cmake : enable curl by default (#12761 )

* cmake : enable curl by default * no curl if no examples * fix build * fix build-linux-cross * add windows-setup-curl * fix * shell * fix path * fix windows-latest-cmake* * run: include_directories * LLAMA_RUN_EXTRA_LIBS * sycl: no llama_curl * no test-arg-parser on windows * clarification * try riscv64 / arm64 * windows: include libcurl inside release binary * add msg * fix mac / ios / android build * will this fix xcode? * try clearing the cache * add bunch of licenses * revert clear cache * fix xcode * fix xcode (2) * fix typo
CANN: fix typo in ggml-cann (#12733 )
2026-05-07 17:44:09 +00:00 · 2025-04-07 13:35:19 +02:00 · 2025-04-07 19:34:14 +08:00 · 2025-04-07 17:10:36 +08:00 · 2025-04-06 15:23:54 +02:00
29 changed files with 727 additions and 1303 deletions
--- a/.devops/cpu.Dockerfile
+++ b/.devops/cpu.Dockerfile
@@ -14,9 +14,9 @@ WORKDIR /app
 COPY . .

 RUN if [ "$TARGETARCH" = "amd64" ]; then \
-        cmake -S . -B build -DCMAKE_BUILD_TYPE=Release -DLLAMA_CURL=ON -DGGML_NATIVE=OFF -DGGML_BACKEND_DL=ON -DGGML_CPU_ALL_VARIANTS=ON; \
+        cmake -S . -B build -DCMAKE_BUILD_TYPE=Release -DGGML_NATIVE=OFF -DGGML_BACKEND_DL=ON -DGGML_CPU_ALL_VARIANTS=ON; \
    elif [ "$TARGETARCH" = "arm64" ]; then \
-        cmake -S . -B build -DCMAKE_BUILD_TYPE=Release -DLLAMA_CURL=ON -DGGML_NATIVE=OFF -DGGML_CPU_ARM_ARCH=${GGML_CPU_ARM_ARCH}; \
+        cmake -S . -B build -DCMAKE_BUILD_TYPE=Release -DGGML_NATIVE=OFF -DGGML_CPU_ARM_ARCH=${GGML_CPU_ARM_ARCH}; \
    else \
        echo "Unsupported architecture"; \
        exit 1; \
--- a/.devops/cuda.Dockerfile
+++ b/.devops/cuda.Dockerfile
@@ -21,7 +21,7 @@ COPY . .
 RUN if [ "${CUDA_DOCKER_ARCH}" != "default" ]; then \
    export CMAKE_ARGS="-DCMAKE_CUDA_ARCHITECTURES=${CUDA_DOCKER_ARCH}"; \
    fi && \
-    cmake -B build -DGGML_NATIVE=OFF -DGGML_CUDA=ON -DLLAMA_CURL=ON ${CMAKE_ARGS} -DCMAKE_EXE_LINKER_FLAGS=-Wl,--allow-shlib-undefined . && \
+    cmake -B build -DGGML_NATIVE=OFF -DGGML_CUDA=ON ${CMAKE_ARGS} -DCMAKE_EXE_LINKER_FLAGS=-Wl,--allow-shlib-undefined . && \
    cmake --build build --config Release -j$(nproc)

 RUN mkdir -p /app/lib && \
--- a/.devops/intel.Dockerfile
+++ b/.devops/intel.Dockerfile
@@ -17,7 +17,7 @@ RUN if [ "${GGML_SYCL_F16}" = "ON" ]; then \
        && export OPT_SYCL_F16="-DGGML_SYCL_F16=ON"; \
    fi && \
    echo "Building with dynamic libs" && \
-    cmake -B build -DGGML_NATIVE=OFF -DGGML_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx -DLLAMA_CURL=ON ${OPT_SYCL_F16} && \
+    cmake -B build -DGGML_NATIVE=OFF -DGGML_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx ${OPT_SYCL_F16} && \
    cmake --build build --config Release -j$(nproc)

 RUN mkdir -p /app/lib && \
--- a/.devops/musa.Dockerfile
+++ b/.devops/musa.Dockerfile
@@ -35,7 +35,7 @@ COPY . .
 RUN if [ "${MUSA_DOCKER_ARCH}" != "default" ]; then \
        export CMAKE_ARGS="-DMUSA_ARCHITECTURES=${MUSA_DOCKER_ARCH}"; \
    fi && \
-    cmake -B build -DGGML_NATIVE=OFF -DGGML_MUSA=ON -DLLAMA_CURL=ON ${CMAKE_ARGS} -DCMAKE_EXE_LINKER_FLAGS=-Wl,--allow-shlib-undefined . && \
+    cmake -B build -DGGML_NATIVE=OFF -DGGML_MUSA=ON ${CMAKE_ARGS} -DCMAKE_EXE_LINKER_FLAGS=-Wl,--allow-shlib-undefined . && \
    cmake --build build --config Release -j$(nproc)

 RUN mkdir -p /app/lib && \
--- a/.devops/rocm.Dockerfile
+++ b/.devops/rocm.Dockerfile
@@ -40,7 +40,7 @@ WORKDIR /app
 COPY . .

 RUN HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" \
-    cmake -S . -B build -DGGML_HIP=ON -DAMDGPU_TARGETS=$ROCM_DOCKER_ARCH -DCMAKE_BUILD_TYPE=Release -DLLAMA_CURL=ON \
+    cmake -S . -B build -DGGML_HIP=ON -DAMDGPU_TARGETS=$ROCM_DOCKER_ARCH -DCMAKE_BUILD_TYPE=Release \
    && cmake --build build --config Release -j$(nproc)

