Hugging Face GGUF editor: [discussion](https://github.com/ggerganov/llama.cpp/discussions/9268) | [tool](https://huggingface.co/spaces/CISCai/gguf-editor)
+- **How to use [MTLResidencySet](https://developer.apple.com/documentation/metal/mtlresidencyset?language=objc) to keep the GPU memory active?** https://github.com/ggml-org/llama.cpp/pull/11427
+- **VS Code extension for FIM completions:** https://github.com/ggml-org/llama.vscode
+- Universal tool call support in `llama-server`: https://github.com/ggml-org/llama.cpp/pull/9639
+- Vim/Neovim plugin for FIM completions: https://github.com/ggml-org/llama.vim
+- Introducing GGUF-my-LoRA https://github.com/ggml-org/llama.cpp/discussions/10123
+- Hugging Face Inference Endpoints now support GGUF out of the box! https://github.com/ggml-org/llama.cpp/discussions/9669
+- Hugging Face GGUF editor: [discussion](https://github.com/ggml-org/llama.cpp/discussions/9268) | [tool](https://huggingface.co/spaces/CISCai/gguf-editor)
 
 ----
 
@@ -35,7 +46,7 @@ range of hardware - locally and in the cloud.
 - Vulkan and SYCL backend support
 - CPU+GPU hybrid inference to partially accelerate models larger than the total VRAM capacity
 
-The `llama.cpp` project is the main playground for developing new features for the [ggml](https://github.com/ggerganov/ggml) library.
+The `llama.cpp` project is the main playground for developing new features for the [ggml](https://github.com/ggml-org/ggml) library.
 
 <details>
 <summary>Models</summary>
@@ -55,23 +66,23 @@ Instructions for adding support for new models: [HOWTO-add-model.md](docs/develo
 - [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] [BERT](https://github.com/ggml-org/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)
-- [X] [Starcoder models](https://github.com/ggerganov/llama.cpp/pull/3187)
+- [X] [Starcoder models](https://github.com/ggml-org/llama.cpp/pull/3187)
 - [X] [Refact](https://huggingface.co/smallcloudai/Refact-1_6B-fim)
-- [X] [MPT](https://github.com/ggerganov/llama.cpp/pull/3417)
-- [X] [Bloom](https://github.com/ggerganov/llama.cpp/pull/3553)
+- [X] [MPT](https://github.com/ggml-org/llama.cpp/pull/3417)
+- [X] [Bloom](https://github.com/ggml-org/llama.cpp/pull/3553)
 - [x] [Yi models](https://huggingface.co/models?search=01-ai/Yi)
 - [X] [StableLM models](https://huggingface.co/stabilityai)
 - [x] [Deepseek models](https://huggingface.co/models?search=deepseek-ai/deepseek)
 - [x] [Qwen models](https://huggingface.co/models?search=Qwen/Qwen)
-- [x] [PLaMo-13B](https://github.com/ggerganov/llama.cpp/pull/3557)
+- [x] [PLaMo-13B](https://github.com/ggml-org/llama.cpp/pull/3557)
 - [x] [Phi models](https://huggingface.co/models?search=microsoft/phi)
-- [x] [PhiMoE](https://github.com/ggerganov/llama.cpp/pull/11003)
+- [x] [PhiMoE](https://github.com/ggml-org/llama.cpp/pull/11003)
 - [x] [GPT-2](https://huggingface.co/gpt2)
-- [x] [Orion 14B](https://github.com/ggerganov/llama.cpp/pull/5118)
+- [x] [Orion 14B](https://github.com/ggml-org/llama.cpp/pull/5118)
 - [x] [InternLM2](https://huggingface.co/models?search=internlm2)
 - [x] [CodeShell](https://github.com/WisdomShell/codeshell)
 - [x] [Gemma](https://ai.google.dev/gemma)
@@ -92,7 +103,7 @@ Instructions for adding support for new models: [HOWTO-add-model.md](docs/develo
 - [x] [Bitnet b1.58 models](https://huggingface.co/1bitLLM)
 - [x] [Flan T5](https://huggingface.co/models?search=flan-t5)
 - [x] [Open Elm models](https://huggingface.co/collections/apple/openelm-instruct-models-6619ad295d7ae9f868b759ca)
-- [x] [ChatGLM3-6b](https://huggingface.co/THUDM/chatglm3-6b) + [ChatGLM4-9b](https://huggingface.co/THUDM/glm-4-9b)
+- [x] [ChatGLM3-6b](https://huggingface.co/THUDM/chatglm3-6b) + [ChatGLM4-9b](https://huggingface.co/THUDM/glm-4-9b) + [GLMEdge-1.5b](https://huggingface.co/THUDM/glm-edge-1.5b-chat) + [GLMEdge-4b](https://huggingface.co/THUDM/glm-edge-4b-chat)
 - [x] [SmolLM](https://huggingface.co/collections/HuggingFaceTB/smollm-6695016cad7167254ce15966)
 - [x] [EXAONE-3.0-7.8B-Instruct](https://huggingface.co/LGAI-EXAONE/EXAONE-3.0-7.8B-Instruct)
 - [x] [FalconMamba Models](https://huggingface.co/collections/tiiuae/falconmamba-7b-66b9a580324dd1598b0f6d4a)
@@ -113,6 +124,7 @@ Instructions for adding support for new models: [HOWTO-add-model.md](docs/develo
 - [x] [Mini CPM](https://huggingface.co/models?search=MiniCPM)
 - [x] [Moondream](https://huggingface.co/vikhyatk/moondream2)
 - [x] [Bunny](https://github.com/BAAI-DCAI/Bunny)
+- [x] [GLM-EDGE](https://huggingface.co/models?search=glm-edge)
 - [x] [Qwen2-VL](https://huggingface.co/collections/Qwen/qwen2-vl-66cee7455501d7126940800d)
 
 </details>
@@ -131,6 +143,7 @@ Instructions for adding support for new models: [HOWTO-add-model.md](docs/develo
 - Rust (more features): [edgenai/llama_cpp-rs](https://github.com/edgenai/llama_cpp-rs)
 - Rust (nicer API): [mdrokz/rust-llama.cpp](https://github.com/mdrokz/rust-llama.cpp)
 - Rust (more direct bindings): [utilityai/llama-cpp-rs](https://github.com/utilityai/llama-cpp-rs)
+- Rust (automated build from crates.io): [ShelbyJenkins/llm_client](https://github.com/ShelbyJenkins/llm_client)
 - C#/.NET: [SciSharp/LLamaSharp](https://github.com/SciSharp/LLamaSharp)
 - C#/VB.NET (more features - community license): [LM-Kit.NET](https://docs.lm-kit.com/lm-kit-net/index.html)
 - Scala 3: [donderom/llm4s](https://github.com/donderom/llm4s)
@@ -140,7 +153,7 @@ Instructions for adding support for new models: [HOWTO-add-model.md](docs/develo
 - Zig: [deins/llama.cpp.zig](https://github.com/Deins/llama.cpp.zig)
 - Flutter/Dart: [netdur/llama_cpp_dart](https://github.com/netdur/llama_cpp_dart)
 - Flutter: [xuegao-tzx/Fllama](https://github.com/xuegao-tzx/Fllama)
-- PHP (API bindings and features built on top of llama.cpp): [distantmagic/resonance](https://github.com/distantmagic/resonance) [(more info)](https://github.com/ggerganov/llama.cpp/pull/6326)
+- PHP (API bindings and features built on top of llama.cpp): [distantmagic/resonance](https://github.com/distantmagic/resonance) [(more info)](https://github.com/ggml-org/llama.cpp/pull/6326)
 - Guile Scheme: [guile_llama_cpp](https://savannah.nongnu.org/projects/guile-llama-cpp)
 - Swift [srgtuszy/llama-cpp-swift](https://github.com/srgtuszy/llama-cpp-swift)
 - Swift [ShenghaiWang/SwiftLlama](https://github.com/ShenghaiWang/SwiftLlama)
@@ -183,6 +196,7 @@ Instructions for adding support for new models: [HOWTO-add-model.md](docs/develo
 - [ramalama](https://github.com/containers/ramalama) (MIT)
 - [semperai/amica](https://github.com/semperai/amica) (MIT)
 - [withcatai/catai](https://github.com/withcatai/catai) (MIT)
+- [Autopen](https://github.com/blackhole89/autopen) (GPL)
 
 </details>
 
@@ -204,6 +218,7 @@ Instructions for adding support for new models: [HOWTO-add-model.md](docs/develo
 - [GPUStack](https://github.com/gpustack/gpustack) - Manage GPU clusters for running LLMs
 - [llama_cpp_canister](https://github.com/onicai/llama_cpp_canister) - llama.cpp as a smart contract on the Internet Computer, using WebAssembly
 - [llama-swap](https://github.com/mostlygeek/llama-swap) - transparent proxy that adds automatic model switching with llama-server
+- [Kalavai](https://github.com/kalavai-net/kalavai-client) - Crowdsource end to end LLM deployment at any scale
 
 </details>
 
@@ -227,6 +242,7 @@ Instructions for adding support for new models: [HOWTO-add-model.md](docs/develo
 | [HIP](docs/build.md#hip) | AMD GPU |
 | [Vulkan](docs/build.md#vulkan) | GPU |
 | [CANN](docs/build.md#cann) | Ascend NPU |
+| [OpenCL](docs/backend/OPENCL.md) | Adreno GPU |
 
 ## Building the project
 
@@ -236,7 +252,7 @@ The project also includes many example programs and tools using the `llama` libr
 - Clone this repository and build locally, see [how to build](docs/build.md)
 - On MacOS or Linux, install `llama.cpp` via [brew, flox or nix](docs/install.md)
 - Use a Docker image, see [documentation for Docker](docs/docker.md)
-- Download pre-built binaries from [releases](https://github.com/ggerganov/llama.cpp/releases)
+- Download pre-built binaries from [releases](https://github.com/ggml-org/llama.cpp/releases)
 
 ## Obtaining and quantizing models
 
@@ -249,14 +265,14 @@ You can either manually download the GGUF file or directly use any `llama.cpp`-c
 
 After downloading a model, use the CLI tools to run it locally - see below.
 
-`llama.cpp` requires the model to be stored in the [GGUF](https://github.com/ggerganov/ggml/blob/master/docs/gguf.md) file format. Models in other data formats can be converted to GGUF using the `convert_*.py` Python scripts in this repo.
+`llama.cpp` requires the model to be stored in the [GGUF](https://github.com/ggml-org/ggml/blob/master/docs/gguf.md) file format. Models in other data formats can be converted to GGUF using the `convert_*.py` Python scripts in this repo.
 
 The Hugging Face platform provides a variety of online tools for converting, quantizing and hosting models with `llama.cpp`:
 
 - Use the [GGUF-my-repo space](https://huggingface.co/spaces/ggml-org/gguf-my-repo) to convert to GGUF format and quantize model weights to smaller sizes
-- Use the [GGUF-my-LoRA space](https://huggingface.co/spaces/ggml-org/gguf-my-lora) to convert LoRA adapters to GGUF format (more info: https://github.com/ggerganov/llama.cpp/discussions/10123)
-- Use the [GGUF-editor space](https://huggingface.co/spaces/CISCai/gguf-editor) to edit GGUF meta data in the browser (more info: https://github.com/ggerganov/llama.cpp/discussions/9268)
-- Use the [Inference Endpoints](https://ui.endpoints.huggingface.co/) to directly host `llama.cpp` in the cloud (more info: https://github.com/ggerganov/llama.cpp/discussions/9669)
+- Use the [GGUF-my-LoRA space](https://huggingface.co/spaces/ggml-org/gguf-my-lora) to convert LoRA adapters to GGUF format (more info: https://github.com/ggml-org/llama.cpp/discussions/10123)
+- Use the [GGUF-editor space](https://huggingface.co/spaces/CISCai/gguf-editor) to edit GGUF meta data in the browser (more info: https://github.com/ggml-org/llama.cpp/discussions/9268)
+- Use the [Inference Endpoints](https://ui.endpoints.huggingface.co/) to directly host `llama.cpp` in the cloud (more info: https://github.com/ggml-org/llama.cpp/discussions/9669)
 
 To learn more about model quantization, [read this documentation](examples/quantize/README.md)
 
@@ -418,7 +434,7 @@ To learn more about model quantization, [read this documentation](examples/quant
 
     </details>
 
-[^1]: [examples/perplexity/README.md](examples/perplexity/README.md)
+[^1]: [examples/perplexity/README.md](./examples/perplexity/README.md)
 [^2]: [https://huggingface.co/docs/transformers/perplexity](https://huggingface.co/docs/transformers/perplexity)
 
 ## [`llama-bench`](examples/llama-bench)
@@ -479,9 +495,9 @@ To learn more about model quantization, [read this documentation](examples/quant
 - Collaborators can push to branches in the `llama.cpp` repo and merge PRs into the `master` branch
 - Collaborators will be invited based on contributions
 - Any help with managing issues, PRs and projects is very appreciated!
-- See [good first issues](https://github.com/ggerganov/llama.cpp/issues?q=is%3Aissue+is%3Aopen+label%3A%22good+first+issue%22) for tasks suitable for first contributions
+- See [good first issues](https://github.com/ggml-org/llama.cpp/issues?q=is%3Aissue+is%3Aopen+label%3A%22good+first+issue%22) for tasks suitable for first contributions
 - Read the [CONTRIBUTING.md](CONTRIBUTING.md) for more information
-- Make sure to read this: [Inference at the edge](https://github.com/ggerganov/llama.cpp/discussions/205)
+- Make sure to read this: [Inference at the edge](https://github.com/ggml-org/llama.cpp/discussions/205)
 - A bit of backstory for those who are interested: [Changelog podcast](https://changelog.com/podcast/532)
 
 ## Other documentation
@@ -496,7 +512,7 @@ To learn more about model quantization, [read this documentation](examples/quant
 - [Running on Docker](docs/docker.md)
 - [Build on Android](docs/android.md)
 - [Performance troubleshooting](docs/development/token_generation_performance_tips.md)
-- [GGML tips & tricks](https://github.com/ggerganov/llama.cpp/wiki/GGML-Tips-&-Tricks)
+- [GGML tips & tricks](https://github.com/ggml-org/llama.cpp/wiki/GGML-Tips-&-Tricks)
 
 #### Seminal papers and background on the models
 
@@ -510,5 +526,18 @@ If your issue is with model generation quality, then please at least scan the fo
     - [Aligning language models to follow instructions](https://openai.com/research/instruction-following)
     - [Training language models to follow instructions with human feedback](https://arxiv.org/abs/2203.02155)
 
-#### References
+## Completions
+Command-line completion is available for some environments.
 
+#### Bash Completion
+```bash
+$ build/bin/llama-cli --completion-bash > ~/.llama-completion.bash
+$ source ~/.llama-completion.bash
+```
+Optionally this can be added to your `.bashrc` or `.bash_profile` to load it
+automatically. For example:
+```console
+$ echo "source ~/.llama-completion.bash" >> ~/.bashrc
+```
+
+## References

SECURITY.md

@@ -62,6 +62,6 @@ Beware that none of the topics under [Using llama.cpp securely](#using-llamacpp-
 <!-- normal version -->
 However, If you have discovered a security vulnerability in this project, please report it privately. **Do not disclose it as a public issue.** This gives us time to work with you to fix the issue before public exposure, reducing the chance that the exploit will be used before a patch is released.
 
-Please disclose it as a private [security advisory](https://github.com/ggerganov/llama.cpp/security/advisories/new).
+Please disclose it as a private [security advisory](https://github.com/ggml-org/llama.cpp/security/advisories/new).
 
 A team of volunteers on a reasonable-effort basis maintains this project. As such, please give us at least 90 days to work on a fix before public exposure.

ci/README.md

@@ -1,11 +1,11 @@
 # CI
 
-In addition to [Github Actions](https://github.com/ggerganov/llama.cpp/actions) `llama.cpp` uses a custom CI framework:
+In addition to [Github Actions](https://github.com/ggml-org/llama.cpp/actions) `llama.cpp` uses a custom CI framework:
 
 https://github.com/ggml-org/ci
 
 It monitors the `master` branch for new commits and runs the
-[ci/run.sh](https://github.com/ggerganov/llama.cpp/blob/master/ci/run.sh) script on dedicated cloud instances. This allows us
+[ci/run.sh](https://github.com/ggml-org/llama.cpp/blob/master/ci/run.sh) script on dedicated cloud instances. This allows us
 to execute heavier workloads compared to just using Github Actions. Also with time, the cloud instances will be scaled
 to cover various hardware architectures, including GPU and Apple Silicon instances.
 

ci/run.sh

@@ -299,7 +299,7 @@ function gg_run_open_llama_7b_v2 {
     (time cmake -DCMAKE_BUILD_TYPE=Release ${CMAKE_EXTRA} .. ) 2>&1 | tee -a $OUT/${ci}-cmake.log
     (time make -j$(nproc)                                    ) 2>&1 | tee -a $OUT/${ci}-make.log
 
-    python ../examples/convert_legacy_llama.py ${path_models} --outfile ${path_models}/ggml-model-f16.gguf
+    python3 ../examples/convert_legacy_llama.py ${path_models} --outfile ${path_models}/ggml-model-f16.gguf
 
     model_f16="${path_models}/ggml-model-f16.gguf"
     model_q8_0="${path_models}/ggml-model-q8_0.gguf"
@@ -433,7 +433,7 @@ function gg_run_pythia_1_4b {
     (time cmake -DCMAKE_BUILD_TYPE=Release ${CMAKE_EXTRA} .. ) 2>&1 | tee -a $OUT/${ci}-cmake.log
     (time make -j$(nproc)                                    ) 2>&1 | tee -a $OUT/${ci}-make.log
 
-    python ../convert_hf_to_gguf.py ${path_models} --outfile ${path_models}/ggml-model-f16.gguf
+    python3 ../convert_hf_to_gguf.py ${path_models} --outfile ${path_models}/ggml-model-f16.gguf
 
     model_f16="${path_models}/ggml-model-f16.gguf"
     model_q8_0="${path_models}/ggml-model-q8_0.gguf"
@@ -564,7 +564,7 @@ function gg_run_pythia_2_8b {
     (time cmake -DCMAKE_BUILD_TYPE=Release ${CMAKE_EXTRA} .. ) 2>&1 | tee -a $OUT/${ci}-cmake.log
     (time make -j$(nproc)                                    ) 2>&1 | tee -a $OUT/${ci}-make.log
 
-    python ../convert_hf_to_gguf.py ${path_models} --outfile ${path_models}/ggml-model-f16.gguf
+    python3 ../convert_hf_to_gguf.py ${path_models} --outfile ${path_models}/ggml-model-f16.gguf
 
     model_f16="${path_models}/ggml-model-f16.gguf"
     model_q8_0="${path_models}/ggml-model-q8_0.gguf"
@@ -699,7 +699,7 @@ function gg_run_embd_bge_small {
     (time cmake -DCMAKE_BUILD_TYPE=Release ${CMAKE_EXTRA} .. ) 2>&1 | tee -a $OUT/${ci}-cmake.log
     (time make -j$(nproc)                                    ) 2>&1 | tee -a $OUT/${ci}-make.log
 
-    python ../convert_hf_to_gguf.py ${path_models} --outfile ${path_models}/ggml-model-f16.gguf
+    python3 ../convert_hf_to_gguf.py ${path_models} --outfile ${path_models}/ggml-model-f16.gguf
 
     model_f16="${path_models}/ggml-model-f16.gguf"
     model_q8_0="${path_models}/ggml-model-q8_0.gguf"
@@ -747,7 +747,7 @@ function gg_run_rerank_tiny {
     (time cmake -DCMAKE_BUILD_TYPE=Release ${CMAKE_EXTRA} .. ) 2>&1 | tee -a $OUT/${ci}-cmake.log
     (time make -j$(nproc)                                    ) 2>&1 | tee -a $OUT/${ci}-make.log
 
-    python ../convert_hf_to_gguf.py ${path_models} --outfile ${path_models}/ggml-model-f16.gguf
+    python3 ../convert_hf_to_gguf.py ${path_models} --outfile ${path_models}/ggml-model-f16.gguf
 
     model_f16="${path_models}/ggml-model-f16.gguf"
 
@@ -814,8 +814,8 @@ if [ -z ${GG_BUILD_LOW_PERF} ]; then
     mkdir -p ${mnt_models}
     ln -sfn ${mnt_models} ${SRC}/models-mnt
 
-    # Create a fresh python venv and enter it
-    if ! python -m venv "$MNT/venv"; then
+    # Create a fresh python3 venv and enter it
+    if ! python3 -m venv "$MNT/venv"; then
         echo "Error: Failed to create Python virtual environment at $MNT/venv."
         exit 1
     fi

cmake/build-info.cmake

@@ -44,7 +44,7 @@ if(MSVC)
     set(BUILD_TARGET ${CMAKE_VS_PLATFORM_NAME})
 else()
     execute_process(
-        COMMAND sh -c "$@ --version | head -1" _ ${CMAKE_C_COMPILER}
+        COMMAND sh -c "\"$@\" --version | head -1" _ ${CMAKE_C_COMPILER}
         OUTPUT_VARIABLE OUT
         OUTPUT_STRIP_TRAILING_WHITESPACE
     )

cmake/llama-config.cmake.in

@@ -3,159 +3,13 @@ set(LLAMA_BUILD_COMMIT @LLAMA_BUILD_COMMIT@)
 set(LLAMA_BUILD_NUMBER @LLAMA_BUILD_NUMBER@)
 set(LLAMA_SHARED_LIB   @BUILD_SHARED_LIBS@)
 
-set(GGML_STATIC @GGML_STATIC@)
-set(GGML_NATIVE @GGML_NATIVE@)
-set(GGML_LTO    @GGML_LTO@)
-set(GGML_CCACHE @GGML_CCACHE@)
-set(GGML_AVX    @GGML_AVX@)
-set(GGML_AVX2   @GGML_AVX2@)
-set(GGML_AVX512 @GGML_AVX512@)
-set(GGML_AVX512_VBMI @GGML_AVX512_VBMI@)
-set(GGML_AVX512_VNNI @GGML_AVX512_VNNI@)
-set(GGML_AVX512_BF16 @GGML_AVX512_BF16@)
-set(GGML_AMX_TILE @GGML_AMX_TILE@)
-set(GGML_AMX_INT8 @GGML_AMX_INT8@)
-set(GGML_AMX_BF16 @GGML_AMX_BF16@)
-set(GGML_FMA  @GGML_FMA@)
-set(GGML_LASX @GGML_LASX@)
-set(GGML_LSX  @GGML_LSX@)
-set(GGML_RVV  @GGML_RVV@)
-set(GGML_SVE  @GGML_SVE@)
-
-set(GGML_ACCELERATE @GGML_ACCELERATE@)
-set(GGML_OPENMP  @GGML_OPENMP@)
-set(GGML_CPU_HBM @GGML_CPU_HBM@)
-set(GGML_BLAS_VENDOR @GGML_BLAS_VENDOR@)
-
-set(GGML_CUDA_FORCE_MMQ    @GGML_CUDA_FORCE_MMQ@)
-set(GGML_CUDA_FORCE_CUBLAS @GGML_CUDA_FORCE_CUBLAS@)
-set(GGML_CUDA_F16          @GGML_CUDA_F16@)
-set(GGML_CUDA_PEER_MAX_BATCH_SIZE @GGML_CUDA_PEER_MAX_BATCH_SIZE@)
-set(GGML_CUDA_NO_PEER_COPY  @GGML_CUDA_NO_PEER_COPY@)
-set(GGML_CUDA_NO_VMM        @GGML_CUDA_NO_VMM@)
-set(GGML_CUDA_FA_ALL_QUANTS @GGML_CUDA_FA_ALL_QUANTS@)
-set(GGML_CUDA_GRAPHS        @GGML_CUDA_GRAPHS@)
-
-set(GGML_HIP_UMA @GGML_HIP_UMA@)
-
-set(GGML_VULKAN_CHECK_RESULTS @GGML_VULKAN_CHECK_RESULTS@)
-set(GGML_VULKAN_DEBUG         @GGML_VULKAN_DEBUG@)
-set(GGML_VULKAN_MEMORY_DEBUG  @GGML_VULKAN_MEMORY_DEBUG@)
-set(GGML_VULKAN_SHADER_DEBUG_INFO @GGML_VULKAN_SHADER_DEBUG_INFO@)
-set(GGML_VULKAN_PERF      @GGML_VULKAN_PERF@)
-set(GGML_VULKAN_VALIDATE  @GGML_VULKAN_VALIDATE@)
-set(GGML_VULKAN_RUN_TESTS @GGML_VULKAN_RUN_TESTS@)
-
-set(GGML_METAL_USE_BF16 @GGML_METAL_USE_BF16@)
-set(GGML_METAL_NDEBUG   @GGML_METAL_NDEBUG@)
-set(GGML_METAL_SHADER_DEBUG  @GGML_METAL_SHADER_DEBUG@)
-set(GGML_METAL_EMBED_LIBRARY @GGML_METAL_EMBED_LIBRARY@)
-set(GGML_METAL_MACOSX_VERSION_MIN @GGML_METAL_MACOSX_VERSION_MIN@)
-set(GGML_METAL_STD @GGML_METAL_STD@)
-
-set(GGML_SYCL_F16    @GGML_SYCL_F16@)
-set(GGML_SYCL_TARGET @GGML_SYCL_TARGET@)
-set(GGML_SYCL_DEVICE_ARCH @GGML_SYCL_DEVICE_ARCH@)
-
-
 @PACKAGE_INIT@
 
 set_and_check(LLAMA_INCLUDE_DIR "@PACKAGE_LLAMA_INCLUDE_INSTALL_DIR@")
 set_and_check(LLAMA_LIB_DIR     "@PACKAGE_LLAMA_LIB_INSTALL_DIR@")
 set_and_check(LLAMA_BIN_DIR     "@PACKAGE_LLAMA_BIN_INSTALL_DIR@")
 
-find_package(Threads REQUIRED)
-
-set(_llama_transient_defines "@GGML_TRANSIENT_DEFINES@")
-set(_llama_link_deps "")
-set(_llama_link_opts "")
-foreach(_ggml_lib ggml ggml-base)
-    string(REPLACE "-" "_" _ggml_lib_var "${_ggml_lib}_LIBRARY")
-    find_library(${_ggml_lib_var} ${_ggml_lib}
-        REQUIRED
-        HINTS ${LLAMA_LIB_DIR}
-        NO_CMAKE_FIND_ROOT_PATH
-    )
-    list(APPEND _llama_link_deps "${${_ggml_lib_var}}")
-    message(STATUS "Found ${${_ggml_lib_var}}")
-endforeach()
-
-foreach(backend amx blas cann cpu cuda hip kompute metal musa rpc sycl vulkan)
-    string(TOUPPER "GGML_${backend}" backend_id)
-    set(_ggml_lib "ggml-${backend}")
-    string(REPLACE "-" "_" _ggml_lib_var "${_ggml_lib}_LIBRARY")
-
-    find_library(${_ggml_lib_var} ${_ggml_lib}
-        HINTS ${LLAMA_LIB_DIR}
-        NO_CMAKE_FIND_ROOT_PATH
-    )
-    if(${_ggml_lib_var})
-        list(APPEND _llama_link_deps "${${_ggml_lib_var}}")
-        set(${backend_id} ON)
-        message(STATUS "Found backend ${${_ggml_lib_var}}")
-    else()
-        set(${backend_id} OFF)
-    endif()
-endforeach()
-
-if (NOT LLAMA_SHARED_LIB)
-    if (APPLE AND GGML_ACCELERATE)
-        find_library(ACCELERATE_FRAMEWORK Accelerate REQUIRED)
-        list(APPEND _llama_link_deps ${ACCELERATE_FRAMEWORK})
-    endif()
-
-    if (GGML_OPENMP)
-        find_package(OpenMP REQUIRED)
-        list(APPEND _llama_link_deps OpenMP::OpenMP_C OpenMP::OpenMP_CXX)
-    endif()
-
-    if (GGML_CPU_HBM)
-        find_library(memkind memkind REQUIRED)
-        list(APPEND _llama_link_deps memkind)
-    endif()
-
-    if (GGML_BLAS)
-        find_package(BLAS REQUIRED)
-        list(APPEND _llama_link_deps ${BLAS_LIBRARIES})
-        list(APPEND _llama_link_opts ${BLAS_LINKER_FLAGS})
-    endif()
-
-    if (GGML_CUDA)
-        find_package(CUDAToolkit REQUIRED)
-    endif()
-
-    if (GGML_METAL)
-        find_library(FOUNDATION_LIBRARY Foundation REQUIRED)
-        find_library(METAL_FRAMEWORK    Metal REQUIRED)
-        find_library(METALKIT_FRAMEWORK MetalKit REQUIRED)
-        list(APPEND _llama_link_deps ${FOUNDATION_LIBRARY}
-                                     ${METAL_FRAMEWORK} ${METALKIT_FRAMEWORK})
-    endif()
-
-    if (GGML_VULKAN)
-        find_package(Vulkan REQUIRED)
-        list(APPEND _llama_link_deps Vulkan::Vulkan)
-    endif()
-
-    if (GGML_HIP)
-        find_package(hip     REQUIRED)
-        find_package(hipblas REQUIRED)
-        find_package(rocblas REQUIRED)
-        list(APPEND _llama_link_deps hip::host roc::rocblas roc::hipblas)
-    endif()
-
-    if (GGML_SYCL)
-        find_package(DNNL)
-        if (${DNNL_FOUND} AND GGML_SYCL_TARGET STREQUAL "INTEL")
-            list(APPEND _llama_link_deps DNNL::dnnl)
-        endif()
-        if (WIN32)
-            find_package(IntelSYCL REQUIRED)
-            find_package(MKL       REQUIRED)
-            list(APPEND _llama_link_deps IntelSYCL::SYCL_CXX MKL::MKL MKL::MKL_SYCL)
-        endif()
-    endif()
-endif()
+find_package(ggml REQUIRED HINTS ${LLAMA_LIB_DIR}/cmake)
 
 find_library(llama_LIBRARY llama
     REQUIRED
@@ -167,12 +21,10 @@ add_library(llama UNKNOWN IMPORTED)
 set_target_properties(llama
     PROPERTIES
         INTERFACE_INCLUDE_DIRECTORIES "${LLAMA_INCLUDE_DIR}"
-        INTERFACE_LINK_LIBRARIES "${_llama_link_deps}"
-        INTERFACE_LINK_OPTIONS   "${_llama_link_opts}"
-        INTERFACE_COMPILE_DEFINITIONS "${_llama_transient_defines}"
+        INTERFACE_LINK_LIBRARIES "ggml::ggml;ggml::ggml-base;"
         IMPORTED_LINK_INTERFACE_LANGUAGES "CXX"
         IMPORTED_LOCATION "${llama_LIBRARY}"
-        INTERFACE_COMPILE_FEATURES cxx_std_11
-        POSITION_INDEPENDENT_CODE ON )
+        INTERFACE_COMPILE_FEATURES c_std_90
+        POSITION_INDEPENDENT_CODE ON)
 
 check_required_components(Llama)

cmake/llama.pc.in

@@ -1,10 +1,10 @@
 prefix=@CMAKE_INSTALL_PREFIX@
-exec_prefix=${prefix}
-libdir=${exec_prefix}/lib
-includedir=${prefix}/include
+exec_prefix=@CMAKE_INSTALL_PREFIX@
+libdir=@CMAKE_INSTALL_FULL_LIBDIR@
+includedir=@CMAKE_INSTALL_FULL_INCLUDEDIR@
 
 Name: llama
 Description: Port of Facebook's LLaMA model in C/C++
-Version: @PROJECT_VERSION@
-Libs: -L${libdir} -lggml  -lggml-base -lllama
+Version: @LLAMA_INSTALL_VERSION@
+Libs: -L${libdir} -lggml -lggml-base -lllama
 Cflags: -I${includedir}

common/CMakeLists.txt

@@ -56,14 +56,19 @@ add_library(${TARGET} STATIC
     arg.cpp
     arg.h
     base64.hpp
+    chat.cpp
+    chat.hpp
+    chat-template.hpp
     common.cpp
     common.h
     console.cpp
     console.h
     json-schema-to-grammar.cpp
     json.hpp
+    llguidance.cpp
     log.cpp
     log.h
+    minja.hpp
     ngram-cache.cpp
     ngram-cache.h
     sampling.cpp
@@ -87,6 +92,50 @@ if (LLAMA_CURL)
     set(LLAMA_COMMON_EXTRA_LIBS ${LLAMA_COMMON_EXTRA_LIBS} ${CURL_LIBRARY})
 endif ()
 
+if (LLAMA_LLGUIDANCE)
+    include(ExternalProject)
+    set(LLGUIDANCE_SRC ${CMAKE_BINARY_DIR}/llguidance/source)
+    set(LLGUIDANCE_PATH ${LLGUIDANCE_SRC}/target/release)
+
+    # Set the correct library file extension based on platform
+    if (WIN32)
+        set(LLGUIDANCE_LIB_NAME "llguidance.lib")
+        # Add Windows-specific libraries
+        set(LLGUIDANCE_PLATFORM_LIBS
+            ws2_32    # Windows Sockets API
+            userenv   # For GetUserProfileDirectoryW
+            ntdll     # For NT functions
+            bcrypt    # For BCryptGenRandom
+        )
+    else()
+        set(LLGUIDANCE_LIB_NAME "libllguidance.a")
+        set(LLGUIDANCE_PLATFORM_LIBS "")
+    endif()
+
+    ExternalProject_Add(llguidance_ext
+        GIT_REPOSITORY https://github.com/guidance-ai/llguidance
+        # v0.6.12:
+        GIT_TAG ced1c9023d47ec194fa977932d35ce65c2ebfc09
+        PREFIX ${CMAKE_BINARY_DIR}/llguidance
+        SOURCE_DIR ${LLGUIDANCE_SRC}
+        BUILD_IN_SOURCE TRUE
+        CONFIGURE_COMMAND ""
+        BUILD_COMMAND cargo build --release
+        INSTALL_COMMAND ""
+        BUILD_BYPRODUCTS ${LLGUIDANCE_PATH}/${LLGUIDANCE_LIB_NAME} ${LLGUIDANCE_PATH}/llguidance.h
+        UPDATE_COMMAND ""
+    )
+    target_compile_definitions(${TARGET} PUBLIC LLAMA_USE_LLGUIDANCE)
+
+    add_library(llguidance STATIC IMPORTED)
+    set_target_properties(llguidance PROPERTIES IMPORTED_LOCATION ${LLGUIDANCE_PATH}/${LLGUIDANCE_LIB_NAME})
+    add_dependencies(llguidance llguidance_ext)
+
+    target_include_directories(${TARGET} PRIVATE ${LLGUIDANCE_PATH})
+    # Add platform libraries to the main target
+    set(LLAMA_COMMON_EXTRA_LIBS ${LLAMA_COMMON_EXTRA_LIBS} llguidance ${LLGUIDANCE_PLATFORM_LIBS})
+endif ()
+
 target_include_directories(${TARGET} PUBLIC .)
 target_compile_features   (${TARGET} PUBLIC cxx_std_17)
 target_link_libraries     (${TARGET} PRIVATE ${LLAMA_COMMON_EXTRA_LIBS} PUBLIC llama Threads::Threads)

common/arg.cpp

@@ -133,7 +133,8 @@ static void common_params_handle_model_default(
         const std::string & model_url,
         std::string & hf_repo,
         std::string & hf_file,
-        const std::string & hf_token) {
+        const std::string & hf_token,
+        const std::string & model_default) {
     if (!hf_repo.empty()) {
         // short-hand to avoid specifying --hf-file -> default it to --model
         if (hf_file.empty()) {
@@ -163,7 +164,7 @@ static void common_params_handle_model_default(
             model = fs_get_cache_file(string_split<std::string>(f, '/').back());
         }
     } else if (model.empty()) {
-        model = DEFAULT_MODEL_PATH;
+        model = model_default;
     }
 }
 
@@ -299,8 +300,9 @@ static bool common_params_parse_ex(int argc, char ** argv, common_params_context
     }
 
