Rebase b9840

.flake8

@@ -15,4 +15,5 @@ exclude =
     build,
     # This contains builds that we don't want to check
     dist  # This is generated with `python build .` for package releases
+    scripts/tune
 # max-complexity = 10

.github/CODEOWNERS

@@ -0,0 +1 @@
+* @tetherto/qvac-internal-merge @tetherto/qvac-internal-mgmt

.github/workflows/build-apple.yml

@@ -100,6 +100,7 @@ jobs:
             -DLLAMA_BUILD_BORINGSSL=ON \
             -DGGML_METAL=OFF \
             -DGGML_RPC=ON \
+            -DGGML_BLAS=OFF \
             -DCMAKE_OSX_DEPLOYMENT_TARGET=13.3
           time cmake --build build --config Release -j $(sysctl -n hw.logicalcpu)
 

.github/workflows/build-cann.yml

@@ -34,71 +34,69 @@ env:
   LLAMA_ARG_LOG_TIMESTAMPS: 1
 
 jobs:
-# TODO: this build is disabled to save Github Actions resources (https://github.com/ggml-org/llama.cpp/pull/23705)
-#       in order to enable it again, we have to provision dedicated runners  to run it
-#  openEuler-latest-cann:
-#    defaults:
-#      run:
-#        shell: bash -el {0}
-#    strategy:
-#      matrix:
-#        arch: [x86, aarch64]
-#        chip_type: ['910b', '310p']
-#        build: ['Release']
-#        use_acl_graph: ['on', 'off']
-#        exclude:
-#          # 310P does not support USE_ACL_GRAPH=on
-#          - chip_type: '310p'
-#            use_acl_graph: 'on'
-#    runs-on: ${{ matrix.arch == 'aarch64' && 'ubuntu-24.04-arm' || 'ubuntu-24.04' }}
-#    steps:
-#      - name: Checkout
-#        uses: actions/checkout@v6
-#        with:
-#          fetch-depth: 0
-#
-#      - name: Free up disk space
-#        uses: ggml-org/free-disk-space@v1.3.1
-#        with:
-#          tool-cache: true
-#
-#      - name: Set container image
-#        id: cann-image
-#        run: |
-#          image="ascendai/cann:${{ matrix.chip_type == '910b' &&  '8.5.0-910b-openeuler24.03-py3.11' || '8.5.0-310p-openeuler24.03-py3.11' }}"
-#          echo "image=${image}" >> "${GITHUB_OUTPUT}"
-#
-#      - name: Pull container image
-#        run: docker pull "${{ steps.cann-image.outputs.image }}"
-#
-#      - name: Build
-#        env:
-#          BUILD_TYPE: ${{ matrix.build }}
-#          SOC_TYPE: ascend${{ matrix.chip_type }}
-#          USE_ACL_GRAPH: ${{ matrix.use_acl_graph }}
-#        run: |
-#          HOST_UID=$(id -u)
-#          HOST_GID=$(id -g)
-#
-#          docker run --rm \
-#            -v "${PWD}:/workspace" \
-#            -w /workspace \
-#            -e SOC_TYPE=${SOC_TYPE} \
-#            -e BUILD_TYPE=${BUILD_TYPE} \
-#            -e USE_ACL_GRAPH=${USE_ACL_GRAPH} \
-#            "${{ steps.cann-image.outputs.image }}" \
-#            bash -lc '
-#              set -e
-#              yum install -y --setopt=install_weak_deps=False --setopt=tsflags=nodocs git gcc gcc-c++ make cmake openssl-devel
-#              yum clean all && rm -rf /var/cache/yum
-#              git config --global --add safe.directory "/workspace"
-#              export LD_LIBRARY_PATH=${ASCEND_TOOLKIT_HOME}/lib64:${ASCEND_TOOLKIT_HOME}/$(uname -m)-linux/devlib/:${LD_LIBRARY_PATH}
-#              cmake -S . -B build \
-#                  -DCMAKE_BUILD_TYPE=${BUILD_TYPE} \
-#                  -DGGML_CANN=on \
-#                  -DSOC_TYPE=${SOC_TYPE} \
-#                  -DUSE_ACL_GRAPH=${USE_ACL_GRAPH}
-#              cmake --build build -j $(nproc)
-#
-#              chown -R '"${HOST_UID}"':'"${HOST_GID}"' /workspace/build
-#            '
+ openEuler-latest-cann:
+   defaults:
+     run:
+       shell: bash -el {0}
+   strategy:
+     matrix:
+       arch: [x86, aarch64]
+       chip_type: ['910b', '310p']
+       build: ['Release']
+       use_acl_graph: ['on', 'off']
+       exclude:
+         # 310P does not support USE_ACL_GRAPH=on
+         - chip_type: '310p'
+           use_acl_graph: 'on'
+   runs-on: ${{ matrix.arch == 'aarch64' && 'ubuntu-24.04-arm' || 'ubuntu-24.04' }}
+   steps:
+     - name: Checkout
+       uses: actions/checkout@v6
+       with:
+         fetch-depth: 0
+
+     - name: Free up disk space
+       uses: ggml-org/free-disk-space@v1.3.1
+       with:
+         tool-cache: true
+
+     - name: Set container image
+       id: cann-image
+       run: |
+         image="ascendai/cann:${{ matrix.chip_type == '910b' &&  '8.5.0-910b-openeuler24.03-py3.11' || '8.5.0-310p-openeuler24.03-py3.11' }}"
+         echo "image=${image}" >> "${GITHUB_OUTPUT}"
+
+     - name: Pull container image
+       run: docker pull "${{ steps.cann-image.outputs.image }}"
+
+     - name: Build
+       env:
+         BUILD_TYPE: ${{ matrix.build }}
+         SOC_TYPE: ascend${{ matrix.chip_type }}
+         USE_ACL_GRAPH: ${{ matrix.use_acl_graph }}
+       run: |
+         HOST_UID=$(id -u)
+         HOST_GID=$(id -g)
+
+         docker run --rm \
+           -v "${PWD}:/workspace" \
+           -w /workspace \
+           -e SOC_TYPE=${SOC_TYPE} \
+           -e BUILD_TYPE=${BUILD_TYPE} \
+           -e USE_ACL_GRAPH=${USE_ACL_GRAPH} \
+           "${{ steps.cann-image.outputs.image }}" \
+           bash -lc '
+             set -e
+             yum install -y --setopt=install_weak_deps=False --setopt=tsflags=nodocs git gcc gcc-c++ make cmake openssl-devel
+             yum clean all && rm -rf /var/cache/yum
+             git config --global --add safe.directory "/workspace"
+             export LD_LIBRARY_PATH=${ASCEND_TOOLKIT_HOME}/lib64:${ASCEND_TOOLKIT_HOME}/$(uname -m)-linux/devlib/:${LD_LIBRARY_PATH}
+             cmake -S . -B build \
+                 -DCMAKE_BUILD_TYPE=${BUILD_TYPE} \
+                 -DGGML_CANN=on \
+                 -DSOC_TYPE=${SOC_TYPE} \
+                 -DUSE_ACL_GRAPH=${USE_ACL_GRAPH}
+             cmake --build build -j $(nproc)
+
+             chown -R '"${HOST_UID}"':'"${HOST_GID}"' /workspace/build
+           '