 RUN mkdir -p /app/lib \
--- a/.github/actions/windows-setup-curl/action.yml
+++ b/.github/actions/windows-setup-curl/action.yml
@@ -0,0 +1,25 @@
+name: 'Windows - Setup CURL'
+description: 'Composite action, to be reused in other workflow'
+inputs:
+  curl_version:
+    description: 'CURL version'
+    required: false
+    default: '8.6.0_6'
+outputs:
+  curl_path:
+    description: "Path to the downloaded libcurl"
+    value: ${{ steps.get_libcurl.outputs.curl_path }}
+
+runs:
+  using: "composite"
+  steps:
+    - name: libCURL
+      id: get_libcurl
+      shell: powershell
+      env:
+        CURL_VERSION: ${{ inputs.curl_version }}
+      run: |
+        curl.exe -o $env:RUNNER_TEMP/curl.zip -L "https://curl.se/windows/dl-${env:CURL_VERSION}/curl-${env:CURL_VERSION}-win64-mingw.zip"
+        mkdir $env:RUNNER_TEMP/libcurl
+        tar.exe -xvf $env:RUNNER_TEMP/curl.zip --strip-components=1 -C $env:RUNNER_TEMP/libcurl
+        echo "curl_path=$env:RUNNER_TEMP/libcurl" >> $env:GITHUB_OUTPUT
--- a/.github/workflows/bench.yml.disabled
+++ b/.github/workflows/bench.yml.disabled
@@ -104,7 +104,6 @@ jobs:
          cmake -B build \
              -DGGML_NATIVE=OFF \
              -DLLAMA_BUILD_SERVER=ON \
-              -DLLAMA_CURL=ON \
              -DLLAMA_CUBLAS=ON \
              -DCUDAToolkit_ROOT=/usr/local/cuda \
              -DCMAKE_CUDA_COMPILER=/usr/local/cuda/bin/nvcc \
--- a/.github/workflows/build-linux-cross.yml
+++ b/.github/workflows/build-linux-cross.yml
@@ -19,7 +19,8 @@ jobs:
          sudo apt-get install -y --no-install-recommends \
                  build-essential \
                  gcc-14-riscv64-linux-gnu \
-                  g++-14-riscv64-linux-gnu
+                  g++-14-riscv64-linux-gnu \
+                  libcurl4-openssl-dev:riscv64

      - name: Build
        run: |
@@ -59,7 +60,8 @@ jobs:
                  glslc \
                  gcc-14-riscv64-linux-gnu \
                  g++-14-riscv64-linux-gnu \
-                  libvulkan-dev:riscv64
+                  libvulkan-dev:riscv64 \
+                  libcurl4-openssl-dev:riscv64

      - name: Build
        run: |
@@ -99,7 +101,8 @@ jobs:
                  build-essential \
                  glslc \
                  crossbuild-essential-arm64 \
-                  libvulkan-dev:arm64
+                  libvulkan-dev:arm64 \
+                  libcurl4-openssl-dev:arm64

      - name: Build
        run: |
--- a/.github/workflows/build.yml
+++ b/.github/workflows/build.yml
@@ -54,6 +54,7 @@ jobs:
        continue-on-error: true
        run: |
          brew update
+          brew install curl

      - name: Build
        id: cmake_build
@@ -62,7 +63,6 @@ jobs:
          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
@@ -92,7 +92,6 @@ 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
@@ -123,6 +122,7 @@ jobs:
        continue-on-error: true
        run: |
          brew update
+          brew install curl

      - name: Build
        id: cmake_build
@@ -133,7 +133,6 @@ jobs:
          cmake -B build \
            -DCMAKE_BUILD_RPATH="@loader_path" \
            -DLLAMA_FATAL_WARNINGS=ON \
-            -DLLAMA_CURL=ON \
            -DGGML_METAL=OFF \
            -DGGML_RPC=ON
          cmake --build build --config Release -j $(sysctl -n hw.logicalcpu)
@@ -162,7 +161,6 @@ 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
@@ -207,7 +205,6 @@ jobs:
        run: |
          cmake -B build \
            -DLLAMA_FATAL_WARNINGS=ON \
-            -DLLAMA_CURL=ON \
            -DGGML_RPC=ON
          cmake --build build --config Release -j $(nproc)

@@ -246,7 +243,6 @@ 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-${{ matrix.build }}.zip ./build/bin/*

      - name: Upload artifacts
@@ -281,7 +277,7 @@ jobs:
        id: depends
        run: |
          sudo apt-get update
-          sudo apt-get install build-essential
+          sudo apt-get install build-essential libcurl4-openssl-dev

      - name: Build
        id: cmake_build
@@ -322,7 +318,7 @@ jobs:
        id: depends
        run: |
          sudo apt-get update
-          sudo apt-get install build-essential
+          sudo apt-get install build-essential libcurl4-openssl-dev

      - name: Build
        id: cmake_build
@@ -360,7 +356,7 @@ jobs:
        id: depends
        run: |
          sudo apt-get update
-          sudo apt-get install build-essential
+          sudo apt-get install build-essential libcurl4-openssl-dev

      - name: Build
        id: cmake_build
@@ -397,7 +393,7 @@ jobs:
          wget -qO - https://packages.lunarg.com/lunarg-signing-key-pub.asc | sudo apt-key add -
          sudo wget -qO /etc/apt/sources.list.d/lunarg-vulkan-jammy.list https://packages.lunarg.com/vulkan/lunarg-vulkan-jammy.list
          sudo apt-get update -y
-          sudo apt-get install -y build-essential mesa-vulkan-drivers vulkan-sdk
+          sudo apt-get install -y build-essential mesa-vulkan-drivers vulkan-sdk libcurl4-openssl-dev

      - name: Build
        id: cmake_build
@@ -431,7 +427,6 @@ 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-vulkan-x64.zip ./build/bin/*

      - name: Upload artifacts
@@ -454,7 +449,7 @@ jobs:
        id: depends
        run: |
          sudo apt-get update
-          sudo apt-get install -y build-essential git cmake rocblas-dev hipblas-dev
+          sudo apt-get install -y build-essential git cmake rocblas-dev hipblas-dev libcurl4-openssl-dev

      - name: ccache
        uses: hendrikmuhs/ccache-action@v1.2.16
@@ -530,7 +525,7 @@ jobs:
        shell: bash
        run: |
          sudo apt update
-          sudo apt install intel-oneapi-compiler-dpcpp-cpp
+          sudo apt install intel-oneapi-compiler-dpcpp-cpp libcurl4-openssl-dev

      - name: install oneAPI MKL library
        shell: bash
@@ -578,7 +573,7 @@ jobs:
        shell: bash
        run: |
          sudo apt update
-          sudo apt install intel-oneapi-compiler-dpcpp-cpp
+          sudo apt install intel-oneapi-compiler-dpcpp-cpp libcurl4-openssl-dev