     // TODO: refactor model params in a common struct
-    common_params_handle_model_default(params.model,         params.model_url,         params.hf_repo,         params.hf_file,         params.hf_token);
-    common_params_handle_model_default(params.vocoder.model, params.vocoder.model_url, params.vocoder.hf_repo, params.vocoder.hf_file, params.hf_token);
+    common_params_handle_model_default(params.model,             params.model_url,             params.hf_repo,             params.hf_file,             params.hf_token, DEFAULT_MODEL_PATH);
+    common_params_handle_model_default(params.speculative.model, params.speculative.model_url, params.speculative.hf_repo, params.speculative.hf_file, params.hf_token, "");
+    common_params_handle_model_default(params.vocoder.model,     params.vocoder.model_url,     params.vocoder.hf_repo,     params.vocoder.hf_file,     params.hf_token, "");
 
     if (params.escape) {
         string_process_escapes(params.prompt);
@@ -323,6 +325,14 @@ static bool common_params_parse_ex(int argc, char ** argv, common_params_context
         throw std::invalid_argument("error: either --embedding or --reranking can be specified, but not both");
     }
 
+    if (!params.chat_template.empty() && !common_chat_verify_template(params.chat_template, params.use_jinja)) {
+        throw std::runtime_error(string_format(
+            "error: the supplied chat template is not supported: %s%s\n",
+            params.chat_template.c_str(),
+            params.use_jinja ? "" : "\nnote: llama.cpp was started without --jinja, we only support commonly used templates"
+        ));
+    }
+
     return true;
 }
 
@@ -355,6 +365,112 @@ static void common_params_print_usage(common_params_context & ctx_arg) {
     print_options(specific_options);
 }
 
+static void common_params_print_completion(common_params_context & ctx_arg) {
+    std::vector<common_arg *> common_options;
+    std::vector<common_arg *> sparam_options;
+    std::vector<common_arg *> specific_options;
+
+    for (auto & opt : ctx_arg.options) {
+        if (opt.is_sparam) {
+            sparam_options.push_back(&opt);
+        } else if (opt.in_example(ctx_arg.ex)) {
+            specific_options.push_back(&opt);
+        } else {
+            common_options.push_back(&opt);
+        }
+    }
+
+    printf("_llama_completions() {\n");
+    printf("    local cur prev opts\n");
+    printf("    COMPREPLY=()\n");
+    printf("    cur=\"${COMP_WORDS[COMP_CWORD]}\"\n");
+    printf("    prev=\"${COMP_WORDS[COMP_CWORD-1]}\"\n\n");
+
+    printf("    opts=\"");
+    auto print_options = [](const std::vector<common_arg *> & options) {
+        for (const common_arg * opt : options) {
+            for (const char * arg : opt->args) {
+                printf("%s ", arg);
+            }
+        }
+    };
+
+    print_options(common_options);
+    print_options(sparam_options);
+    print_options(specific_options);
+    printf("\"\n\n");
+
+    printf("    case \"$prev\" in\n");
+    printf("        --model)\n");
+    printf("            COMPREPLY=( $(compgen -f -X '!*.gguf' -- \"$cur\") $(compgen -d -- \"$cur\") )\n");
+    printf("            return 0\n");
+    printf("            ;;\n");
+    printf("        --grammar-file)\n");
+    printf("            COMPREPLY=( $(compgen -f -X '!*.gbnf' -- \"$cur\") $(compgen -d -- \"$cur\") )\n");
+    printf("            return 0\n");
+    printf("            ;;\n");
+    printf("        --chat-template-file)\n");
+    printf("            COMPREPLY=( $(compgen -f -X '!*.jinja' -- \"$cur\") $(compgen -d -- \"$cur\") )\n");
+    printf("            return 0\n");
+    printf("            ;;\n");
+    printf("        *)\n");
+    printf("            COMPREPLY=( $(compgen -W \"${opts}\" -- \"$cur\") )\n");
+    printf("            return 0\n");
+    printf("            ;;\n");
+    printf("    esac\n");
+    printf("}\n\n");
+
+    std::set<std::string> executables = {
+        "llama-batched",
+        "llama-batched-bench",
+        "llama-bench",
+        "llama-cli",
+        "llama-convert-llama2c-to-ggml",
+        "llama-cvector-generator",
+        "llama-embedding",
+        "llama-eval-callback",
+        "llama-export-lora",
+        "llama-gbnf-validator",
+        "llama-gen-docs",
+        "llama-gguf",
+        "llama-gguf-hash",
+        "llama-gguf-split",
+        "llama-gritlm",
+        "llama-imatrix",
+        "llama-infill",
+        "llama-llava-cli",
+        "llama-llava-clip-quantize-cli",
+        "llama-lookahead",
+        "llama-lookup",
+        "llama-lookup-create",
+        "llama-lookup-merge",
+        "llama-lookup-stats",
+        "llama-minicpmv-cli",
+        "llama-parallel",
+        "llama-passkey",
+        "llama-perplexity",
+        "llama-q8dot",
+        "llama-quantize",
+        "llama-quantize-stats",
+        "llama-qwen2vl-cli",
+        "llama-retrieval",
+        "llama-run",
+        "llama-save-load-state",
+        "llama-server",
+        "llama-simple",
+        "llama-simple-chat",
+        "llama-speculative",
+        "llama-speculative-simple",
+        "llama-tokenize",
+        "llama-tts",
+        "llama-vdot"
+    };
+
+    for (const auto& exe : executables) {
+        printf("complete -F _llama_completions %s\n", exe.c_str());
+    }
+}
+
 static std::vector<ggml_backend_dev_t> parse_device_list(const std::string & value) {
     std::vector<ggml_backend_dev_t> devices;
     auto dev_names = string_split<std::string>(value, ',');
@@ -376,6 +492,30 @@ static std::vector<ggml_backend_dev_t> parse_device_list(const std::string & val
     return devices;
 }
 
+static void add_rpc_devices(std::string servers) {
+    auto rpc_servers = string_split<std::string>(servers, ',');
+    if (rpc_servers.empty()) {
+        throw std::invalid_argument("no RPC servers specified");
+    }
+    ggml_backend_reg_t rpc_reg = ggml_backend_reg_by_name("RPC");
+    if (!rpc_reg) {
+        throw std::invalid_argument("failed to find RPC backend");
+    }
+    typedef ggml_backend_dev_t (*ggml_backend_rpc_add_device_t)(const char * endpoint);
+    ggml_backend_rpc_add_device_t ggml_backend_rpc_add_device_fn = (ggml_backend_rpc_add_device_t) ggml_backend_reg_get_proc_address(rpc_reg, "ggml_backend_rpc_add_device");
+    if (!ggml_backend_rpc_add_device_fn) {
+        throw std::invalid_argument("failed to find RPC device add function");
+    }
+    for (const auto & server : rpc_servers) {
+        ggml_backend_dev_t dev = ggml_backend_rpc_add_device_fn(server.c_str());
+        if (dev) {
+            ggml_backend_device_register(dev);
+        } else {
+            throw std::invalid_argument("failed to register RPC device");
+        }
+    }
+}
+
 bool common_params_parse(int argc, char ** argv, common_params & params, llama_example ex, void(*print_usage)(int, char **)) {
     auto ctx_arg = common_params_parser_init(params, ex, print_usage);
     const common_params params_org = ctx_arg.params; // the example can modify the default params
@@ -392,6 +532,10 @@ bool common_params_parse(int argc, char ** argv, common_params & params, llama_e
             }
             exit(0);
         }
+        if (ctx_arg.params.completion) {
+            common_params_print_completion(ctx_arg);
+            exit(0);
+        }
     } catch (const std::invalid_argument & ex) {
         fprintf(stderr, "%s\n", ex.what());
         ctx_arg.params = params_org;
@@ -460,6 +604,13 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
             exit(0);
         }
     ));
+    add_opt(common_arg(
+        {"--completion-bash"},
+        "print source-able bash completion script for llama.cpp",
+        [](common_params & params) {
+            params.completion = true;
+        }
+    ));
     add_opt(common_arg(
         {"--verbose-prompt"},
         string_format("print a verbose prompt before generation (default: %s)", params.verbose_prompt ? "true" : "false"),
@@ -640,7 +791,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
     ));
     add_opt(common_arg(
         {"--no-context-shift"},
-        string_format("disables context shift on inifinite text generation (default: %s)", params.ctx_shift ? "disabled" : "enabled"),
+        string_format("disables context shift on infinite text generation (default: %s)", params.ctx_shift ? "disabled" : "enabled"),
         [](common_params & params) {
             params.ctx_shift = false;
         }
@@ -843,7 +994,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
         [](common_params & params) {
             params.warmup = false;
         }
-    ).set_examples({LLAMA_EXAMPLE_MAIN, LLAMA_EXAMPLE_SERVER}));
+    ).set_examples({LLAMA_EXAMPLE_MAIN, LLAMA_EXAMPLE_SERVER, LLAMA_EXAMPLE_EMBEDDING}));
     add_opt(common_arg(
         {"--spm-infill"},
         string_format(
@@ -912,6 +1063,13 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
             params.sampling.min_p = std::stof(value);
         }
     ).set_sparam());
+    add_opt(common_arg(
+        {"--top-nsigma"}, "N",
+        string_format("top-n-sigma sampling (default: %.1f, -1.0 = disabled)", params.sampling.top_n_sigma),
+        [](common_params & params, const std::string & value) {
+            params.sampling.top_n_sigma = std::stof(value);
+        }
+    ).set_examples({LLAMA_EXAMPLE_MAIN}).set_sparam());
     add_opt(common_arg(
         {"--xtc-probability"}, "N",
         string_format("xtc probability (default: %.1f, 0.0 = disabled)", (double)params.sampling.xtc_probability),
@@ -1385,7 +1543,8 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
             {"--rpc"}, "SERVERS",
             "comma separated list of RPC servers",
             [](common_params & params, const std::string & value) {
-                params.rpc_servers = value;
+                add_rpc_devices(value);
+                GGML_UNUSED(params);
             }
         ).set_env("LLAMA_ARG_RPC"));
     }
@@ -1410,7 +1569,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
         "- isolate: only spawn threads on CPUs on the node that execution started on\n"
         "- numactl: use the CPU map provided by numactl\n"
         "if run without this previously, it is recommended to drop the system page cache before using this\n"
-        "see https://github.com/ggerganov/llama.cpp/issues/1437",
+        "see https://github.com/ggml-org/llama.cpp/issues/1437",
         [](common_params & params, const std::string & value) {
             /**/ if (value == "distribute" || value == "") { params.numa = GGML_NUMA_STRATEGY_DISTRIBUTE; }
             else if (value == "isolate") { params.numa = GGML_NUMA_STRATEGY_ISOLATE; }
@@ -1430,15 +1589,28 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
         {"--list-devices"},
         "print list of available devices and exit",
         [](common_params &) {
-            printf("Available devices:\n");
+            std::vector<ggml_backend_dev_t> rpc_devices;
+            std::vector<ggml_backend_dev_t> all_devices;
             for (size_t i = 0; i < ggml_backend_dev_count(); ++i) {
                 auto * dev = ggml_backend_dev_get(i);
                 if (ggml_backend_dev_type(dev) == GGML_BACKEND_DEVICE_TYPE_GPU) {
-                    size_t free, total;
-                    ggml_backend_dev_memory(dev, &free, &total);
-                    printf("  %s: %s (%zu MiB, %zu MiB free)\n", ggml_backend_dev_name(dev), ggml_backend_dev_description(dev), total / 1024 / 1024, free / 1024 / 1024);
+                    ggml_backend_reg_t reg = ggml_backend_dev_backend_reg(dev);
+                    if (ggml_backend_reg_name(reg) == std::string("RPC")) {
+                        rpc_devices.push_back(dev);
+                    } else {
+                        all_devices.push_back(dev);
+                    }
                 }
             }
+            // insert RPC devices in front
+            all_devices.insert(all_devices.begin(), rpc_devices.begin(), rpc_devices.end());
+            printf("Available devices:\n");
+            for (size_t i = 0; i < all_devices.size(); ++i) {
+                auto * dev = all_devices[i];
+                size_t free, total;
+                ggml_backend_dev_memory(dev, &free, &total);
+                printf("  %s: %s (%zu MiB, %zu MiB free)\n", ggml_backend_dev_name(dev), ggml_backend_dev_description(dev), total / 1024 / 1024, free / 1024 / 1024);
+            }
             exit(0);
         }
     ));
@@ -1604,6 +1776,13 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
             params.hf_repo = value;
         }
     ).set_env("LLAMA_ARG_HF_REPO"));
+    add_opt(common_arg(
+        {"-hfd", "-hfrd", "--hf-repo-draft"}, "<user>/<model>[:quant]",
+        "Same as --hf-repo, but for the draft model (default: unused)",
+        [](common_params & params, const std::string & value) {
+            params.speculative.hf_repo = value;
+        }
+    ).set_env("LLAMA_ARG_HFD_REPO"));
     add_opt(common_arg(
         {"-hff", "--hf-file"}, "FILE",
         "Hugging Face model file. If specified, it will override the quant in --hf-repo (default: unused)",
@@ -1913,24 +2092,55 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
             }
         }
     ).set_examples({LLAMA_EXAMPLE_SERVER}));
+    add_opt(common_arg(
+        {"--jinja"},
+        "use jinja template for chat (default: disabled)",
+        [](common_params & params) {
+            params.use_jinja = true;
+        }
+    ).set_examples({LLAMA_EXAMPLE_SERVER, LLAMA_EXAMPLE_MAIN}).set_env("LLAMA_ARG_JINJA"));
+    add_opt(common_arg(
+        {"--reasoning-format"}, "FORMAT",
+        "reasoning format (default: deepseek; allowed values: deepseek, none)\n"
+        "controls whether thought tags are extracted from the response, and in which format they're returned. 'none' leaves thoughts unparsed in `message.content`, 'deepseek' puts them in `message.reasoning_content` (for DeepSeek R1 & Command R7B only).\n"
+        "only supported for non-streamed responses",
+        [](common_params & params, const std::string & value) {
+            /**/ if (value == "deepseek") { params.reasoning_format = COMMON_REASONING_FORMAT_DEEPSEEK; }
+            else if (value == "none") {     params.reasoning_format = COMMON_REASONING_FORMAT_NONE; }
+            else { std::invalid_argument("invalid value"); }
+        }
+    ).set_examples({LLAMA_EXAMPLE_SERVER, LLAMA_EXAMPLE_MAIN}).set_env("LLAMA_ARG_THINK"));
     add_opt(common_arg(
         {"--chat-template"}, "JINJA_TEMPLATE",
         string_format(
             "set custom jinja chat template (default: template taken from model's metadata)\n"
             "if suffix/prefix are specified, template will be disabled\n"
+            "only commonly used templates are accepted (unless --jinja is set before this flag):\n"
             "list of built-in templates:\n%s", list_builtin_chat_templates().c_str()
         ),
         [](common_params & params, const std::string & value) {
-            if (!common_chat_verify_template(value)) {
-                throw std::runtime_error(string_format(
-                    "error: the supplied chat template is not supported: %s\n"
-                    "note: llama.cpp does not use jinja parser, we only support commonly used templates\n",
-                    value.c_str()
-                ));
-            }
             params.chat_template = value;
         }
     ).set_examples({LLAMA_EXAMPLE_MAIN, LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_CHAT_TEMPLATE"));
+    add_opt(common_arg(
+        {"--chat-template-file"}, "JINJA_TEMPLATE_FILE",
+        string_format(
+            "set custom jinja chat template file (default: template taken from model's metadata)\n"
+            "if suffix/prefix are specified, template will be disabled\n"
+            "only commonly used templates are accepted (unless --jinja is set before this flag):\n"
+            "list of built-in templates:\n%s", list_builtin_chat_templates().c_str()
+        ),
+        [](common_params & params, const std::string & value) {
+            std::ifstream file(value);
+            if (!file) {
+                throw std::runtime_error(string_format("error: failed to open file '%s'\n", value.c_str()));
+            }
+            std::copy(
+                std::istreambuf_iterator<char>(file),
+                std::istreambuf_iterator<char>(),
+                std::back_inserter(params.chat_template));
+        }
+    ).set_examples({LLAMA_EXAMPLE_MAIN, LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_CHAT_TEMPLATE_FILE"));
     add_opt(common_arg(
         {"-sps", "--slot-prompt-similarity"}, "SIMILARITY",
         string_format("how much the prompt of a request must match the prompt of a slot in order to use that slot (default: %.2f, 0.0 = disabled)\n", params.slot_prompt_similarity),
@@ -2037,7 +2247,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
     ).set_env("LLAMA_LOG_VERBOSITY"));
     add_opt(common_arg(
         {"--log-prefix"},
-        "Enable prefx in log messages",
+        "Enable prefix in log messages",
         [](common_params &) {
             common_log_set_prefix(common_log_main(), true);
         }
@@ -2229,6 +2439,13 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
             params.vocoder.model = value;
         }
     ).set_examples({LLAMA_EXAMPLE_TTS, LLAMA_EXAMPLE_SERVER}));
+     add_opt(common_arg(
+        {"--tts-use-guide-tokens"},
+        "Use guide tokens to improve TTS word recall",
+        [](common_params & params) {
+            params.vocoder.use_guide_tokens = true;
+        }
+    ).set_examples({LLAMA_EXAMPLE_TTS, LLAMA_EXAMPLE_SERVER}));
 
     // model-specific
     add_opt(common_arg(
@@ -2242,5 +2459,47 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
         }
     ).set_examples({LLAMA_EXAMPLE_TTS}));
 
+    add_opt(common_arg(
+        {"--embd-bge-small-en-default"},
+        string_format("use default bge-small-en-v1.5 model (note: can download weights from the internet)"),
+        [](common_params & params) {
+            params.hf_repo = "ggml-org/bge-small-en-v1.5-Q8_0-GGUF";
+            params.hf_file = "bge-small-en-v1.5-q8_0.gguf";
+            params.pooling_type = LLAMA_POOLING_TYPE_NONE;
+            params.embd_normalize = 2;
+            params.n_ctx = 512;
+            params.verbose_prompt = true;
+            params.embedding = true;
+        }
+    ).set_examples({LLAMA_EXAMPLE_EMBEDDING, LLAMA_EXAMPLE_SERVER}));
+
+    add_opt(common_arg(
+        {"--embd-e5-small-en-default"},
+        string_format("use default e5-small-v2 model (note: can download weights from the internet)"),
+        [](common_params & params) {
+            params.hf_repo = "ggml-org/e5-small-v2-Q8_0-GGUF";
+            params.hf_file = "e5-small-v2-q8_0.gguf";
+            params.pooling_type = LLAMA_POOLING_TYPE_NONE;
+            params.embd_normalize = 2;
+            params.n_ctx = 512;
+            params.verbose_prompt = true;
+            params.embedding = true;
+        }
+    ).set_examples({LLAMA_EXAMPLE_EMBEDDING, LLAMA_EXAMPLE_SERVER}));
+
+    add_opt(common_arg(
+        {"--embd-gte-small-default"},
+        string_format("use default gte-small model (note: can download weights from the internet)"),
+        [](common_params & params) {
+            params.hf_repo = "ggml-org/gte-small-Q8_0-GGUF";
+            params.hf_file = "gte-small-q8_0.gguf";
+            params.pooling_type = LLAMA_POOLING_TYPE_NONE;
+            params.embd_normalize = 2;
+            params.n_ctx = 512;
+            params.verbose_prompt = true;
+            params.embedding = true;
+        }
+    ).set_examples({LLAMA_EXAMPLE_EMBEDDING, LLAMA_EXAMPLE_SERVER}));
+
     return ctx_arg;
 }

common/chat-template.hpp

@@ -0,0 +1,529 @@
+/*
+    Copyright 2024 Google LLC
+
+    Use of this source code is governed by an MIT-style
+    license that can be found in the LICENSE file or at
+    https://opensource.org/licenses/MIT.
+*/
+// SPDX-License-Identifier: MIT
+#pragma once
+
+#include "minja.hpp"
+#include <json.hpp>
+#include <string>
+#include <vector>
+
+using json = nlohmann::ordered_json;
+
+namespace minja {
+
+struct chat_template_caps {
+    bool supports_tools = false;
+    bool supports_tool_calls = false;
+    bool supports_tool_responses = false;
+    bool supports_system_role = false;
+    bool supports_parallel_tool_calls = false;
+    bool supports_tool_call_id = false;
+    // meta-llama/Llama-3.1-8B-Instruct expects arguments to be an object.
+    // Most other templates (and OpenAI's API) expect the arguments object to be stringified.
+    bool requires_object_arguments = false;
+    // CohereForAI/c4ai-command-r-plus simple variant
+    bool requires_non_null_content = false;
+    // MiniMaxAI/MiniMax-Text-01 special
+    bool requires_typed_content = false;
+};
+
+struct chat_template_inputs {
+    nlohmann::ordered_json messages;
+    nlohmann::ordered_json tools;
+    bool add_generation_prompt = true;
+    nlohmann::ordered_json extra_context;
+    std::chrono::system_clock::time_point now = std::chrono::system_clock::now();
+};
+
+struct chat_template_options {
+    bool apply_polyfills = true;
+    bool use_bos_token = true;
+    bool use_eos_token = true;
+    bool define_strftime_now = true;
+
+    bool polyfill_tools = true;
+    bool polyfill_tool_call_examples = true;
+    bool polyfill_tool_calls = true;
+    bool polyfill_tool_responses = true;
+    bool polyfill_system_role = true;
+    bool polyfill_object_arguments = true;
+    bool polyfill_typed_content = true;
+};
+
+class chat_template {
+
+  private:
+    chat_template_caps caps_;
+    std::string source_;
+    std::string bos_token_;
+    std::string eos_token_;
+    std::shared_ptr<minja::TemplateNode> template_root_;
+    std::string tool_call_example_;
+
+    std::string try_raw_render(
+        const nlohmann::ordered_json & messages,
+        const nlohmann::ordered_json & tools,
+        bool add_generation_prompt,
+        const nlohmann::ordered_json & extra_context = nlohmann::ordered_json()) const
+    {
+        try {
+            chat_template_inputs inputs;
+            inputs.messages = messages;
+            inputs.tools = tools;
+            inputs.add_generation_prompt = add_generation_prompt;
+            inputs.extra_context = extra_context;
+            // Use fixed date for tests
+            inputs.now = std::chrono::system_clock::from_time_t(0);
+
+            chat_template_options opts;
+            opts.apply_polyfills = false;
+
+            auto prompt = apply(inputs, opts);
+            // fprintf(stderr, "try_raw_render: %s\n", prompt.c_str());
+            return prompt;
+        } catch (const std::exception & e) {
+            // fprintf(stderr, "try_raw_render error: %s\n", e.what());
+            return "";
+        }
+    }
+
+  public:
+
+    chat_template(const std::string & source, const std::string & bos_token, const std::string & eos_token)
+        : source_(source), bos_token_(bos_token), eos_token_(eos_token)
+    {
+        template_root_ = minja::Parser::parse(source_, {
+            /* .trim_blocks = */ true,
+            /* .lstrip_blocks = */ true,
+            /* .keep_trailing_newline = */ false,
+        });
+
+        auto contains = [](const std::string & haystack, const std::string & needle) {
+            return haystack.find(needle) != std::string::npos;
+        };
+
+        const std::string user_needle = "<User Needle>";
+        const std::string sys_needle = "<System Needle>";
+        const json dummy_str_user_msg = {{"role", "user"}, {"content", user_needle}};
+        const json dummy_typed_user_msg = {{"role", "user"}, {"content", json::array({{{"type", "text"}, {"text", user_needle}}})}};
+
+        caps_.requires_typed_content =
+            !contains(try_raw_render(json::array({dummy_str_user_msg}), {}, false), user_needle)
+            && contains(try_raw_render(json::array({dummy_typed_user_msg}), {}, false), user_needle);
+
+        const auto dummy_user_msg = caps_.requires_typed_content
+            ? dummy_typed_user_msg
+            : dummy_str_user_msg;
+        const json needle_system_msg = {
+            {"role", "system"},
+            {"content", caps_.requires_typed_content ? json::array({{{"type", "text"}, {"text", sys_needle}}}) : json(sys_needle)},
+        };
+
+        caps_.supports_system_role = contains(try_raw_render({needle_system_msg, dummy_user_msg,}, {}, false), sys_needle);
+
+        auto out = try_raw_render(json::array({
+            dummy_user_msg
+        }), json::array({
+            {
+                {"name", "some_tool"},
+                {"type", "function"},
+                {"function", {
+                    {"name", "some_tool"},
+                    {"description", "Some tool."},
+                    {"parameters", {
+                        {"type", "object"},
+                        {"properties", {
+                            {"arg", {
+                                {"type", "string"},
+                                {"description", "Some argument."},
+                            }},
+                        }},
+                        {"required", json::array({ "arg" })},
+                    }},
+                }},
+            },
+        }), false);
+        caps_.supports_tools = contains(out, "some_tool");
+
+        auto make_tool_calls_msg = [&](const json & tool_calls) {
+            return json {
+                {"role", "assistant"},
+                {"content", nullptr},
+                {"tool_calls", tool_calls},
+            };
+        };
+        auto make_tool_call = [](const std::string & tool_name, const json & arguments) {
+            return json {
+                {"id", "call_1___"},
+                {"type", "function"},
+                {"function", {
+                    {"arguments", arguments},
+                    {"name", tool_name},
+                }},
+            };
+        };
+        const json dummy_args_obj {{"argument_needle", "print('Hello, World!')"}};
+
+        // Note: the arguments are rendered in both cases, but may be double-escaped, which we don't want.
+        out = try_raw_render(json::array({
+            dummy_user_msg,
+            make_tool_calls_msg(json::array({make_tool_call("ipython", dummy_args_obj.dump())})),
+        }), {}, false);
+        auto tool_call_renders_str_arguments = contains(out, "\"argument_needle\":") || contains(out, "'argument_needle':");
+        out = try_raw_render(json::array({
+            dummy_user_msg,
+            make_tool_calls_msg(json::array({make_tool_call("ipython", dummy_args_obj)})),
+        }), {}, false);
+        auto tool_call_renders_obj_arguments = contains(out, "\"argument_needle\":") || contains(out, "'argument_needle':");
+
+        caps_.supports_tool_calls = tool_call_renders_str_arguments || tool_call_renders_obj_arguments;
+        caps_.requires_object_arguments = !tool_call_renders_str_arguments && tool_call_renders_obj_arguments;
+        auto out_empty = try_raw_render(json::array({dummy_user_msg, {{"role", "assistant"}, {"content", ""}}}), {}, false);
+        auto out_null = try_raw_render(json::array({dummy_user_msg, {{"role", "assistant"}, {"content", nullptr}}}), {}, false);
+        caps_.requires_non_null_content = contains(out_empty, user_needle) && !contains(out_null, user_needle);
+
+        if (caps_.supports_tool_calls) {
+            auto dummy_args = caps_.requires_object_arguments ? dummy_args_obj : json(dummy_args_obj.dump());
+            auto tc1 = make_tool_call("test_tool1", dummy_args);
+            auto tc2 = make_tool_call("test_tool2", dummy_args);
+            auto out = try_raw_render(json::array({
+                dummy_user_msg,
+                make_tool_calls_msg(json::array({tc1, tc2})),
+            }), {}, false);
+            caps_.supports_parallel_tool_calls = contains(out, "test_tool1") && contains(out, "test_tool2");
+
+            out = try_raw_render(json::array({
+                dummy_user_msg,
+                make_tool_calls_msg(json::array({tc1})),
+                {
+                    {"role", "tool"},
+                    {"name", "test_tool1"},
+                    {"content", "Some response!"},
+                    {"tool_call_id", "call_911_"},
+                }
+            }), {}, false);
+            caps_.supports_tool_responses = contains(out, "Some response!");
+            caps_.supports_tool_call_id = contains(out, "call_911_");
+        }
+
+        try {
+            if (!caps_.supports_tools) {
+                const json user_msg {
+                    {"role", "user"},
+                    {"content", "Hey"},
+                };
+                const json args {
+                    {"arg1", "some_value"},
+                };
+                const json tool_call_msg {
+                    {"role", "assistant"},
+                    {"content", nullptr},
+                    {"tool_calls", json::array({
+                        {
+                            // TODO: detect if requires numerical id or fixed length == 6 like Nemo
+                            {"id", "call_1___"},
+                            {"type", "function"},
+                            {"function", {
+                                {"name", "tool_name"},
+                                {"arguments", (caps_.requires_object_arguments ? args : json(minja::Value(args).dump(-1, /* to_json= */ true)))},
+                            }},
+                        },
+                    })},
+                };
+                std::string prefix, full;
+                {
+                    chat_template_inputs inputs;
+                    inputs.messages = json::array({user_msg});
+                    inputs.add_generation_prompt = true;
+                    prefix = apply(inputs);
+                }
+                {
+                    chat_template_inputs inputs;
+                    inputs.messages = json::array({user_msg, tool_call_msg});
+                    inputs.add_generation_prompt = false;
+                    full = apply(inputs);
+                }
+                auto eos_pos_last = full.rfind(eos_token_);
+                if (eos_pos_last == prefix.size() - eos_token_.size() ||
+                      (full[full.size() - 1] == '\n' && (eos_pos_last == full.size() - eos_token_.size() - 1))) {
+                    full = full.substr(0, eos_pos_last);
+                }
+                size_t common_prefix_length = 0;
+                for (size_t i = 0; i < prefix.size() && i < full.size(); ++i) {
+                    if (prefix[i] != full[i]) {
+                        break;
+                    }
+                    if (prefix[i] == '<') {
+                        // DeepSeek R1's template (as of 20250209) adds a trailing <think> if add_generation_prompt,
+                        // but it removes thinking tags for past messages.
+                        // The prefix and full strings diverge at <think> vs. <｜tool▁calls▁begin｜>, we avoid consuming the leading <.
+                        continue;
+                    }
+                    common_prefix_length = i + 1;
+                }
+                auto example = full.substr(common_prefix_length);
+                if (example.find("tool_name") == std::string::npos && example.find("some_value") == std::string::npos) {
+                    fprintf(stderr, "Failed to infer a tool call example (possible template bug)\n");
+                } else {
+                    tool_call_example_ = example;
+                }
+            }
+        } catch (const std::exception & e) {
+            fprintf(stderr, "Failed to generate tool call example: %s\n", e.what());
+        }
+    }
+
+    const std::string & source() const { return source_; }
+    const std::string & bos_token() const { return bos_token_; }
+    const std::string & eos_token() const { return eos_token_; }
+    const chat_template_caps & original_caps() const { return caps_; }
+
+    // Deprecated, please use the form with chat_template_inputs and chat_template_options
+    std::string apply(
+        const nlohmann::ordered_json & messages,
+        const nlohmann::ordered_json & tools,
+        bool add_generation_prompt,
+        const nlohmann::ordered_json & extra_context = nlohmann::ordered_json(),
+        bool apply_polyfills = true)
+    {
+        fprintf(stderr, "[%s] Deprecated!\n", __func__);
+        chat_template_inputs inputs;
+        inputs.messages = messages;
+        inputs.tools = tools;
+        inputs.add_generation_prompt = add_generation_prompt;
+        inputs.extra_context = extra_context;
+        inputs.now = std::chrono::system_clock::now();
+
+        chat_template_options opts;
+        opts.apply_polyfills = apply_polyfills;
+
+        return apply(inputs, opts);
+    }
+
+    std::string apply(
+        const chat_template_inputs & inputs,
+        const chat_template_options & opts = chat_template_options()) const
+    {
+        json actual_messages;
+
+        auto has_tools = inputs.tools.is_array() && !inputs.tools.empty();
+        auto has_tool_calls = false;
+        auto has_tool_responses = false;
+        auto has_string_content = false;
+        for (const auto & message : inputs.messages) {
+            if (message.contains("tool_calls") && !message["tool_calls"].is_null()) {
+                has_tool_calls = true;
+            }
+            if (message.contains("role") && message["role"] == "tool") {
+                has_tool_responses = true;
+            }
+            if (message.contains("content") && message["content"].is_string()) {
+                has_string_content = true;
+            }
+        }
+
+        auto polyfill_system_role = opts.polyfill_system_role && !caps_.supports_system_role;
+        auto polyfill_tools = opts.polyfill_tools && has_tools && !caps_.supports_tools;
+        auto polyfill_tool_call_example = polyfill_tools && opts.polyfill_tool_call_examples;
+        auto polyfill_tool_calls = opts.polyfill_tool_calls && has_tool_calls && !caps_.supports_tool_calls;
+        auto polyfill_tool_responses = opts.polyfill_tool_responses && has_tool_responses && !caps_.supports_tool_responses;
+        auto polyfill_object_arguments = opts.polyfill_object_arguments && has_tool_calls && caps_.requires_object_arguments;
+        auto polyfill_typed_content = opts.polyfill_typed_content && has_string_content && caps_.requires_typed_content;
+
+        auto needs_polyfills = opts.apply_polyfills && (false
+            || polyfill_system_role
+            || polyfill_tools
+            || polyfill_tool_calls
+            || polyfill_tool_responses
+            || polyfill_object_arguments
+            || polyfill_typed_content
+        );
+
+        if (needs_polyfills) {
+            actual_messages = json::array();
+
+            auto add_message = [&](const json & msg) {
+                if (polyfill_typed_content && msg.contains("content") && !msg.at("content").is_null() && msg.at("content").is_string()) {
+                    actual_messages.push_back({
+                        {"role", msg.at("role")},
+                        {"content", {{
+                            {"type", "text"},
+                            {"text", msg.at("content")},
+                        }}},
+                    });
+                } else {
+                    actual_messages.push_back(msg);
+                }
+            };
+
+            std::string pending_system;
+            auto flush_sys = [&]() {
+                if (!pending_system.empty()) {
+                    add_message({
+                        {"role", "user"},
+                        {"content", pending_system},
+                    });
+                    pending_system.clear();
+                }
+            };
+
+            json adjusted_messages;
+            if (polyfill_tools) {
+                adjusted_messages = add_system(inputs.messages,
+                    "You can call any of the following tools to satisfy the user's requests: " + minja::Value(inputs.tools).dump(2, /* to_json= */ true) +
+                    (!polyfill_tool_call_example || tool_call_example_.empty() ? "" : "\n\nExample tool call syntax:\n\n" + tool_call_example_ + "\n\n"));
+            } else {
+                adjusted_messages = inputs.messages;
+            }
+
+            for (const auto & message_ : adjusted_messages) {
+                auto message = message_;
+                if (!message.contains("role") || !message.contains("content")) {
+                    throw std::runtime_error("message must have 'role' and 'content' fields: " + message.dump());
+                }
+                std::string role = message.at("role");
+
+                if (message.contains("tool_calls")) {
+                    if (polyfill_object_arguments || polyfill_tool_calls) {
+                        for (auto & tool_call : message.at("tool_calls")) {
+                            if (tool_call["type"] == "function") {
+                                auto & function = tool_call.at("function");
+                                auto & arguments = function.at("arguments");
+                                if (arguments.is_string()) {
+                                    try {
+                                        arguments = json::parse(arguments.get<std::string>());
+                                    } catch (const std::exception & ecvt) {
+                                        fprintf(stderr, "Failed to parse arguments: %s\n", ecvt.what());
+                                    }
+                                }
+                            }
+                        }
+                    }
+                    if (polyfill_tool_calls) {
+                        auto content = message.at("content");
+                        auto tool_calls = json::array();
+                        for (const auto & tool_call : message.at("tool_calls")) {
+                            if (tool_call.at("type") != "function") {
+                                continue;
+                            }
+                            const auto & function = tool_call.at("function");
+                            auto tc = json {
+                                {"name", function.at("name")},
+                                {"arguments", function.at("arguments")},
+                            };
+                            if (tool_call.contains("id")) {
+                                tc["id"] = tool_call["id"];
+                            }
+                            tool_calls.push_back(tc);
+                        }
+                        auto obj = json {
+                            {"tool_calls", tool_calls},
+                        };
+                        if (!content.is_null() && content != "") {
+                            obj["content"] = content;
+                        }
+                        message["content"] = obj.dump(2);
+                        message.erase("tool_calls");
+                    }
+                }
+                if (polyfill_tool_responses && role == "tool") {
+                    message["role"] = "user";
+                    auto obj = json {
+                        {"tool_response", {
+                            {"content", message.at("content")},
+                        }},
+                    };
+                    if (message.contains("name")) {
+                        obj["tool_response"]["name"] = message.at("name");
+                    }
+                    if (message.contains("tool_call_id")) {
+                        obj["tool_response"]["tool_call_id"] = message.at("tool_call_id");
+                    }
+                    message["content"] = obj.dump(2);
+                    message.erase("name");
+                }
+
+                if (!message["content"].is_null() && polyfill_system_role) {
+                    std::string content = message.at("content");
+                    if (role == "system") {
+                        if (!pending_system.empty()) pending_system += "\n";
+                        pending_system += content;
+                        continue;
+                    } else {
+                        if (role == "user") {
+                            if (!pending_system.empty()) {
+                                message["content"] = pending_system + (content.empty() ? "" : "\n" + content);
+                                pending_system.clear();
+                            }
+                        } else {
+                            flush_sys();
+                        }
+                    }
+                }
+                add_message(message);
+            }
+            flush_sys();
+        } else {
+            actual_messages = inputs.messages;
+        }
+
+        auto context = minja::Context::make(json({
+            {"messages", actual_messages},
+            {"add_generation_prompt", inputs.add_generation_prompt},
+        }));
+        context->set("bos_token", opts.use_bos_token ? bos_token_ : "");
+        context->set("eos_token", opts.use_eos_token ? eos_token_ : "");
+        if (opts.define_strftime_now) {
+            auto now = inputs.now;
+            context->set("strftime_now", Value::callable([now](const std::shared_ptr<minja::Context> &, minja::ArgumentsValue & args) {
+                args.expectArgs("strftime_now", {1, 1}, {0, 0});
+                auto format = args.args[0].get<std::string>();
+
+                auto time = std::chrono::system_clock::to_time_t(now);
+                auto local_time = *std::localtime(&time);
+                std::ostringstream ss;
+                ss << std::put_time(&local_time, format.c_str());
+                return ss.str();
+            }));
+        }
+        if (!inputs.tools.is_null()) {
+            context->set("tools", minja::Value(inputs.tools));
+        }
+        if (!inputs.extra_context.is_null()) {
+            for (auto & kv : inputs.extra_context.items()) {
+                context->set(kv.key(), minja::Value(kv.value()));
+            }
+        }
+
+        auto ret = template_root_->render(context);
+        // fprintf(stderr, "actual_messages: %s\n", actual_messages.dump(2).c_str());
+        // fprintf(stderr, "apply: %s\n\n", ret.c_str());
+        return ret;
+    }
+
+    static nlohmann::ordered_json add_system(const nlohmann::ordered_json & messages, const std::string & system_prompt) {
+        json messages_with_system = messages;
+
+        if (messages_with_system.size() > 0 && messages_with_system[0].at("role") == "system") {
+            std::string existing_system = messages_with_system.at(0).at("content");
+            messages_with_system[0] = json {
+                {"role", "system"},
+                {"content", existing_system + "\n\n" + system_prompt},
+            };
+        } else {
+            messages_with_system.insert(messages_with_system.begin(), json {
+                {"role", "system"},
+                {"content", system_prompt},
+            });
+        }
+        return messages_with_system;
+    }
+};
+
+}  // namespace minja