.github/workflows/build-cmake-pkg.yml

@@ -25,7 +25,7 @@ jobs:
           cmake --build build --config Release
           cmake --install build --prefix "$PREFIX" --config Release
 
-          export LLAMA_CONFIG="$PREFIX"/lib/cmake/llama/llama-config.cmake
+          export LLAMA_CONFIG="$PREFIX"/share/llama/llama-config.cmake
           tclsh <<'EOF'
           set build(commit)  [string trim [exec git rev-parse --short HEAD]]
           set build(number)  [string trim [exec git rev-list  --count HEAD]]
@@ -47,5 +47,5 @@ jobs:
           EOF
 
           cd examples/simple-cmake-pkg
-          cmake -S . -B build -DCMAKE_PREFIX_PATH="$PREFIX"/lib/cmake
+          cmake -S . -B build -DCMAKE_PREFIX_PATH="$PREFIX"/share
           cmake --build build

.github/workflows/build-cpu.yml

@@ -7,7 +7,10 @@ on:
       - master
     paths: [
       '.github/workflows/build-cpu.yml',
-      '.github/workflows/build-cmake-pkg.yml',
+      # Disable. There are some modifications in the fork
+      # so that ggml dynamic builds work with vcpkg.
+      # '.github/workflows/build-cmake-pkg.yml',
+      '.github/workflows/build-linux-cross.yml',
       '**/CMakeLists.txt',
       '**/.cmake',
       '**/*.h',
@@ -20,7 +23,10 @@ on:
     types: [opened, synchronize, reopened]
     paths: [
       '.github/workflows/build-cpu.yml',
-      '.github/workflows/build-cmake-pkg.yml',
+      # Disable. There are some modifications in the fork
+      # so that ggml dynamic builds work with vcpkg.
+      # '.github/workflows/build-cmake-pkg.yml',
+      '.github/workflows/build-linux-cross.yml',
       '**/CMakeLists.txt',
       '**/.cmake',
       '**/*.h',
@@ -41,8 +47,8 @@ env:
   LLAMA_ARG_LOG_TIMESTAMPS: 1
 