      - name: install oneAPI MKL library
        shell: bash
@@ -636,6 +631,7 @@ jobs:
          cmake -B build -G Xcode \
            -DGGML_METAL_USE_BF16=ON \
            -DGGML_METAL_EMBED_LIBRARY=ON \
+            -DLLAMA_BUILD_COMMON=OFF \
            -DLLAMA_BUILD_EXAMPLES=OFF \
            -DLLAMA_BUILD_TESTS=OFF \
            -DLLAMA_BUILD_SERVER=OFF \
@@ -671,6 +667,7 @@ jobs:
          cmake -B build -G Xcode \
            -DGGML_METAL_USE_BF16=ON \
            -DGGML_METAL_EMBED_LIBRARY=ON \
+            -DLLAMA_BUILD_COMMON=OFF \
            -DLLAMA_BUILD_EXAMPLES=OFF \
            -DLLAMA_BUILD_TESTS=OFF \
            -DLLAMA_BUILD_SERVER=OFF \
@@ -700,6 +697,7 @@ jobs:
          cmake -B build -G Xcode \
            -DGGML_METAL_USE_BF16=ON \
            -DGGML_METAL_EMBED_LIBRARY=ON \
+            -DLLAMA_BUILD_COMMON=OFF \
            -DLLAMA_BUILD_EXAMPLES=OFF \
            -DLLAMA_BUILD_TESTS=OFF \
            -DLLAMA_BUILD_SERVER=OFF \
@@ -739,6 +737,7 @@ jobs:
          cmake -B build -G Xcode \
            -DGGML_METAL_USE_BF16=ON \
            -DGGML_METAL_EMBED_LIBRARY=ON \
+            -DLLAMA_CURL=OFF \
            -DLLAMA_BUILD_EXAMPLES=OFF \
            -DLLAMA_BUILD_TESTS=OFF \
            -DLLAMA_BUILD_SERVER=OFF \
@@ -899,10 +898,17 @@ jobs:
            -DCMAKE_INSTALL_PREFIX="$env:RUNNER_TEMP/opencl-arm64-release"
          cmake --build build-arm64-release --target install --config release

+      - name: libCURL
+        id: get_libcurl
+        uses: ./.github/actions/windows-setup-curl
+
      - name: Build
        id: cmake_build
+        env:
+          CURL_PATH: ${{ steps.get_libcurl.outputs.curl_path }}
        run: |
-          cmake -S . -B build ${{ matrix.defines }}
+          cmake -S . -B build ${{ matrix.defines }} `
+            -DCURL_LIBRARY="$env:CURL_PATH/lib/libcurl.dll.a" -DCURL_INCLUDE_DIR="$env:CURL_PATH/include"
          cmake --build build --config Release -j ${env:NUMBER_OF_PROCESSORS}

      - name: Add libopenblas.dll
@@ -962,9 +968,10 @@ jobs:
      - name: Pack artifacts
        id: pack_artifacts
        if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }}
+        env:
+          CURL_PATH: ${{ steps.get_libcurl.outputs.curl_path }}
        run: |
-          Copy-Item LICENSE .\build\bin\Release\llama.cpp.txt
-          Copy-Item .\examples\run\linenoise.cpp\LICENSE .\build\bin\Release\linenoise.cpp.txt
+          Copy-Item $env:CURL_PATH\bin\libcurl-x64.dll .\build\bin\Release\libcurl-x64.dll
          7z a llama-${{ steps.tag.outputs.name }}-bin-win-${{ matrix.build }}.zip .\build\bin\Release\*

      - name: Upload artifacts
@@ -990,7 +997,7 @@ jobs:
            DEBIAN_FRONTEND: noninteractive
          run: |
              apt update
-              apt install -y cmake build-essential ninja-build libgomp1 git
+              apt install -y cmake build-essential ninja-build libgomp1 git libcurl4-openssl-dev

        - name: ccache
          uses: hendrikmuhs/ccache-action@v1.2.16
@@ -1092,16 +1099,23 @@ jobs:
        run: |
          choco install ninja

+      - name: libCURL
+        id: get_libcurl
+        uses: ./.github/actions/windows-setup-curl
+
      - name: Build
        id: cmake_build
        shell: cmd
+        env:
+          CURL_PATH: ${{ steps.get_libcurl.outputs.curl_path }}
        run: |
          call "C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\VC\Auxiliary\Build\vcvars64.bat"
          cmake -S . -B build -G "Ninja Multi-Config" ^
            -DLLAMA_BUILD_SERVER=ON ^
            -DGGML_NATIVE=OFF ^
            -DGGML_CUDA=ON ^
-            -DGGML_RPC=ON
+            -DGGML_RPC=ON ^
+            -DCURL_LIBRARY="%CURL_PATH%/lib/libcurl.dll.a" -DCURL_INCLUDE_DIR="%CURL_PATH%/include"
          set /A NINJA_JOBS=%NUMBER_OF_PROCESSORS%-1
          cmake --build build --config Release -j %NINJA_JOBS% -t ggml
          cmake --build build --config Release
@@ -1122,7 +1136,10 @@ jobs:
      - name: Pack artifacts
        id: pack_artifacts
        if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }}
+        env:
+          CURL_PATH: ${{ steps.get_libcurl.outputs.curl_path }}
        run: |
+          cp $env:CURL_PATH\bin\libcurl-x64.dll .\build\bin\Release\libcurl-x64.dll
          7z a llama-${{ steps.tag.outputs.name }}-bin-win-${{ matrix.build }}-cu${{ matrix.cuda }}-x64.zip .\build\bin\Release\*

      - name: Upload artifacts
@@ -1177,6 +1194,8 @@ jobs:
        run:  |
          scripts/install-oneapi.bat $WINDOWS_BASEKIT_URL $WINDOWS_DPCPP_MKL

+      # TODO: add libcurl support ; we will also need to modify win-build-sycl.bat to accept user-specified args
+
      - name: Build
        id: cmake_build
        run:  examples/sycl/win-build-sycl.bat
@@ -1262,8 +1281,14 @@ jobs:
          key: ${{ github.job }}
          evict-old-files: 1d

+      - name: libCURL
+        id: get_libcurl
+        uses: ./.github/actions/windows-setup-curl
+
      - name: Build
        id: cmake_build
+        env:
+          CURL_PATH: ${{ steps.get_libcurl.outputs.curl_path }}
        run: |
          $env:HIP_PATH=$(Resolve-Path 'C:\Program Files\AMD\ROCm\*\bin\clang.exe' | split-path | split-path)
          $env:CMAKE_PREFIX_PATH="${env:HIP_PATH}"
@@ -1274,9 +1299,11 @@ jobs:
            -DCMAKE_BUILD_TYPE=Release `
            -DGGML_HIP=ON `
            -DGGML_HIP_ROCWMMA_FATTN=ON `
-            -DGGML_RPC=ON
+            -DGGML_RPC=ON `
+            -DCURL_LIBRARY="$env:CURL_PATH/lib/libcurl.dll.a" -DCURL_INCLUDE_DIR="$env:CURL_PATH/include"
          cmake --build build -j ${env:NUMBER_OF_PROCESSORS}