common/chat.hpp

@@ -0,0 +1,55 @@
+// Chat support (incl. tool call grammar constraining & output parsing) w/ generic & custom template handlers.
+
+#pragma once
+
+#include "common.h"
+#include <json.hpp>
+#include <optional>
+#include <string>
+#include <vector>
+
+using json = nlohmann::ordered_json;
+
+struct common_chat_inputs {
+    json messages;
+    json tools;
+    json tool_choice;
+    json json_schema;
+    bool parallel_tool_calls;
+    bool stream;
+    std::string grammar;
+    bool add_generation_prompt = true;
+    bool extract_reasoning     = true;
+};
+
+enum common_chat_format {
+    COMMON_CHAT_FORMAT_CONTENT_ONLY,
+    COMMON_CHAT_FORMAT_GENERIC,
+    COMMON_CHAT_FORMAT_MISTRAL_NEMO,
+    COMMON_CHAT_FORMAT_LLAMA_3_X,
+    COMMON_CHAT_FORMAT_LLAMA_3_X_WITH_BUILTIN_TOOLS,
+    COMMON_CHAT_FORMAT_DEEPSEEK_R1,
+    COMMON_CHAT_FORMAT_DEEPSEEK_R1_EXTRACT_REASONING,
+    COMMON_CHAT_FORMAT_FIREFUNCTION_V2,
+    COMMON_CHAT_FORMAT_FUNCTIONARY_V3_2,
+    COMMON_CHAT_FORMAT_FUNCTIONARY_V3_1_LLAMA_3_1,
+    COMMON_CHAT_FORMAT_HERMES_2_PRO,
+    COMMON_CHAT_FORMAT_COMMAND_R7B,
+    COMMON_CHAT_FORMAT_COMMAND_R7B_EXTRACT_REASONING,
+
+    COMMON_CHAT_FORMAT_COUNT, // Not a format, just the # formats
+};
+
+struct common_chat_params {
+    common_chat_format                  format = COMMON_CHAT_FORMAT_CONTENT_ONLY;
+    json                                prompt;
+    std::string                         grammar;
+    bool                                grammar_lazy = false;
+    std::vector<common_grammar_trigger> grammar_triggers;
+    std::vector<std::string>            preserved_tokens;
+    std::vector<std::string>            additional_stops;
+};
+
+struct common_chat_params common_chat_params_init(const common_chat_template & tmpl, const struct common_chat_inputs & params);
+std::string               common_chat_format_name(common_chat_format format);
+common_chat_msg           common_chat_parse(      const std::string & input, common_chat_format format);

common/common.cpp

@@ -12,6 +12,8 @@
 #include "json.hpp"
 #include "json-schema-to-grammar.h"
 #include "llama.h"
+#include "chat.hpp"
+#include "chat-template.hpp"
 
 #include <algorithm>
 #include <cinttypes>
@@ -483,6 +485,48 @@ void string_replace_all(std::string & s, const std::string & search, const std::
     s = std::move(builder);
 }
 
+std::string string_join(const std::vector<std::string> & values, const std::string & separator) {
+    std::ostringstream result;
+    for (size_t i = 0; i < values.size(); ++i) {
+        if (i > 0) {
+            result << separator;
+        }
+        result << values[i];
+    }
+    return result.str();
+}
+
+std::vector<std::string> string_split(const std::string & str, const std::string & delimiter) {
+    std::vector<std::string> parts;
+    size_t start = 0;
+    size_t end = str.find(delimiter);
+
+    while (end != std::string::npos) {
+        parts.push_back(str.substr(start, end - start));
+        start = end + delimiter.length();
+        end = str.find(delimiter, start);
+    }
+
+    parts.push_back(str.substr(start));
+
+    return parts;
+}
+
+std::string string_repeat(const std::string & str, size_t n) {
+    if (n == 0) {
+        return "";
+    }
+
+    std::string result;
+    result.reserve(str.length() * n);
+
+    for (size_t i = 0; i < n; ++i) {
+        result += str;
+    }
+
+    return result;
+}
+
 std::string string_from(bool value) {
     return value ? "true" : "false";
 }
@@ -1043,7 +1087,6 @@ struct llama_model_params common_model_params_to_llama(common_params & params) {
     if (params.n_gpu_layers != -1) {
         mparams.n_gpu_layers = params.n_gpu_layers;
     }
-    mparams.rpc_servers     = params.rpc_servers.c_str();
     mparams.main_gpu        = params.main_gpu;
     mparams.split_mode      = params.split_mode;
     mparams.tensor_split    = params.tensor_split;
@@ -1729,67 +1772,80 @@ std::string common_detokenize(const struct llama_vocab * vocab, const std::vecto
 // Chat template utils
 //
 
-std::string common_get_builtin_chat_template(const struct llama_model * model) {
-    const char * ptr_tmpl = llama_model_chat_template(model);
-    return ptr_tmpl == nullptr ? "" : ptr_tmpl;
-}
-
-bool common_chat_verify_template(const std::string & tmpl) {
+bool common_chat_verify_template(const std::string & tmpl, bool use_jinja) {
+    if (use_jinja) {
+        try {
+            auto chat_template = common_chat_template(tmpl, "<s>", "</s>");
+            common_chat_inputs inputs;
+            inputs.messages = json::array({{
+                {"role", "user"},
+                {"content", "test"},
+            }});
+            common_chat_params_init(chat_template, inputs);
+            return true;
+        } catch (const std::exception & e) {
+            LOG_ERR("%s: failed to apply template: %s\n", __func__, e.what());
+            return false;
+        }
+    }
     llama_chat_message chat[] = {{"user", "test"}};
     const int res = llama_chat_apply_template(tmpl.c_str(), chat, 1, true, nullptr, 0);
     return res >= 0;
 }
 
-std::string common_chat_apply_template(const struct llama_model * model,
-        const std::string & tmpl,
+std::string common_chat_apply_template(
+        const common_chat_template & tmpl,
         const std::vector<common_chat_msg> & msgs,
-        bool add_ass) {
+        bool add_ass,
+        bool use_jinja) {
+    if (use_jinja) {
+        auto messages = json::array();
+        for (const auto & msg : msgs) {
+            messages.push_back({{"role", msg.role}, {"content", msg.content}});
+        }
+        common_chat_inputs inputs;
+        inputs.messages = messages;
+        inputs.add_generation_prompt = add_ass;
+        return common_chat_params_init(tmpl, inputs).prompt;
+    }
+
     int alloc_size = 0;
-    bool fallback = false; // indicate if we must fallback to default chatml
     std::vector<llama_chat_message> chat;
     for (const auto & msg : msgs) {
         chat.push_back({msg.role.c_str(), msg.content.c_str()});
         alloc_size += (msg.role.size() + msg.content.size()) * 1.25;
     }
 
-    const char * ptr_tmpl = tmpl.empty() ? llama_model_chat_template(model) : tmpl.c_str();
     std::vector<char> buf(alloc_size);
 
     // run the first time to get the total output length
-    int32_t res = llama_chat_apply_template(ptr_tmpl, chat.data(), chat.size(), add_ass, buf.data(), buf.size());
+    int32_t res = llama_chat_apply_template(tmpl.source().c_str(), 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");
-        }
-
-        // If the built-in template is not supported, we default to chatml
-        res = llama_chat_apply_template("chatml", chat.data(), chat.size(), add_ass, buf.data(), buf.size());
-        fallback = true;
+        // 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");
     }
 
     // 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 ? "chatml" : ptr_tmpl,
-            chat.data(), chat.size(), add_ass, buf.data(), buf.size());
+        res = llama_chat_apply_template(tmpl.source().c_str(), chat.data(), chat.size(), add_ass, buf.data(), buf.size());
     }
 
     std::string formatted_chat(buf.data(), res);
     return formatted_chat;
 }
 
-std::string common_chat_format_single(const struct llama_model * model,
-        const std::string & tmpl,
+std::string common_chat_format_single(
+        const common_chat_template & tmpl,
         const std::vector<common_chat_msg> & past_msg,
         const common_chat_msg & new_msg,
-        bool add_ass) {
+        bool add_ass,
+        bool use_jinja) {
     std::ostringstream ss;
-    auto fmt_past_msg = past_msg.empty() ? "" : common_chat_apply_template(model, tmpl, past_msg, false);
+    auto fmt_past_msg = past_msg.empty() ? "" : common_chat_apply_template(tmpl, past_msg, false, use_jinja);
     std::vector<common_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') {
@@ -1797,21 +1853,87 @@ std::string common_chat_format_single(const struct llama_model * model,
     };
     // format chat with new_msg
     chat_new.push_back(new_msg);
-    auto fmt_new_msg = common_chat_apply_template(model, tmpl, chat_new, add_ass);
+    auto fmt_new_msg = common_chat_apply_template(tmpl, chat_new, add_ass, use_jinja);
     // get the diff part
     ss << fmt_new_msg.substr(fmt_past_msg.size(), fmt_new_msg.size() - fmt_past_msg.size());
     return ss.str();
 }
 
-std::string common_chat_format_example(const struct llama_model * model,
-        const std::string & tmpl) {
+std::string common_chat_format_example(const common_chat_template & tmpl, bool use_jinja) {
     std::vector<common_chat_msg> msgs = {
-        {"system",    "You are a helpful assistant"},
-        {"user",      "Hello"},
-        {"assistant", "Hi there"},
-        {"user",      "How are you?"},
+        {"system",    "You are a helpful assistant", {}},
+        {"user",      "Hello", {}},
+        {"assistant", "Hi there", {}},
+        {"user",      "How are you?", {}},
     };
-    return common_chat_apply_template(model, tmpl, msgs, true);
+    return common_chat_apply_template(tmpl, msgs, true, use_jinja);
+}
+
+#define CHATML_TEMPLATE_SRC \
+    "{%- for message in messages -%}\n" \
+    "  {{- '<|im_start|>' + message.role + '\n' + message.content + '<|im_end|>\n' -}}\n" \
+    "{%- endfor -%}\n" \
+    "{%- if add_generation_prompt -%}\n" \
+    "  {{- '<|im_start|>assistant\n' -}}\n" \
+    "{%- endif -%}"
+
+common_chat_templates common_chat_templates_from_model(const struct llama_model * model, const std::string & chat_template_override)
+{
+    std::string default_template_src;
+    std::string template_tool_use_src;
+
+    bool has_explicit_template = !chat_template_override.empty();
+    if (chat_template_override.empty()) {
+        auto str = llama_model_chat_template(model, /* name */ nullptr);
+        if (str) {
+            default_template_src = str;
+            has_explicit_template = true;
+        }
+        str = llama_model_chat_template(model, /* name */ "tool_use");
+        if (str) {
+            template_tool_use_src = str;
+            has_explicit_template = true;
+        }
+    } else {
+        default_template_src = chat_template_override;
+    }
+    if (default_template_src.empty() || default_template_src == "chatml") {
+        if (!template_tool_use_src.empty()) {
+            default_template_src = template_tool_use_src;
+        } else {
+            default_template_src = CHATML_TEMPLATE_SRC;
+        }
+    }
+    auto vocab = llama_model_get_vocab(model);
+    const auto get_token = [&](llama_token token, const char * name, const char * jinja_variable_name) {
+        if (token == LLAMA_TOKEN_NULL) {
+            if (default_template_src.find(jinja_variable_name) != std::string::npos
+                || template_tool_use_src.find(jinja_variable_name) != std::string::npos) {
+                LOG_WRN("%s: warning: vocab does not have a %s token, jinja template won't work as intended.\n", __func__, name);
+            }
+            return std::string();
+        } else {
+            return common_token_to_piece(vocab, token, true);
+        }
+    };
+    auto token_bos = get_token(llama_vocab_bos(vocab), "BOS", "bos_token");
+    auto token_eos = get_token(llama_vocab_eos(vocab), "EOS", "eos_token");
+    try {
+        return {
+            has_explicit_template,
+            std::make_unique<minja::chat_template>(default_template_src, token_bos, token_eos),
+            template_tool_use_src.empty()
+                ? nullptr
+                : std::make_unique<minja::chat_template>(template_tool_use_src, token_bos, token_eos),
+        };
+    } catch (const std::exception & e) {
+        LOG_ERR("%s: failed to parse chat template: %s\n", __func__, e.what());
+        return {
+            has_explicit_template,
+            std::make_unique<minja::chat_template>(CHATML_TEMPLATE_SRC, token_bos, token_eos),
+            nullptr,
+        };
+    }
 }
 
 //

common/common.h

@@ -4,6 +4,7 @@
 
 #include "llama-cpp.h"
 
+#include <set>
 #include <string>
 #include <vector>
 #include <sstream>
@@ -109,6 +110,11 @@ enum common_conversation_mode {
     COMMON_CONVERSATION_MODE_AUTO     = 2,
 };
 
+struct common_grammar_trigger {
+    std::string word;
+    bool at_start;
+};
+
 // sampling parameters
 struct common_params_sampling {
     uint32_t seed = LLAMA_DEFAULT_SEED; // the seed used to initialize llama_sampler
@@ -134,6 +140,7 @@ struct common_params_sampling {
     int32_t dry_allowed_length = 2;     // tokens extending repetitions beyond this receive penalty
     int32_t dry_penalty_last_n = -1;    // how many tokens to scan for repetitions (0 = disable penalty, -1 = context size)
     int32_t mirostat           = 0;     // 0 = disabled, 1 = mirostat, 2 = mirostat 2.0
+    float   top_n_sigma        = -1.00f;// -1.0 = disabled
     float   mirostat_tau       = 5.00f; // target entropy
     float   mirostat_eta       = 0.10f; // learning rate
     bool    ignore_eos         = false;
@@ -154,7 +161,11 @@ struct common_params_sampling {
         COMMON_SAMPLER_TYPE_TEMPERATURE,
     };
 
-    std::string grammar; // optional BNF-like grammar to constrain sampling
+    std::string                         grammar; // optional BNF-like grammar to constrain sampling
+    bool                                grammar_lazy = false;
+    std::vector<common_grammar_trigger> grammar_trigger_words;  // optional trigger words to trigger lazy grammar
+    std::vector<llama_token>            grammar_trigger_tokens; // optional trigger tokens to trigger lazy grammar and print trigger special tokens.
+    std::set<llama_token>               preserved_tokens;
 
     std::vector<llama_logit_bias> logit_bias; // logit biases to apply
 
@@ -175,7 +186,11 @@ struct common_params_speculative {
     struct cpu_params cpuparams;
     struct cpu_params cpuparams_batch;
 
-    std::string model = ""; // draft model for speculative decoding                          // NOLINT
+    std::string hf_repo = ""; // HF repo                                                     // NOLINT
+    std::string hf_file = ""; // HF file                                                     // NOLINT
+
+    std::string model = "";     // draft model for speculative decoding                      // NOLINT
+    std::string model_url = ""; // model url to download                                     // NOLINT
 };
 
 struct common_params_vocoder {
@@ -184,6 +199,13 @@ struct common_params_vocoder {
 
     std::string model     = ""; // model path                                                // NOLINT
     std::string model_url = ""; // model url to download                                     // NOLINT
+
+    bool use_guide_tokens = false; // enable guide tokens to improve TTS accuracy            // NOLINT
+};
+
+enum common_reasoning_format {
+    COMMON_REASONING_FORMAT_NONE,
+    COMMON_REASONING_FORMAT_DEEPSEEK, // Extract thinking tag contents and return as `message.reasoning_content`
 };
 
 struct common_params {
@@ -246,7 +268,6 @@ struct common_params {
     std::string lookup_cache_static  = ""; // path of static ngram cache file for lookup decoding           // NOLINT
     std::string lookup_cache_dynamic = ""; // path of dynamic ngram cache file for lookup decoding          // NOLINT
     std::string logits_file          = ""; // file for saving *all* logits                                  // NOLINT
-    std::string rpc_servers          = ""; // comma separated list of RPC servers                           // NOLINT
 
     std::vector<std::string> in_files;   // all input files
     std::vector<std::string> antiprompt; // strings upon which more user input is prompted (a.k.a. reverse prompts)
@@ -277,6 +298,7 @@ struct common_params {
     bool   kl_divergence    = false; // compute KL divergence
 
     bool usage             = false; // print usage
+    bool completion        = false; // print source-able completion script
     bool use_color         = false; // use color to distinguish generations and inputs
     bool special           = false; // enable special token output
     bool interactive       = false; // interactive mode
@@ -329,7 +351,9 @@ struct common_params {
     std::string hostname      = "127.0.0.1";
     std::string public_path   = "";                                                                         // NOLINT
     std::string chat_template = "";                                                                         // NOLINT
+    bool use_jinja = false;                                                                                 // NOLINT
     bool enable_chat_template = true;
+    common_reasoning_format reasoning_format = COMMON_REASONING_FORMAT_DEEPSEEK;
 
     std::vector<std::string> api_keys;
 
@@ -408,13 +432,13 @@ bool set_process_priority(enum ggml_sched_priority prio);
 //
 
 #ifdef __GNUC__
-#ifdef __MINGW32__
-#define LLAMA_COMMON_ATTRIBUTE_FORMAT(...) __attribute__((format(gnu_printf, __VA_ARGS__)))
+#    if defined(__MINGW32__) && !defined(__clang__)
+#        define LLAMA_COMMON_ATTRIBUTE_FORMAT(...) __attribute__((format(gnu_printf, __VA_ARGS__)))
+#    else
+#        define LLAMA_COMMON_ATTRIBUTE_FORMAT(...) __attribute__((format(printf, __VA_ARGS__)))
+#    endif
 #else
-#define LLAMA_COMMON_ATTRIBUTE_FORMAT(...) __attribute__((format(printf, __VA_ARGS__)))
-#endif
-#else
-#define LLAMA_COMMON_ATTRIBUTE_FORMAT(...)
+#    define LLAMA_COMMON_ATTRIBUTE_FORMAT(...)
 #endif
 
 LLAMA_COMMON_ATTRIBUTE_FORMAT(1, 2)
@@ -423,6 +447,10 @@ std::string string_format(const char * fmt, ...);
 std::string string_strip(const std::string & str);
 std::string string_get_sortable_timestamp();
 
+std::string string_join(const std::vector<std::string> & values, const std::string & separator);
+std::vector<std::string> string_split(const std::string & str, const std::string & delimiter);
+std::string string_repeat(const std::string & str, size_t n);
+
 void string_replace_all(std::string & s, const std::string & search, const std::string & replace);
 
 template<class T>
@@ -507,12 +535,14 @@ struct llama_model * common_load_model_from_url(
     const std::string & local_path,
     const std::string & hf_token,
     const struct llama_model_params & params);
+
 struct llama_model * common_load_model_from_hf(
     const std::string & repo,
     const std::string & remote_path,
     const std::string & local_path,
     const std::string & hf_token,
     const struct llama_model_params & params);
+
 std::pair<std::string, std::string> common_get_hf_file(
     const std::string & hf_repo_with_tag,
     const std::string & hf_token);
@@ -590,36 +620,57 @@ std::string common_detokenize(
 // Chat template utils
 //
 
+struct common_tool_call {
+    std::string name;
+    std::string arguments;
+    std::string id;
+};
+
 // same with llama_chat_message, but uses std::string
 struct common_chat_msg {
     std::string role;
     std::string content;
+    std::vector<common_tool_call> tool_calls;
+    std::string reasoning_content = "";
 };
 
-// Get the built-in chat template for the model. Return empty string if not present.
-std::string common_get_builtin_chat_template(const struct llama_model * model);
-
 // Check if the template supplied via "--chat-template" is supported or not. Returns true if it's valid
-bool common_chat_verify_template(const std::string & tmpl);
+bool common_chat_verify_template(const std::string & tmpl, bool use_jinja);
+
+namespace minja {
+    class chat_template;
+}
+
+typedef minja::chat_template common_chat_template;
+
+struct common_chat_templates {
+    bool has_explicit_template; // Model had builtin template or template overridde was specified.
+    std::unique_ptr<common_chat_template> template_default; // always set (defaults to chatml)
+    std::unique_ptr<common_chat_template> template_tool_use;
+};
 
 // 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 common_chat_apply_template(const struct llama_model * model,
-        const std::string & tmpl,
+std::string common_chat_apply_template(
+        const common_chat_template & tmpl,
         const std::vector<common_chat_msg> & chat,
-        bool add_ass);
+        bool add_ass,
+        bool use_jinja);
 
 // Format single message, while taking into account the position of that message in chat history
-std::string common_chat_format_single(const struct llama_model * model,
-        const std::string & tmpl,
+std::string common_chat_format_single(
+        const common_chat_template & tmpl,
         const std::vector<common_chat_msg> & past_msg,
         const common_chat_msg & new_msg,
-        bool add_ass);
+        bool add_ass,
+        bool use_jinja);
 
 // Returns an example of formatted chat
-std::string common_chat_format_example(const struct llama_model * model,
-        const std::string & tmpl);
+std::string common_chat_format_example(
+    const common_chat_template & tmpl, bool use_jinja);
+
+common_chat_templates common_chat_templates_from_model(const struct llama_model * model, const std::string & chat_template_override);
 
 //
 // KV cache utils

common/json-schema-to-grammar.cpp

@@ -1,4 +1,6 @@
 #include "json-schema-to-grammar.h"
+#include "common.h"
+
 #include <algorithm>
 #include <fstream>
 #include <map>
@@ -11,11 +13,6 @@
 
 using json = nlohmann::ordered_json;
 
-template <typename Iterator>
-static std::string join(Iterator begin, Iterator end, const std::string & separator);
-
-static std::string repeat(const std::string & str, size_t n);
-
 static std::string build_repetition(const std::string & item_rule, int min_items, int max_items, const std::string & separator_rule = "") {
     auto has_max = max_items != std::numeric_limits<int>::max();
 
@@ -128,8 +125,8 @@ static void _build_min_max_int(int min_value, int max_value, std::stringstream &
                 if (sub_len > 0) {
                     auto from_sub = from.substr(i + 1);
                     auto to_sub = to.substr(i + 1);
-                    auto sub_zeros = repeat("0", sub_len);
-                    auto sub_nines = repeat("9", sub_len);
+                    auto sub_zeros = string_repeat("0", sub_len);
+                    auto sub_nines = string_repeat("9", sub_len);
 
                     auto to_reached = false;
                     out << "(";
@@ -188,8 +185,8 @@ static void _build_min_max_int(int min_value, int max_value, std::stringstream &
         auto max_digits = max_s.length();
 
         for (auto digits = min_digits; digits < max_digits; digits++) {
-            uniform_range(min_s, repeat("9", digits));
-            min_s = "1" + repeat("0", digits);
+            uniform_range(min_s, string_repeat("9", digits));
+            min_s = "1" + string_repeat("0", digits);
             out << " | ";
         }
         uniform_range(min_s, max_s);
@@ -318,49 +315,6 @@ 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 = {'^', '$', '.', '[', ']', '(', ')', '|', '{', '}', '*', '+', '?'};
 
-template <typename Iterator>
-std::string join(Iterator begin, Iterator end, const std::string & separator) {
-    std::ostringstream result;
-    if (begin != end) {
-        result << *begin;
-        for (Iterator it = begin + 1; it != end; ++it) {
-            result << separator << *it;
-        }
-    }
-    return result.str();
-}
-
-static std::vector<std::string> split(const std::string & str, const std::string & delimiter) {
-    std::vector<std::string> tokens;
-    size_t start = 0;
-    size_t end = str.find(delimiter);
-
-    while (end != std::string::npos) {
-        tokens.push_back(str.substr(start, end - start));
-        start = end + delimiter.length();
-        end = str.find(delimiter, start);
-    }
-
-    tokens.push_back(str.substr(start));
-
-    return tokens;
-}
-
-static std::string repeat(const std::string & str, size_t n) {
-    if (n == 0) {
-        return "";
-    }
-
-    std::string result;
-    result.reserve(str.length() * n);
-
-    for (size_t i = 0; i < n; ++i) {
-        result += str;
-    }
-
-    return result;
-}
-
 static std::string replacePattern(const std::string & input, const std::regex & regex, const std::function<std::string(const std::smatch  &)> & replacement) {
     std::smatch match;
     std::string result;
@@ -389,6 +343,7 @@ static std::string format_literal(const std::string & literal) {
 
 class SchemaConverter {
 private:
+    friend std::string build_grammar(const std::function<void(const common_grammar_builder &)> & cb, const common_grammar_options & options);
     std::function<json(const std::string &)> _fetch_json;
     bool _dotall;
     std::map<std::string, std::string> _rules;
@@ -418,7 +373,7 @@ private:
         for (size_t i = 0; i < alt_schemas.size(); i++) {
             rules.push_back(visit(alt_schemas[i], name + (name.empty() ? "alternative-" : "-") + std::to_string(i)));
         }
-        return join(rules.begin(), rules.end(), " | ");
+        return string_join(rules, " | ");
     }
 
     std::string _visit_pattern(const std::string & pattern, const std::string & name) {
@@ -481,7 +436,7 @@ private:
                 for (const auto & item : ret) {
                     results.push_back(to_rule(item));
                 }
-                return std::make_pair(join(results.begin(), results.end(), " "), false);
+                return std::make_pair(string_join(results, " "), false);
             };
 
             while (i < length) {
@@ -539,7 +494,7 @@ private:
                     }
                     curly_brackets += '}';
                     i++;
-                    auto nums = split(curly_brackets.substr(1, curly_brackets.length() - 2), ",");
+                    auto nums = string_split(curly_brackets.substr(1, curly_brackets.length() - 2), ",");
                     int min_times = 0;
                     int max_times = std::numeric_limits<int>::max();
                     try {
@@ -809,10 +764,11 @@ private:
 public:
     SchemaConverter(
         const std::function<json(const std::string &)> & fetch_json,
-        bool dotall)
+        bool dotall,
+        bool compact_spaces)
           : _fetch_json(fetch_json), _dotall(dotall)
     {
-        _rules["space"] = SPACE_RULE;
+        _rules["space"] = compact_spaces ? "\" \"?" : SPACE_RULE;
     }
 
     void resolve_refs(json & schema, const std::string & url) {
@@ -854,7 +810,7 @@ public:
                             return;
                         }
                         std::string pointer = ref.substr(ref.find('#') + 1);
-                        std::vector<std::string> tokens = split(pointer, "/");
+                        std::vector<std::string> tokens = string_split(pointer, "/");
                         for (size_t i = 1; i < tokens.size(); ++i) {
                             std::string sel = tokens[i];
                             if (target.is_null() || !target.contains(sel)) {
@@ -905,7 +861,7 @@ public:
             for (const auto & v : schema["enum"]) {
                 enum_values.push_back(_generate_constant_rule(v));
             }
-            return _add_rule(rule_name, "(" + join(enum_values.begin(), enum_values.end(), " | ") + ") space");
+            return _add_rule(rule_name, "(" + string_join(enum_values, " | ") + ") space");
         } else if ((schema_type.is_null() || schema_type == "object")
                 && (schema.contains("properties") ||
                     (schema.contains("additionalProperties") && schema["additionalProperties"] != true))) {
@@ -1019,10 +975,10 @@ public:
 
     void check_errors() {
         if (!_errors.empty()) {
-            throw std::runtime_error("JSON schema conversion failed:\n" + join(_errors.begin(), _errors.end(), "\n"));
+            throw std::runtime_error("JSON schema conversion failed:\n" + string_join(_errors, "\n"));
         }
         if (!_warnings.empty()) {
-            fprintf(stderr, "WARNING: JSON schema conversion was incomplete: %s\n", join(_warnings.begin(), _warnings.end(), "; ").c_str());
+            fprintf(stderr, "WARNING: JSON schema conversion was incomplete: %s\n", string_join(_warnings, "; ").c_str());
         }
     }
 
@@ -1035,11 +991,35 @@ public:
     }
 };
 
-std::string json_schema_to_grammar(const json & schema) {
-    SchemaConverter converter([](const std::string &) { return json::object(); }, /* dotall= */ false);
-    auto copy = schema;
-    converter.resolve_refs(copy, "input");
-    converter.visit(copy, "");
+std::string json_schema_to_grammar(const json & schema, bool force_gbnf) {
+#ifdef LLAMA_USE_LLGUIDANCE
+    if (!force_gbnf) {
+        return "%llguidance {}\nstart: %json " + schema.dump();
+    }
+#else
+    (void)force_gbnf;
+#endif // LLAMA_USE_LLGUIDANCE
+    return build_grammar([&](const common_grammar_builder & callbacks) {
+        auto copy = schema;
+        callbacks.resolve_refs(copy);
+        callbacks.add_schema("", copy);
+    });
+}
+
+std::string build_grammar(const std::function<void(const common_grammar_builder &)> & cb, const common_grammar_options & options) {
+    SchemaConverter converter([&](const std::string &) { return json(); }, options.dotall, options.compact_spaces);
+    common_grammar_builder builder {
+        /* .add_rule = */ [&](const std::string & name, const std::string & rule) {
+            return converter._add_rule(name, rule);
+        },
+        /* .add_schema = */ [&](const std::string & name, const nlohmann::ordered_json & schema) {
+            return converter.visit(schema, name == "root" ? "" : name);
+        },
+        /* .resolve_refs = */ [&](nlohmann::ordered_json & schema) {
+            converter.resolve_refs(schema, "");
+        }
+    };
+    cb(builder);
     converter.check_errors();
     return converter.format_grammar();
 }

common/json-schema-to-grammar.h

@@ -5,4 +5,18 @@
 #define JSON_ASSERT GGML_ASSERT
 #include "json.hpp"
 