 jobs:
-  build-cmake-pkg:
-    uses: ./.github/workflows/build-cmake-pkg.yml
+  # build-cmake-pkg:
+    # uses: ./.github/workflows/build-cmake-pkg.yml
 
   ubuntu:
     strategy:
@@ -60,10 +66,17 @@ jobs:
         id: checkout
         uses: actions/checkout@v6
 
+      - name: CPU feature hash
+        id: cpu_info
+        run: |
+          CPU_HASH=$(echo | gcc -march=native -dM -E - 2>/dev/null | sort | md5sum | cut -d' ' -f1)
+          echo "cpu_hash=${CPU_HASH}" >> "$GITHUB_OUTPUT"
+          echo "CPU feature hash: ${CPU_HASH}"
+
       - name: ccache
         uses: ggml-org/ccache-action@v1.2.21
         with:
-          key: cpu-${{ matrix.os }}
+          key: cpu-${{ matrix.os }}-${{ steps.cpu_info.outputs.cpu_hash }}
           evict-old-files: 1d
           save: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
 

.github/workflows/build-cross.yml

@@ -21,6 +21,7 @@ concurrency:
 
 
 jobs:
+  # Disabled. Fails to install some dependencies from arch-specific repositories.
   # ubuntu-24-riscv64-cpu-cross:
   #   runs-on: ubuntu-24.04
 

.github/workflows/build-cuda-ubuntu.yml

@@ -45,6 +45,11 @@ jobs:
         id: checkout
         uses: actions/checkout@v6
 
+      - name: Setup Node.js (build the embedded web UI from source)
+        uses: actions/setup-node@v4
+        with:
+          node-version: 22
+
       - name: Install dependencies
         env:
           DEBIAN_FRONTEND: noninteractive
@@ -81,6 +86,11 @@ jobs:
         id: checkout
         uses: actions/checkout@v6
 
+      - name: Setup Node.js (build the embedded web UI from source)
+        uses: actions/setup-node@v4
+        with:
+          node-version: 22
+
       - name: Dependencies
         id: depends
         run: |
@@ -113,6 +123,11 @@ jobs:
         id: checkout
         uses: actions/checkout@v6
 
+      - name: Setup Node.js (build the embedded web UI from source)
+        uses: actions/setup-node@v4
+        with:
+          node-version: 22
+
       - name: Dependencies
         id: depends
         run: |

.github/workflows/build-opencl.yml

@@ -80,3 +80,57 @@ jobs:
         run: |
           cmake -S . -B build -G "Ninja Multi-Config" -D CMAKE_TOOLCHAIN_FILE=cmake/arm64-windows-llvm.cmake -DCMAKE_PREFIX_PATH="$env:RUNNER_TEMP/opencl-arm64-release" -DGGML_OPENCL=ON -DGGML_OPENCL_USE_ADRENO_KERNELS=ON -DLLAMA_BUILD_BORINGSSL=ON
           cmake --build build --config Release -j ${env:NUMBER_OF_PROCESSORS}
+
+  ubuntu-24-opencl:
+    runs-on: ubuntu-24.04
+
+    steps:
+      - name: Clone
+        id: checkout
+        uses: actions/checkout@v6
+
+      - name: ccache
+        uses: ggml-org/ccache-action@v1.2.21
+        with:
+          key: ubuntu-24-opencl
+          evict-old-files: 1d
+
+      - name: Dependencies
+        id: depends
+        run: |
+          sudo apt-get update
+          # intel-opencl-icd provides a CPU OpenCL device (same package set the
+          # OpenVINO job installs); the rest is the ICD loader + CL headers.
+          sudo apt-get install -y --no-install-recommends \
+            build-essential cmake clinfo \
+            ocl-icd-opencl-dev opencl-headers opencl-clhpp-headers intel-opencl-icd
+
+      - name: Build (GGML_OPENCL, portable kernels)
+        id: cmake_build
+        run: |
+          # NOTE: no -DLLAMA_FATAL_WARNINGS=ON — upstream ggml-opencl.cpp has
+          # intentional (un-annotated) case fall-throughs that GCC's
+          # -Wimplicit-fallthrough would turn into errors. This job's purpose is
+          # to run test-backend-ops, not to gate upstream warning-cleanliness.
+          cmake -B build \
+            -DGGML_OPENCL=ON \
+            -DLLAMA_BUILD_TOOLS=OFF \
+            -DLLAMA_BUILD_EXAMPLES=OFF \
+            -DLLAMA_BUILD_SERVER=OFF
+          time cmake --build build --config Release -j $(nproc) --target test-backend-ops
+
+      - name: Test backend ops (OpenCL)
+        id: test_backend_ops
+        run: |
+          # Functional coverage for the ggml-opencl ops DocTR uses on Adreno:
+          # CONV_2D (regular direct conv), CONV_2D_DW (depthwise), POOL_2D,
+          # HARDSWISH, HARDSIGMOID. These kernels load unconditionally (no
+          # GGML_OPENCL_USE_ADRENO_KERNELS gate), so they are portable OpenCL C
+          # and run here on the Intel CPU ICD, compared against the CPU
+          # reference. NOTE: a CPU ICD does NOT reproduce Adreno-compiler bugs
+          # (e.g. the char->float type-punning miscompile this PR works around) —
+          # that requires the manual Firebase Test Lab device run. This job is a
+          # regression net for kernel math / indexing / dtype correctness that a
+          # compile-only job cannot catch.
+          clinfo -l || true
+          ./build/bin/test-backend-ops test -o CONV_2D,CONV_2D_DW,POOL_2D,HARDSWISH,HARDSIGMOID