+  # TODO: reuse windows-latest-cmake-hip instead of duplicating this job
  windows-latest-cmake-hip-release:
    if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }}
    runs-on: windows-latest
@@ -1318,8 +1345,14 @@ jobs:
        run: |
          & 'C:\Program Files\AMD\ROCm\*\bin\clang.exe' --version

+      - name: libCURL
+        id: get_libcurl
+        uses: ./.github/actions/windows-setup-curl
+
      - name: Build
        id: cmake_build
+        env:
+          CURL_PATH: ${{ steps.get_libcurl.outputs.curl_path }}
        run: |
          $env:HIP_PATH=$(Resolve-Path 'C:\Program Files\AMD\ROCm\*\bin\clang.exe' | split-path | split-path)
          $env:CMAKE_PREFIX_PATH="${env:HIP_PATH}"
@@ -1331,7 +1364,8 @@ jobs:
            -DAMDGPU_TARGETS=${{ matrix.gpu_target }} `
            -DGGML_HIP_ROCWMMA_FATTN=ON `
            -DGGML_HIP=ON `
-            -DGGML_RPC=ON
+            -DGGML_RPC=ON `
+            -DCURL_LIBRARY="$env:CURL_PATH/lib/libcurl.dll.a" -DCURL_INCLUDE_DIR="$env:CURL_PATH/include"
          cmake --build build -j ${env:NUMBER_OF_PROCESSORS}
          md "build\bin\rocblas\library\"
          cp "${env:HIP_PATH}\bin\hipblas.dll" "build\bin\"
@@ -1353,7 +1387,10 @@ jobs:

      - name: Pack artifacts
        id: pack_artifacts
+        env:
+          CURL_PATH: ${{ steps.get_libcurl.outputs.curl_path }}
        run: |
+          cp $env:CURL_PATH\bin\libcurl-x64.dll .\build\bin\libcurl-x64.dll
          7z a llama-${{ steps.tag.outputs.name }}-bin-win-hip-x64-${{ matrix.gpu_target }}.zip .\build\bin\*

      - name: Upload artifacts
@@ -1378,6 +1415,7 @@ jobs:
          cmake -B build -G Xcode \
            -DGGML_METAL_USE_BF16=ON \
            -DGGML_METAL_EMBED_LIBRARY=ON \
+            -DLLAMA_CURL=OFF \
            -DLLAMA_BUILD_EXAMPLES=OFF \
            -DLLAMA_BUILD_TESTS=OFF \
            -DLLAMA_BUILD_SERVER=OFF \
--- a/.github/workflows/server.yml
+++ b/.github/workflows/server.yml
@@ -129,7 +129,6 @@ jobs:
          cmake -B build \
              -DGGML_NATIVE=OFF \
              -DLLAMA_BUILD_SERVER=ON \
-              -DLLAMA_CURL=ON \
              -DCMAKE_BUILD_TYPE=${{ matrix.build_type }} \
              -DLLAMA_SANITIZE_${{ matrix.sanitizer }}=ON \
              -DGGML_OPENMP=OFF ;
@@ -142,7 +141,6 @@ jobs:
          cmake -B build \
              -DGGML_NATIVE=OFF \
              -DLLAMA_BUILD_SERVER=ON \
-              -DLLAMA_CURL=ON \
              -DCMAKE_BUILD_TYPE=${{ matrix.build_type }} \
              -DLLAMA_SANITIZE_${{ matrix.sanitizer }}=ON ;
          cmake --build build --config ${{ matrix.build_type }} -j $(nproc) --target llama-server
@@ -154,7 +152,6 @@ jobs:
          cmake -B build \
              -DGGML_NATIVE=OFF \
              -DLLAMA_BUILD_SERVER=ON \
-              -DLLAMA_CURL=ON \
              -DCMAKE_BUILD_TYPE=${{ matrix.build_type }} ;
          cmake --build build --config ${{ matrix.build_type }} -j $(nproc) --target llama-server

@@ -195,17 +192,14 @@ jobs:

      - name: libCURL
        id: get_libcurl
-        env:
-          CURL_VERSION: 8.6.0_6
-        run: |
-          curl.exe -o $env:RUNNER_TEMP/curl.zip -L "https://curl.se/windows/dl-${env:CURL_VERSION}/curl-${env:CURL_VERSION}-win64-mingw.zip"
-          mkdir $env:RUNNER_TEMP/libcurl
-          tar.exe -xvf $env:RUNNER_TEMP/curl.zip --strip-components=1 -C $env:RUNNER_TEMP/libcurl
+        uses: ./.github/actions/windows-setup-curl

      - name: Build
        id: cmake_build
+        env:
+          CURL_PATH: ${{ steps.get_libcurl.outputs.curl_path }}
        run: |
-          cmake -B build -DLLAMA_CURL=ON -DCURL_LIBRARY="$env:RUNNER_TEMP/libcurl/lib/libcurl.dll.a" -DCURL_INCLUDE_DIR="$env:RUNNER_TEMP/libcurl/include"
+          cmake -B build -DCURL_LIBRARY="$env:CURL_PATH/lib/libcurl.dll.a" -DCURL_INCLUDE_DIR="$env:CURL_PATH/include"
          cmake --build build --config Release -j ${env:NUMBER_OF_PROCESSORS} --target llama-server

      - name: Python setup
@@ -221,8 +215,10 @@ jobs:

      - name: Copy Libcurl
        id: prepare_libcurl
+        env:
+          CURL_PATH: ${{ steps.get_libcurl.outputs.curl_path }}
        run: |
-          cp $env:RUNNER_TEMP/libcurl/bin/libcurl-x64.dll ./build/bin/Release/libcurl-x64.dll
+          cp $env:CURL_PATH/bin/libcurl-x64.dll ./build/bin/Release/libcurl-x64.dll