-std::string json_schema_to_grammar(const nlohmann::ordered_json& schema);
+std::string json_schema_to_grammar(const nlohmann::ordered_json & schema,
+                                   bool force_gbnf = false);
+
+struct common_grammar_builder {
+    std::function<std::string(const std::string &, const std::string &)> add_rule;
+    std::function<std::string(const std::string &, const nlohmann::ordered_json &)> add_schema;
+    std::function<void(nlohmann::ordered_json &)> resolve_refs;
+};
+
+struct common_grammar_options {
+    bool dotall = false;
+    bool compact_spaces = false;
+};
+
+std::string build_grammar(const std::function<void(const common_grammar_builder &)> & cb, const common_grammar_options & options = {});

common/llguidance.cpp

@@ -0,0 +1,270 @@
+#include "sampling.h"
+#include "log.h"
+
+#ifdef LLAMA_USE_LLGUIDANCE
+
+#    include "llguidance.h"
+#    include <cmath>
+
+struct llama_sampler_llg {
+    const llama_vocab * vocab;
+    std::string         grammar_kind;
+    std::string         grammar_data;
+    LlgTokenizer *      tokenizer;
+    LlgConstraint *     grammar;
+    LlgMaskResult       llg_res;
+    bool                has_llg_res;
+};
+
+static LlgConstraint * llama_sampler_llg_new(LlgTokenizer * tokenizer, const char * grammar_kind,
+                                             const char * grammar_data) {
+    LlgConstraintInit cinit;
+    llg_constraint_init_set_defaults(&cinit, tokenizer);
+    const char * log_level = getenv("LLGUIDANCE_LOG_LEVEL");
+    if (log_level && *log_level) {
+        cinit.log_stderr_level = atoi(log_level);
+    }
+    auto c = llg_new_constraint_any(&cinit, grammar_kind, grammar_data);
+    if (llg_get_error(c)) {
+        LOG_ERR("llg error: %s\n", llg_get_error(c));
+        llg_free_constraint(c);
+        return nullptr;
+    }
+    return c;
+}
+
+static const char * llama_sampler_llg_name(const llama_sampler * /*smpl*/) {
+    return "llguidance";
+}
+
+static void llama_sampler_llg_accept_impl(llama_sampler * smpl, llama_token token) {
+    auto * ctx = (llama_sampler_llg *) smpl->ctx;
+    if (ctx->grammar) {
+        LlgCommitResult res;
+        llg_commit_token(ctx->grammar, token, &res);
+        ctx->has_llg_res = false;
+    }
+}
+
+static void llama_sampler_llg_apply(llama_sampler * smpl, llama_token_data_array * cur_p) {
+    auto * ctx = (llama_sampler_llg *) smpl->ctx;
+    if (ctx->grammar) {
+        if (!ctx->has_llg_res) {
+            if (llg_compute_mask(ctx->grammar, &ctx->llg_res) == 0) {
+                ctx->has_llg_res = true;
+            } else {
+                LOG_ERR("llg error: %s\n", llg_get_error(ctx->grammar));
+                llg_free_constraint(ctx->grammar);
+                ctx->grammar = nullptr;
+            }
+        }
+        if (ctx->has_llg_res) {
+            if (ctx->llg_res.is_stop) {
+                for (size_t i = 0; i < cur_p->size; ++i) {
+                    if (!llama_vocab_is_eog(ctx->vocab, cur_p->data[i].id)) {
+                        cur_p->data[i].logit = -INFINITY;
+                    }
+                }
+            } else {
+                const uint32_t * mask = ctx->llg_res.sample_mask;
+                for (size_t i = 0; i < cur_p->size; ++i) {
+                    auto token = cur_p->data[i].id;
+                    if ((mask[token / 32] & (1 << (token % 32))) == 0) {
+                        cur_p->data[i].logit = -INFINITY;
+                    }
+                }
+            }
+        }
+    }
+}
+
+static void llama_sampler_llg_reset(llama_sampler * smpl) {
+    auto * ctx = (llama_sampler_llg *) smpl->ctx;
+    if (!ctx->grammar) {
+        return;
+    }
+
+    auto * grammar_new = llama_sampler_llg_new(ctx->tokenizer, ctx->grammar_kind.c_str(), ctx->grammar_data.c_str());
+    llg_free_constraint(ctx->grammar);
+    ctx->grammar     = grammar_new;
+    ctx->has_llg_res = false;
+}
+
+static llama_sampler * llama_sampler_llg_clone(const llama_sampler * smpl) {
+    const auto * ctx = (const llama_sampler_llg *) smpl->ctx;
+
+    auto * result = llama_sampler_init_llg(ctx->vocab, nullptr, nullptr);
+
+    // copy the state
+    {
+        auto * result_ctx = (llama_sampler_llg *) result->ctx;
+
+        if (ctx->grammar) {
+            result_ctx->grammar_kind = ctx->grammar_kind;
+            result_ctx->grammar_data = ctx->grammar_data;
+            result_ctx->grammar      = llg_clone_constraint(ctx->grammar);
+            result_ctx->tokenizer    = llg_clone_tokenizer(ctx->tokenizer);
+        }
+    }
+
+    return result;
+}
+
+static void llama_sampler_llg_free(llama_sampler * smpl) {
+    const auto * ctx = (llama_sampler_llg *) smpl->ctx;
+
+    if (ctx->grammar) {
+        llg_free_constraint(ctx->grammar);
+        llg_free_tokenizer(ctx->tokenizer);
+    }
+
+    delete ctx;
+}
+
+static llama_sampler_i llama_sampler_llg_i = {
+    /* .name   = */ llama_sampler_llg_name,
+    /* .accept = */ llama_sampler_llg_accept_impl,
+    /* .apply  = */ llama_sampler_llg_apply,
+    /* .reset  = */ llama_sampler_llg_reset,
+    /* .clone  = */ llama_sampler_llg_clone,
+    /* .free   = */ llama_sampler_llg_free,
+};
+
+static size_t llama_sampler_llg_tokenize_fn(const void * user_data, const uint8_t * bytes, size_t bytes_len,
+                                            uint32_t * output_tokens, size_t output_tokens_len) {
+    const llama_vocab * vocab = (const llama_vocab *) user_data;
+    int                 r     = 0;
+    try {
+        r = llama_tokenize(vocab, (const char *) bytes, bytes_len, (int32_t *) output_tokens, output_tokens_len, false,
+                           true);
+    } catch (const std::exception & e) {
+        GGML_ABORT("llama_tokenize failed: %s\n", e.what());
+    }
+    if (r < 0) {
+        return -r;
+    }
+    return r;
+}
+
+static LlgTokenizer * llama_sampler_llg_new_tokenizer(const llama_vocab * vocab) {
+    // TODO store the tokenizer in the vocab somehow
+    static const llama_vocab * vocab_cache;
+    static LlgTokenizer *      tokenizer_cache;
+
+    if (vocab_cache == vocab) {
+        return llg_clone_tokenizer(tokenizer_cache);
+    }
+
+    auto tok_eos = llama_vocab_eot(vocab);
+    if (tok_eos == LLAMA_TOKEN_NULL) {
+        tok_eos = llama_vocab_eos(vocab);
+    }
+
+    size_t vocab_size = llama_vocab_n_tokens(vocab);
+
+    auto token_lens       = new uint32_t[vocab_size];
+    // we typically have ~7 bytes per token; let's go on the safe side here
+    auto token_bytes_size = vocab_size * 16 + 1024 * 1024;
+    auto token_bytes      = new uint8_t[token_bytes_size];
+
+    size_t offset = 0;
+    for (size_t i = 0; i < vocab_size; i++) {
+        size_t max_token = 1024;
+        if (token_bytes_size - offset < max_token) {
+            GGML_ABORT("token_bytes buffer too small\n");
+        }
+
+        llama_token token = i;
+        auto        dp    = (char *) token_bytes + offset;
+        auto        size  = llama_detokenize(vocab, &token, 1, dp, max_token, false, false);
+        if (size < 0) {
+            GGML_ABORT("llama_detokenize failed\n");
+        }
+        if (size == 0) {
+            size = llama_detokenize(vocab, &token, 1, dp + 1, max_token - 1, false, true);
+            if (size < 0) {
+                GGML_ABORT("llama_detokenize failed\n");
+            }
+            if (size != 0) {
+                *dp = '\xff';  // special token prefix marker
+                size += 1;
+            }
+        }
+
+        token_lens[i] = size;
+        offset += size;
+    }
+
+    LlgTokenizerInit tinit = {
+        /* .vocab_size                         = */ (uint32_t) vocab_size,
+        /* .tok_eos                            = */ (uint32_t) tok_eos,
+        /* .token_lens                         = */ token_lens,
+        /* .token_bytes                        = */ token_bytes,
+        /* .tokenizer_json                     = */ nullptr,
+        /* .tokenize_assumes_string            = */ true,
+        /* .tokenize_fn                        = */ llama_sampler_llg_tokenize_fn,
+        /* .use_approximate_greedy_tokenize_fn = */ false,
+        /* .tokenize_user_data                 = */ vocab,
+    };
+
+    char           error_buffer[1024];
+    LlgTokenizer * tokenizer = llg_new_tokenizer(&tinit, error_buffer, sizeof(error_buffer));
+
+    delete[] token_bytes;
+    delete[] token_lens;
+
+    if (tokenizer == nullptr) {
+        LOG_ERR("llg tokenizer error: %s\n", error_buffer);
+        return tokenizer;
+    }
+
+    if (tokenizer_cache) {
+        llg_free_tokenizer(tokenizer_cache);
+    }
+    vocab_cache     = vocab;
+    tokenizer_cache = tokenizer;
+
+    return llg_clone_tokenizer(tokenizer_cache);
+}
+
+llama_sampler * llama_sampler_init_llg(const llama_vocab * vocab, const char * grammar_kind,
+                                       const char * grammar_data) {
+    auto * ctx = new llama_sampler_llg;
+
+    if (grammar_kind != nullptr && grammar_kind[0] != '\0') {
+        auto tokenizer = llama_sampler_llg_new_tokenizer(vocab);
+        *ctx           = {
+            /* .vocab        = */ vocab,
+            /* .grammar_kind = */ grammar_kind,
+            /* .grammar_data = */ grammar_data,
+            /* .tokenizer    = */ tokenizer,
+            /* .grammar      = */ llama_sampler_llg_new(tokenizer, grammar_kind, grammar_data),
+            /* .llg_res      = */ {},
+            /* .has_llg_res  = */ false,
+        };
+    } else {
+        *ctx = {
+            /* .vocab        = */ vocab,
+            /* .grammar_kind = */ {},
+            /* .grammar_data = */ {},
+            /* .tokenizer    = */ nullptr,
+            /* .grammar      = */ nullptr,
+            /* .llg_res      = */ {},
+            /* .has_llg_res  = */ false,
+        };
+    }
+
+    return llama_sampler_init(
+        /* .iface = */ &llama_sampler_llg_i,
+        /* .ctx   = */ ctx
+    );
+}
+
+#else
+
+llama_sampler * llama_sampler_init_llg(const llama_vocab *, const char *, const char *) {
+    LOG_WRN("llguidance (cmake -DLLAMA_LLGUIDANCE=ON) is not enabled");
+    return nullptr;
+}
+
+#endif  // LLAMA_USE_LLGUIDANCE

common/log.cpp

@@ -1,5 +1,6 @@
 #include "log.h"
 
+#include <chrono>
 #include <condition_variable>
 #include <cstdarg>
 #include <cstdio>
@@ -14,16 +15,6 @@ void common_log_set_verbosity_thold(int verbosity) {
     common_log_verbosity_thold = verbosity;
 }
 
-#define LOG_COL_DEFAULT "\033[0m"
-#define LOG_COL_BOLD    "\033[1m"
-#define LOG_COL_RED     "\033[31m"
-#define LOG_COL_GREEN   "\033[32m"
-#define LOG_COL_YELLOW  "\033[33m"
-#define LOG_COL_BLUE    "\033[34m"
-#define LOG_COL_MAGENTA "\033[35m"
-#define LOG_COL_CYAN    "\033[36m"
-#define LOG_COL_WHITE   "\033[37m"
-
 static int64_t t_us() {
     return std::chrono::duration_cast<std::chrono::microseconds>(std::chrono::system_clock::now().time_since_epoch()).count();
 }
@@ -206,6 +197,7 @@ public:
                 vsnprintf(entry.msg.data(), entry.msg.size(), ss.str().c_str(), args_copy);
             }
 #endif
+            va_end(args_copy);
         }
 
         entry.level = level;

common/log.h

@@ -2,9 +2,20 @@
 
 #include "ggml.h" // for ggml_log_level
 
+#define LOG_CLR_TO_EOL  "\033[K\r"
+#define LOG_COL_DEFAULT "\033[0m"
+#define LOG_COL_BOLD    "\033[1m"
+#define LOG_COL_RED     "\033[31m"
+#define LOG_COL_GREEN   "\033[32m"
+#define LOG_COL_YELLOW  "\033[33m"
+#define LOG_COL_BLUE    "\033[34m"
+#define LOG_COL_MAGENTA "\033[35m"
+#define LOG_COL_CYAN    "\033[36m"
+#define LOG_COL_WHITE   "\033[37m"
+
 #ifndef __GNUC__
 #    define LOG_ATTRIBUTE_FORMAT(...)
-#elif defined(__MINGW32__)
+#elif defined(__MINGW32__) && !defined(__clang__)
 #    define LOG_ATTRIBUTE_FORMAT(...) __attribute__((format(gnu_printf, __VA_ARGS__)))
 #else
 #    define LOG_ATTRIBUTE_FORMAT(...) __attribute__((format(printf, __VA_ARGS__)))

common/sampling.cpp

@@ -134,11 +134,11 @@ std::string common_params_sampling::print() const {
     snprintf(result, sizeof(result),
             "\trepeat_last_n = %d, repeat_penalty = %.3f, frequency_penalty = %.3f, presence_penalty = %.3f\n"
             "\tdry_multiplier = %.3f, dry_base = %.3f, dry_allowed_length = %d, dry_penalty_last_n = %d\n"
-            "\ttop_k = %d, top_p = %.3f, min_p = %.3f, xtc_probability = %.3f, xtc_threshold = %.3f, typical_p = %.3f, temp = %.3f\n"
+            "\ttop_k = %d, top_p = %.3f, min_p = %.3f, xtc_probability = %.3f, xtc_threshold = %.3f, typical_p = %.3f, top_n_sigma = %.3f, temp = %.3f\n"
             "\tmirostat = %d, mirostat_lr = %.3f, mirostat_ent = %.3f",
             penalty_last_n, penalty_repeat, penalty_freq, penalty_present,
             dry_multiplier, dry_base, dry_allowed_length, dry_penalty_last_n,
-            top_k, top_p, min_p, xtc_probability, xtc_threshold, typ_p, temp,
+            top_k, top_p, min_p, xtc_probability, xtc_threshold, typ_p, top_n_sigma, temp,
             mirostat, mirostat_eta, mirostat_tau);
 
     return std::string(result);
@@ -151,9 +151,30 @@ struct common_sampler * common_sampler_init(const struct llama_model * model, co
 
     lparams.no_perf = params.no_perf;
 
+    struct llama_sampler * grmr;
+    if (params.grammar.compare(0, 11, "%llguidance") == 0) {
+#ifdef LLAMA_USE_LLGUIDANCE
+        grmr = llama_sampler_init_llg(vocab, "lark", params.grammar.c_str());
+#else
+        GGML_ABORT("llguidance (cmake -DLLAMA_LLGUIDANCE=ON) is not enabled");
+#endif // LLAMA_USE_LLGUIDANCE
+    } else {
+        std::vector<const char *> trigger_words;
+        trigger_words.reserve(params.grammar_trigger_words.size());
+        for (const auto & str : params.grammar_trigger_words) {
+            trigger_words.push_back(str.word.c_str());
+        }
+
+        grmr = params.grammar_lazy
+             ? llama_sampler_init_grammar_lazy(vocab, params.grammar.c_str(), "root",
+                                               trigger_words.data(), trigger_words.size(),
+                                               params.grammar_trigger_tokens.data(), params.grammar_trigger_tokens.size())
+             :      llama_sampler_init_grammar(vocab, params.grammar.c_str(), "root");
+    }
+
     auto * result = new common_sampler {
         /* .params = */ params,
-        /* .grmr   = */ llama_sampler_init_grammar(vocab, params.grammar.c_str(), "root"),
+        /* .grmr   = */ grmr,
         /* .chain  = */ llama_sampler_chain_init(lparams),
         /* .prev   = */ ring_buffer<llama_token>(std::max(32, params.n_prev)),
         /* .cur    = */ {},
@@ -167,45 +188,51 @@ struct common_sampler * common_sampler_init(const struct llama_model * model, co
                 params.logit_bias.data()));
 
     if (params.mirostat == 0) {
-        for (const auto & cnstr : params.samplers) {
-            switch (cnstr) {
-                case COMMON_SAMPLER_TYPE_DRY:
-                    {
-                        std::vector<const char *> c_breakers;
-                        c_breakers.reserve(params.dry_sequence_breakers.size());
-                        for (const auto & str : params.dry_sequence_breakers) {
-                            c_breakers.push_back(str.c_str());
-                        }
+        if (params.top_n_sigma >= 0) {
+            llama_sampler_chain_add(result->chain, llama_sampler_init_top_k        (params.top_k));
+            llama_sampler_chain_add(result->chain, llama_sampler_init_temp         (params.temp));
+            llama_sampler_chain_add(result->chain, llama_sampler_init_top_n_sigma  (params.top_n_sigma));
+        } else {
+            for (const auto & cnstr : params.samplers) {
+                switch (cnstr) {
+                    case COMMON_SAMPLER_TYPE_DRY:
+                        {
+                            std::vector<const char *> c_breakers;
+                            c_breakers.reserve(params.dry_sequence_breakers.size());
+                            for (const auto & str : params.dry_sequence_breakers) {
+                                c_breakers.push_back(str.c_str());
+                            }
 
-                        llama_sampler_chain_add(result->chain, llama_sampler_init_dry      (vocab, llama_model_n_ctx_train(model), params.dry_multiplier, params.dry_base, params.dry_allowed_length, params.dry_penalty_last_n, c_breakers.data(), c_breakers.size()));
-                    }
-                    break;
-                case COMMON_SAMPLER_TYPE_TOP_K:
-                    llama_sampler_chain_add(result->chain, llama_sampler_init_top_k    (params.top_k));
-                    break;
-                case COMMON_SAMPLER_TYPE_TOP_P:
-                    llama_sampler_chain_add(result->chain, llama_sampler_init_top_p    (params.top_p, params.min_keep));
-                    break;
-                case COMMON_SAMPLER_TYPE_MIN_P:
-                    llama_sampler_chain_add(result->chain, llama_sampler_init_min_p    (params.min_p, params.min_keep));
-                    break;
-                case COMMON_SAMPLER_TYPE_XTC:
-                    llama_sampler_chain_add(result->chain, llama_sampler_init_xtc      (params.xtc_probability, params.xtc_threshold, params.min_keep, params.seed));
-                    break;
-                case COMMON_SAMPLER_TYPE_TYPICAL_P:
-                    llama_sampler_chain_add(result->chain, llama_sampler_init_typical  (params.typ_p, params.min_keep));
-                    break;
-                case COMMON_SAMPLER_TYPE_TEMPERATURE:
-                    llama_sampler_chain_add(result->chain, llama_sampler_init_temp_ext (params.temp, params.dynatemp_range, params.dynatemp_exponent));
-                    break;
-                case COMMON_SAMPLER_TYPE_INFILL:
-                    llama_sampler_chain_add(result->chain, llama_sampler_init_infill   (vocab));
-                    break;
-                case COMMON_SAMPLER_TYPE_PENALTIES:
-                    llama_sampler_chain_add(result->chain, llama_sampler_init_penalties(params.penalty_last_n, params.penalty_repeat, params.penalty_freq, params.penalty_present));
-                    break;
-                default:
-                    GGML_ASSERT(false && "unknown sampler type");
+                            llama_sampler_chain_add(result->chain, llama_sampler_init_dry      (vocab, llama_model_n_ctx_train(model), params.dry_multiplier, params.dry_base, params.dry_allowed_length, params.dry_penalty_last_n, c_breakers.data(), c_breakers.size()));
+                        }
+                        break;
+                    case COMMON_SAMPLER_TYPE_TOP_K:
+                        llama_sampler_chain_add(result->chain, llama_sampler_init_top_k    (params.top_k));
+                        break;
+                    case COMMON_SAMPLER_TYPE_TOP_P:
+                        llama_sampler_chain_add(result->chain, llama_sampler_init_top_p    (params.top_p, params.min_keep));
+                        break;
+                    case COMMON_SAMPLER_TYPE_MIN_P:
+                        llama_sampler_chain_add(result->chain, llama_sampler_init_min_p    (params.min_p, params.min_keep));
+                        break;
+                    case COMMON_SAMPLER_TYPE_XTC:
+                        llama_sampler_chain_add(result->chain, llama_sampler_init_xtc      (params.xtc_probability, params.xtc_threshold, params.min_keep, params.seed));
+                        break;
+                    case COMMON_SAMPLER_TYPE_TYPICAL_P:
+                        llama_sampler_chain_add(result->chain, llama_sampler_init_typical  (params.typ_p, params.min_keep));
+                        break;
+                    case COMMON_SAMPLER_TYPE_TEMPERATURE:
+                        llama_sampler_chain_add(result->chain, llama_sampler_init_temp_ext (params.temp, params.dynatemp_range, params.dynatemp_exponent));
+                        break;
+                    case COMMON_SAMPLER_TYPE_INFILL:
+                        llama_sampler_chain_add(result->chain, llama_sampler_init_infill   (vocab));
+                        break;
+                    case COMMON_SAMPLER_TYPE_PENALTIES:
+                        llama_sampler_chain_add(result->chain, llama_sampler_init_penalties(params.penalty_last_n, params.penalty_repeat, params.penalty_freq, params.penalty_present));
+                        break;
+                    default:
+                        GGML_ASSERT(false && "unknown sampler type");
+                }
             }
         }
         llama_sampler_chain_add(result->chain, llama_sampler_init_dist(params.seed));

common/sampling.h

@@ -102,3 +102,6 @@ std::string common_sampler_type_to_str(enum common_sampler_type cnstr);
 
 std::vector<enum common_sampler_type> common_sampler_types_from_names(const std::vector<std::string> & names, bool allow_alt_names);
 std::vector<enum common_sampler_type> common_sampler_types_from_chars(const std::string & chars);
+
+llama_sampler * llama_sampler_init_llg(const llama_vocab * vocab,
+                const char * grammar_kind, const char * grammar_data);

common/speculative.h

@@ -9,7 +9,7 @@ struct common_speculative_params {
     int n_draft = 16;  // max drafted tokens
     int n_reuse = 256;
 
-    float p_min = 0.9f; // min probabiliy required to accept a token in the draft
+    float p_min = 0.9f; // min probability required to accept a token in the draft
 };
 
 struct common_speculative * common_speculative_init(struct llama_context * ctx_dft);

convert_hf_to_gguf.py

@@ -558,7 +558,7 @@ class Model:
 
     # NOTE: this function is generated by convert_hf_to_gguf_update.py
     #       do not modify it manually!
-    # ref:  https://github.com/ggerganov/llama.cpp/pull/6920
+    # ref:  https://github.com/ggml-org/llama.cpp/pull/6920
     # Marker: Start get_vocab_base_pre
     def get_vocab_base_pre(self, tokenizer) -> str:
         # encoding this string and hashing the resulting tokens would (hopefully) give us a unique identifier that
@@ -648,7 +648,7 @@ class Model:
         if chkhsh == "7967bfa498ade6b757b064f31e964dddbb80f8f9a4d68d4ba7998fcf281c531a":
             # ref: https://huggingface.co/jinaai/jina-embeddings-v2-base-code
             res = "jina-v2-code"
-        if chkhsh == "b6e8e1518dc4305be2fe39c313ed643381c4da5db34a98f6a04c093f8afbe99b":
+        if chkhsh == "b6e8e1518dc4305be2fe39c313ed643381c4da5db34a98f6a04c093f8afbe99b" or chkhsh == "81d72c7348a9f0ebe86f23298d37debe0a5e71149e29bd283904c02262b27516":
             # ref: https://huggingface.co/THUDM/glm-4-9b-chat
             res = "chatglm-bpe"
         if chkhsh == "7fc505bd3104ca1083b150b17d088b59534ede9bde81f0dd2090967d7fe52cee":
@@ -696,6 +696,9 @@ class Model:
         if chkhsh == "877081d19cf6996e2c4ff0e1236341e9b7bde288f5311a56a937f0afbbb3aeb5":
             # ref: https://huggingface.co/deepseek-ai/DeepSeek-V3
             res = "deepseek-v3"
+        if chkhsh == "b3f499bb4255f8ca19fccd664443283318f2fd2414d5e0b040fbdd0cc195d6c5":
+            # ref: https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B
+            res = "deepseek-r1-qwen"
 
         if res is None:
             logger.warning("\n")
@@ -705,7 +708,7 @@ class Model:
             logger.warning("**          - the model has not been added to convert_hf_to_gguf_update.py yet")
             logger.warning("**          - the pre-tokenization config has changed upstream")
             logger.warning("**          Check your model files and convert_hf_to_gguf_update.py and update them accordingly.")
-            logger.warning("** ref:     https://github.com/ggerganov/llama.cpp/pull/6920")
+            logger.warning("** ref:     https://github.com/ggml-org/llama.cpp/pull/6920")
             logger.warning("**")
             logger.warning(f"** chkhsh:  {chkhsh}")
             logger.warning("**************************************************************************************")
@@ -2832,7 +2835,7 @@ class InternLM2Model(Model):
         if chat_eos_token_id is not None:
             # For the chat model, we replace the eos with '<|im_end|>'.
             # TODO: this is a hack, should be fixed
-            #       https://github.com/ggerganov/llama.cpp/pull/6745#issuecomment-2067687048
+            #       https://github.com/ggml-org/llama.cpp/pull/6745#issuecomment-2067687048
             special_vocab.special_token_ids["eos"] = chat_eos_token_id
             logger.warning(f"Replace eos:{old_eos} with a special token:{chat_eos_token_id}"
                            " in chat mode so that the conversation can end normally.")
@@ -2882,6 +2885,66 @@ class InternLM2Model(Model):
             return [(self.map_tensor_name(name), data_torch)]
 
 
+@Model.register("InternLM3ForCausalLM")
+class InternLM3Model(Model):
+    model_arch = gguf.MODEL_ARCH.LLAMA
+
+    def set_vocab(self):
+        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))
+
+        tokenizer_config_file = self.dir_model / 'tokenizer_config.json'
+        if tokenizer_config_file.is_file():
+            with open(tokenizer_config_file, "r", encoding="utf-8") as f:
+                tokenizer_config_json = json.load(f)
+                if "add_prefix_space" in tokenizer_config_json:
+                    self.gguf_writer.add_add_space_prefix(tokenizer_config_json["add_prefix_space"])
+
+                if "added_tokens_decoder" in tokenizer_config_json:
+                    for token_id, token_data in tokenizer_config_json["added_tokens_decoder"].items():
+                        if token_data.get("special"):
+                            token_id = int(token_id)
+                            token = token_data["content"]
+                            special_vocab._set_special_token(token, token_id)
+                            # update eos token
+                            if token == '<|im_end|>' and "eos" in special_vocab.special_token_ids:
+                                special_vocab.special_token_ids["eos"] = token_id
+
+        special_vocab.add_to_gguf(self.gguf_writer)
+
+    def set_gguf_parameters(self):
+        super().set_gguf_parameters()
+        hparams = self.hparams
+        self.gguf_writer.add_vocab_size(hparams["vocab_size"])
+
+        if "head_dim" in hparams:
+            rope_dim = hparams["head_dim"]
+        else:
+            rope_dim = hparams["hidden_size"] // hparams["num_attention_heads"]
+        self.gguf_writer.add_rope_dimension_count(rope_dim)
+
+        if self.hparams.get("rope_scaling") is not None and "factor" in self.hparams["rope_scaling"]:
+            if self.hparams["rope_scaling"].get("type") == "linear" or self.hparams["rope_scaling"].get("rope_type") == "linear":
+                self.gguf_writer.add_rope_scaling_type(gguf.RopeScalingType.LINEAR)
+                self.gguf_writer.add_rope_scaling_factor(self.hparams["rope_scaling"]["factor"])
+
+    def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
+        n_head = self.hparams["num_attention_heads"]
+        n_kv_head = self.hparams.get("num_key_value_heads")
+        if name.endswith(("q_proj.weight", "q_proj.bias")):
+            data_torch = LlamaModel.permute(data_torch, n_head, n_head)
+        if name.endswith(("k_proj.weight", "k_proj.bias")):
+            data_torch = LlamaModel.permute(data_torch, n_head, n_kv_head)
+        return [(self.map_tensor_name(name), data_torch)]
+
+
 @Model.register("BertModel", "BertForMaskedLM", "CamembertModel")
 class BertModel(Model):
     model_arch = gguf.MODEL_ARCH.BERT
@@ -4450,7 +4513,7 @@ class JaisModel(Model):
         self.gguf_writer.add_max_alibi_bias(self.max_alibi_bias)
 
 
-@Model.register("ChatGLMModel", "ChatGLMForConditionalGeneration")
+@Model.register("GlmForCausalLM", "ChatGLMModel", "ChatGLMForConditionalGeneration")
 class ChatGLMModel(Model):
     model_arch = gguf.MODEL_ARCH.CHATGLM
 
@@ -4556,47 +4619,15 @@ class ChatGLMModel(Model):
 
         from transformers import AutoTokenizer
         tokenizer = AutoTokenizer.from_pretrained(dir_model, trust_remote_code=True)
-        vocab_size = hparams["padded_vocab_size"]
+        vocab_size = hparams.get("padded_vocab_size",hparams["vocab_size"])
         assert max(tokenizer.get_vocab().values()) < vocab_size
 
-        tokpre = self.get_vocab_base_pre(tokenizer)
-
-        merges = []
-        vocab = {}
-        mergeable_ranks = tokenizer.mergeable_ranks
-        for token, rank in mergeable_ranks.items():
-            vocab[ChatGLMModel.token_bytes_to_string(token)] = rank
-            if len(token) == 1:
-                continue
-            merged = ChatGLMModel.bpe(mergeable_ranks, token, max_rank=rank)
-            assert len(merged) >= 2 and len(merged) <= 7
-            merges.append(' '.join(map(ChatGLMModel.token_bytes_to_string, merged)))
-
-        # for this kind of tokenizer, added_vocab is not a subset of vocab, so they need to be combined
-        added_vocab = tokenizer.get_added_vocab()
-        reverse_vocab = {id_ : encoded_tok for encoded_tok, id_ in {**vocab, **added_vocab}.items()}
-
-        for i in range(vocab_size):
-            if i not in reverse_vocab:
-                tokens.append(f"[PAD{i}]")
-                toktypes.append(gguf.TokenType.UNUSED)
-            elif reverse_vocab[i] in added_vocab:
-                tokens.append(reverse_vocab[i])
-                if tokenizer.added_tokens_decoder[i].special:
-                    toktypes.append(gguf.TokenType.CONTROL)
-                else:
-                    toktypes.append(gguf.TokenType.USER_DEFINED)
-            else:
-                tokens.append(reverse_vocab[i])
-                toktypes.append(gguf.TokenType.NORMAL)
-
+        tokens, toktypes, tokpre = self.get_vocab_base()
         self.gguf_writer.add_tokenizer_model("gpt2")
         self.gguf_writer.add_tokenizer_pre(tokpre)
         self.gguf_writer.add_token_list(tokens)
         self.gguf_writer.add_token_types(toktypes)
-
-        special_vocab = gguf.SpecialVocab(dir_model, load_merges=False)
-        special_vocab.merges = merges
+        special_vocab = gguf.SpecialVocab(self.dir_model, load_merges=True)
         # only add special tokens when they were not already loaded from config.json
         special_vocab._set_special_token("eos", tokenizer.get_added_vocab()["<|endoftext|>"])
         special_vocab._set_special_token("eot", tokenizer.get_added_vocab()["<|user|>"])
@@ -4607,16 +4638,20 @@ class ChatGLMModel(Model):
     def set_gguf_parameters(self):
         n_embed = self.hparams.get("hidden_size", self.hparams.get("n_embed"))
         n_head = self.hparams.get("n_head", self.hparams.get("num_attention_heads"))
-        n_head_kv = self.hparams.get("multi_query_group_num", n_head)
+        n_head_kv = self.hparams.get("multi_query_group_num", self.hparams.get("num_key_value_heads", n_head))
         self.gguf_writer.add_context_length(self.hparams.get("seq_length", n_embed))
         self.gguf_writer.add_embedding_length(n_embed)
-        self.gguf_writer.add_feed_forward_length(self.hparams.get("ffn_hidden_size", 4 * n_embed))
-        self.gguf_writer.add_block_count(self.hparams["num_layers"])
+        self.gguf_writer.add_feed_forward_length(self.hparams.get("ffn_hidden_size", self.hparams.get("intermediate_size", 4 * n_embed)))
+        self.gguf_writer.add_block_count(self.hparams.get("num_layers", self.hparams["num_hidden_layers"]))
         self.gguf_writer.add_head_count(n_head)
         self.gguf_writer.add_head_count_kv(n_head_kv)
-        self.gguf_writer.add_layer_norm_rms_eps(self.hparams["layernorm_epsilon"])
+        self.gguf_writer.add_layer_norm_rms_eps(self.hparams.get("layernorm_epsilon",1e-5))
         self.gguf_writer.add_file_type(self.ftype)
-        self.gguf_writer.add_rope_dimension_count(64)
+        if "attention_dim" in self.hparams:
+            rope_dim = self.hparams["attention_dim"]
+        else:
+            rope_dim = self.hparams["hidden_size"] // self.hparams["num_attention_heads"]
+        self.gguf_writer.add_rope_dimension_count(int(rope_dim * self.hparams.get("partial_rotary_factor", 0.5)))
         self.gguf_writer.add_add_bos_token(False)
         rope_freq = 10000
         if "rope_ratio" in self.hparams:
@@ -4626,7 +4661,7 @@ class ChatGLMModel(Model):
     def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
         del bid  # unused
 
-        if name.endswith(".rotary_pos_emb.inv_freq"):
+        if name.endswith(".rotary_pos_emb.inv_freq") or name.startswith("model.vision."):
             return []
 
         name = name.removeprefix("transformer.")

convert_hf_to_gguf_update.py

@@ -8,7 +8,7 @@
 # provide the necessary information to llama.cpp via the GGUF header in order to implement
 # the same pre-tokenizer.
 #
-# ref: https://github.com/ggerganov/llama.cpp/pull/6920
+# ref: https://github.com/ggml-org/llama.cpp/pull/6920
 #
 # Instructions:
 #
@@ -65,49 +65,50 @@ else:
 
 # TODO: add models here, base models preferred
 models = [
-    {"name": "llama-spm",      "tokt": TOKENIZER_TYPE.SPM, "repo": "https://huggingface.co/meta-llama/Llama-2-7b-hf", },
-    {"name": "llama-bpe",      "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/meta-llama/Meta-Llama-3-8B", },
-    {"name": "phi-3",          "tokt": TOKENIZER_TYPE.SPM, "repo": "https://huggingface.co/microsoft/Phi-3-mini-4k-instruct", },
-    {"name": "deepseek-llm",   "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/deepseek-ai/deepseek-llm-7b-base", },
-    {"name": "deepseek-coder", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/deepseek-ai/deepseek-coder-6.7b-base", },
-    {"name": "falcon",         "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/tiiuae/falcon-7b", },
-    {"name": "bert-bge",       "tokt": TOKENIZER_TYPE.WPM, "repo": "https://huggingface.co/BAAI/bge-small-en-v1.5", },
-    {"name": "falcon3",        "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/tiiuae/Falcon3-7B-Base", },
-    {"name": "bert-bge-large", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/BAAI/bge-large-zh-v1.5", },
-    {"name": "mpt",            "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/mosaicml/mpt-7b", },
-    {"name": "starcoder",      "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/bigcode/starcoder2-3b", },
-    {"name": "gpt-2",          "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/openai-community/gpt2", },
-    {"name": "stablelm2",      "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/stabilityai/stablelm-2-zephyr-1_6b", },
-    {"name": "refact",         "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/smallcloudai/Refact-1_6-base", },
-    {"name": "command-r",      "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/CohereForAI/c4ai-command-r-v01", },
-    {"name": "qwen2",          "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/Qwen/Qwen1.5-7B", },
-    {"name": "olmo",           "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/allenai/OLMo-1.7-7B-hf", },
-    {"name": "dbrx",           "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/databricks/dbrx-base", },
-    {"name": "jina-v1-en",     "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/jinaai/jina-reranker-v1-tiny-en", },
-    {"name": "jina-v2-en",     "tokt": TOKENIZER_TYPE.WPM, "repo": "https://huggingface.co/jinaai/jina-embeddings-v2-base-en", }, # WPM!
-    {"name": "jina-v2-es",     "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/jinaai/jina-embeddings-v2-base-es", },
-    {"name": "jina-v2-de",     "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/jinaai/jina-embeddings-v2-base-de", },
-    {"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", },
-    {"name": "t5",             "tokt": TOKENIZER_TYPE.UGM, "repo": "https://huggingface.co/google-t5/t5-small", },
-    {"name": "codeshell",      "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/WisdomShell/CodeShell-7B", },
-    {"name": "tekken",         "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/mistralai/Mistral-Nemo-Base-2407", },
-    {"name": "smollm",         "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/HuggingFaceTB/SmolLM-135M", },
-    {'name': "bloom",          "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/bigscience/bloom", },
-    {'name': "gpt3-finnish",   "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/TurkuNLP/gpt3-finnish-small", },
-    {"name": "exaone",         "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/LGAI-EXAONE/EXAONE-3.0-7.8B-Instruct", },
-    {"name": "phi-2",          "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/microsoft/phi-2", },
-    {"name": "chameleon",      "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/facebook/chameleon-7b", },
-    {"name": "minerva-7b",     "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/sapienzanlp/Minerva-7B-base-v1.0", },
-    {"name": "roberta-bpe",    "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/sentence-transformers/stsb-roberta-base"},
-    {"name": "gigachat",       "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/ai-sage/GigaChat-20B-A3B-instruct"},
-    {"name": "megrez",         "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/Infinigence/Megrez-3B-Instruct"},
-    {"name": "deepseek-v3",    "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/deepseek-ai/DeepSeek-V3"},
+    {"name": "llama-spm",        "tokt": TOKENIZER_TYPE.SPM, "repo": "https://huggingface.co/meta-llama/Llama-2-7b-hf", },
+    {"name": "llama-bpe",        "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/meta-llama/Meta-Llama-3-8B", },
+    {"name": "phi-3",            "tokt": TOKENIZER_TYPE.SPM, "repo": "https://huggingface.co/microsoft/Phi-3-mini-4k-instruct", },
+    {"name": "deepseek-llm",     "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/deepseek-ai/deepseek-llm-7b-base", },
+    {"name": "deepseek-coder",   "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/deepseek-ai/deepseek-coder-6.7b-base", },
+    {"name": "falcon",           "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/tiiuae/falcon-7b", },
+    {"name": "bert-bge",         "tokt": TOKENIZER_TYPE.WPM, "repo": "https://huggingface.co/BAAI/bge-small-en-v1.5", },
+    {"name": "falcon3",          "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/tiiuae/Falcon3-7B-Base", },
+    {"name": "bert-bge-large",   "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/BAAI/bge-large-zh-v1.5", },
+    {"name": "mpt",              "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/mosaicml/mpt-7b", },
+    {"name": "starcoder",        "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/bigcode/starcoder2-3b", },
+    {"name": "gpt-2",            "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/openai-community/gpt2", },
+    {"name": "stablelm2",        "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/stabilityai/stablelm-2-zephyr-1_6b", },
+    {"name": "refact",           "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/smallcloudai/Refact-1_6-base", },
+    {"name": "command-r",        "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/CohereForAI/c4ai-command-r-v01", },
+    {"name": "qwen2",            "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/Qwen/Qwen1.5-7B", },
+    {"name": "olmo",             "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/allenai/OLMo-1.7-7B-hf", },
+    {"name": "dbrx",             "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/databricks/dbrx-base", },
+    {"name": "jina-v1-en",       "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/jinaai/jina-reranker-v1-tiny-en", },
+    {"name": "jina-v2-en",       "tokt": TOKENIZER_TYPE.WPM, "repo": "https://huggingface.co/jinaai/jina-embeddings-v2-base-en", }, # WPM!
+    {"name": "jina-v2-es",       "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/jinaai/jina-embeddings-v2-base-es", },
+    {"name": "jina-v2-de",       "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/jinaai/jina-embeddings-v2-base-de", },
+    {"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", },
+    {"name": "t5",               "tokt": TOKENIZER_TYPE.UGM, "repo": "https://huggingface.co/google-t5/t5-small", },
+    {"name": "codeshell",        "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/WisdomShell/CodeShell-7B", },
+    {"name": "tekken",           "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/mistralai/Mistral-Nemo-Base-2407", },
+    {"name": "smollm",           "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/HuggingFaceTB/SmolLM-135M", },
+    {'name': "bloom",            "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/bigscience/bloom", },
+    {'name': "gpt3-finnish",     "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/TurkuNLP/gpt3-finnish-small", },
+    {"name": "exaone",           "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/LGAI-EXAONE/EXAONE-3.0-7.8B-Instruct", },
+    {"name": "phi-2",            "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/microsoft/phi-2", },
+    {"name": "chameleon",        "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/facebook/chameleon-7b", },
+    {"name": "minerva-7b",       "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/sapienzanlp/Minerva-7B-base-v1.0", },
+    {"name": "roberta-bpe",      "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/sentence-transformers/stsb-roberta-base"},
+    {"name": "gigachat",         "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/ai-sage/GigaChat-20B-A3B-instruct"},
+    {"name": "megrez",           "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/Infinigence/Megrez-3B-Instruct"},
+    {"name": "deepseek-v3",      "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/deepseek-ai/DeepSeek-V3"},
+    {"name": "deepseek-r1-qwen", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B"},
 ]
 