.github/workflows/build-vulkan.yml

@@ -48,7 +48,7 @@ jobs:
         id: depends
         run: |
           sudo apt-get update
-          sudo apt-get install -y gcc-14 g++-14 build-essential glslc libvulkan-dev spirv-headers libssl-dev ninja-build
+          sudo apt-get install -y gcc-14 g++-14 build-essential glslc libvulkan-dev spirv-headers libssl-dev ninja-build cmake
           echo "CC=gcc-14" >> "$GITHUB_ENV"
           echo "CXX=g++-14" >> "$GITHUB_ENV"
 
@@ -129,6 +129,33 @@ jobs:
           export GGML_VK_VISIBLE_DEVICES=0
           export GGML_VK_DISABLE_F16=1
           export GGML_VK_DISABLE_COOPMAT=1
+          # NVIDIA Tesla T4 with driver 570.x has a known bug where any test
+          # that initializes the Vulkan backend can non-deterministically
+          # segfault during exit/cleanup inside libnvidia-gpucomp.so atexit
+          # handlers racing with the driver's [vkps] Update thread. All test
+          # cases pass but the process may crash on teardown. Upstream llama.cpp
+          # disabled their T4 Vulkan CI node entirely for the same reason.
+          # Workaround: on T4, tolerate SegFault-only failures since the crash
+          # is in the driver's exit path, not in the test logic.
+          # Refs: ggml-org/llama.cpp#10528, ggml-org/llama.cpp#10989
+          IS_T4=false
+          if nvidia-smi --query-gpu=name --format=csv,noheader 2>/dev/null | grep -q "T4"; then
+            IS_T4=true
+            echo "::notice::Detected NVIDIA T4 -- will tolerate segfault-on-exit (known driver bug)"
+            # FLASH_ATTN_EXT causes GPU shader hangs on T4 (Turing) FlashAttention path
+            export GGML_SKIP_FLASH_ATTN=1
+          fi
+          # Old cmdline for ctest
+          # ctest -L main --verbose --timeout 4200 2>&1 | tee ctest_output.log
           # This is using llvmpipe and runs slower than other backends
           # test-backend-ops is too slow on llvmpipe, skip it
-          ctest -L main -E test-backend-ops --verbose --timeout 900
+          ctest -L main -E test-backend-ops --verbose --timeout 900 2>&1 | tee ctest_output.log
+          CTEST_EXIT=${PIPESTATUS[0]}
+          if [ "$CTEST_EXIT" -ne 0 ] && [ "$IS_T4" = true ]; then
+            NON_SEGFAULT=$(sed -n '/The following tests FAILED:/,$p' ctest_output.log | grep -E '^\s+[0-9]' | grep -vic 'segfault' || true)
+            if [ "$NON_SEGFAULT" -eq 0 ]; then
+              echo "::warning::All test failures are SegFault-on-exit (NVIDIA T4 driver bug) -- treating as success"
+              exit 0
+            fi
+          fi
+          exit $CTEST_EXIT