      - name: Tests
        id: server_integration_tests
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -81,7 +81,7 @@ option(LLAMA_BUILD_EXAMPLES "llama: build examples"       ${LLAMA_STANDALONE})
 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_CURL       "llama: use libcurl to download model from an URL" ON)
 option(LLAMA_LLGUIDANCE "llama-common: include LLGuidance library for structured output in common utils" OFF)

 # Required for relocatable CMake package
@@ -168,6 +168,11 @@ add_subdirectory(src)
 # utils, programs, examples and tests
 #

+if (NOT LLAMA_BUILD_COMMON)
+    message(STATUS "LLAMA_BUILD_COMMON is OFF, disabling LLAMA_CURL")
+    set(LLAMA_CURL OFF)
+endif()
+
 if (LLAMA_BUILD_COMMON)
    add_subdirectory(common)
 endif()
@@ -242,3 +247,20 @@ configure_file(cmake/llama.pc.in

 install(FILES "${CMAKE_CURRENT_BINARY_DIR}/llama.pc"
        DESTINATION ${CMAKE_INSTALL_LIBDIR}/pkgconfig)
+
+#
+# copy the license files
+#
+
+# Check if running in GitHub Actions
+if(DEFINED ENV{GITHUB_ACTIONS} AND "$ENV{GITHUB_ACTIONS}" STREQUAL "true")
+    message(STATUS "Running inside GitHub Actions - copying license files")
+
+    # Copy all files from licenses/ to build/bin/
+    file(GLOB LICENSE_FILES "${CMAKE_SOURCE_DIR}/licenses/*")
+    foreach(LICENSE_FILE ${LICENSE_FILES})
+        get_filename_component(FILENAME ${LICENSE_FILE} NAME)
+        configure_file(${LICENSE_FILE} "${CMAKE_BINARY_DIR}/bin/${FILENAME}" COPYONLY)
+    endforeach()
+endif()
+
--- a/build-xcframework.sh
+++ b/build-xcframework.sh
@@ -399,6 +399,7 @@ cmake -B build-ios-sim -G Xcode \
    -DCMAKE_XCODE_ATTRIBUTE_SUPPORTED_PLATFORMS=iphonesimulator \
    -DCMAKE_C_FLAGS="${COMMON_C_FLAGS}" \
    -DCMAKE_CXX_FLAGS="${COMMON_CXX_FLAGS}" \
+    -DLLAMA_CURL=OFF \
    -S .
 cmake --build build-ios-sim --config Release -- -quiet

@@ -411,6 +412,7 @@ cmake -B build-ios-device -G Xcode \
    -DCMAKE_XCODE_ATTRIBUTE_SUPPORTED_PLATFORMS=iphoneos \
    -DCMAKE_C_FLAGS="${COMMON_C_FLAGS}" \
    -DCMAKE_CXX_FLAGS="${COMMON_CXX_FLAGS}" \
+    -DLLAMA_CURL=OFF \
    -S .
 cmake --build build-ios-device --config Release -- -quiet

@@ -421,6 +423,7 @@ cmake -B build-macos -G Xcode \
    -DCMAKE_OSX_ARCHITECTURES="arm64;x86_64" \
    -DCMAKE_C_FLAGS="${COMMON_C_FLAGS}" \
    -DCMAKE_CXX_FLAGS="${COMMON_CXX_FLAGS}" \
+    -DLLAMA_CURL=OFF \
    -S .
 cmake --build build-macos --config Release -- -quiet

@@ -434,6 +437,7 @@ cmake -B build-visionos -G Xcode \
    -DCMAKE_XCODE_ATTRIBUTE_SUPPORTED_PLATFORMS=xros \
    -DCMAKE_C_FLAGS="-D_XOPEN_SOURCE=700 ${COMMON_C_FLAGS}" \
    -DCMAKE_CXX_FLAGS="-D_XOPEN_SOURCE=700 ${COMMON_CXX_FLAGS}" \
+    -DLLAMA_CURL=OFF \
    -S .
 cmake --build build-visionos --config Release -- -quiet

@@ -447,6 +451,7 @@ cmake -B build-visionos-sim -G Xcode \
    -DCMAKE_XCODE_ATTRIBUTE_SUPPORTED_PLATFORMS=xrsimulator \
    -DCMAKE_C_FLAGS="-D_XOPEN_SOURCE=700 ${COMMON_C_FLAGS}" \
    -DCMAKE_CXX_FLAGS="-D_XOPEN_SOURCE=700 ${COMMON_CXX_FLAGS}" \
+    -DLLAMA_CURL=OFF \
    -S .
 cmake --build build-visionos-sim --config Release -- -quiet

@@ -462,6 +467,7 @@ cmake -B build-tvos-sim -G Xcode \
    -DCMAKE_XCODE_ATTRIBUTE_SUPPORTED_PLATFORMS=appletvsimulator \
    -DCMAKE_C_FLAGS="${COMMON_C_FLAGS}" \
    -DCMAKE_CXX_FLAGS="${COMMON_CXX_FLAGS}" \
+    -DLLAMA_CURL=OFF \
    -S .
 cmake --build build-tvos-sim --config Release -- -quiet

@@ -476,6 +482,7 @@ cmake -B build-tvos-device -G Xcode \
    -DCMAKE_XCODE_ATTRIBUTE_SUPPORTED_PLATFORMS=appletvos \
    -DCMAKE_C_FLAGS="${COMMON_C_FLAGS}" \
    -DCMAKE_CXX_FLAGS="${COMMON_CXX_FLAGS}" \
+    -DLLAMA_CURL=OFF \
    -S .
 cmake --build build-tvos-device --config Release -- -quiet

--- a/common/CMakeLists.txt
+++ b/common/CMakeLists.txt
@@ -85,7 +85,10 @@ set(LLAMA_COMMON_EXTRA_LIBS build_info)