 
@@ -245,7 +246,7 @@ src_func = f"""
             logger.warning("**          - the model has not been added to convert_hf_to_gguf_update.py yet")
             logger.warning("**          - the pre-tokenization config has changed upstream")
             logger.warning("**          Check your model files and convert_hf_to_gguf_update.py and update them accordingly.")
-            logger.warning("** ref:     https://github.com/ggerganov/llama.cpp/pull/6920")
+            logger.warning("** ref:     https://github.com/ggml-org/llama.cpp/pull/6920")
             logger.warning("**")
             logger.warning(f"** chkhsh:  {{chkhsh}}")
             logger.warning("**************************************************************************************")

convert_lora_to_gguf.py

@@ -395,7 +395,7 @@ if __name__ == '__main__':
                         logger.error(f"Unexpected name '{name}': Not a lora_A or lora_B tensor")
                         if ".embed_tokens.weight" in name or ".lm_head.weight" in name:
                             logger.error("Embeddings is present in the adapter. This can be due to new tokens added during fine tuning")
-                            logger.error("Please refer to https://github.com/ggerganov/llama.cpp/pull/9948")
+                            logger.error("Please refer to https://github.com/ggml-org/llama.cpp/pull/9948")
                         sys.exit(1)
 
                     if base_name in tensor_map:
@@ -419,7 +419,7 @@ if __name__ == '__main__':
                 # some archs may have the same tensor for lm_head and output (tie word embeddings)
                 # in this case, adapters targeting lm_head will fail when using llama-export-lora
                 # therefore, we ignore them for now
-                # see: https://github.com/ggerganov/llama.cpp/issues/9065
+                # see: https://github.com/ggml-org/llama.cpp/issues/9065
                 if name == "lm_head.weight" and len(dest) == 0:
                     raise ValueError("lm_head is present in adapter, but is ignored in base model")
                 for dest_name, dest_data in dest:

docs/android.md

@@ -12,7 +12,7 @@ $ apt update && apt upgrade -y
 $ apt install git cmake
 ```
 
-Then, follow the [build instructions](https://github.com/ggerganov/llama.cpp/blob/master/docs/build.md), specifically for CMake.
+Then, follow the [build instructions](https://github.com/ggml-org/llama.cpp/blob/master/docs/build.md), specifically for CMake.
 
 Once the binaries are built, download your model of choice (e.g., from Hugging Face). It's recommended to place it in the `~/` directory for best performance:
 

docs/backend/OPENCL.md

@@ -0,0 +1,205 @@
+# llama.cpp for OpenCL
+
+- [Background](#background)
+- [OS](#os)
+- [Hardware](#hardware)
+- [DataType Supports](#datatype-supports)
+- [Model Preparation](#model-preparation)
+- [CMake Options](#cmake-options)
+- [Android](#android)
+- [Windows 11 Arm64](#windows-11-arm64)
+- [Known Issue](#known-issues)
+- [TODO](#todo)
+
+## Background
+
+OpenCL (Open Computing Language) is an open, royalty-free standard for cross-platform, parallel programming of diverse accelerators found in supercomputers, cloud servers, personal computers, mobile devices and embedded platforms. OpenCL specifies a programming language (based on C99) for programming these devices and application programming interfaces (APIs) to control the platform and execute programs on the compute devices. Similar to CUDA, OpenCL has been widely used to program GPUs and is supported by most GPU vendors.
+
+### Llama.cpp + OpenCL
+
+The llama.cpp OpenCL backend is designed to enable llama.cpp on **Qualcomm Adreno GPU** firstly via OpenCL. Thanks to the portabilty of OpenCL, the OpenCL backend can also run on certain Intel GPUs although the performance is not optimal.
+
+## OS
+
+| OS      | Status  | Verified                                       |
+|---------|---------|------------------------------------------------|
+| Android | Support | Snapdragon 8 Gen 3, Snapdragon 8 Elite         |
+| Windows | Support | Windows 11 Arm64 with Snapdragon X Elite       |
+| Linux   | Support | Ubuntu 22.04 WSL2 with Intel 12700H            |
+
+## Hardware
+
+### Adreno GPU
+
+**Verified devices**
+
+| Adreno GPU                           | Status  |
+|:------------------------------------:|:-------:|
+| Adreno 750 (Snapdragon 8 Gen 3)      | Support |
+| Adreno 830 (Snapdragon 8 Elite)      | Support |
+| Adreno X85 (Snapdragon X Elite)      | Support |
+
+## DataType Supports
+
+| DataType               | Status                     |
+|:----------------------:|:--------------------------:|
+| Q4_0                   | Support                    |
+| Q6_K                   | Support, but not optimized |
+
+## Model Preparation
+
+You can refer to the general [*Prepare and Quantize*](README.md#prepare-and-quantize) guide for model prepration.
+
+Currently we support `Q4_0` quantization and have optimize for it. To achieve best performance on Adreno GPU, add `--pure` to `llama-quantize`. For example,
+
+```sh
+./llama-quantize --pure ggml-model-qwen2.5-3b-f16.gguf ggml-model-qwen-3b-Q4_0.gguf Q4_0
+```
+
+Since `Q6_K` is also supported, `Q4_0` quantization without `--pure` will also work. However, the performance will be worse compared to pure `Q4_0` quantization.
+
+## CMake Options
+
+The OpenCL backend has the following CMake options that control the behavior of the backend.
+
+| CMake options                     | Default value  | Description                               |
+|:---------------------------------:|:--------------:|:------------------------------------------|
+| `GGML_OPENCL_EMBED_KERNELS`       | `ON`           | Embed OpenCL kernels into the executable. |
+| `GGML_OPENCL_USE_ADRENO_KERNELS`  | `ON`           | Use kernels optimized for Adreno.         |
+
+## Android
+
+Ubuntu 22.04 is used for targeting Android. Make sure the following tools are accessible from command line,
+
+* Git
+* CMake 3.29
+* Ninja
+* Python3
+
+### I. Setup Environment
+
+1. **Install NDK**
+
+```sh
+cd ~
+wget https://dl.google.com/android/repository/commandlinetools-linux-8512546_latest.zip && \
+unzip commandlinetools-linux-8512546_latest.zip && \
+mkdir -p ~/android-sdk/cmdline-tools && \
+mv cmdline-tools latest && \
+mv latest ~/android-sdk/cmdline-tools/ && \
+rm -rf commandlinetools-linux-8512546_latest.zip
+
+yes | ~/android-sdk/cmdline-tools/latest/bin/sdkmanager "ndk;26.3.11579264"
+```
+
+2. **Install OpenCL Headers and Library**
+
+```sh
+mkdir -p ~/dev/llm
+cd ~/dev/llm
+
+git clone https://github.com/KhronosGroup/OpenCL-Headers && \
+cd OpenCL-Headers && \
+cp -r CL ~/android-sdk/ndk/26.3.11579264/toolchains/llvm/prebuilt/linux-x86_64/sysroot/usr/include
+
+cd ~/dev/llm
+
+git clone https://github.com/KhronosGroup/OpenCL-ICD-Loader && \
+cd OpenCL-ICD-Loader && \
+mkdir build_ndk26 && cd build_ndk26 && \
+cmake .. -G Ninja -DCMAKE_BUILD_TYPE=Release \
+  -DCMAKE_TOOLCHAIN_FILE=$HOME/android-sdk/ndk/26.3.11579264/build/cmake/android.toolchain.cmake \
+  -DOPENCL_ICD_LOADER_HEADERS_DIR=$HOME/android-sdk/ndk/26.3.11579264/toolchains/llvm/prebuilt/linux-x86_64/sysroot/usr/include \
+  -DANDROID_ABI=arm64-v8a \
+  -DANDROID_PLATFORM=24 \
+  -DANDROID_STL=c++_shared && \
+ninja && \
+cp libOpenCL.so ~/android-sdk/ndk/26.3.11579264/toolchains/llvm/prebuilt/linux-x86_64/sysroot/usr/lib/aarch64-linux-android
+```
+
+### II. Build llama.cpp
+
+```sh
+cd ~/dev/llm
+
+git clone https://github.com/ggml-org/llama.cpp && \
+cd llama.cpp && \
+mkdir build-android && cd build-android
+
+cmake .. -G Ninja \
+  -DCMAKE_TOOLCHAIN_FILE=$HOME/android-sdk/ndk/26.3.11579264/build/cmake/android.toolchain.cmake \
+  -DANDROID_ABI=arm64-v8a \
+  -DANDROID_PLATFORM=android-28 \
+  -DBUILD_SHARED_LIBS=OFF \
+  -DGGML_OPENCL=ON
+
+ninja
+```
+
+## Windows 11 Arm64
+
+A Snapdragon X Elite device with Windows 11 Arm64 is used. Make sure the following tools are accessible from command line,
+
+* Git
+* CMake 3.29
+* Clang 19
+* Ninja
+* Visual Studio 2022
+
+Powershell is used for the following instructions.
+
+### I. Setup Environment
+
+1. **Install OpenCL Headers and Library**
+
+```powershell
+mkdir -p ~/dev/llm
+
+cd ~/dev/llm
+git clone https://github.com/KhronosGroup/OpenCL-Headers && cd OpenCL-Headers
+mkdir build && cd build
+cmake .. -G Ninja `
+  -DBUILD_TESTING=OFF `
+  -DOPENCL_HEADERS_BUILD_TESTING=OFF `
+  -DOPENCL_HEADERS_BUILD_CXX_TESTS=OFF `
+  -DCMAKE_INSTALL_PREFIX="$HOME/dev/llm/opencl"
+cmake --build . --target install
+
+cd ~/dev/llm
+git clone https://github.com/KhronosGroup/OpenCL-ICD-Loader && cd OpenCL-ICD-Loader
+mkdir build && cd build
+cmake .. -G Ninja `
+  -DCMAKE_BUILD_TYPE=Release `
+  -DCMAKE_PREFIX_PATH="$HOME/dev/llm/opencl" `
+  -DCMAKE_INSTALL_PREFIX="$HOME/dev/llm/opencl"
+cmake --build . --target install
+```
+
+### II. Build llama.cpp
+
+```powershell
+
+mkdir -p ~/dev/llm
+cd ~/dev/llm
+
+git clone https://github.com/ggml-org/llama.cpp && cd llama.cpp
+mkdir build && cd build
+
+cmake .. -G Ninja `
+  -DCMAKE_TOOLCHAIN_FILE="$HOME/dev/llm/llama.cpp/cmake/arm64-windows-llvm.cmake" `
+  -DCMAKE_BUILD_TYPE=Release `
+  -DCMAKE_PREFIX_PATH="$HOME/dev/llm/opencl" `
+  -DBUILD_SHARED_LIBS=OFF `
+  -DGGML_OPENCL=ON
+ninja
+```
+
+## Known Issues
+
+- Qwen2.5 0.5B model produces gibberish output with Adreno kernels.
+
+## TODO
+
+- Fix Qwen2.5 0.5B
+- Optimization for Q6_K
+- Support and optimization for Q4_K

docs/backend/SYCL.md

@@ -36,8 +36,8 @@ The following release is verified with good quality:
 
 |Commit ID|Tag|Release|Verified  Platform| Update date|
 |-|-|-|-|-|
-|3bcd40b3c593d14261fb2abfabad3c0fb5b9e318|b4040 |[llama-b4040-bin-win-sycl-x64.zip](https://github.com/ggerganov/llama.cpp/releases/download/b4040/llama-b4040-bin-win-sycl-x64.zip) |Arc770/Linux/oneAPI 2024.1<br>MTL Arc GPU/Windows 11/oneAPI 2024.1| 2024-11-19|
-|fb76ec31a9914b7761c1727303ab30380fd4f05c|b3038 |[llama-b3038-bin-win-sycl-x64.zip](https://github.com/ggerganov/llama.cpp/releases/download/b3038/llama-b3038-bin-win-sycl-x64.zip) |Arc770/Linux/oneAPI 2024.1<br>MTL Arc GPU/Windows 11/oneAPI 2024.1||
+|3bcd40b3c593d14261fb2abfabad3c0fb5b9e318|b4040 |[llama-b4040-bin-win-sycl-x64.zip](https://github.com/ggml-org/llama.cpp/releases/download/b4040/llama-b4040-bin-win-sycl-x64.zip) |Arc770/Linux/oneAPI 2024.1<br>MTL Arc GPU/Windows 11/oneAPI 2024.1| 2024-11-19|
+|fb76ec31a9914b7761c1727303ab30380fd4f05c|b3038 |[llama-b3038-bin-win-sycl-x64.zip](https://github.com/ggml-org/llama.cpp/releases/download/b3038/llama-b3038-bin-win-sycl-x64.zip) |Arc770/Linux/oneAPI 2024.1<br>MTL Arc GPU/Windows 11/oneAPI 2024.1||
 
 
 ## News
@@ -58,7 +58,7 @@ The following release is verified with good quality:
 - 2024.3
   - Release binary files of Windows.
   - A blog is published: **Run LLM on all Intel GPUs Using llama.cpp**: [intel.com](https://www.intel.com/content/www/us/en/developer/articles/technical/run-llm-on-all-gpus-using-llama-cpp-artical.html) or [medium.com](https://medium.com/@jianyu_neo/run-llm-on-all-intel-gpus-using-llama-cpp-fd2e2dcbd9bd).
-  - New base line is ready: [tag b2437](https://github.com/ggerganov/llama.cpp/tree/b2437).
+  - New base line is ready: [tag b2437](https://github.com/ggml-org/llama.cpp/tree/b2437).
   - Support multiple cards: **--split-mode**: [none|layer]; not support [row], it's on developing.
   - Support to assign main GPU by **--main-gpu**, replace $GGML_SYCL_DEVICE.
   - Support detecting all GPUs with level-zero and same top **Max compute units**.
@@ -133,7 +133,7 @@ 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="GGML_SYCL_F16=ON" --target light -f .devops/intel.Dockerfile .
 ```
 
 *Notes*:

docs/build.md

@@ -3,7 +3,7 @@
 **To get the Code:**
 
 ```bash
-git clone https://github.com/ggerganov/llama.cpp
+git clone https://github.com/ggml-org/llama.cpp
 cd llama.cpp
 ```
 
@@ -46,7 +46,7 @@ cmake --build build --config Release
   ```
 
 - Building for Windows (x86, x64 and arm64) with MSVC or clang as compilers:
-    - Install Visual Studio 2022, e.g. via the [Community Edition](https://visualstudio.microsoft.com/de/vs/community/). In the installer, select at least the following options (this also automatically installs the required additional tools like CMake,...):
+    - Install Visual Studio 2022, e.g. via the [Community Edition](https://visualstudio.microsoft.com/vs/community/). In the installer, select at least the following options (this also automatically installs the required additional tools like CMake,...):
     - Tab Workload: Desktop-development with C++
     - Tab Components (select quickly via search): C++-_CMake_ Tools for Windows, _Git_ for Windows, C++-_Clang_ Compiler for Windows, MS-Build Support for LLVM-Toolset (clang)
     - Please remember to always use a Developer Command Prompt / PowerShell for VS2022 for git, build, test
@@ -125,21 +125,66 @@ For detailed info, please refer to [llama.cpp for SYCL](./backend/SYCL.md).
 
 ## CUDA
 
-This provides GPU acceleration using an NVIDIA GPU. Make sure to have the CUDA toolkit installed. You can download it from your Linux distro's package manager (e.g. `apt install nvidia-cuda-toolkit`) or from the [NVIDIA developer site](https://developer.nvidia.com/cuda-downloads).
+This provides GPU acceleration using an NVIDIA GPU. Make sure to have the [CUDA toolkit](https://developer.nvidia.com/cuda-toolkit) installed.
 
-If you are using Fedora (using Fedora Workstation, or an 'Atomic' variant such as Silverblue), or would like to set up CUDA in a toolbox, please consider our [Fedora CUDA guide](./cuda-fedora.md). Unfortunately, the process is not as simple as one might expect.
+#### Download directly from NVIDIA
+You may find the official downloads here: [NVIDIA developer site](https://developer.nvidia.com/cuda-downloads).
 
-- Using `CMake`:
 
-  ```bash
-  cmake -B build -DGGML_CUDA=ON
-  cmake --build build --config Release
-  ```
+#### Compile and run inside a Fedora Toolbox Container
+We also have a [guide](./cuda-fedora.md) for setting up CUDA toolkit in a Fedora [toolbox container](https://containertoolbx.org/).
 
-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.
+**Recommended for:**
+
+- ***Particularly*** *convenient* for users of [Atomic Desktops for Fedora](https://fedoraproject.org/atomic-desktops/); such as: [Silverblue](https://fedoraproject.org/atomic-desktops/silverblue/) and [Kinoite](https://fedoraproject.org/atomic-desktops/kinoite/).
+- Toolbox is installed by default: [Fedora Workstation](https://fedoraproject.org/workstation/) or [Fedora KDE Plasma Desktop](https://fedoraproject.org/spins/kde).
+- *Optionally* toolbox packages are available: [Arch Linux](https://archlinux.org/), [Red Hat Enterprise Linux >= 8.5](https://www.redhat.com/en/technologies/linux-platforms/enterprise-linux), or [Ubuntu](https://ubuntu.com/download)
+
+
+### Compilation
+```bash
+cmake -B build -DGGML_CUDA=ON
+cmake --build build --config Release
+```
+
+### Override Compute Capability Specifications
+
+If `nvcc` cannot detect your gpu, you may get compile-warnings such as:
+ ```text
+nvcc warning : Cannot find valid GPU for '-arch=native', default arch is used
+```
+
+To override the `native` GPU detection:
+
+#### 1. Take note of the `Compute Capability` of your NVIDIA devices: ["CUDA: Your GPU Compute > Capability"](https://developer.nvidia.com/cuda-gpus).
+
+```text
+GeForce RTX 4090      8.9
+GeForce RTX 3080 Ti   8.6
+GeForce RTX 3070      8.6
+```
+
+#### 2. Manually list each varying `Compute Capability` in the `CMAKE_CUDA_ARCHITECTURES` list.
+
+```bash
+cmake -B build -DGGML_CUDA=ON -DCMAKE_CUDA_ARCHITECTURES="86;89"
+```
+
+### Runtime CUDA environmental variables
+
+You may set the [cuda environmental variables](https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#env-vars) at runtime.
+
+```bash
+# Use `CUDA_VISIBLE_DEVICES` to hide the first compute device.
+CUDA_VISIBLE_DEVICES="-0" ./build/bin/llama-server --model /srv/models/llama.gguf
+```
+
+### Unified Memory
 
 The environment variable `GGML_CUDA_ENABLE_UNIFIED_MEMORY=1` can be used to enable unified memory in Linux. This allows swapping to system RAM instead of crashing when the GPU VRAM is exhausted. In Windows this setting is available in the NVIDIA control panel as `System Memory Fallback`.
 
+### Performance Tuning
+
 The following compilation options are also available to tweak performance:
 
 | Option                        | Legal values           | Default | Description                                                                                                                                                                                                                                                                             |
@@ -286,7 +331,7 @@ You don't need to install Vulkan SDK. It will be installed inside the container.
 
 ```sh
 # Build the image
-docker build -t llama-cpp-vulkan -f .devops/llama-cli-vulkan.Dockerfile .
+docker build -t llama-cpp-vulkan --target light -f .devops/vulkan.Dockerfile .
 
 # Then, use it:
 docker run -it --rm -v "$(pwd):/app:Z" --device /dev/dri/renderD128:/dev/dri/renderD128 --device /dev/dri/card1:/dev/dri/card1 llama-cpp-vulkan -m "/app/models/YOUR_MODEL_FILE" -p "Building a website can be done in 10 simple steps:" -n 400 -e -ngl 33

docs/cuda-fedora.md

@@ -1,17 +1,16 @@
 # Setting Up CUDA on Fedora
 
 In this guide we setup [Nvidia CUDA](https://docs.nvidia.com/cuda/) in a toolbox container. This guide is applicable for:
+
 - [Fedora Workstation](https://fedoraproject.org/workstation/)
 - [Atomic Desktops for Fedora](https://fedoraproject.org/atomic-desktops/)
 - [Fedora Spins](https://fedoraproject.org/spins)
-- [Other Distributions](https://containertoolbx.org/distros/), including `Red Hat Enterprise Linux >= 8.`, `Arch Linux`, and `Ubuntu`.
-
+- [Other Distributions](https://containertoolbx.org/distros/), including `Red Hat Enterprise Linux >= 8.5`, `Arch Linux`, and `Ubuntu`.
 
 ## Table of Contents
 
 - [Prerequisites](#prerequisites)
-- [Monitoring NVIDIA CUDA Repositories](#monitoring-nvidia-cuda-repositories)
-- [Using the Fedora 39 CUDA Repository](#using-the-fedora-39-cuda-repository)
+- [Using the Fedora 41 CUDA Repository](#using-the-fedora-41-cuda-repository)
 - [Creating a Fedora Toolbox Environment](#creating-a-fedora-toolbox-environment)
 - [Installing Essential Development Tools](#installing-essential-development-tools)
 - [Adding the CUDA Repository](#adding-the-cuda-repository)
@@ -29,44 +28,33 @@ In this guide we setup [Nvidia CUDA](https://docs.nvidia.com/cuda/) in a toolbox
 ## Prerequisites
 
 - **Toolbox Installed on the Host System** `Fedora Silverblue` and `Fedora Workstation` both have toolbox by default, other distributions may need to install the [toolbox package](https://containertoolbx.org/install/).
-- **NVIDIA Drivers and Graphics Card installed on Host System (optional)** To run CUDA program, such as `llama.cpp`, the host should be setup to access your NVIDIA hardware. Fedora Hosts can use the [RPM Fusion Repository](https://rpmfusion.org/Howto/NVIDIA).
+- **NVIDIA Drivers and Graphics Card installed on Host System (recommended)** To run CUDA program, such as `llama.cpp`, the host should be setup to access your NVIDIA hardware. Fedora Hosts can use the [RPM Fusion Repository](https://rpmfusion.org/Howto/NVIDIA).
 - **Internet connectivity** to download packages.
 
-### Monitoring NVIDIA CUDA Repositories
+### Using the Fedora 41 CUDA Repository
 
-Before proceeding, it is advisable to check if NVIDIA has updated their CUDA repositories for your Fedora version. NVIDIA's repositories can be found at:
+The latest release is 41.
 
-- [Fedora 40 CUDA Repository](https://developer.download.nvidia.com/compute/cuda/repos/fedora40/x86_64/)
 - [Fedora 41 CUDA Repository](https://developer.download.nvidia.com/compute/cuda/repos/fedora41/x86_64/)
 
-As of the latest update, these repositories do not contain the `cuda` meta-package or are missing essential components.
-
-### Using the Fedora 39 CUDA Repository
-
-Since the newer repositories are incomplete, we'll use the Fedora 39 repository:
-
-- [Fedora 39 CUDA Repository](https://developer.download.nvidia.com/compute/cuda/repos/fedora39/x86_64/)
-
-**Note:** Fedora 39 is no longer maintained, so we recommend using a toolbox environment to prevent system conflicts.
+**Note:** We recommend using a toolbox environment to prevent system conflicts.
 
 ## Creating a Fedora Toolbox Environment
 
-This guide focuses on Fedora hosts, but with small adjustments, it can work for other hosts. Using a Fedora 39 toolbox allows us to install the necessary packages without affecting the host system.
+This guide focuses on Fedora hosts, but with small adjustments, it can work for other hosts. Using the Fedora Toolbox allows us to install the necessary packages without affecting the host system.
 
 **Note:** Toolbox is available for other systems, and even without Toolbox, it is possible to use Podman or Docker.
 
-We do not recommend installing on the host system, as Fedora 39 is out-of-maintenance, and instead you should upgrade to a maintained version of Fedora for your host.
-
-1. **Create a Fedora 39 Toolbox:**
+1. **Create a Fedora 41 Toolbox:**
 
    ```bash
-   toolbox create --image registry.fedoraproject.org/fedora-toolbox:39 --container fedora-toolbox-39-cuda
+   toolbox create --image registry.fedoraproject.org/fedora-toolbox:41 --container fedora-toolbox-41-cuda
    ```
 
 2. **Enter the Toolbox:**
 
    ```bash
-   toolbox enter --container fedora-toolbox-39-cuda
+   toolbox enter --container fedora-toolbox-41-cuda
    ```
 
    Inside the toolbox, you have root privileges and can install packages without affecting the host system.
@@ -85,7 +73,7 @@ We do not recommend installing on the host system, as Fedora 39 is out-of-mainte
    sudo dnf install vim-default-editor --allowerasing
    ```
 
-   The `--allowerasing` flag resolves any package conflicts.
+   The `--allowerasing` flag will allow the removal of the conflicting `nano-default-editor` package.
 
 3. **Install Development Tools and Libraries:**
 
@@ -100,7 +88,7 @@ We do not recommend installing on the host system, as Fedora 39 is out-of-mainte
 Add the NVIDIA CUDA repository to your DNF configuration:
 
 ```bash
-sudo dnf config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/fedora39/x86_64/cuda-fedora39.repo
+sudo dnf config-manager addrepo --from-repofile=https://developer.download.nvidia.com/compute/cuda/repos/fedora41/x86_64/cuda-fedora41.repo
 ```
 
 After adding the repository, synchronize the package manager again:
@@ -109,106 +97,62 @@ After adding the repository, synchronize the package manager again:
 sudo dnf distro-sync
 ```
 
-## Installing `nvidia-driver-libs`
+## Installing `nvidia-driver-libs` and `nvidia-driver-cuda-libs`
 
-Attempt to install `nvidia-driver-libs`:
+We need to detect if the host is supplying the [NVIDIA driver libraries into the toolbox](https://github.com/containers/toolbox/blob/main/src/pkg/nvidia/nvidia.go).
 
 ```bash
-sudo dnf install nvidia-driver-libs
+ls -la /usr/lib64/libcuda.so.1
 ```
 
 **Explanation:**
 
-- `nvidia-driver-libs` contains necessary NVIDIA driver libraries required by CUDA.
-- This step might fail due to conflicts with existing NVIDIA drivers on the host system.
+- `nvidia-driver-libs` and `nvidia-driver-cuda-libs` contains necessary NVIDIA driver libraries required by CUDA,
+  on hosts with NVIDIA drivers installed the Fedora Container will supply the host libraries.
 
-## Manually Resolving Package Conflicts
+### Install Nvidia Driver Libraries on Guest (if `libcuda.so.1` was NOT found).
+
+```bash
+sudo dnf install nvidia-driver-libs nvidia-driver-cuda-libs
+```
+
+### Manually Updating the RPM database for host-supplied NVIDIA drivers (if `libcuda.so.1` was found).
 
 If the installation fails due to conflicts, we'll manually download and install the required packages, excluding conflicting files.
 
-### 1. Download the `nvidia-driver-libs` RPM
+#### 1. Download `nvidia-driver-libs` and `nvidia-driver-cuda-libs` RPM's (with dependencies)
 
 ```bash
-sudo dnf download --arch x86_64 nvidia-driver-libs
+sudo dnf download --destdir=/tmp/nvidia-driver-libs --resolve --arch x86_64 nvidia-driver-libs nvidia-driver-cuda-libs
 ```
 
-You should see a file similar to:
-
-```
-nvidia-driver-libs-560.35.05-1.fc39.x86_64.rpm
-```
-
-### 2. Attempt to Install the RPM
+#### 2. Update the RPM database to assume the installation of these packages.
 
 ```bash
-sudo dnf install nvidia-driver-libs-560.35.05-1.fc39.x86_64.rpm
-```
-
-**Expected Error:**
-
-Installation may fail with errors pointing to conflicts with `egl-gbm` and `egl-wayland`.
-
-**Note: It is important to carefully read the error messages to identify the exact paths that need to be excluded.**
-
-### 3. Download Dependencies
-
-```bash
-sudo dnf download --arch x86_64 egl-gbm egl-wayland
-```
-
-### 4. Install `egl-gbm` with Excluded Paths
-
-Exclude conflicting files during installation:
-
-```bash
-sudo rpm --install --verbose --hash \
-  --excludepath=/usr/lib64/libnvidia-egl-gbm.so.1.1.2 \
-  --excludepath=/usr/share/egl/egl_external_platform.d/15_nvidia_gbm.json \
-  egl-gbm-1.1.2^20240919gitb24587d-3.fc39.x86_64.rpm
-```
-
-**Explanation:**
-
-- The `--excludepath` option skips installing files that conflict with existing files.
-- Adjust the paths based on the error messages you receive.
-
-### 5. Install `egl-wayland` with Excluded Paths
-
-```bash
-sudo rpm --install --verbose --hash \
-  --excludepath=/usr/share/egl/egl_external_platform.d/10_nvidia_wayland.json \
-  egl-wayland-1.1.17^20241118giteeb29e1-5.fc39.x86_64.rpm
-```
-
-### 6. Install `nvidia-driver-libs` with Excluded Paths
-
-```bash
-sudo rpm --install --verbose --hash \
-  --excludepath=/usr/share/glvnd/egl_vendor.d/10_nvidia.json \
-  --excludepath=/usr/share/nvidia/nvoptix.bin \
-  nvidia-driver-libs-560.35.05-1.fc39.x86_64.rpm
+sudo rpm --install --verbose --hash --justdb /tmp/nvidia-driver-libs/*
 ```
 
 **Note:**
 
-- Replace the paths with the ones causing conflicts in your installation if they differ.
-- The `--verbose` and `--hash` options provide detailed output during installation.
+- The `--justdb` option only updates the RPM database, without touching the filesystem.
 
-## Finalizing the Installation of `nvidia-driver-libs`
+#### Finalizing the Installation of `nvidia-driver-libs` and `nvidia-driver-cuda-libs`
 
 After manually installing the dependencies, run:
 
 ```bash
-sudo dnf install nvidia-driver-libs
+sudo dnf install nvidia-driver-libs nvidia-driver-cuda-libs
 ```
 
 You should receive a message indicating the package is already installed:
 
 ```
-Package nvidia-driver-libs-3:560.35.05-1.fc39.x86_64 is already installed.
-Dependencies resolved.
+Updating and loading repositories:
+Repositories loaded.
+Package "nvidia-driver-libs-3:570.86.10-1.fc41.x86_64" is already installed.
+Package "nvidia-driver-cuda-libs-3:570.86.10-1.fc41.x86_64" is already installed.
+
 Nothing to do.
-Complete!
 ```
 
 ## Installing the CUDA Meta-Package
@@ -233,7 +177,7 @@ To use CUDA, add its binary directory to your system's `PATH`.
 
    **Explanation:**
 
-   - We add to  `/etc/profile.d/` as the `/etc/` folder is unique to this particular container, and is not shared with other containers or the host system.
+   - We add to `/etc/profile.d/` as the `/etc/` folder is unique to this particular container, and is not shared with other containers or the host system.
    - The backslash `\` before `$PATH` ensures the variable is correctly written into the script.
 
 2. **Make the Script Executable:**
@@ -262,26 +206,33 @@ You should see output similar to:
 
 ```
 nvcc: NVIDIA (R) Cuda compiler driver
-Copyright (c) 2005-2024 NVIDIA Corporation
-Built on Tue_Oct_29_23:50:19_PDT_2024
-Cuda compilation tools, release 12.6, V12.6.85
-Build cuda_12.6.r12.6/compiler.35059454_0
+Copyright (c) 2005-2025 NVIDIA Corporation
+Built on Wed_Jan_15_19:20:09_PST_2025
+Cuda compilation tools, release 12.8, V12.8.61
+Build cuda_12.8.r12.8/compiler.35404655_0
 ```
 
 This output confirms that the CUDA compiler is accessible and indicates the installed version.
 
 ## Conclusion
 
-You have successfully set up CUDA on Fedora within a toolbox environment using the Fedora 39 CUDA repository. By manually resolving package conflicts and configuring the environment, you can develop CUDA applications without affecting your host system.
+You have successfully set up CUDA on Fedora within a toolbox environment using the Fedora 41 CUDA repository. By manually updating the RPM db and configuring the environment, you can develop CUDA applications without affecting your host system.
 
 ## Troubleshooting
 
 - **Installation Failures:**
-  - If you encounter errors during installation, carefully read the error messages. They often indicate conflicting files or missing dependencies.
-  - Use the `--excludepath` option with `rpm` to exclude conflicting files during manual installations.
 
-- **Driver Conflicts:**
-  - Since the host system may already have NVIDIA drivers installed, conflicts can arise. Using the toolbox environment helps isolate these issues.
+  - If you encounter errors during installation, carefully read the error messages. They often indicate conflicting files or missing dependencies.
+  - You may use the `--excludepath` option with `rpm` to exclude conflicting files during manual RPM installations.
+
+- **Rebooting the Container:**
+
+  - Sometimes there may be a bug in the NVIDIA driver host passthrough (such as missing a shared library). Rebooting the container may solve this issue:
+
+  ```bash
+  # on the host system
+  podman container restart --all
+  ```
 
 - **Environment Variables Not Set:**
   - If `nvcc` is not found after installation, ensure that `/usr/local/cuda/bin` is in your `PATH`.
@@ -291,11 +242,13 @@ You have successfully set up CUDA on Fedora within a toolbox environment using t
 ## Additional Notes
 
 - **Updating CUDA in the Future:**
+
   - Keep an eye on the official NVIDIA repositories for updates to your Fedora version.
   - When an updated repository becomes available, adjust your `dnf` configuration accordingly.
 
 - **Building `llama.cpp`:**
-  - With CUDA installed, you can follow these [build instructions for `llama.cpp`](https://github.com/ggerganov/llama.cpp/blob/master/docs/build.md) to compile it with CUDA support.
+
+  - With CUDA installed, you can follow these [build instructions for `llama.cpp`](https://github.com/ggml-org/llama.cpp/blob/master/docs/build.md) to compile it with CUDA support.
   - Ensure that any CUDA-specific build flags or paths are correctly set in your build configuration.
 