 # Use curl to download model url
 if (LLAMA_CURL)
-    find_package(CURL REQUIRED)
+    find_package(CURL)
+    if (NOT CURL_FOUND)
+        message(FATAL_ERROR "Could NOT find CURL. Hint: to disable this feature, set -DLLAMA_CURL=OFF")
+    endif()
    target_compile_definitions(${TARGET} PUBLIC LLAMA_USE_CURL)
    include_directories(${CURL_INCLUDE_DIRS})
    find_library(CURL_LIBRARY curl REQUIRED)
--- a/examples/llama.android/llama/build.gradle.kts
+++ b/examples/llama.android/llama/build.gradle.kts
@@ -18,6 +18,7 @@ android {
        }
        externalNativeBuild {
            cmake {
+                arguments += "-DLLAMA_CURL=OFF"
                arguments += "-DLLAMA_BUILD_COMMON=ON"
                arguments += "-DGGML_LLAMAFILE=OFF"
                arguments += "-DCMAKE_BUILD_TYPE=Release"
--- a/examples/run/CMakeLists.txt
+++ b/examples/run/CMakeLists.txt
@@ -1,5 +1,16 @@
 set(TARGET llama-run)
 add_executable(${TARGET} run.cpp linenoise.cpp/linenoise.cpp)
+
+# TODO: avoid copying this code block from common/CMakeLists.txt
+set(LLAMA_RUN_EXTRA_LIBS "")
+if (LLAMA_CURL)
+    find_package(CURL REQUIRED)
+    target_compile_definitions(${TARGET} PUBLIC LLAMA_USE_CURL)
+    include_directories(${CURL_INCLUDE_DIRS})
+    find_library(CURL_LIBRARY curl REQUIRED)
+    set(LLAMA_RUN_EXTRA_LIBS ${LLAMA_RUN_EXTRA_LIBS} ${CURL_LIBRARY})
+endif ()
+
 install(TARGETS ${TARGET} RUNTIME)
-target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
+target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT} ${LLAMA_RUN_EXTRA_LIBS})
 target_compile_features(${TARGET} PRIVATE cxx_std_17)
--- a/examples/server/tests/README.md
+++ b/examples/server/tests/README.md
@@ -17,7 +17,7 @@ To mitigate it, you can increase values in `n_predict`, `kv_size`.

 ```shell
 cd ../../..
-cmake -B build -DLLAMA_CURL=ON
+cmake -B build
 cmake --build build --target llama-server
 ```

--- a/examples/sycl/win-build-sycl.bat
+++ b/examples/sycl/win-build-sycl.bat
@@ -13,10 +13,10 @@ if %errorlevel% neq 0 goto ERROR

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

 ::  for FP32
-cmake -G "Ninja" ..  -DGGML_SYCL=ON -DCMAKE_C_COMPILER=cl -DCMAKE_CXX_COMPILER=icx -DBUILD_SHARED_LIBS=ON -DCMAKE_BUILD_TYPE=Release
+cmake -G "Ninja" .. -DLLAMA_CURL=OFF -DGGML_SYCL=ON -DCMAKE_C_COMPILER=cl -DCMAKE_CXX_COMPILER=icx -DBUILD_SHARED_LIBS=ON -DCMAKE_BUILD_TYPE=Release
 if %errorlevel% neq 0 goto ERROR
 ::  build example/main only
 ::  make main
--- a/ggml/src/ggml-cann/aclnn_ops.cpp
+++ b/ggml/src/ggml-cann/aclnn_ops.cpp
--- a/ggml/src/ggml-cann/aclnn_ops.h
+++ b/ggml/src/ggml-cann/aclnn_ops.h
@@ -31,20 +31,25 @@
 * IN THE SOFTWARE.
 */

-#include <aclnnop/aclnn_add.h>
+#include <aclnnop/aclnn_abs.h>
+#include <aclnnop/aclnn_neg.h>
+#include <aclnnop/aclnn_exp.h>
 #include <aclnnop/aclnn_arange.h>
 #include <aclnnop/aclnn_argsort.h>
 #include <aclnnop/aclnn_cat.h>
 #include <aclnnop/aclnn_clamp.h>
-#include <aclnnop/aclnn_div.h>
 #include <aclnnop/aclnn_gelu.h>
+#include <aclnnop/aclnn_gelu_v2.h>
+#include <aclnnop/aclnn_sigmoid.h>
 #include <aclnnop/aclnn_hardsigmoid.h>
 #include <aclnnop/aclnn_hardswish.h>
 #include <aclnnop/aclnn_leaky_relu.h>
-#include <aclnnop/aclnn_mul.h>
 #include <aclnnop/aclnn_relu.h>
 #include <aclnnop/aclnn_silu.h>
 #include <aclnnop/aclnn_tanh.h>
+#include <aclnnop/aclnn_sqrt.h>
+#include <aclnnop/aclnn_sin.h>
+#include <aclnnop/aclnn_cos.h>
 #include "acl_tensor.h"
 #include "common.h"

@@ -63,23 +68,6 @@
 */
 void ggml_cann_repeat(ggml_backend_cann_context& ctx, ggml_tensor* dst);

-/**
- * @brief   Adds two ggml tensors using the CANN backend.
- *
- * @details This function performs an element-wise addition of two tensors. In
- *          case the tensors do not have the same shape, one or both tensors
- *          will be broadcasted to match the shape of the other before the
- *          addition is performed.The formula for the operation is given by:
- *          \f[
- *              \text{dst} = \text{acl_src0} + \alpha \cdot \text{acl_src1}
- *          \f]
- *
- * @param ctx The CANN context used for operations.
- * @param dst The ggml tensor representing the destination, result of the
- *            addition is stored at dst->data, and dst->op is `GGML_OP_ADD`
- */
-void ggml_cann_add(ggml_backend_cann_context& ctx, ggml_tensor* dst);
-
 /**
 * @brief   Applies the Leaky ReLU activation function to a tensor using the CANN
 *          backend.
@@ -131,19 +119,6 @@ void ggml_cann_concat(ggml_backend_cann_context& ctx, ggml_tensor* dst);
 */
 void ggml_cann_arange(ggml_backend_cann_context& ctx, ggml_tensor* dst);

-/**
- * @brief   Computes the square of the elements of a ggml tensor using the CANN
- *          backend.
- * @details The function sets the second source tensor of the destination
- *          tensor `dst` to be equal to the first source tensor. This is
- *          effectively squaring the elements since the multiplication becomes
- *          `element * element`.
- * @param ctx The CANN context used for operations.
- * @param dst The destination tensor where the squared values will be stored，
- *            which dst->op is `GGML_OP_SQR`.
- */
-void ggml_cann_sqr(ggml_backend_cann_context& ctx, ggml_tensor* dst);
-
 /**
 * @brief   Applies a clamp operation to the elements of a ggml tensor using the
 *          CANN backend.
@@ -275,6 +250,20 @@ void ggml_cann_acc(ggml_backend_cann_context& ctx, ggml_tensor* dst);
 */
 void ggml_cann_sum_rows(ggml_backend_cann_context& ctx, ggml_tensor* dst);

+/**
+ * @brief   Computes the sum of elements in a ggml tensor.
+ *
+ * @details This function performs a reduction sum operation along the last
+ *          dimension of the input tensor `src`. The result of the sum is stored
+ *          in the destination tensor `dst`.
+ *
+ * @param ctx The CANN context used for operations.
+ * @param dst The destination tensor where the reduced values will be stored。
+ *
+ */
+
+void ggml_cann_sum(ggml_backend_cann_context& ctx, ggml_tensor* dst);
+
 /**
 * @brief   Upsamples a ggml tensor using nearest neighbor interpolation using
 *          the CANN backend.
@@ -500,128 +489,247 @@ void ggml_cann_rope(ggml_backend_cann_context& ctx, ggml_tensor* dst);
 void ggml_cann_argmax(ggml_backend_cann_context& ctx, ggml_tensor* dst);

 /**
- * @brief   Computes the cosine of each element in a ggml tensor using the CANN backend.
+ * @brief Adds two tensors element-wise and stores the result in a destination
+ * tensor.
 *
- * @details This function applies the cosine function element-wise to the input tensor.
- *          The computed cosine values are stored in the destination tensor `dst`.
- *          The operation is optimized using the CANN backend for improved performance.
+ * This function performs the operation:
+ * \f[
+ *    dst = acl\_src0 + alpha \times acl\_src1
+ * \f]
+ * where alpha is a scalar value and defaults to 1.0f.
 *
- * @param ctx The CANN context used for operations.
- * @param dst The destination tensor where the cosine values will be stored.
- *            dst->op is `GGML_OP_COS`.
+ * @param ctx The context for the CANN backend operations.
+ * @param acl_src0 The first source tensor.
+ * @param acl_src1 The second source tensor.
+ * @param acl_dst The destination tensor where the result will be stored.
 */
-void ggml_cann_cos(ggml_backend_cann_context& ctx, ggml_tensor* dst);
+void aclnn_add(ggml_backend_cann_context& ctx, aclTensor* acl_src0,
+    aclTensor* acl_src1, aclTensor* acl_dst = nullptr);

 /**
- * @brief   Computes the sine of each element in a ggml tensor using the CANN backend.
+ * @brief Sub two tensors element-wise and stores the result in a destination
+ * tensor.
 *
- * @details This function applies the sine function element-wise to the input tensor.
- *          The computed sine values are stored in the destination tensor `dst`.
- *          The operation is optimized using the CANN backend for improved performance.
+ * This function performs the operation:
+ * \f[
+ *    dst = acl\_src0 - alpha \times acl\_src1
+ * \f]
+ * where alpha is a scalar value and defaults to 1.0f.
 *
- * @param ctx The CANN context used for operations.
- * @param dst The destination tensor where the sine values will be stored.
- *            dst->op is `GGML_OP_SIN`.
+ * @param ctx The context for the CANN backend operations.
+ * @param acl_src0 The first source tensor.
+ * @param acl_src1 The second source tensor.
+ * @param acl_dst The destination tensor where the result will be stored.
 */
-void ggml_cann_sin(ggml_backend_cann_context& ctx, ggml_tensor* dst);
+void aclnn_sub(ggml_backend_cann_context& ctx, aclTensor* acl_src0,
+    aclTensor* acl_src1, aclTensor* acl_dst = nullptr);