 - **Using the Toolbox Environment:**

docs/development/HOWTO-add-model.md

@@ -104,16 +104,16 @@ Note: to debug the inference graph: you can use [llama-eval-callback](/examples/
 
 ## GGUF specification
 
-https://github.com/ggerganov/ggml/blob/master/docs/gguf.md
+https://github.com/ggml-org/ggml/blob/master/docs/gguf.md
 
 ## Resources
 
-- YaRN RoPE scaling https://github.com/ggerganov/llama.cpp/pull/2268
-- support Baichuan serial models https://github.com/ggerganov/llama.cpp/pull/3009
-- support attention bias https://github.com/ggerganov/llama.cpp/pull/4283
-- Mixtral support https://github.com/ggerganov/llama.cpp/pull/4406
-- BERT embeddings https://github.com/ggerganov/llama.cpp/pull/5423
-- Grok-1 support https://github.com/ggerganov/llama.cpp/pull/6204
-- Command R Plus support https://github.com/ggerganov/llama.cpp/pull/6491
-- support arch DBRX https://github.com/ggerganov/llama.cpp/pull/6515
-- How to convert HuggingFace model to GGUF format https://github.com/ggerganov/llama.cpp/discussions/2948
+- YaRN RoPE scaling https://github.com/ggml-org/llama.cpp/pull/2268
+- support Baichuan serial models https://github.com/ggml-org/llama.cpp/pull/3009
+- support attention bias https://github.com/ggml-org/llama.cpp/pull/4283
+- Mixtral support https://github.com/ggml-org/llama.cpp/pull/4406
+- BERT embeddings https://github.com/ggml-org/llama.cpp/pull/5423
+- Grok-1 support https://github.com/ggml-org/llama.cpp/pull/6204
+- Command R Plus support https://github.com/ggml-org/llama.cpp/pull/6491
+- support arch DBRX https://github.com/ggml-org/llama.cpp/pull/6515
+- How to convert HuggingFace model to GGUF format https://github.com/ggml-org/llama.cpp/discussions/2948

docs/docker.md

@@ -7,21 +7,21 @@
 ## Images
 We have three Docker images available for this project:
 
-1. `ghcr.io/ggerganov/llama.cpp:full`: This image includes both the main executable file and the tools to convert LLaMA models into ggml and convert into 4-bit quantization. (platforms: `linux/amd64`, `linux/arm64`)
-2. `ghcr.io/ggerganov/llama.cpp:light`: This image only includes the main executable file. (platforms: `linux/amd64`, `linux/arm64`)
-3. `ghcr.io/ggerganov/llama.cpp:server`: This image only includes the server executable file. (platforms: `linux/amd64`, `linux/arm64`)
+1. `ghcr.io/ggml-org/llama.cpp:full`: This image includes both the main executable file and the tools to convert LLaMA models into ggml and convert into 4-bit quantization. (platforms: `linux/amd64`, `linux/arm64`)
+2. `ghcr.io/ggml-org/llama.cpp:light`: This image only includes the main executable file. (platforms: `linux/amd64`, `linux/arm64`)
+3. `ghcr.io/ggml-org/llama.cpp:server`: This image only includes the server executable file. (platforms: `linux/amd64`, `linux/arm64`)
 
 Additionally, there the following images, similar to the above:
 
-- `ghcr.io/ggerganov/llama.cpp:full-cuda`: Same as `full` but compiled with CUDA support. (platforms: `linux/amd64`)
-- `ghcr.io/ggerganov/llama.cpp:light-cuda`: Same as `light` but compiled with CUDA support. (platforms: `linux/amd64`)
-- `ghcr.io/ggerganov/llama.cpp:server-cuda`: Same as `server` but compiled with CUDA support. (platforms: `linux/amd64`)
-- `ghcr.io/ggerganov/llama.cpp:full-rocm`: Same as `full` but compiled with ROCm support. (platforms: `linux/amd64`, `linux/arm64`)
-- `ghcr.io/ggerganov/llama.cpp:light-rocm`: Same as `light` but compiled with ROCm support. (platforms: `linux/amd64`, `linux/arm64`)
-- `ghcr.io/ggerganov/llama.cpp:server-rocm`: Same as `server` but compiled with ROCm support. (platforms: `linux/amd64`, `linux/arm64`)
-- `ghcr.io/ggerganov/llama.cpp:full-musa`: Same as `full` but compiled with MUSA support. (platforms: `linux/amd64`)
-- `ghcr.io/ggerganov/llama.cpp:light-musa`: Same as `light` but compiled with MUSA support. (platforms: `linux/amd64`)
-- `ghcr.io/ggerganov/llama.cpp:server-musa`: Same as `server` but compiled with MUSA support. (platforms: `linux/amd64`)
+- `ghcr.io/ggml-org/llama.cpp:full-cuda`: Same as `full` but compiled with CUDA support. (platforms: `linux/amd64`)
+- `ghcr.io/ggml-org/llama.cpp:light-cuda`: Same as `light` but compiled with CUDA support. (platforms: `linux/amd64`)
+- `ghcr.io/ggml-org/llama.cpp:server-cuda`: Same as `server` but compiled with CUDA support. (platforms: `linux/amd64`)
+- `ghcr.io/ggml-org/llama.cpp:full-rocm`: Same as `full` but compiled with ROCm support. (platforms: `linux/amd64`, `linux/arm64`)
+- `ghcr.io/ggml-org/llama.cpp:light-rocm`: Same as `light` but compiled with ROCm support. (platforms: `linux/amd64`, `linux/arm64`)
+- `ghcr.io/ggml-org/llama.cpp:server-rocm`: Same as `server` but compiled with ROCm support. (platforms: `linux/amd64`, `linux/arm64`)
+- `ghcr.io/ggml-org/llama.cpp:full-musa`: Same as `full` but compiled with MUSA support. (platforms: `linux/amd64`)
+- `ghcr.io/ggml-org/llama.cpp:light-musa`: Same as `light` but compiled with MUSA support. (platforms: `linux/amd64`)
+- `ghcr.io/ggml-org/llama.cpp:server-musa`: Same as `server` but compiled with MUSA support. (platforms: `linux/amd64`)
 
 The GPU enabled images are not currently tested by CI beyond being built. They are not built with any variation from the ones in the Dockerfiles defined in [.devops/](../.devops/) and the GitHub Action defined in [.github/workflows/docker.yml](../.github/workflows/docker.yml). If you need different settings (for example, a different CUDA, ROCm or MUSA library, you'll need to build the images locally for now).
 
@@ -32,25 +32,25 @@ The easiest way to download the models, convert them to ggml and optimize them i
 Replace `/path/to/models` below with the actual path where you downloaded the models.
 
 ```bash
-docker run -v /path/to/models:/models ghcr.io/ggerganov/llama.cpp:full --all-in-one "/models/" 7B
+docker run -v /path/to/models:/models ghcr.io/ggml-org/llama.cpp:full --all-in-one "/models/" 7B
 ```
 
 On completion, you are ready to play!
 
 ```bash
-docker run -v /path/to/models:/models ghcr.io/ggerganov/llama.cpp:full --run -m /models/7B/ggml-model-q4_0.gguf -p "Building a website can be done in 10 simple steps:" -n 512
+docker run -v /path/to/models:/models ghcr.io/ggml-org/llama.cpp:full --run -m /models/7B/ggml-model-q4_0.gguf -p "Building a website can be done in 10 simple steps:" -n 512
 ```
 
 or with a light image:
 
 ```bash
-docker run -v /path/to/models:/models ghcr.io/ggerganov/llama.cpp:light -m /models/7B/ggml-model-q4_0.gguf -p "Building a website can be done in 10 simple steps:" -n 512
+docker run -v /path/to/models:/models ghcr.io/ggml-org/llama.cpp:light -m /models/7B/ggml-model-q4_0.gguf -p "Building a website can be done in 10 simple steps:" -n 512
 ```
 
 or with a server image:
 
 ```bash
-docker run -v /path/to/models:/models -p 8000:8000 ghcr.io/ggerganov/llama.cpp:server -m /models/7B/ggml-model-q4_0.gguf --port 8000 --host 0.0.0.0 -n 512
+docker run -v /path/to/models:/models -p 8000:8000 ghcr.io/ggml-org/llama.cpp:server -m /models/7B/ggml-model-q4_0.gguf --port 8000 --host 0.0.0.0 -n 512
 ```
 
 ## Docker With CUDA
@@ -60,16 +60,16 @@ Assuming one has the [nvidia-container-toolkit](https://github.com/NVIDIA/nvidia
 ## Building Docker locally
 
 ```bash
-docker build -t local/llama.cpp:full-cuda -f .devops/full-cuda.Dockerfile .
-docker build -t local/llama.cpp:light-cuda -f .devops/llama-cli-cuda.Dockerfile .
-docker build -t local/llama.cpp:server-cuda -f .devops/llama-server-cuda.Dockerfile .
+docker build -t local/llama.cpp:full-cuda --target full -f .devops/cuda.Dockerfile .
+docker build -t local/llama.cpp:light-cuda --target light -f .devops/cuda.Dockerfile .
+docker build -t local/llama.cpp:server-cuda --target server -f .devops/cuda.Dockerfile .
 ```
 
 You may want to pass in some different `ARGS`, depending on the CUDA environment supported by your container host, as well as the GPU architecture.
 
 The defaults are:
 
-- `CUDA_VERSION` set to `12.6.0`
+- `CUDA_VERSION` set to `12.4.0`
 - `CUDA_DOCKER_ARCH` set to the cmake build default, which includes all the supported architectures
 
 The resulting images, are essentially the same as the non-CUDA images:
@@ -95,16 +95,16 @@ Assuming one has the [mt-container-toolkit](https://developer.mthreads.com/musa/
 ## Building Docker locally
 
 ```bash
-docker build -t local/llama.cpp:full-musa -f .devops/full-musa.Dockerfile .
-docker build -t local/llama.cpp:light-musa -f .devops/llama-cli-musa.Dockerfile .
-docker build -t local/llama.cpp:server-musa -f .devops/llama-server-musa.Dockerfile .
+docker build -t local/llama.cpp:full-musa --target full -f .devops/musa.Dockerfile .
+docker build -t local/llama.cpp:light-musa --target light -f .devops/musa.Dockerfile .
+docker build -t local/llama.cpp:server-musa --target server -f .devops/musa.Dockerfile .
 ```
 
 You may want to pass in some different `ARGS`, depending on the MUSA environment supported by your container host, as well as the GPU architecture.
 
 The defaults are:
 
-- `MUSA_VERSION` set to `rc3.1.0`
+- `MUSA_VERSION` set to `rc3.1.1`
 
 The resulting images, are essentially the same as the non-MUSA images:
 

docs/install.md

@@ -7,7 +7,7 @@ On Mac and Linux, the homebrew package manager can be used via
 ```sh
 brew install llama.cpp
 ```
-The formula is automatically updated with new `llama.cpp` releases. More info: https://github.com/ggerganov/llama.cpp/discussions/7668
+The formula is automatically updated with new `llama.cpp` releases. More info: https://github.com/ggml-org/llama.cpp/discussions/7668
 
 ## Nix
 

docs/llguidance.md

@@ -0,0 +1,53 @@
+# LLGuidance Support in llama.cpp
+
+[LLGuidance](https://github.com/guidance-ai/llguidance) is a library for constrained decoding (also called constrained sampling or structured outputs) for Large Language Models (LLMs). Initially developed as the backend for the [Guidance](https://github.com/guidance-ai/guidance) library, it can also be used independently.
+
+LLGuidance supports JSON Schemas and arbitrary context-free grammars (CFGs) written in a [variant](https://github.com/guidance-ai/llguidance/blob/main/docs/syntax.md) of Lark syntax. It is [very fast](https://github.com/guidance-ai/jsonschemabench/tree/main/maskbench) and has [excellent](https://github.com/guidance-ai/llguidance/blob/main/docs/json_schema.md) JSON Schema coverage but requires the Rust compiler, which complicates the llama.cpp build process.
+
+## Building
+
+To enable LLGuidance support, build llama.cpp with the `LLAMA_LLGUIDANCE` option:
+
+```sh
+cmake -B build -DLLAMA_LLGUIDANCE=ON
+make -C build -j
+```
+
+For Windows use `cmake --build build --config Release` instead of `make`.
+
+This requires the Rust compiler and the `cargo` tool to be [installed](https://www.rust-lang.org/tools/install).
+
+## Interface
+
+There are no new command-line arguments or modifications to `common_params`. When enabled, grammars starting with `%llguidance` are passed to LLGuidance instead of the [current](../grammars/README.md) llama.cpp grammars. Additionally, JSON Schema requests (e.g., using the `-j` argument in `llama-cli`) are also passed to LLGuidance.
+
+For your existing GBNF grammars, you can use [gbnf_to_lark.py script](https://github.com/guidance-ai/llguidance/blob/main/python/llguidance/gbnf_to_lark.py) to convert them to LLGuidance Lark-like format.
+
+## Performance
+
+Computing a "token mask" (i.e., the set of allowed tokens) for a llama3 tokenizer with 128k tokens takes, on average, 50μs of single-core CPU time for the [JSON Schema Bench](https://github.com/guidance-ai/jsonschemabench). The p99 time is 0.5ms, and the p100 time is 20ms. These results are due to the lexer/parser split and several [optimizations](https://github.com/guidance-ai/llguidance/blob/main/docs/optimizations.md).
+
+## JSON Schema
+
+LLGuidance adheres closely to the JSON Schema specification. For example:
+
+- `additionalProperties` defaults to `true`, unlike current grammars, though you can set `"additionalProperties": false` if needed.
+- any whitespace is allowed.
+- The definition order in the `"properties": {}` object is maintained, regardless of whether properties are required (current grammars always puts required properties first).
+
+Unsupported schemas result in an error message—no keywords are silently ignored.
+
+## Why Not Reuse GBNF Format?
+
+GBNF lacks the concept of a lexer.
+
+Most programming languages, including JSON, use a two-step process: a lexer (built with regular expressions) converts a byte stream into lexemes, which are then processed by a CFG parser. This approach is faster because lexers are cheaper to evaluate, and there is ~10x fewer lexemes than bytes.
+LLM tokens often align with lexemes, so the parser is engaged in under 0.5% of tokens, with the lexer handling the rest.
+
+However, the user has to provide the distinction between lexemes and CFG symbols. In [Lark](https://github.com/lark-parser/lark), lexeme names are uppercase, while CFG symbols are lowercase.
+The [gbnf_to_lark.py script](https://github.com/guidance-ai/llguidance/blob/main/scripts/gbnf_to_lark.py) can often take care of this automatically.
+See [LLGuidance syntax docs](https://github.com/guidance-ai/llguidance/blob/main/docs/syntax.md#terminals-vs-rules) for more details.
+
+## Error Handling
+
+Errors are currently printed to `stderr`, and generation continues. Improved error handling may be added in the future.

examples/batched.swift/Sources/main.swift

@@ -31,6 +31,11 @@ defer {
     llama_model_free(model)
 }
 
+guard let vocab = llama_model_get_vocab(model) else {
+    print("Failed to get vocab")
+    exit(1)
+}
+
 var tokens = tokenize(text: prompt, add_bos: true)
 
 let n_kv_req = UInt32(tokens.count) + UInt32((n_len - Int(tokens.count)) * n_parallel)
@@ -41,7 +46,7 @@ context_params.n_batch = UInt32(max(n_len, n_parallel))
 context_params.n_threads = 8
 context_params.n_threads_batch = 8
 
-let context = llama_new_context_with_model(model, context_params)
+let context = llama_init_from_model(model, context_params)
 guard context != nil else {
     print("Failed to initialize context")
     exit(1)
@@ -141,7 +146,7 @@ while n_cur <= n_len {
         let new_token_id = llama_sampler_sample(smpl, context, i_batch[i])
 
         // is it an end of stream? -> mark the stream as finished
-        if llama_vocab_is_eog(model, new_token_id) || n_cur == n_len {
+        if llama_vocab_is_eog(vocab, new_token_id) || n_cur == n_len {
             i_batch[i] = -1
             // print("")
             if n_parallel > 1 {
@@ -207,7 +212,7 @@ private func tokenize(text: String, add_bos: Bool) -> [llama_token] {
     let utf8Count = text.utf8.count
     let n_tokens = utf8Count + (add_bos ? 1 : 0)
     let tokens = UnsafeMutablePointer<llama_token>.allocate(capacity: n_tokens)
-    let tokenCount = llama_tokenize(model, text, Int32(utf8Count), tokens, Int32(n_tokens), add_bos, /*special tokens*/ false)
+    let tokenCount = llama_tokenize(vocab, text, Int32(utf8Count), tokens, Int32(n_tokens), add_bos, /*special tokens*/ false)
     var swiftTokens: [llama_token] = []
     for i in 0 ..< tokenCount {
         swiftTokens.append(tokens[Int(i)])
@@ -218,12 +223,12 @@ private func tokenize(text: String, add_bos: Bool) -> [llama_token] {
 
 private func token_to_piece(token: llama_token, buffer: inout [CChar]) -> String? {
     var result = [CChar](repeating: 0, count: 8)
-    let nTokens = llama_token_to_piece(model, token, &result, Int32(result.count), 0, false)
+    let nTokens = llama_token_to_piece(vocab, token, &result, Int32(result.count), 0, false)
     if nTokens < 0 {
         let actualTokensCount = -Int(nTokens)
         result = .init(repeating: 0, count: actualTokensCount)
         let check = llama_token_to_piece(
-            model,
+            vocab,
             token,
             &result,
             Int32(result.count),

examples/cvector-generator/README.md

@@ -3,9 +3,9 @@
 This example demonstrates how to generate a control vector using gguf models.
 
 Related PRs:
-- [Add support for control vectors](https://github.com/ggerganov/llama.cpp/pull/5970)
-- (Issue) [Generate control vector using llama.cpp](https://github.com/ggerganov/llama.cpp/issues/6880)
-- [Add cvector-generator example](https://github.com/ggerganov/llama.cpp/pull/7514)
+- [Add support for control vectors](https://github.com/ggml-org/llama.cpp/pull/5970)
+- (Issue) [Generate control vector using llama.cpp](https://github.com/ggml-org/llama.cpp/issues/6880)
+- [Add cvector-generator example](https://github.com/ggml-org/llama.cpp/pull/7514)
 
 ## Examples
 

examples/export-lora/export-lora.cpp

@@ -345,8 +345,18 @@ struct lora_merge_ctx {
             gf = ggml_new_graph(ctx0);
             struct ggml_tensor * cur = inp_base;
             for (size_t i = 0; i < adapters.size(); ++i) {
-                struct ggml_tensor * a_T = ggml_cont(ctx0, ggml_transpose(ctx0, ggml_cast(ctx0, inp_a[i], GGML_TYPE_F32)));
-                struct ggml_tensor * delta = ggml_mul_mat(ctx0, a_T, ggml_cast(ctx0, inp_b[i], GGML_TYPE_F32));
+                struct ggml_tensor * delta;
+                bool is_tok_embd = string_starts_with(name_base, "token_embd");
+                if (is_tok_embd) {
+                    printf("%s :     detected token embeddings tensor\n", __func__);
+                    delta = ggml_mul_mat(ctx0,
+                        ggml_cast(ctx0, inp_b[i], GGML_TYPE_F32),
+                        ggml_cast(ctx0, inp_a[i], GGML_TYPE_F32));
+                } else {
+                    delta = ggml_mul_mat(ctx0,
+                        ggml_cont(ctx0, ggml_transpose(ctx0, ggml_cast(ctx0, inp_a[i], GGML_TYPE_F32))),
+                        ggml_cast(ctx0, inp_b[i], GGML_TYPE_F32));
+                }
                 // scale
                 const float alpha = adapters[i]->alpha;
                 const float rank  = (float) inp_b[i]->ne[0];

examples/gbnf-validator/gbnf-validator.cpp

@@ -76,7 +76,7 @@ int main(int argc, char** argv) {
         grammar_str = buffer.str();
     }
 
-    llama_grammar * grammar = llama_grammar_init_impl(nullptr, grammar_str.c_str(), "root");
+    llama_grammar * grammar = llama_grammar_init_impl(nullptr, grammar_str.c_str(), "root", false, nullptr, 0, nullptr, 0);
     if (grammar == nullptr) {
         fprintf(stdout, "Failed to initialize llama_grammar\n");
         return 1;

examples/gguf-split/tests.sh

@@ -41,7 +41,7 @@ echo PASS
 echo
 
 # 2b. Test the sharded model is loading properly
-$MAIN --model $WORK_PATH/ggml-model-split-00001-of-00006.gguf --n-predict 32
+$MAIN -no-cnv --model $WORK_PATH/ggml-model-split-00001-of-00006.gguf --n-predict 32
 echo PASS
 echo
 
@@ -51,7 +51,7 @@ echo PASS
 echo
 
 # 3b. Test the merged model is loading properly
-$MAIN --model $WORK_PATH/ggml-model-merge.gguf --n-predict 32
+$MAIN -no-cnv --model $WORK_PATH/ggml-model-merge.gguf --n-predict 32
 echo PASS
 echo
 
@@ -61,7 +61,7 @@ echo PASS
 echo
 
 # 4b. Test the sharded model is loading properly
-$MAIN --model $WORK_PATH/ggml-model-split-32-tensors-00001-of-00007.gguf --n-predict 32
+$MAIN -no-cnv --model $WORK_PATH/ggml-model-split-32-tensors-00001-of-00007.gguf --n-predict 32
 echo PASS
 echo
 
@@ -71,7 +71,7 @@ echo
 #echo
 
 # 5b. Test the merged model is loading properly
-#$MAIN --model $WORK_PATH/ggml-model-merge-2.gguf --n-predict 32
+#$MAIN -no-cnv --model $WORK_PATH/ggml-model-merge-2.gguf --n-predict 32
 #echo PASS
 #echo
 
@@ -81,7 +81,7 @@ echo PASS
 echo
 
 # 6b. Test the sharded model is loading properly
-$MAIN --model $WORK_PATH/ggml-model-split-2G-00001-of-00002.gguf --n-predict 32
+$MAIN -no-cnv --model $WORK_PATH/ggml-model-split-2G-00001-of-00002.gguf --n-predict 32
 echo PASS
 echo
 

examples/imatrix/README.md

@@ -1,7 +1,7 @@
 # llama.cpp/examples/imatrix
 
-Compute an importance matrix for a model and given text dataset. Can be used during quantization to enchance the quality of the quantized models.
-More information is available here: https://github.com/ggerganov/llama.cpp/pull/4861
+Compute an importance matrix for a model and given text dataset. Can be used during quantization to enhance the quality of the quantized models.
+More information is available here: https://github.com/ggml-org/llama.cpp/pull/4861
 
 ## Usage
 

examples/imatrix/imatrix.cpp

@@ -3,6 +3,7 @@
 #include "log.h"
 #include "llama.h"
 
+#include <chrono>
 #include <cmath>
 #include <cstdio>
 #include <cstring>
@@ -99,7 +100,7 @@ bool IMatrixCollector::collect_imatrix(struct ggml_tensor * t, bool ask, void *
     const float * data = is_host ? (const float *) src1->data : m_src1_data.data();
 
     // this has been adapted to the new format of storing merged experts in a single 3d tensor
-    // ref: https://github.com/ggerganov/llama.cpp/pull/6387
+    // ref: https://github.com/ggml-org/llama.cpp/pull/6387
     if (t->op == GGML_OP_MUL_MAT_ID) {
         //   ids  -> [n_experts_used, n_tokens]
         //   src1 -> [cols, n_expert_used, n_tokens]

examples/llama-bench/llama-bench.cpp

@@ -683,7 +683,7 @@ struct cmd_params_instance {
     bool               cpu_strict;
     int                poll;
     int                n_gpu_layers;
-    std::string        rpc_servers;
+    std::string        rpc_servers_str;
     llama_split_mode   split_mode;
     int                main_gpu;
     bool               no_kv_offload;
@@ -696,8 +696,37 @@ struct cmd_params_instance {
         llama_model_params mparams = llama_model_default_params();
 
         mparams.n_gpu_layers = n_gpu_layers;
-        if (!rpc_servers.empty()) {
-            mparams.rpc_servers = rpc_servers.c_str();
+        if (!rpc_servers_str.empty()) {
+            auto rpc_servers = string_split<std::string>(rpc_servers_str, ',');
+
+            // add RPC devices
+            if (!rpc_servers.empty()) {
+                ggml_backend_reg_t rpc_reg = ggml_backend_reg_by_name("RPC");
+                if (!rpc_reg) {
+                    fprintf(stderr, "%s: failed to find RPC backend\n", __func__);
+                    exit(1);
+                }
+
+                typedef ggml_backend_dev_t (*ggml_backend_rpc_add_device_t)(const char * endpoint);
+                ggml_backend_rpc_add_device_t ggml_backend_rpc_add_device_fn = (ggml_backend_rpc_add_device_t) ggml_backend_reg_get_proc_address(rpc_reg, "ggml_backend_rpc_add_device");
+                if (!ggml_backend_rpc_add_device_fn) {
+                    fprintf(stderr, "%s: failed to find RPC device add function\n", __func__);
+                    exit(1);
+                }
+                static std::vector<ggml_backend_dev_t> devices;
+                devices.clear();
+                for (const std::string & server : rpc_servers) {
+                    ggml_backend_dev_t dev = ggml_backend_rpc_add_device_fn(server.c_str());
+                    if (dev) {
+                        devices.push_back(dev);
+                    } else {
+                        fprintf(stderr, "%s: failed to add RPC device for server '%s'\n", __func__, server.c_str());
+                        exit(1);
+                    }
+                }
+                devices.push_back(nullptr);
+                mparams.devices = devices.data();
+            }
         }
         mparams.split_mode   = split_mode;
         mparams.main_gpu     = main_gpu;
@@ -708,7 +737,7 @@ struct cmd_params_instance {
     }
 
     bool equal_mparams(const cmd_params_instance & other) const {
-        return model == other.model && n_gpu_layers == other.n_gpu_layers && rpc_servers == other.rpc_servers &&
+        return model == other.model && n_gpu_layers == other.n_gpu_layers && rpc_servers_str == other.rpc_servers_str &&
                split_mode == other.split_mode && main_gpu == other.main_gpu && use_mmap == other.use_mmap &&
                tensor_split == other.tensor_split;
     }
@@ -847,8 +876,8 @@ static std::vector<cmd_params_instance> get_cmd_params_instances(const cmd_param
 struct test {
     static const std::string build_commit;
     static const int         build_number;
-    static const std::string cpu_info;
-    static const std::string gpu_info;
+    const std::string        cpu_info;
+    const std::string        gpu_info;
     std::string              model_filename;
     std::string              model_type;
     uint64_t                 model_size;
@@ -874,7 +903,10 @@ struct test {
     std::string              test_time;
     std::vector<uint64_t>    samples_ns;
 
-    test(const cmd_params_instance & inst, const llama_model * lmodel, const llama_context * ctx) {
+    test(const cmd_params_instance & inst, const llama_model * lmodel, const llama_context * ctx) :
+        cpu_info(get_cpu_info()),
+        gpu_info(get_gpu_info()) {
+
         model_filename = inst.model;
         char buf[128];
         llama_model_desc(lmodel, buf, sizeof(buf));
@@ -1029,8 +1061,6 @@ struct test {
 
 const std::string test::build_commit = LLAMA_COMMIT;
 const int         test::build_number = LLAMA_BUILD_NUMBER;
-const std::string test::cpu_info     = get_cpu_info();
-const std::string test::gpu_info     = get_gpu_info();
 
 struct printer {
     virtual ~printer() {}

examples/llama.android/llama/src/main/cpp/CMakeLists.txt

@@ -14,7 +14,7 @@ project("llama-android")
 #include(FetchContent)
 #FetchContent_Declare(
 #        llama
-#        GIT_REPOSITORY https://github.com/ggerganov/llama.cpp
+#        GIT_REPOSITORY https://github.com/ggml-org/llama.cpp
 #        GIT_TAG        master
 #)
 

examples/llama.android/llama/src/main/cpp/llama-android.cpp

@@ -347,6 +347,7 @@ Java_android_llama_cpp_LLamaAndroid_completion_1init(
         jlong context_pointer,
         jlong batch_pointer,
         jstring jtext,
+        jboolean format_chat,
         jint n_len
     ) {
 
@@ -356,7 +357,8 @@ Java_android_llama_cpp_LLamaAndroid_completion_1init(
     const auto context = reinterpret_cast<llama_context *>(context_pointer);
     const auto batch = reinterpret_cast<llama_batch *>(batch_pointer);
 
-    const auto tokens_list = common_tokenize(context, text, 1);
+    bool parse_special = (format_chat == JNI_TRUE);
+    const auto tokens_list = common_tokenize(context, text, true, parse_special);
 
     auto n_ctx = llama_n_ctx(context);
     auto n_kv_req = tokens_list.size() + (n_len - tokens_list.size());
@@ -368,7 +370,7 @@ Java_android_llama_cpp_LLamaAndroid_completion_1init(
     }
 
     for (auto id : tokens_list) {
-        LOGi("%s", common_token_to_piece(context, id).c_str());
+        LOGi("token: `%s`-> %d ", common_token_to_piece(context, id).c_str(), id);
     }
 
     common_batch_clear(*batch);

examples/llama.android/llama/src/main/java/android/llama/cpp/LLamaAndroid.kt

@@ -65,6 +65,7 @@ class LLamaAndroid {
         context: Long,
         batch: Long,
         text: String,
+        formatChat: Boolean,
         nLen: Int
     ): Int
 
@@ -115,10 +116,10 @@ class LLamaAndroid {
         }
     }
 
-    fun send(message: String): Flow<String> = flow {
+    fun send(message: String, formatChat: Boolean = false): Flow<String> = flow {
         when (val state = threadLocalState.get()) {
             is State.Loaded -> {
-                val ncur = IntVar(completion_init(state.context, state.batch, message, nlen))
+                val ncur = IntVar(completion_init(state.context, state.batch, message, formatChat, nlen))
                 while (ncur.value <= nlen) {
                     val str = completion_loop(state.context, state.batch, state.sampler, nlen, ncur)
                     if (str == null) {

examples/llama.swiftui/README.md

@@ -3,9 +3,9 @@
 Local inference of llama.cpp on an iPhone. This is a sample app that can be used as a starting
 point for more advanced projects.
 
-For usage instructions and performance stats, check the following discussion: https://github.com/ggerganov/llama.cpp/discussions/4508
+For usage instructions and performance stats, check the following discussion: https://github.com/ggml-org/llama.cpp/discussions/4508
 
-![image](https://github.com/ggerganov/llama.cpp/assets/1991296/2b40284f-8421-47a2-b634-74eece09a299)
+![image](https://github.com/ggml-org/llama.cpp/assets/1991296/2b40284f-8421-47a2-b634-74eece09a299)
 
 Video demonstration:
 

examples/llama.swiftui/llama.cpp.swift/LibLlama.swift

@@ -24,6 +24,7 @@ func llama_batch_add(_ batch: inout llama_batch, _ id: llama_token, _ pos: llama
 actor LlamaContext {
     private var model: OpaquePointer
     private var context: OpaquePointer
+    private var vocab: OpaquePointer
     private var sampling: UnsafeMutablePointer<llama_sampler>
     private var batch: llama_batch
     private var tokens_list: [llama_token]
@@ -47,6 +48,7 @@ actor LlamaContext {
         self.sampling = llama_sampler_chain_init(sparams)
         llama_sampler_chain_add(self.sampling, llama_sampler_init_temp(0.4))
         llama_sampler_chain_add(self.sampling, llama_sampler_init_dist(1234))
+        vocab = llama_model_get_vocab(model)
     }
 
     deinit {
@@ -79,7 +81,7 @@ actor LlamaContext {
         ctx_params.n_threads       = Int32(n_threads)
         ctx_params.n_threads_batch = Int32(n_threads)
 
-        let context = llama_new_context_with_model(model, ctx_params)
+        let context = llama_init_from_model(model, ctx_params)
         guard let context else {
             print("Could not load context!")
             throw LlamaError.couldNotInitializeContext
@@ -151,7 +153,7 @@ actor LlamaContext {
 
         new_token_id = llama_sampler_sample(sampling, context, batch.n_tokens - 1)
 
-        if llama_vocab_is_eog(model, new_token_id) || n_cur == n_len {
+        if llama_vocab_is_eog(vocab, new_token_id) || n_cur == n_len {
             print("\n")
             is_done = true
             let new_token_str = String(cString: temporary_invalid_cchars + [0])
@@ -297,7 +299,7 @@ actor LlamaContext {
         let utf8Count = text.utf8.count
         let n_tokens = utf8Count + (add_bos ? 1 : 0) + 1
         let tokens = UnsafeMutablePointer<llama_token>.allocate(capacity: n_tokens)
-        let tokenCount = llama_tokenize(model, text, Int32(utf8Count), tokens, Int32(n_tokens), add_bos, false)
+        let tokenCount = llama_tokenize(vocab, text, Int32(utf8Count), tokens, Int32(n_tokens), add_bos, false)
 
         var swiftTokens: [llama_token] = []
         for i in 0..<tokenCount {
@@ -316,7 +318,7 @@ actor LlamaContext {
         defer {
             result.deallocate()
         }
-        let nTokens = llama_token_to_piece(model, token, result, 8, 0, false)
+        let nTokens = llama_token_to_piece(vocab, token, result, 8, 0, false)
 
         if nTokens < 0 {
             let newResult = UnsafeMutablePointer<Int8>.allocate(capacity: Int(-nTokens))
@@ -324,7 +326,7 @@ actor LlamaContext {
             defer {
                 newResult.deallocate()
             }
-            let nNewTokens = llama_token_to_piece(model, token, newResult, -nTokens, 0, false)
+            let nNewTokens = llama_token_to_piece(vocab, token, newResult, -nTokens, 0, false)
             let bufferPointer = UnsafeBufferPointer(start: newResult, count: Int(nNewTokens))
             return Array(bufferPointer)
         } else {

examples/llama.vim

@@ -39,7 +39,7 @@
 "
 "   :call llama#init()
 "
-" more info: https://github.com/ggerganov/llama.cpp/pull/9787
+" more info: https://github.com/ggml-org/llama.cpp/pull/9787
 "
 
 " colors (adjust to your liking)

examples/llava/CMakeLists.txt

@@ -50,3 +50,10 @@ set_target_properties(${TARGET} PROPERTIES OUTPUT_NAME llama-qwen2vl-cli)
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE common llava ${CMAKE_THREAD_LIBS_INIT})
 target_compile_features(${TARGET} PRIVATE cxx_std_17)
+
+set(TARGET llama-llava-clip-quantize-cli)
+add_executable(${TARGET} clip-quantize-cli.cpp)
+set_target_properties(${TARGET} PROPERTIES OUTPUT_NAME llama-llava-clip-quantize-cli)
+install(TARGETS ${TARGET} RUNTIME)
+target_link_libraries(${TARGET} PRIVATE common llava ${CMAKE_THREAD_LIBS_INIT})
+target_compile_features(${TARGET} PRIVATE cxx_std_17)

examples/llava/README-glmedge.md

@@ -0,0 +1,43 @@
+# GLMV-EDGE
+
+Currently this implementation supports [glm-edge-v-2b](https://huggingface.co/THUDM/glm-edge-v-2b) and [glm-edge-v-5b](https://huggingface.co/THUDM/glm-edge-v-5b).
+
+## Usage
+Build with cmake or run `make llama-llava-cli` to build it.
+
+After building, run: `./llama-llava-cli` to see the usage. For example:
+
+```sh
+./llama-llava-cli -m model_path/ggml-model-f16.gguf --mmproj model_path/mmproj-model-f16.gguf --image img_path/image.jpg -p "<|system|>\n system prompt <image><|user|>\n prompt <|assistant|>\n"
+```
+
+**note**: A lower temperature like 0.1 is recommended for better quality. add `--temp 0.1` to the command to do so.
+**note**: For GPU offloading ensure to use the `-ngl` flag just like usual
+
+## GGUF conversion
+
+1. Clone a GLMV-EDGE model ([2B](https://huggingface.co/THUDM/glm-edge-v-2b) or [5B](https://huggingface.co/THUDM/glm-edge-v-5b)). For example:
+
+```sh
+git clone https://huggingface.co/THUDM/glm-edge-v-5b or https://huggingface.co/THUDM/glm-edge-v-2b
+```
+
+2. Use `glmedge-surgery.py` to split the GLMV-EDGE model to LLM and multimodel projector constituents:
+
+```sh
+python ./examples/llava/glmedge-surgery.py -m ../model_path
+```
+
+4. Use `glmedge-convert-image-encoder-to-gguf.py` to convert the GLMV-EDGE image encoder to GGUF:
+
+```sh
+python ./examples/llava/glmedge-convert-image-encoder-to-gguf.py -m ../model_path --llava-projector ../model_path/glm.projector --output-dir ../model_path
+```
+
+5. Use `examples/convert_hf_to_gguf.py` to convert the LLM part of GLMV-EDGE to GGUF:
+
+```sh
+python convert_hf_to_gguf.py ../model_path
+```
+
+Now both the LLM part and the image encoder are in the `model_path` directory.

examples/llava/README-minicpmo2.6.md

@@ -0,0 +1,46 @@
+## MiniCPM-o 2.6
+Currently, this readme only supports minicpm-omni's image capabilities, and we will update the full-mode support as soon as possible.
+
+### Prepare models and code
+
+Download [MiniCPM-o-2_6](https://huggingface.co/openbmb/MiniCPM-o-2_6) PyTorch model from huggingface to "MiniCPM-o-2_6" folder.
+
+Clone llama.cpp:
+```bash
+git clone git@github.com:OpenBMB/llama.cpp.git
+cd llama.cpp
+git checkout minicpm-omni
+```
+
+### Usage of MiniCPM-o 2.6
+
+Convert PyTorch model to gguf files (You can also download the converted [gguf](https://huggingface.co/openbmb/MiniCPM-o-2_6-gguf) by us)
+
+```bash
+python ./examples/llava/minicpmv-surgery.py -m ../MiniCPM-o-2_6
+python ./examples/llava/minicpmv-convert-image-encoder-to-gguf.py -m ../MiniCPM-o-2_6 --minicpmv-projector ../MiniCPM-o-2_6/minicpmv.projector --output-dir ../MiniCPM-o-2_6/ --image-mean 0.5 0.5 0.5 --image-std 0.5 0.5 0.5 --minicpmv_version 4
+python ./convert_hf_to_gguf.py ../MiniCPM-o-2_6/model
+
+# quantize int4 version
+./llama-quantize ../MiniCPM-o-2_6/model/ggml-model-f16.gguf ../MiniCPM-o-2_6/model/ggml-model-Q4_K_M.gguf Q4_K_M
+```
+
+Build llama.cpp using `CMake`:
+https://github.com/ggml-org/llama.cpp/blob/master/docs/build.md
+
+```bash
+cmake -B build
+cmake --build build --config Release
+```
+
+Inference on Linux or Mac
+```
+# run f16 version
+./llama-minicpmv-cli -m ../MiniCPM-o-2_6/model/ggml-model-f16.gguf --mmproj ../MiniCPM-o-2_6/mmproj-model-f16.gguf -c 4096 --temp 0.7 --top-p 0.8 --top-k 100 --repeat-penalty 1.05 --image xx.jpg -p "What is in the image?"
+
+# run quantized int4 version
+./llama-minicpmv-cli -m ../MiniCPM-o-2_6/model/ggml-model-Q4_K_M.gguf --mmproj ../MiniCPM-o-2_6/mmproj-model-f16.gguf -c 4096 --temp 0.7 --top-p 0.8 --top-k 100 --repeat-penalty 1.05 --image xx.jpg  -p "What is in the image?"
+
+# or run in interactive mode
+./llama-minicpmv-cli -m ../MiniCPM-o-2_6/model/ggml-model-Q4_K_M.gguf --mmproj ../MiniCPM-o-2_6/mmproj-model-f16.gguf -c 4096 --temp 0.7 --top-p 0.8 --top-k 100 --repeat-penalty 1.05 --image xx.jpg -i
+```

examples/llava/README-minicpmv2.5.md

@@ -6,7 +6,7 @@ Download [MiniCPM-Llama3-V-2_5](https://huggingface.co/openbmb/MiniCPM-Llama3-V-
 
 Clone llama.cpp:
 ```bash
-git clone https://github.com/ggerganov/llama.cpp
+git clone https://github.com/ggml-org/llama.cpp
 cd llama.cpp
 ```
 

examples/llava/README-quantize.md

@@ -0,0 +1,44 @@
+# Quantizing CLIP Visual Projector
+
+This is the tool for quantizing the CLIP visual projector model. Quantization reduces the precision of the model's weights, which can significantly decrease the model size and improve inference speed, often with minimal impact on performance.
+
+## Usage
+
+To quantize a CLIP visual projector model, use the following command:
+
+```sh
+./bin/llama-llava-clip-quantize-cli /path/to/ggml-model-f32.gguf /path/to/ggml-model-quantized.gguf <type>
+```
+
+After the quantization, the visual projector can be used freely with the existing LLAVA cli (LLAVA, Qwen2VL, etc).
+
+### Arguments
+
+- `/path/to/ggml-model-f32.gguf`: The path to the input model file in FP32 or FP16 format.
+- `/path/to/ggml-model-quantized.gguf`: The path where the quantized model will be saved.
+- `<type>`: The quantization type to apply. This should be an integer corresponding to one of the quantization types defined in the `enum ggml_type`.
+
+### Quantization Types
+
+The following quantization types are supported, based on the `enum ggml_type` definition:
+
+- `2` - `q4_0`: 4-bit quantization with a single scale value.
+- `3` - `q4_1`: 4-bit quantization with a separate scale value for each block.
+- `6` - `q5_0`: 5-bit quantization with a single scale value.
+- `7` - `q5_1`: 5-bit quantization with a separate scale value for each block.
+- `8` - `q8_0`: 8-bit quantization with a single scale value.
+
+### Example
+
+To quantize a model using the `q4_0` quantization type, you would run:
+
+```sh
+./bin/llama-llava-clip-quantize-cli /path/to/ggml-model-f32.gguf /path/to/ggml-model-quantized.gguf 2
+```
+
+This command will generate a quantized model at `/path/to/ggml-model-quantized.gguf` using the `q4_0` quantization method.
+
+## Notes
+
+- Quantization can lead to a loss in model accuracy, depending on the chosen quantization type. It is recommended to evaluate the quantized model's performance on your specific task to ensure it meets your requirements.
+- The quantized model will typically be smaller in size and faster to run, making it more suitable for deployment in resource-constrained environments.

examples/llava/clip-quantize-cli.cpp

@@ -0,0 +1,59 @@
+#include "arg.h"
+#include "base64.hpp"
+#include "log.h"
+#include "common.h"
+#include "sampling.h"
+#include "clip.h"
+#include "llava.h"
+#include "llama.h"
+#include "ggml.h"
+
+static void print_usage(int argc, char ** argv) {
+    (void) argc;
+
+    fprintf(stderr, "usage: %s /path/to/ggml-model-f32.gguf /path/to/ggml-model-quantized.gguf type\n", argv[0]);
+    fprintf(stderr, "  type = 2 - q4_0\n");
+    fprintf(stderr, "  type = 3 - q4_1\n");
+    fprintf(stderr, "  type = 6 - q5_0\n");
+    fprintf(stderr, "  type = 7 - q5_1\n");
+    fprintf(stderr, "  type = 8 - q8_0\n");
+}
+
+int main(int argc, char ** argv) {
+    if (argc != 4) {
+        print_usage(argc, argv);
+        return 1;
+    }
+
+    const std::string fname_inp = argv[1];
+    const std::string fname_out = argv[2];
+
+    const int itype = atoi(argv[3]);
+
+    const int64_t t_main_start_us = ggml_time_us();
+
+    int64_t t_quantize_us = 0;
+
+    // load the model
+    {
+        const int64_t t_start_us = ggml_time_us();
+
+        if (!clip_model_quantize(fname_inp.c_str(), fname_out.c_str(), itype)) {
+            fprintf(stderr, "%s: failed to quantize model from '%s'\n", __func__, fname_inp.c_str());
+            return 1;
+        }
+
+        t_quantize_us = ggml_time_us() - t_start_us;
+    }
+
+    // report timing
+    {
+        const int64_t t_main_end_us = ggml_time_us();
+
+        printf("\n");
+        printf("%s: quantize time = %8.2f ms\n", __func__, t_quantize_us / 1000.0f);
+        printf("%s:    total time = %8.2f ms\n", __func__, (t_main_end_us - t_main_start_us) / 1000.0f);
+    }
+
+    return 0;
+}

examples/llava/clip.cpp

@@ -102,6 +102,7 @@ static std::string format(const char * fmt, ...) {
 #define KEY_HAS_VIS_ENC         "clip.has_vision_encoder"
 #define KEY_HAS_LLAVA_PROJ      "clip.has_llava_projector"
 #define KEY_HAS_MINICPMV_PROJ   "clip.has_minicpmv_projector"
+#define KEY_HAS_GLM_PROJ        "clip.has_glm_projector"
 #define KEY_MINICPMV_VERSION    "clip.minicpmv_version"
 #define KEY_HAS_QWEN2VL_MERGER  "clip.has_qwen2vl_merger"
 #define KEY_USE_GELU            "clip.use_gelu"
@@ -160,6 +161,15 @@ static std::string format(const char * fmt, ...) {
 #define TN_MINICPMV_ATTN "resampler.attn.%s.%s"
 #define TN_MINICPMV_LN "resampler.ln_%s.%s"
 
+#define TN_GLM_ADAPER_CONV "adapter.conv.%s"
+#define TN_GLM_ADAPTER_LINEAR "adapter.linear.linear.%s"
+#define TN_GLM_ADAPTER_NORM_1 "adapter.linear.norm1.%s"
+#define TN_GLM_ADAPTER_D_H_2_4H "adapter.linear.dense_h_to_4h.%s"
+#define TN_GLM_ADAPTER_GATE "adapter.linear.gate.%s"
+#define TN_GLM_ADAPTER_D_4H_2_H "adapter.linear.dense_4h_to_h.%s"
+#define TN_GLM_BOI_W "adapter.boi"
+#define TN_GLM_EOI_W "adapter.eoi"
+
 
 enum projector_type {
     PROJECTOR_TYPE_MLP,
@@ -167,6 +177,7 @@ enum projector_type {
     PROJECTOR_TYPE_LDP,
     PROJECTOR_TYPE_LDPV2,
     PROJECTOR_TYPE_RESAMPLER,
+    PROJECTOR_TYPE_GLM_EDGE,
     PROJECTOR_TYPE_MERGER,
     PROJECTOR_TYPE_UNKNOWN,
 };
@@ -176,6 +187,7 @@ static std::map<projector_type, std::string> PROJECTOR_TYPE_NAMES = {
     { PROJECTOR_TYPE_LDP, "ldp" },
     { PROJECTOR_TYPE_LDPV2, "ldpv2"},
     { PROJECTOR_TYPE_RESAMPLER, "resampler"},
+    { PROJECTOR_TYPE_GLM_EDGE, "adapter"},
     { PROJECTOR_TYPE_MERGER, "qwen2vl_merger"},
 };
 
@@ -500,6 +512,12 @@ struct clip_vision_model {
     struct ggml_tensor * mm_4_w = NULL;
     struct ggml_tensor * mm_4_b = NULL;
 
+    //GLMV-Edge projection
+    struct ggml_tensor * mm_model_adapter_conv_w;
+    struct ggml_tensor * mm_model_adapter_conv_b;
+    struct ggml_tensor * boi_w;
+    struct ggml_tensor * eoi_w;
+
     // MobileVLM projection
     struct ggml_tensor * mm_model_mlp_1_w;
     struct ggml_tensor * mm_model_mlp_1_b;
@@ -560,6 +578,7 @@ struct clip_ctx {
     bool has_vision_encoder  = false;
     bool has_llava_projector = false;
     bool has_minicpmv_projector = false;
+    bool has_glm_projector = false;
     bool has_qwen2vl_merger = false;
     int minicpmv_version = 2;
 
@@ -638,7 +657,7 @@ static ggml_cgraph * clip_image_build_graph(clip_ctx * ctx, const clip_image_f32
 
     const int batch_size = imgs->size;
 
-    if (ctx->has_llava_projector || ctx->has_minicpmv_projector) {
+    if (ctx->has_llava_projector || ctx->has_minicpmv_projector || ctx->has_glm_projector) {
         GGML_ASSERT(batch_size == 1);
     }
 
@@ -718,6 +737,9 @@ static ggml_cgraph * clip_image_build_graph(clip_ctx * ctx, const clip_image_f32
         else if (ctx->minicpmv_version == 3) {
             pos_embed = ggml_new_tensor_3d(ctx0, GGML_TYPE_F32, 3584, pos_w * pos_h, 1);
         }
+        else if (ctx->minicpmv_version == 4) {
+            pos_embed = ggml_new_tensor_3d(ctx0, GGML_TYPE_F32, 3584, pos_w * pos_h, 1);
+        }
         ggml_set_name(pos_embed, "pos_embed");
         ggml_set_input(pos_embed);
     }
@@ -731,8 +753,7 @@ static ggml_cgraph * clip_image_build_graph(clip_ctx * ctx, const clip_image_f32
     }
 
     // loop over layers
-    if (ctx->has_minicpmv_projector || ctx->has_qwen2vl_merger) {
-        // TODO: figure out why we doing thing in this way ???
+    if (ctx->has_minicpmv_projector || ctx->has_glm_projector || ctx->has_qwen2vl_merger) {
         n_layer += 1;
     }
     for (int il = 0; il < n_layer - 1; il++) {
@@ -1053,6 +1074,11 @@ static ggml_cgraph * clip_image_build_graph(clip_ctx * ctx, const clip_image_f32
                     n_head = hidden_size/d_head;
                     num_query = 64;
                 }
+                else if (ctx->minicpmv_version == 4) {
+                    hidden_size = 3584;
+                    n_head = hidden_size/d_head;
+                    num_query = 64;
+                }
 
                 struct ggml_tensor * Q = ggml_add(ctx0, ggml_mul_mat(ctx0, model.mm_model_attn_q_w, q), model.mm_model_attn_q_b);
                 Q = ggml_scale_inplace(ctx0, Q, 1.0f / sqrt((float)d_head));
@@ -1087,7 +1113,33 @@ static ggml_cgraph * clip_image_build_graph(clip_ctx * ctx, const clip_image_f32
             GGML_ASSERT(false);
         }
     }
-    else if (ctx->proj_type == PROJECTOR_TYPE_MERGER) {
+    // glm projector
+    else if (ctx->has_glm_projector) {
+        if (ctx->proj_type == PROJECTOR_TYPE_GLM_EDGE) {
+            size_t gridsz = (size_t)sqrt(embeddings->ne[1]);
+            embeddings = ggml_cont(ctx0, ggml_permute(ctx0,embeddings,1,0,2,3));
+            embeddings = ggml_reshape_3d(ctx0, embeddings, gridsz, gridsz, embeddings->ne[1]);
+            embeddings = ggml_conv_2d(ctx0, model.mm_model_adapter_conv_w, embeddings, 2, 2, 0, 0, 1, 1);
+            embeddings = ggml_reshape_3d(ctx0, embeddings,embeddings->ne[0]*embeddings->ne[1] , embeddings->ne[2], batch_size);
+            embeddings = ggml_cont(ctx0, ggml_permute(ctx0,embeddings, 1, 0, 2, 3));
+            embeddings = ggml_add(ctx0, embeddings, model.mm_model_adapter_conv_b);
+            //GLU
+            {
+                embeddings = ggml_mul_mat(ctx0, model.mm_model_mlp_0_w, embeddings);
+                embeddings = ggml_norm(ctx0, embeddings, eps);
+                embeddings = ggml_add(ctx0, ggml_mul(ctx0, embeddings, model.mm_model_ln_q_w), model.mm_model_ln_q_b);
+                embeddings = ggml_gelu_inplace(ctx0, embeddings);
+                struct ggml_tensor * x = embeddings;
+                embeddings = ggml_mul_mat(ctx0, model.mm_model_mlp_2_w, embeddings);
+                x = ggml_mul_mat(ctx0, model.mm_model_mlp_1_w,x);
+                embeddings = ggml_silu_inplace(ctx0, embeddings);
+                embeddings = ggml_mul(ctx0, embeddings,x);
+                embeddings = ggml_mul_mat(ctx0, model.mm_model_mlp_3_w, embeddings);
+            }
+        } else {
+            GGML_ABORT("fatel error");
+        }
+    } else if (ctx->proj_type == PROJECTOR_TYPE_MERGER) {
         embeddings = ggml_reshape_3d(ctx0, embeddings, hidden_size * 4, num_positions / 4, batch_size);
 
         embeddings = ggml_mul_mat(ctx0, model.mm_0_w, embeddings);
@@ -1276,6 +1328,11 @@ struct clip_ctx * clip_model_load(const char * fname, const int verbosity = 1) {
             new_clip->minicpmv_version = gguf_get_val_i32(ctx, idx);
         }
 
+        idx = gguf_find_key(ctx, KEY_HAS_GLM_PROJ);
+        if (idx != -1) {
+            new_clip->has_glm_projector = gguf_get_val_bool(ctx, idx);
+        }
+
         idx = gguf_find_key(ctx, KEY_HAS_QWEN2VL_MERGER);
         if (idx != -1) {
             new_clip->has_qwen2vl_merger = gguf_get_val_bool(ctx, idx);
@@ -1300,6 +1357,7 @@ struct clip_ctx * clip_model_load(const char * fname, const int verbosity = 1) {
             LOG_INF("%s: vision_encoder: %d\n", __func__, new_clip->has_vision_encoder);
             LOG_INF("%s: llava_projector:  %d\n", __func__, new_clip->has_llava_projector);
             LOG_INF("%s: minicpmv_projector:  %d\n", __func__, new_clip->has_minicpmv_projector);
+            LOG_INF("%s: glm_projector:  %d\n", __func__, new_clip->has_glm_projector);
             LOG_INF("%s: model size:     %.2f MB\n", __func__, model_size / 1024.0 / 1024.0);
             LOG_INF("%s: metadata size:  %.2f MB\n", __func__, ggml_get_mem_size(meta) / 1024.0 / 1024.0);
         }
@@ -1567,6 +1625,18 @@ struct clip_ctx * clip_model_load(const char * fname, const int verbosity = 1) {
             vision_model.mm_model_ln_post_w = get_tensor(new_clip->ctx_data, format(TN_MINICPMV_LN, "post", "weight"));
             vision_model.mm_model_ln_post_b = get_tensor(new_clip->ctx_data, format(TN_MINICPMV_LN, "post", "bias"));
         }
+        else if (new_clip->proj_type == PROJECTOR_TYPE_GLM_EDGE) {
+            vision_model.mm_model_adapter_conv_w = get_tensor(new_clip->ctx_data, format(TN_GLM_ADAPER_CONV, "weight"));
+            vision_model.mm_model_adapter_conv_b = get_tensor(new_clip->ctx_data, format(TN_GLM_ADAPER_CONV, "bias"));
+            vision_model.mm_model_mlp_0_w = get_tensor(new_clip->ctx_data, format(TN_GLM_ADAPTER_LINEAR,"weight"));
+            vision_model.mm_model_ln_q_w = get_tensor(new_clip->ctx_data, format(TN_GLM_ADAPTER_NORM_1,"weight"));
+            vision_model.mm_model_ln_q_b = get_tensor(new_clip->ctx_data, format(TN_GLM_ADAPTER_NORM_1,"bias"));
+            vision_model.mm_model_mlp_1_w =  get_tensor(new_clip->ctx_data, format(TN_GLM_ADAPTER_D_H_2_4H,"weight"));
+            vision_model.mm_model_mlp_2_w =  get_tensor(new_clip->ctx_data, format(TN_GLM_ADAPTER_GATE,"weight"));
+            vision_model.mm_model_mlp_3_w =  get_tensor(new_clip->ctx_data, format(TN_GLM_ADAPTER_D_4H_2_H,"weight"));
+            vision_model.boi_w = get_tensor(new_clip->ctx_data, TN_GLM_BOI_W);
+            vision_model.eoi_w = get_tensor(new_clip->ctx_data, TN_GLM_EOI_W);
+        }
         else if (new_clip->proj_type == PROJECTOR_TYPE_MERGER) {
             vision_model.mm_0_w = get_tensor(new_clip->ctx_data, format(TN_LLAVA_PROJ, 0, "weight"));
             vision_model.mm_0_b = get_tensor(new_clip->ctx_data, format(TN_LLAVA_PROJ, 0, "bias"));
@@ -2041,6 +2111,7 @@ static std::vector<std::vector<clip_image_u8 *>> uhd_slice_image(const clip_imag
                 images[images.size()-1].push_back(patch);
             }
         }
+        clip_image_u8_free(refine_image);
     }
     return images;
 }
@@ -2079,6 +2150,13 @@ bool clip_image_preprocess(struct clip_ctx * ctx, const clip_image_u8 * img, cli
                 clip_image_f32_free(res);
             }
         }
+        for (size_t i = 0; i < imgs.size(); ++i) {
+            for (size_t j = 0; j < imgs[i].size(); ++j) {
+                if (imgs[i][j] != nullptr) {
+                    clip_image_u8_free(imgs[i][j]);
+                }
+            }
+        }
         return true;
     }
     else if (ctx->has_qwen2vl_merger) {
@@ -2099,6 +2177,20 @@ bool clip_image_preprocess(struct clip_ctx * ctx, const clip_image_u8 * img, cli
         return true;
     }
 
+    if (ctx->has_glm_projector) {
+        res_imgs->size = 1;
+        res_imgs->data = new clip_image_f32[res_imgs->size];
+        clip_image_u8 resized_image;
+        int32_t sz=ctx->vision_model.hparams.image_size;
+        bicubic_resize(*img, resized_image,sz,sz);
+        clip_image_f32 * res = clip_image_f32_init();
+        //clip_image_save_to_bmp(resized_image, "resized.bmp");
+        normalize_image_u8_to_f32(&resized_image, res, ctx->image_mean, ctx->image_std);
+        res_imgs->data[0] = *res;
+        clip_image_f32_free(res);
+        return true;
+    }
+
     bool pad_to_square = true;
     if (!ctx->has_vision_encoder) {
         LOG_ERR("This gguf file seems to have no vision encoder\n");
@@ -2284,7 +2376,8 @@ void clip_free(clip_ctx * ctx) {
 }
 
 size_t clip_embd_nbytes(const struct clip_ctx * ctx) {
-    return clip_n_patches(ctx) * clip_n_mmproj_embd(ctx) * sizeof(float);
+    int extra_tokens = ctx->has_glm_projector ? 2 : 0;
+    return (clip_n_patches(ctx) + extra_tokens) * clip_n_mmproj_embd(ctx) * sizeof(float);
 }
 
 size_t clip_embd_nbytes_by_img(const struct clip_ctx * ctx, int img_h, int img_w) {
@@ -2326,7 +2419,7 @@ int clip_n_patches_by_img(const struct clip_ctx * ctx, struct clip_image_f32 * i
 
     int n_patches = (params.image_size / params.patch_size) * (params.image_size / params.patch_size);
 
-    if (ctx->proj_type == PROJECTOR_TYPE_LDP || ctx->proj_type == PROJECTOR_TYPE_LDPV2) {
+    if (ctx->proj_type == PROJECTOR_TYPE_LDP || ctx->proj_type == PROJECTOR_TYPE_LDPV2 || ctx->proj_type == PROJECTOR_TYPE_GLM_EDGE) {
         n_patches /= 4;
     } else if (ctx->proj_type == PROJECTOR_TYPE_RESAMPLER) {
         if (ctx->minicpmv_version == 2) {
@@ -2335,6 +2428,9 @@ int clip_n_patches_by_img(const struct clip_ctx * ctx, struct clip_image_f32 * i
         else if (ctx->minicpmv_version == 3) {
             n_patches = 64;
         }
+        else if (ctx->minicpmv_version == 4) {
+            n_patches = 64;
+        }
     } else if (ctx->proj_type == PROJECTOR_TYPE_MERGER) {
         int patch_size = params.patch_size * 2;
         int x_patch = img->nx / patch_size + (int)(img->nx % patch_size > 0);
@@ -2456,6 +2552,12 @@ bool clip_image_batch_encode(clip_ctx * ctx, const int n_threads, const clip_ima
     if (ctx->has_minicpmv_projector) {
         GGML_ASSERT(batch_size == 1);
     }
+    if (ctx->has_glm_projector) {
+        GGML_ASSERT(batch_size == 1);
+        ggml_tensor * boi = ctx->vision_model.boi_w;
+        ggml_backend_tensor_get(boi,vec,0,ggml_nbytes(boi));
+        vec = (float*)(vec+ggml_nelements(boi)); //offset for boi
+    }
 
     // build the inference graph
     ggml_cgraph * gf = clip_image_build_graph(ctx, imgs, ctx->load_image_size, true);
@@ -2514,8 +2616,8 @@ bool clip_image_batch_encode(clip_ctx * ctx, const int n_threads, const clip_ima
             //    -> https://huggingface.co/HuggingFaceM4/siglip-so400m-14-980-flash-attn2-navit/blob/d66538faeba44480d0bfaa42145eef26f9423199/modeling_siglip.py#L316
             struct ggml_tensor * positions = ggml_graph_get_tensor(gf, "positions");
             int* positions_data = (int*)malloc(ggml_nbytes(positions));
-            int bucket_coords_h[70];
-            int bucket_coords_w[70];
+            int bucket_coords_h[1024];
+            int bucket_coords_w[1024];
             for (int i = 0; i < pos_h; i++){
                 bucket_coords_h[i] = std::floor(70.0*i/pos_h);
             }
@@ -2543,6 +2645,9 @@ bool clip_image_batch_encode(clip_ctx * ctx, const int n_threads, const clip_ima
             else if (ctx->minicpmv_version == 3) {
                 embed_dim = 3584;
             }
+            else if (ctx->minicpmv_version == 4) {
+                embed_dim = 3584;
+            }
             auto pos_embed_t = get_2d_sincos_pos_embed(embed_dim, std::make_pair(pos_w, pos_h));
 
             float * pos_embed_data = (float *)malloc(ggml_nbytes(pos_embed));
@@ -2605,7 +2710,7 @@ bool clip_image_batch_encode(clip_ctx * ctx, const int n_threads, const clip_ima
             ggml_backend_tensor_set(positions, positions_data, 0, ggml_nbytes(positions));
             free(positions_data);
 
-            {
+            if (!ctx->has_glm_projector) {
                 struct ggml_tensor * patches = ggml_graph_get_tensor(gf, "patches");
                 int* patches_data = (int*)malloc(ggml_nbytes(patches));
                 for (int i = 0; i < num_patches; i++) {
@@ -2629,14 +2734,19 @@ bool clip_image_batch_encode(clip_ctx * ctx, const int n_threads, const clip_ima
     // copy the embeddings to the location passed by the user
     ggml_backend_tensor_get(embeddings, vec, 0, ggml_nbytes(embeddings));
 
+    if (ctx->has_glm_projector) {
+        //eoi
+        ggml_tensor * eoi = ctx->vision_model.eoi_w;
+        int offset = ggml_nelements(embeddings);
+        ggml_backend_tensor_get(eoi, vec+offset, 0, ggml_nbytes(eoi));
+    }
+
     return true;
 }
 
 bool clip_model_quantize(const char * fname_inp, const char * fname_out, const int itype) {
-    ggml_type type = GGML_TYPE_Q4_1;
-
     assert(itype < GGML_TYPE_COUNT);
-    type = static_cast<ggml_type>(itype);
+    ggml_type type = static_cast<ggml_type>(itype);
 
     auto * ctx_clip = clip_model_load(fname_inp, 2);
 
@@ -2689,8 +2799,8 @@ bool clip_model_quantize(const char * fname_inp, const char * fname_out, const i
             }
         }
 
-        // quantize only 2D tensors
-        quantize &= (ggml_n_dims(cur) == 2);
+        // quantize only 2D tensors and bigger than block size
+        quantize &= (ggml_n_dims(cur) == 2) && cur->ne[0] > ggml_blck_size(type);
 
         if (quantize) {
             new_type = type;
@@ -2786,6 +2896,12 @@ int clip_n_mmproj_embd(const struct clip_ctx * ctx) {
         else if (ctx->minicpmv_version == 3) {
             return 3584;
         }
+        else if (ctx->minicpmv_version == 4) {
+            return 3584;
+        }
+    }
+    if (ctx->proj_type == PROJECTOR_TYPE_GLM_EDGE){
+        return ctx->vision_model.mm_model_mlp_3_w->ne[1];
     }
     if (ctx->proj_type == PROJECTOR_TYPE_MERGER) {
         return ctx->vision_model.mm_1_b->ne[0];
@@ -2802,6 +2918,9 @@ int clip_is_minicpmv(const struct clip_ctx * ctx) {
     return 0;
 }
 
+bool clip_is_glm(const struct clip_ctx * ctx) {
+    return ctx->has_glm_projector;
+}
 bool clip_is_qwen2vl(const struct clip_ctx * ctx) {
     return ctx->has_qwen2vl_merger;
 }

examples/llava/clip.h

@@ -93,6 +93,8 @@ CLIP_API bool clip_is_qwen2vl(const struct clip_ctx * ctx);
 
 CLIP_API bool clip_encode_float_image (struct clip_ctx * ctx, int n_threads, float * img, int h, int w, float * vec);
 
+CLIP_API bool clip_is_glm(const struct clip_ctx * ctx);
+
 #ifdef __cplusplus
 }
 #endif

examples/llava/glmedge-convert-image-encoder-to-gguf.py

@@ -0,0 +1,280 @@
+import argparse
+import os
+import json
+import re
+
+import torch
+import numpy as np
+from gguf import *
+
+TEXT = "clip.text"
+VISION = "clip.vision"
+from transformers import SiglipVisionModel, SiglipVisionConfig
+
+def k(raw_key: str, arch: str) -> str:
+    return raw_key.format(arch=arch)
+
+
+def should_skip_tensor(name: str, has_text: bool, has_vision: bool, has_llava: bool) -> bool:
+    if name in (
+        "logit_scale",
+        "text_model.embeddings.position_ids",
+        "vision_model.embeddings.position_ids",
+    ):
+        return True
+
+    if name in (
+        "vision_model.head.probe",
+        "vision_model.head.attention.in_proj_weight",
+        "vision_model.head.attention.in_proj_bias",
+        "vision_model.head.attention.out_proj.weight",
+        "vision_model.head.attention.out_proj.bias",
+        "vision_model.head.layernorm.weight",
+        "vision_model.head.layernorm.bias",
+        "vision_model.head.mlp.fc1.weight",
+        "vision_model.head.mlp.fc1.bias",
+        "vision_model.head.mlp.fc2.weight",
+        "vision_model.head.mlp.fc2.bias"
+    ):
+        return True
+
+    if name.startswith("v") and not has_vision:
+        return True
+
+    if name.startswith("t") and not has_text:
+        return True
+
+    return False
+
+
+def get_tensor_name(name: str) -> str:
+    if "projection" in name:
+        return name
+    if "mm_projector" in name:
+        name = name.replace("model.mm_projector", "mm")
+        name = re.sub(r'mm\.mlp\.mlp', 'mm.model.mlp', name, count=1)
+        name = re.sub(r'mm\.peg\.peg', 'mm.model.peg', name, count=1)
+        return name
+
+    return name.replace("text_model", "t").replace("vision_model", "v").replace("encoder.layers", "blk").replace("embeddings.", "").replace("_proj", "").replace("self_attn.", "attn_").replace("layer_norm", "ln").replace("layernorm", "ln").replace("mlp.fc1", "ffn_down").replace("mlp.fc2", "ffn_up").replace("embedding", "embd").replace("final", "post").replace("layrnorm", "ln")
+
+
+def bytes_to_unicode():
+    """
+    Returns list of utf-8 byte and a corresponding list of unicode strings.
+    The reversible bpe codes work on unicode strings.
+    This means you need a large # of unicode characters in your vocab if you want to avoid UNKs.
+    When you're at something like a 10B token dataset you end up needing around 5K for decent coverage.
+    This is a significant percentage of your normal, say, 32K bpe vocab.
+    To avoid that, we want lookup tables between utf-8 bytes and unicode strings.
+    And avoids mapping to whitespace/control characters the bpe code barfs on.
+    """
+    bs = (
+        list(range(ord("!"), ord("~") + 1))
+        + list(range(ord("¡"), ord("¬") + 1))
+        + list(range(ord("®"), ord("ÿ") + 1))
+    )
+    cs = bs[:]
+    n = 0
+    for b in range(2**8):
+        if b not in bs:
+            bs.append(b)
+            cs.append(2**8 + n)
+            n += 1
+    cs = [chr(n) for n in cs]
+    return dict(zip(bs, cs))
+
+
+ap = argparse.ArgumentParser()
+ap.add_argument("-m", "--model-dir", help="Path to model directory cloned from HF Hub", required=True)
+ap.add_argument("--use-f32", action="store_true", default=False, help="Use f32 instead of f16")
+ap.add_argument("--text-only", action="store_true", required=False,
+                help="Save a text-only model. It can't be used to encode images")
+ap.add_argument("--vision-only", action="store_true", required=False,
+                help="Save a vision-only model. It can't be used to encode texts")
+ap.add_argument("--clip-model-is-vision", action="store_true", required=False,
+                help="The clip model is a pure vision model (ShareGPT4V vision extract for example)")
+ap.add_argument("--clip-model-is-openclip", action="store_true", required=False,
+                help="The clip model is from openclip (for ViT-SO400M type))")
+ap.add_argument("--llava-projector", help="Path to llava.projector file. If specified, save an image encoder for LLaVA models.")
+ap.add_argument("--projector-type", help="Type of projector. Possible values: mlp, ldp, ldpv2", choices=["mlp", "ldp", "ldpv2","adapter"], default="adapter")
+ap.add_argument("-o", "--output-dir", help="Directory to save GGUF files. Default is the original model directory", default=None)
+# Example --image_mean 0.48145466 0.4578275 0.40821073 --image_std 0.26862954 0.26130258 0.27577711
+# Example --image_mean 0.5 0.5 0.5 --image_std 0.5 0.5 0.5
+default_image_mean = [0.5, 0.5, 0.5]
+default_image_std = [0.5, 0.5, 0.5]
+ap.add_argument('--image-mean', type=float, nargs='+', help='Mean of the images for normalization (overrides processor) ', default=None)
+ap.add_argument('--image-std', type=float, nargs='+', help='Standard deviation of the images for normalization (overrides processor)', default=None)
+
+# with proper
+args = ap.parse_args()
+
+
+if args.text_only and args.vision_only:
+    print("--text-only and --image-only arguments cannot be specified at the same time.")
+    exit(1)
+
+if args.use_f32:
+    print("WARNING: Weights for the convolution op is always saved in f16, as the convolution op in GGML does not support 32-bit kernel weights yet.")
+
+# output in the same directory as the model if output_dir is None
+dir_model = args.model_dir
+
+if args.clip_model_is_vision or not os.path.exists(dir_model + "/vocab.json") or args.clip_model_is_openclip:
+    vocab = None
+    tokens = None
+else:
+    with open(dir_model + "/vocab.json", "r", encoding="utf-8") as f:
+        vocab = json.load(f)
+        tokens = [key for key in vocab]
+
+with open(dir_model + "/config.json", "r", encoding="utf-8") as f:
+    config = json.load(f)
+    if args.clip_model_is_vision:
+        v_hparams = config
+        t_hparams = None
+    else:
+        v_hparams = config["vision_config"]
+        t_hparams = None
+
+# possible data types
+#   ftype == 0 -> float32
+#   ftype == 1 -> float16
+#
+# map from ftype to string
+ftype_str = ["f32", "f16"]
+
+ftype = 1
+if args.use_f32:
+    ftype = 0
+
+vision_config = SiglipVisionConfig(**v_hparams)
+model = SiglipVisionModel(vision_config)
+model.load_state_dict(torch.load(os.path.join(dir_model, "glm.clip")))
+
+fname_middle = None
+has_text_encoder = False
+has_vision_encoder = True
+has_glm_projector = True
+if args.text_only:
+    fname_middle = "text-"
+    has_vision_encoder = False
+elif args.llava_projector is not None:
+    fname_middle = "mmproj-"
+    has_text_encoder = False
+    has_glm_projector = True
+elif args.vision_only:
+    fname_middle = "vision-"
+    has_text_encoder = False
+else:
+    fname_middle = ""
+
+output_dir = args.output_dir if args.output_dir is not None else dir_model
+os.makedirs(output_dir, exist_ok=True)
+output_prefix = os.path.basename(output_dir).replace("ggml_", "")
+fname_out = os.path.join(output_dir, f"{fname_middle}model-{ftype_str[ftype]}.gguf")
+fout = GGUFWriter(path=fname_out, arch="clip")
+
+fout.add_bool("clip.has_text_encoder", has_text_encoder)
+fout.add_bool("clip.has_vision_encoder", has_vision_encoder)
+fout.add_bool("clip.has_glm_projector", has_glm_projector)
+fout.add_file_type(ftype)
+model_name = config["_name_or_path"] if "_name_or_path" in config else os.path.basename(dir_model)
+fout.add_name(model_name)
+if has_glm_projector:
+    fout.add_description("image encoder for glm4v")
+    fout.add_string("clip.projector_type", "adapter")
+else:
+    fout.add_description("two-tower CLIP model")
+
+if has_text_encoder:
+    assert t_hparams is not None
+    assert tokens is not None
+    # text_model hparams
+    fout.add_uint32(k(KEY_CONTEXT_LENGTH, TEXT), t_hparams["max_position_embeddings"])
+    fout.add_uint32(k(KEY_EMBEDDING_LENGTH, TEXT), t_hparams["hidden_size"])
+    fout.add_uint32(k(KEY_FEED_FORWARD_LENGTH, TEXT), t_hparams["intermediate_size"])
+    fout.add_uint32("clip.text.projection_dim", t_hparams.get("projection_dim", config["projection_dim"]))
+    fout.add_uint32(k(KEY_ATTENTION_HEAD_COUNT, TEXT), t_hparams["num_attention_heads"])
+    fout.add_float32(k(KEY_ATTENTION_LAYERNORM_EPS, TEXT), t_hparams["layer_norm_eps"])
+    fout.add_uint32(k(KEY_BLOCK_COUNT, TEXT), t_hparams["num_hidden_layers"])
+    fout.add_token_list(tokens)
+
+if has_vision_encoder:
+    # vision_model hparams
+    fout.add_uint32("clip.vision.image_size", v_hparams["image_size"])
+    fout.add_uint32("clip.vision.patch_size", v_hparams["patch_size"])
+    fout.add_uint32(k(KEY_EMBEDDING_LENGTH, VISION), v_hparams["hidden_size"])
+    fout.add_uint32(k(KEY_FEED_FORWARD_LENGTH, VISION), v_hparams["intermediate_size"])
+    fout.add_uint32("clip.vision.projection_dim", 0)
+    fout.add_uint32(k(KEY_ATTENTION_HEAD_COUNT, VISION), v_hparams["num_attention_heads"])
+    fout.add_float32(k(KEY_ATTENTION_LAYERNORM_EPS, VISION), 1e-6)
+    fout.add_uint32(k(KEY_BLOCK_COUNT, VISION), v_hparams["num_hidden_layers"])
+
+    image_mean = args.image_mean if args.image_mean is not None else default_image_mean
+    image_std = args.image_std if args.image_std is not None else default_image_std
+    fout.add_array("clip.vision.image_mean", image_mean)
+    fout.add_array("clip.vision.image_std", image_std)
+
+fout.add_bool("clip.use_gelu", True)
+
+
+if has_glm_projector:
+    # model.vision_model.encoder.layers.pop(-1)  # pyright: ignore[reportAttributeAccessIssue]
+    projector = torch.load(args.llava_projector)
+    for name, data in projector.items():
+        name = get_tensor_name(name)
+        # pw and dw conv ndim==4
+        if data.ndim == 2 or data.ndim == 4:
+            data = data.squeeze().numpy().astype(np.float16)
+        else:
+            data = data.squeeze().numpy().astype(np.float32)
+        if name.startswith("vision."):
+            name=name.replace("vision.","")
+        fout.add_tensor(name, data)
+        print(f"Projector {name} - {data.dtype} - shape = {data.shape}")
+        # print(f"Projector {name} tensors added\n")
+
+state_dict = model.state_dict()  # pyright: ignore[reportAttributeAccessIssue]
+for name, data in state_dict.items():
+    if should_skip_tensor(name, has_text_encoder, has_vision_encoder, has_glm_projector):
+        # we don't need this
+        print(f"skipping parameter: {name}")
+        continue
+
+    name = get_tensor_name(name)
+    data = data.squeeze().numpy()
+
+    n_dims = len(data.shape)
+
+    # ftype == 0 -> float32, ftype == 1 -> float16
+    ftype_cur = 0
+    if n_dims == 4:
+        print(f"tensor {name} is always saved in f16")
+        data = data.astype(np.float16)
+        ftype_cur = 1
+    elif ftype == 1:
+        if name[-7:] == ".weight" and n_dims == 2:
+            # print("  Converting to float16")
+            data = data.astype(np.float16)
+            ftype_cur = 1
+        else:
+            # print("  Converting to float32")
+            data = data.astype(np.float32)
+            ftype_cur = 0
+    else:
+        if data.dtype != np.float32:
+            # print("  Converting to float32")
+            data = data.astype(np.float32)
+            ftype_cur = 0
+    print(f"siglip {name} - {data.dtype} - shape = {data.shape}")
+    # print(f"{name} - {ftype_str[ftype_cur]} - shape = {data.shape}")
+    fout.add_tensor(name, data)
+
+
+fout.write_header_to_file()
+fout.write_kv_data_to_file()
+fout.write_tensors_to_file()
+fout.close()
+
+print("Done. Output file: " + fname_out)

examples/llava/glmedge-surgery.py

@@ -0,0 +1,33 @@
+import argparse
+import os
+import torch
+from transformers import AutoModel
+
+ap = argparse.ArgumentParser()
+ap.add_argument("-m", "--model", help="Path to GLM model")
+args = ap.parse_args()
+
+# find the model part that includes the the multimodal projector weights
+model = AutoModel.from_pretrained(args.model, trust_remote_code=True, local_files_only=True)
+checkpoint = model.state_dict()
+
+# get a list of mm tensor names
+mm_tensors = [k for k, v in checkpoint.items() if k.startswith("vision.adapter.")]
+
+# store these tensors in a new dictionary and torch.save them
+projector = {name: checkpoint[name].float() for name in mm_tensors}
+torch.save(projector, f"{args.model}/glm.projector")
+
+clip_tensors = [k for k, v in checkpoint.items() if k.startswith("vision.vit.model.vision_model.")]
+if len(clip_tensors) > 0:
+    clip = {name.replace("vision.vit.model.", ""): checkpoint[name].float() for name in clip_tensors}
+    torch.save(clip, f"{args.model}/glm.clip")
+
+    # added tokens should be removed to be able to convert Mistral models
+    if os.path.exists(f"{args.model}/added_tokens.json"):
+        with open(f"{args.model}/added_tokens.json", "w") as f:
+            f.write("{}\n")
+
+print("Done!")
+print(f"Now you can convert {args.model} to a regular LLaMA GGUF file.")
+print(f"Also, use {args.model}glm.projector to prepare a glm-encoder.gguf file.")

examples/llava/llava.cpp

@@ -216,7 +216,7 @@ static bool clip_llava_handle_patches(clip_ctx * ctx_clip, std::vector<float *>
     return true;
 }
 
-static clip_image_f32 * only_v2_5_reshape_by_patch(clip_image_f32 * image, int patch_size) {
+static clip_image_f32 * reshape_by_patch(clip_image_f32 * image, int patch_size) {
     int width = image->nx;
     int height = image->ny;
     int num_patches = (height / patch_size) * (width / patch_size);
@@ -277,13 +277,7 @@ static bool encode_image_with_clip(clip_ctx * ctx_clip, int n_threads, const cli
                 encoded = clip_image_encode(ctx_clip, n_threads, &img_res_v.data[i], image_embd_v[i]);
             }
             else {
-                int has_minicpmv_projector = clip_is_minicpmv(ctx_clip);
-                if (has_minicpmv_projector == 2) {
-                    encoded = clip_image_encode(ctx_clip, n_threads, only_v2_5_reshape_by_patch(&img_res_v.data[i], patch_size), image_embd_v[i]);
-                }
-                else if (has_minicpmv_projector == 3) {
-                    encoded = clip_image_encode(ctx_clip, n_threads, &img_res_v.data[i], image_embd_v[i]);
-                }
+                encoded = clip_image_encode(ctx_clip, n_threads, reshape_by_patch(&img_res_v.data[i], patch_size), image_embd_v[i]);
             }
 
             if (!encoded) {
@@ -313,6 +307,23 @@ static bool encode_image_with_clip(clip_ctx * ctx_clip, int n_threads, const cli
         load_image_size->height = img->ny;
         clip_add_load_image_size(ctx_clip, load_image_size);
         LOG_INF("%s: load_image_size %d %d\n", __func__, load_image_size->width, load_image_size->height);
+        delete[] img_res_v.data;
+        img_res_v.size = 0;
+        img_res_v.data = nullptr;
+    }
+    else if (clip_is_glm(ctx_clip)){
+        struct clip_image_size * load_image_size = clip_image_size_init();
+        load_image_size->width = img_res_v.data[0].nx;
+        load_image_size->height = img_res_v.data[0].ny;
+        clip_add_load_image_size(ctx_clip, load_image_size);
+
+        bool encoded = clip_image_encode(ctx_clip, n_threads, &img_res_v.data[0], image_embd);
+        int pos = int(load_image_size->width/clip_patch_size(ctx_clip)/2);
+        *n_img_pos = (pos * pos + 2);
+        if (!encoded){
+            LOG_ERR("Unable to encode image \n");
+            return false;
+        }
     }
     else if (strcmp(mm_patch_merge_type, "spatial_unpad") != 0) {
         // flat / default llava-1.5 type embedding
@@ -398,6 +409,9 @@ bool llava_image_embed_make_with_clip_img(clip_ctx * ctx_clip, int n_threads, co
     if (clip_is_minicpmv(ctx_clip)) {
         num_max_patches = 10;
     }
+    if (clip_is_glm(ctx_clip)) {
+        num_max_patches = 1;
+    }
     float * image_embd;
     if (clip_is_qwen2vl(ctx_clip)) {
         // qwen2vl don't split image into chunks, so `num_max_patches` is not needed.

examples/llava/minicpmv-cli.cpp

@@ -140,6 +140,9 @@ static void process_image(struct llava_context * ctx_llava, struct llava_image_e
     else if (has_minicpmv_projector == 3) {
         system_prompt = "<|im_start|>user\n";
     }
+    else if (has_minicpmv_projector == 4) {
+        system_prompt = "<|im_start|>user\n";
+    }
     LOG_INF("%s: image token past: %d\n", __func__, n_past);
     eval_string(ctx_llava->ctx_llama, (system_prompt+"<image>").c_str(), params->n_batch, &n_past, false);
     process_eval_image_embed(ctx_llava, embeds, params->n_batch, &n_past, idx++);
@@ -227,6 +230,9 @@ static struct common_sampler * llama_init(struct llava_context * ctx_llava, comm
         else if (has_minicpmv_projector == 3) {
             user_prompt = "<|im_start|>user\n" + prompt;
         }
+        else if (has_minicpmv_projector == 4) {
+            user_prompt = "<|im_start|>user\n" + prompt;
+        }
     }
 
     eval_string(ctx_llava->ctx_llama, user_prompt.c_str(), params->n_batch, &n_past, false);
@@ -236,6 +242,9 @@ static struct common_sampler * llama_init(struct llava_context * ctx_llava, comm
     else if (has_minicpmv_projector == 3) {
         eval_string(ctx_llava->ctx_llama, "<|im_end|><|im_start|>assistant\n", params->n_batch, &n_past, false);
     }
+    else if (has_minicpmv_projector == 4) {
+        eval_string(ctx_llava->ctx_llama, "<|im_end|><|im_start|>assistant\n", params->n_batch, &n_past, false);
+    }
 
     // generate the response
 
@@ -308,7 +317,6 @@ int main(int argc, char ** argv) {
                     const auto * tmp = llama_loop(ctx_llava, smpl, n_past);
                     response += tmp;
                     if (strcmp(tmp, "</s>") == 0) break;
-                    if (strstr(tmp, "###")) break; // Yi-VL behavior
                     printf("%s", tmp);// mistral llava-1.6
                     if (strstr(response.c_str(), "<user>")) break; // minicpm-v
                     fflush(stdout);

examples/llava/minicpmv-convert-image-encoder-to-gguf.py

@@ -501,7 +501,7 @@ default_image_mean = [0.48145466, 0.4578275, 0.40821073]
 default_image_std = [0.26862954, 0.26130258, 0.27577711]
 ap.add_argument('--image-mean', type=float, nargs='+', help='Mean of the images for normalization (overrides processor) ', default=None)
 ap.add_argument('--image-std', type=float, nargs='+', help='Standard deviation of the images for normalization (overrides processor)', default=None)
-ap.add_argument('--minicpmv_version', type=int, help='minicpmv_version: MiniCPM-V-2 use 1; MiniCPM-V-2.5 use 2; MiniCPM-V-2.6 use 3', default=2)
+ap.add_argument('--minicpmv_version', type=int, help='minicpmv_version: MiniCPM-V-2 use 1; MiniCPM-V-2.5 use 2; MiniCPM-V-2.6 use 3; MiniCPM-o-2.6 use 4', default=2)
 
 # with proper
 args = ap.parse_args()
@@ -545,12 +545,19 @@ if args.use_f32:
 
 minicpmv_version = args.minicpmv_version
 emb_dim = 4096
+block_count = 26
 if minicpmv_version == 1:
     emb_dim = 2304
+    block_count = 26
 elif minicpmv_version == 2:
     emb_dim = 4096
+    block_count = 27
 elif minicpmv_version == 3:
     emb_dim = 3584
+    block_count = 27
+elif minicpmv_version == 4:
+    emb_dim = 3584
+    block_count = 27
 
 default_vision_config = {
         "hidden_size": 1152,
@@ -567,6 +574,9 @@ model = Idefics2VisionTransformer(vision_config)
 if minicpmv_version == 3:
     vision_config = SiglipVisionConfig(**default_vision_config)
     model = SiglipVisionTransformer(vision_config)
+elif minicpmv_version == 4:
+    vision_config = SiglipVisionConfig(**default_vision_config)
+    model = SiglipVisionTransformer(vision_config)
 
 processor = None
 # if model.attn_pool is not None:
@@ -587,7 +597,7 @@ elif args.minicpmv_projector is not None:
     fname_middle = "mmproj-"
     has_text_encoder = False
     has_minicpmv_projector = True
-    minicpmv_version = 3
+    minicpmv_version = 4
 elif args.vision_only:
     fname_middle = "vision-"
     has_text_encoder = False
@@ -625,7 +635,6 @@ if has_vision_encoder:
     fout.add_uint32("clip.vision.projection_dim", 0)
     fout.add_uint32(add_key_str(KEY_ATTENTION_HEAD_COUNT, VISION), 16)
     fout.add_float32(add_key_str(KEY_ATTENTION_LAYERNORM_EPS, VISION), 1e-6)
-    block_count = 26
     fout.add_uint32(add_key_str(KEY_BLOCK_COUNT, VISION), block_count)
 
     if processor is not None:

examples/llava/minicpmv-surgery.py

@@ -8,7 +8,7 @@ ap.add_argument("-m", "--model", help="Path to MiniCPM-V model")
 args = ap.parse_args()
 
 # find the model part that includes the the multimodal projector weights
-model = AutoModel.from_pretrained(args.model, trust_remote_code=True, local_files_only=True)
+model = AutoModel.from_pretrained(args.model, trust_remote_code=True, local_files_only=True, torch_dtype=torch.bfloat16)
 checkpoint = model.state_dict()
 
 # get a list of mm tensor names

examples/lookahead/README.md

@@ -4,4 +4,4 @@ Demonstration of lookahead decoding technique:
 
 https://lmsys.org/blog/2023-11-21-lookahead-decoding/
 
-More info: https://github.com/ggerganov/llama.cpp/pull/4207
+More info: https://github.com/ggml-org/llama.cpp/pull/4207

examples/lookup/README.md

@@ -8,5 +8,5 @@ The key parameters for lookup decoding are `ngram_min`, `ngram_max` and `n_draft
 
 More info:
 
-https://github.com/ggerganov/llama.cpp/pull/4484
-https://github.com/ggerganov/llama.cpp/issues/4226
+https://github.com/ggml-org/llama.cpp/pull/4484
+https://github.com/ggml-org/llama.cpp/issues/4226

examples/main-cmake-pkg/CMakeLists.txt

@@ -1,32 +0,0 @@
-cmake_minimum_required(VERSION 3.12)
-project("llama-cli-cmake-pkg" C CXX)
-set(TARGET llama-cli-cmake-pkg)
-
-find_package(Llama 0.0.1 REQUIRED)
-
-# Bake common functionality in with target. Because applications
-# using the relocatable Llama package should be outside of the
-# source tree, llama-cli-cmake-pkg pretends the dependencies are built-in.
-set(_common_path "${CMAKE_CURRENT_LIST_DIR}/../../common")
-add_library(common OBJECT)
-file(GLOB _common_files
-    "${_common_path}/*.h"
-    "${_common_path}/*.cpp"
-)
-target_sources(common PRIVATE ${_common_files})
-
-# If the common project was part of "llama-cli-cmake-pkg" the transient
-# defines would automatically be attached. Because the common func-
-# tionality is separate, but dependent upon the defines, it must be
-# explicitly extracted from the "llama" target.
-#
-get_target_property(_llama_transient_defines llama
-    INTERFACE_COMPILE_DEFINITIONS)
-
-target_compile_definitions(common PRIVATE "${_llama_transient_defines}")
-
-add_executable(${TARGET} ${CMAKE_CURRENT_LIST_DIR}/../main/main.cpp)
-target_include_directories(${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_17)

examples/main-cmake-pkg/README.md

@@ -1,31 +0,0 @@
-# llama.cpp/example/main-cmake-pkg
-
-This program builds [llama-cli](../main) using a relocatable CMake package. It serves as an example of using the `find_package()` CMake command to conveniently include [llama.cpp](https://github.com/ggerganov/llama.cpp) in projects which live outside of the source tree.
-
-## Building
-
-Because this example is "outside of the source tree", it is important to first build/install llama.cpp using CMake. An example is provided here, but please see the [llama.cpp build instructions](../..) for more detailed build instructions.
-
-### Considerations
-
-When hardware acceleration libraries are used (e.g. CUDA, Metal, etc.), CMake must be able to locate the associated CMake package.
-
-### Build llama.cpp and install to C:\LlamaCPP directory
-
-```cmd
-git clone https://github.com/ggerganov/llama.cpp
-cd llama.cpp
-cmake -B build -DBUILD_SHARED_LIBS=OFF -G "Visual Studio 17 2022" -A x64
-cmake --build build --config Release
-cmake --install build --prefix C:/LlamaCPP
-```
-
-### Build llama-cli-cmake-pkg
-
-
-```cmd
-cd ..\examples\main-cmake-pkg
-cmake -B build -DBUILD_SHARED_LIBS=OFF -DCMAKE_PREFIX_PATH="C:/LlamaCPP/lib/cmake/Llama" -G "Visual Studio 17 2022" -A x64
-cmake --build build --config Release
-cmake --install build --prefix C:/MyLlamaApp
-```

examples/main/README.md

@@ -1,6 +1,6 @@
 # llama.cpp/examples/main
 
-This example program allows you to use various LLaMA language models easily and efficiently. It is specifically designed to work with the [llama.cpp](https://github.com/ggerganov/llama.cpp) project, which provides a plain C/C++ implementation with optional 4-bit quantization support for faster, lower memory inference, and is optimized for desktop CPUs. This program can be used to perform various inference tasks with LLaMA models, including generating text based on user-provided prompts and chat-like interactions with reverse prompts.
+This example program allows you to use various LLaMA language models easily and efficiently. It is specifically designed to work with the [llama.cpp](https://github.com/ggml-org/llama.cpp) project, which provides a plain C/C++ implementation with optional 4-bit quantization support for faster, lower memory inference, and is optimized for desktop CPUs. This program can be used to perform various inference tasks with LLaMA models, including generating text based on user-provided prompts and chat-like interactions with reverse prompts.
 
 ## Table of Contents
 
@@ -37,7 +37,7 @@ Once downloaded, place your model in the models folder in llama.cpp.
 
 ##### Infinite text from a starting prompt (you can use `Ctrl-C` to stop it):
 ```bash
-./llama-cli -m models\gemma-1.1-7b-it.Q4_K_M.gguf --ignore-eos -n -1
+./llama-cli -m models/gemma-1.1-7b-it.Q4_K_M.gguf --ignore-eos -n -1
 ```
 
 ### Windows:
@@ -121,7 +121,7 @@ When --in-prefix or --in-suffix options are enabled the chat template ( --chat-t
 
 ### Chat templates
 
- `--chat-template JINJA_TEMPLATE`: This option sets a custom jinja chat template. It accepts a string, not a file name.  Default: template taken from model's metadata. Llama.cpp only supports [some pre-defined templates](https://github.com/ggerganov/llama.cpp/wiki/Templates-supported-by-llama_chat_apply_template). These include llama2, llama3, gemma, monarch, chatml, orion, vicuna, vicuna-orca, deepseek, command-r, zephyr. When --in-prefix or --in-suffix options are enabled the chat template ( --chat-template ) is disabled.
+ `--chat-template JINJA_TEMPLATE`: This option sets a custom jinja chat template. It accepts a string, not a file name.  Default: template taken from model's metadata. Llama.cpp only supports [some pre-defined templates](https://github.com/ggml-org/llama.cpp/wiki/Templates-supported-by-llama_chat_apply_template). These include llama2, llama3, gemma, monarch, chatml, orion, vicuna, vicuna-orca, deepseek, command-r, zephyr. When --in-prefix or --in-suffix options are enabled the chat template ( --chat-template ) is disabled.
 
  Example usage: `--chat-template gemma`
 
@@ -265,6 +265,14 @@ Being experimental and unique, XTC is disabled by default. The recommended combi
 
 Example usage: `--xtc-probability 0.5 --xtc-threshold 0.1`
 
+### Top-nσ Sampling
+
+-   `--top-nsigma N`: Limit the next token selection to a subset of tokens with pre-softmax logits that are within n * σ less than the max logit (default: -1, -1 = disabled).
+
+Top-nσ sampling is a text generation method that selects tokens based on a statistical threshold in pre-softmax logits. It works by only sampling from tokens with logits that are within n * σ of the maximum logit. This method helps maintain a stable sampling space regardless of temperature scaling, allowing it to perform well on reasoning tasks even in high temperatures. Without complex probability manipulation, it efficiently filters tokens directly on the pre-softmax logits. A higher value for top-nsigma (e.g., 5) will take more noisy tokens into consideration, while a lower value (e.g., 1) will focous on the more informative region of the sampling space.
+
+Example usage: `--top-nsigma 1`
+
 ### Logit Bias
 
 -   `-l TOKEN_ID(+/-)BIAS, --logit-bias TOKEN_ID(+/-)BIAS`: Modify the likelihood of a token appearing in the generated text completion.
@@ -310,9 +318,9 @@ These options help improve the performance and memory usage of the LLaMA models.
 
 ### Batch Size
 
--   `-b N, --batch-size N`: Set the batch size for prompt processing (default: `2048`). This large batch size benefits users who have BLAS installed and enabled it during the build. If you don't have BLAS enabled ("BLAS=0"), you can use a smaller number, such as 8, to see the prompt progress as it's evaluated in some situations.
+- `-ub N`, `--ubatch-size N`: Physical batch size. This is the maximum number of tokens that may be processed at a time. Increasing this value may improve performance during prompt processing, at the expense of higher memory usage. Default: `512`.
 
-- `-ub N`, `--ubatch-size N`: physical maximum batch size. This is for pipeline parallelization. Default: `512`.
+- `-b N`, `--batch-size N`: Logical batch size. Increasing this value above the value of the physical batch size may improve prompt processing performance when using multiple GPUs with pipeline parallelism. Default: `2048`.
 
 ### Prompt Caching
 

examples/main/main.cpp

@@ -4,6 +4,7 @@
 #include "log.h"
 #include "sampling.h"
 #include "llama.h"
+#include "chat-template.hpp"
 
 #include <cstdio>
 #include <cstring>
@@ -84,14 +85,6 @@ static void sigint_handler(int signo) {
 }
 #endif
 
-static std::string chat_add_and_format(struct llama_model * model, std::vector<common_chat_msg> & chat_msgs, const std::string & role, const std::string & content) {
-    common_chat_msg new_msg{role, content};
-    auto formatted = common_chat_format_single(model, g_params->chat_template, chat_msgs, new_msg, role == "user");
-    chat_msgs.push_back({role, content});
-    LOG_DBG("formatted: '%s'\n", formatted.c_str());
-    return formatted;
-}
-
 int main(int argc, char ** argv) {
     common_params params;
     g_params = &params;
@@ -165,6 +158,7 @@ int main(int argc, char ** argv) {
     }
 
     const llama_vocab * vocab = llama_model_get_vocab(model);
+    auto chat_templates = common_chat_templates_from_model(model, params.chat_template);
 
     LOG_INF("%s: llama threadpool init, n_threads = %d\n", __func__, (int) params.cpuparams.n_threads);
 
@@ -207,7 +201,7 @@ int main(int argc, char ** argv) {
     }
 
     // auto enable conversation mode if chat template is available
-    const bool has_chat_template = !common_get_builtin_chat_template(model).empty() || !params.chat_template.empty();
+    const bool has_chat_template = chat_templates.has_explicit_template && chat_templates.template_default;
     if (params.conversation_mode == COMMON_CONVERSATION_MODE_AUTO) {
         if (has_chat_template) {
             LOG_INF("%s: chat template is available, enabling conversation mode (disable it with -no-cnv)\n", __func__);
@@ -225,7 +219,7 @@ int main(int argc, char ** argv) {
     // print chat template example in conversation mode
     if (params.conversation_mode) {
         if (params.enable_chat_template) {
-            LOG_INF("%s: chat template example:\n%s\n", __func__, common_chat_format_example(model, params.chat_template).c_str());
+            LOG_INF("%s: chat template example:\n%s\n", __func__, common_chat_format_example(*chat_templates.template_default, params.use_jinja).c_str());
         } else {
             LOG_INF("%s: in-suffix/prefix is specified, chat template will be disabled\n", __func__);
         }
@@ -260,7 +254,7 @@ int main(int argc, char ** argv) {
         }
     }
 
-    const bool add_bos = llama_vocab_get_add_bos(vocab);
+    const bool add_bos = llama_vocab_get_add_bos(vocab) && !params.use_jinja;
     if (!llama_model_has_encoder(model)) {
         GGML_ASSERT(!llama_vocab_get_add_eos(vocab));
     }
@@ -269,10 +263,18 @@ int main(int argc, char ** argv) {
 
     std::vector<llama_token> embd_inp;
 
+    auto chat_add_and_format = [&chat_msgs, &chat_templates](const std::string & role, const std::string & content) {
+        common_chat_msg new_msg{role, content, {}};
+        auto formatted = common_chat_format_single(*chat_templates.template_default, chat_msgs, new_msg, role == "user", g_params->use_jinja);
+        chat_msgs.push_back({role, content, {}});
+        LOG_DBG("formatted: '%s'\n", formatted.c_str());
+        return formatted;
+    };
+
     {
         auto prompt = (params.conversation_mode && params.enable_chat_template)
             // format the system prompt in conversation mode (fallback to default if empty)
-            ? chat_add_and_format(model, chat_msgs, "system", params.prompt.empty() ? DEFAULT_SYSTEM_MESSAGE : params.prompt)
+            ? chat_add_and_format("system", params.prompt.empty() ? DEFAULT_SYSTEM_MESSAGE : params.prompt)
             // otherwise use the prompt as is
             : params.prompt;
         if (params.interactive_first || !params.prompt.empty() || session_tokens.empty()) {
@@ -501,12 +503,14 @@ int main(int argc, char ** argv) {
 
     std::vector<llama_token> embd;
 
-    // tokenized antiprompts
-    std::vector<std::vector<llama_token>> antiprompt_ids;
+    // single-token antiprompts
+    std::vector<llama_token> antiprompt_token;
 
-    antiprompt_ids.reserve(params.antiprompt.size());
     for (const std::string & antiprompt : params.antiprompt) {
-        antiprompt_ids.emplace_back(::common_tokenize(ctx, antiprompt, false, true));
+        auto ids = ::common_tokenize(ctx, antiprompt, false, true);
+        if (ids.size() == 1) {
+            antiprompt_token.push_back(ids[0]);
+        }
     }
 
     if (llama_model_has_encoder(model)) {
@@ -751,14 +755,11 @@ int main(int argc, char ** argv) {
 
                 // check for reverse prompt using special tokens
                 llama_token last_token = common_sampler_last(smpl);
-                for (std::vector<llama_token> ids : antiprompt_ids) {
-                    if (ids.size() == 1 && last_token == ids[0]) {
-                        if (params.interactive) {
-                            is_interacting = true;
-                        }
-                        is_antiprompt = true;
-                        break;
+                if (std::find(antiprompt_token.begin(), antiprompt_token.end(), last_token) != antiprompt_token.end()) {
+                    if (params.interactive) {
+                        is_interacting = true;
                     }
+                    is_antiprompt = true;
                 }
 
                 if (is_antiprompt) {
@@ -779,7 +780,7 @@ int main(int argc, char ** argv) {
                     }
 
                     if (params.enable_chat_template) {
-                        chat_add_and_format(model, chat_msgs, "assistant", assistant_ss.str());
+                        chat_add_and_format("assistant", assistant_ss.str());
                     }
                     is_interacting = true;
                     LOG("\n");
@@ -844,7 +845,7 @@ int main(int argc, char ** argv) {
 
                     bool format_chat = params.conversation_mode && params.enable_chat_template;
                     std::string user_inp = format_chat
-                        ? chat_add_and_format(model, chat_msgs, "user", std::move(buffer))
+                        ? chat_add_and_format("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 = common_tokenize(ctx, params.input_prefix, false, true);

examples/passkey/README.md

@@ -5,8 +5,8 @@ models ability to recall information from long contexts.
 
 See the following PRs for more info:
 
-- https://github.com/ggerganov/llama.cpp/pull/3856
-- https://github.com/ggerganov/llama.cpp/pull/4810
+- https://github.com/ggml-org/llama.cpp/pull/3856
+- https://github.com/ggml-org/llama.cpp/pull/4810
 
 ### Usage
 

examples/perplexity/perplexity.cpp

@@ -3,6 +3,7 @@
 #include "log.h"
 #include "llama.h"
 
+#include <chrono>
 #include <algorithm>
 #include <array>
 #include <atomic>

examples/pydantic_models_to_grammar_examples.py

@@ -23,7 +23,7 @@ def create_completion(host, prompt, gbnf_grammar):
     """Calls the /completion API on llama-server.
 
     See
-    https://github.com/ggerganov/llama.cpp/tree/HEAD/examples/server#api-endpoints
+    https://github.com/ggml-org/llama.cpp/tree/HEAD/examples/server#api-endpoints
     """
     print(f"  Request:\n    Grammar:\n{textwrap.indent(gbnf_grammar, '      ')}\n    Prompt:\n{textwrap.indent(prompt.rstrip(), '      ')}")
     headers = {"Content-Type": "application/json"}

examples/quantize/README.md

@@ -69,22 +69,22 @@ Several quantization methods are supported. They differ in the resulting model d
 |   13B | ms/tok @ 8th |      - |     73 |     82 |     98 |    105 |    128 |
 |   13B | bits/weight  |   16.0 |    4.5 |    5.0 |    5.5 |    6.0 |    8.5 |
 
-- [k-quants](https://github.com/ggerganov/llama.cpp/pull/1684)
+- [k-quants](https://github.com/ggml-org/llama.cpp/pull/1684)
 - recent k-quants improvements and new i-quants
-  - [#2707](https://github.com/ggerganov/llama.cpp/pull/2707)
-  - [#2807](https://github.com/ggerganov/llama.cpp/pull/2807)
-  - [#4773 - 2-bit i-quants (inference)](https://github.com/ggerganov/llama.cpp/pull/4773)
-  - [#4856 - 2-bit i-quants (inference)](https://github.com/ggerganov/llama.cpp/pull/4856)
-  - [#4861 - importance matrix](https://github.com/ggerganov/llama.cpp/pull/4861)
-  - [#4872 - MoE models](https://github.com/ggerganov/llama.cpp/pull/4872)
-  - [#4897 - 2-bit quantization](https://github.com/ggerganov/llama.cpp/pull/4897)
-  - [#4930 - imatrix for all k-quants](https://github.com/ggerganov/llama.cpp/pull/4930)
-  - [#4951 - imatrix on the GPU](https://github.com/ggerganov/llama.cpp/pull/4957)
-  - [#4969 - imatrix for legacy quants](https://github.com/ggerganov/llama.cpp/pull/4969)
-  - [#4996 - k-quants tuning](https://github.com/ggerganov/llama.cpp/pull/4996)
-  - [#5060 - Q3_K_XS](https://github.com/ggerganov/llama.cpp/pull/5060)
-  - [#5196 - 3-bit i-quants](https://github.com/ggerganov/llama.cpp/pull/5196)
-  - [quantization tuning](https://github.com/ggerganov/llama.cpp/pull/5320), [another one](https://github.com/ggerganov/llama.cpp/pull/5334), and [another one](https://github.com/ggerganov/llama.cpp/pull/5361)
+  - [#2707](https://github.com/ggml-org/llama.cpp/pull/2707)
+  - [#2807](https://github.com/ggml-org/llama.cpp/pull/2807)
+  - [#4773 - 2-bit i-quants (inference)](https://github.com/ggml-org/llama.cpp/pull/4773)
+  - [#4856 - 2-bit i-quants (inference)](https://github.com/ggml-org/llama.cpp/pull/4856)
+  - [#4861 - importance matrix](https://github.com/ggml-org/llama.cpp/pull/4861)
+  - [#4872 - MoE models](https://github.com/ggml-org/llama.cpp/pull/4872)
+  - [#4897 - 2-bit quantization](https://github.com/ggml-org/llama.cpp/pull/4897)
+  - [#4930 - imatrix for all k-quants](https://github.com/ggml-org/llama.cpp/pull/4930)
+  - [#4951 - imatrix on the GPU](https://github.com/ggml-org/llama.cpp/pull/4957)
+  - [#4969 - imatrix for legacy quants](https://github.com/ggml-org/llama.cpp/pull/4969)
+  - [#4996 - k-quants tuning](https://github.com/ggml-org/llama.cpp/pull/4996)
+  - [#5060 - Q3_K_XS](https://github.com/ggml-org/llama.cpp/pull/5060)
+  - [#5196 - 3-bit i-quants](https://github.com/ggml-org/llama.cpp/pull/5196)
+  - [quantization tuning](https://github.com/ggml-org/llama.cpp/pull/5320), [another one](https://github.com/ggml-org/llama.cpp/pull/5334), and [another one](https://github.com/ggml-org/llama.cpp/pull/5361)
 
 **Llama 2 7B**
 

examples/quantize/tests.sh

@@ -47,7 +47,7 @@ echo PASS
 echo
 
 # 3a. Test the requanted model is loading properly
-$MAIN --model $WORK_PATH/ggml-model-requant-00001-of-00006.gguf --n-predict 32
+$MAIN -no-cnv --model $WORK_PATH/ggml-model-requant-00001-of-00006.gguf --n-predict 32
 echo PASS
 echo
 
@@ -57,7 +57,7 @@ echo PASS
 echo
 
 # 4b. Test the requanted model is loading properly
-$MAIN --model $WORK_PATH/ggml-model-requant-merge.gguf --n-predict 32
+$MAIN -no-cnv --model $WORK_PATH/ggml-model-requant-merge.gguf --n-predict 32
 echo PASS
 echo