-template <aclnnStatus getWorkspaceSize(const aclTensor*, const aclTensor*,
-                                       aclTensor*, uint64_t*, aclOpExecutor**),
-          aclnnStatus execute(void*, uint64_t, aclOpExecutor*, aclrtStream)>
-void ggml_cann_mul_div(ggml_backend_cann_context& ctx, ggml_tensor* dst) {
+/**
+ * @brief Performs element-wise multiplication of two tensors and stores the
+ * result in a destination tensor.
+ *
+ * This function performs element-wise multiplication of the tensors `acl_src`
+ * and `acl_other` and stores the result in the destination tensor `acl_dst`.
+ * The operation is defined as:
+ * \f[
+ *     \text {acl_dst }_i=\text {acl_src }_i \times \text {acl_other }_i
+ * \f]
+ *
+ * @param ctx The context for the CANN backend operations.
+ * @param acl_src The first tensor for element-wise multiplication.
+ * @param acl_other The second tensor for element-wise multiplication.
+ * @param acl_dst The destination tensor where the result will be stored.
+ */
+void aclnn_mul(ggml_backend_cann_context& ctx, aclTensor* acl_src,
+    aclTensor* acl_other, aclTensor* acl_dst = nullptr);
+
+/**
+ * @brief Matrix division, optionally in-place.
+ *
+ * This function division each element of the source tensor `acl_src` by the
+ * tensor `acl_other` and stores the result in the destination tensor `acl_dst`.
+ * If `inplace` is true, `acl_dst` will not be used and the operation is
+ * performed in-place on `acl_src`. The operation is defined as: \f[
+ *     \text{dst}_i = \frac{\text{acl_src}_i}{\text{acl_other}_i}
+ * \f]
+ *
+ * @param ctx The context for the CANN backend operations.
+ * @param acl_src Numerator tensor..
+ * @param acl_other Denominator tensor.
+ * @param acl_dst The destination tensor where the result will be stored if
+ * `inplace` is false.
+ * @param inplace Flag indicating whether to perform the operation in-place on
+ * `acl_src`.
+ */
+void aclnn_div(ggml_backend_cann_context& ctx, aclTensor* acl_src,
+    aclTensor* acl_other, aclTensor* acl_dst = nullptr);
+
+/**
+ * @brief Applies element-wise cosine function to the elements of a tensor.
+ *
+ * This function computes the cosine of each element in the source tensor
+ * `acl_src` and stores the result in the destination tensor `acl_dst`. The
+ * operation is defined as: \f[ \text {acl_dst }_i=\cos \left(\text {acl_src
+ * }_i\right) \f]
+ *
+ * @param ctx The context for the CANN backend operations.
+ * @param acl_src The source tensor on which the cosine function will be
+ * applied.
+ * @param acl_dst The destination tensor where the cosine results will be
+ * stored.
+ */
+void aclnn_cos(ggml_backend_cann_context& ctx, aclTensor* acl_src,
+    aclTensor* acl_dst);
+
+/**
+ * @brief Applies element-wise sine function to the elements of a tensor.
+ *
+ * This function computes the sine of each element in the source tensor
+ `acl_src`
+ * and stores the result in the destination tensor `acl_dst`.
+ * The operation is defined as:
+ * \f[
+ *     \text {acl_dst }_i=\sin \left(\text {acl_src }_i\right)
+ * \f]
+
+ * @param ctx The context for the CANN backend operations.
+ * @param acl_src The source tensor on which the sine function will be applied.
+ * @param acl_dst The destination tensor where the sine results will be stored.
+ */
+void aclnn_sin(ggml_backend_cann_context& ctx, aclTensor* acl_src,
+    aclTensor* acl_dst);
+
+/**
+ * @brief Launches an asynchronous task using the memory allocator.
+ *
+ * This macro submit an asynchronous task on the specified stream.
+ * The task uses memory allocated by the allocator. It is guaranteed
+ * that the memory will not be accessed by other tasks until this task
+ * completes, due to the sequential execution order within the same stream.
+ *
+ * @param OP_NAME aclnn operator name.
+ * @param args Additional arguments required by the task.
+ *
+ * @note
+ * Memory from the allocator will be "freed" immediately and can be
+ * reallocated to other pointers. However, it won't be accessed by any
+ * other task before this asynchronous task ends, because all tasks in the
+ * same stream are executed in queue order.
+ */
+#define GGML_CANN_CALL_ACLNN_OP(OP_NAME, ...)                                                \
+    do {                                                                                     \
+        uint64_t        workspaceSize = 0;                                                   \
+        aclOpExecutor * executor;                                                            \
+        void *          workspaceAddr = nullptr;                                             \
+                                                                                             \
+        ACL_CHECK(aclnn##OP_NAME##GetWorkspaceSize(__VA_ARGS__, &workspaceSize, &executor)); \
+                                                                                             \
+        if (workspaceSize > 0) {                                                             \
+            ggml_cann_pool_alloc workspace_allocator(ctx.pool(), workspaceSize);             \
+            workspaceAddr = workspace_allocator.get();                                       \
+        }                                                                                    \
+        ACL_CHECK(aclnn##OP_NAME(workspaceAddr, workspaceSize, executor, ctx.stream()));     \
+    } while (0)
+
+
+/**
+ * @brief Prepares broadcast-compatible ACL tensors for two input tensors and one output tensor.
+ *
+ * This function checks whether broadcasting is needed between `src0` and `src1`.
+ * If broadcasting is required, it calculates the proper shapes and creates
+ * ACL tensors with broadcast parameters. Otherwise, it directly creates ACL tensors
+ * based on the original tensor shapes.
+ *
+ * @param src0     The first input tensor (reference shape).
+ * @param src1     The second input tensor (possibly broadcasted).
+ * @param dst      The destination/output tensor.
+ * @param acl_src0 Output pointer to the created ACL tensor corresponding to src0.
+ * @param acl_src1 Output pointer to the created ACL tensor corresponding to src1.
+ * @param acl_dst  Output pointer to the created ACL tensor corresponding to dst.
+ */
+void bcast_shape(ggml_tensor * src0, ggml_tensor * src1, ggml_tensor * dst, aclTensor ** acl_src0,
+                        aclTensor ** acl_src1, aclTensor ** acl_dst);
+
+/**
+ * @brief Applies a element-wise operation to two input tensors using the CANN backend.
+ *
+ * This templated function takes a binary operator and applies it to two source tensors
+ * associated with the destination tensor. The function handles broadcasting as needed.
+ *
+ * @tparam binary_op A callable object (e.g., lambda or function pointer) representing
+ *         the binary operation to be performed. It must take three arguments:
+ *         (ggml_backend_cann_context&, aclTensor*, aclTensor*, aclTensor*).
+ *
+ * @param ctx The CANN backend context used to manage execution and resources.
+ * @param dst The destination tensor.
+ */
+template <auto binary_op>
+void ggml_cann_binary_op(ggml_backend_cann_context& ctx, ggml_tensor* dst) {
    ggml_tensor* src0 = dst->src[0];
    ggml_tensor* src1 = dst->src[1];
-    GGML_ASSERT(ggml_can_repeat(src1, src0) && ggml_are_same_shape(src0, dst));

    aclTensor* acl_src0;
    aclTensor* acl_src1;
    aclTensor* acl_dst;

    // Need bcast
-    if (!ggml_are_same_shape(src0, src1) && ggml_cann_need_bcast(src0, src1)) {
-        BCAST_SHAPE(src0, src1)
-        acl_src0 = ggml_cann_create_tensor(src0, BCAST_PARAM(src0));
-        acl_src1 = ggml_cann_create_tensor(src1, BCAST_PARAM(src1));
-        acl_dst = ggml_cann_create_tensor(dst, BCAST_PARAM(src0));
-    } else {
-        acl_src0 = ggml_cann_create_tensor(src0);
-        acl_src1 = ggml_cann_create_tensor(src1);
-        acl_dst = ggml_cann_create_tensor(dst);
-    }
-
-    uint64_t workspaceSize = 0;
-    aclOpExecutor* executor;
-    void* workspaceAddr = nullptr;
-
-    ACL_CHECK(getWorkspaceSize(acl_src0, acl_src1, acl_dst, &workspaceSize,
-                               &executor));
-    if (workspaceSize > 0) {
-        ggml_cann_pool_alloc workspace_allocator(ctx.pool(), workspaceSize);
-        workspaceAddr = workspace_allocator.get();
-    }
-
-    aclrtStream main_stream = ctx.stream();
-    ACL_CHECK(execute(workspaceAddr, workspaceSize, executor, main_stream));
+    bcast_shape(src0, src1, dst, &acl_src0, &acl_src1, &acl_dst);
+    binary_op(ctx, acl_src0, acl_src1, acl_dst);

    ACL_CHECK(aclDestroyTensor(acl_src0));
    ACL_CHECK(aclDestroyTensor(acl_src1));
    ACL_CHECK(aclDestroyTensor(acl_dst));
 }

-// Activation functions template.
-template <aclnnStatus getWorkspaceSize(const aclTensor*, aclTensor*, uint64_t*,
-                                       aclOpExecutor**),
-          aclnnStatus execute(void*, uint64_t, aclOpExecutor*,
-                              const aclrtStream)>
-void ggml_cann_activation(ggml_backend_cann_context& ctx, ggml_tensor* dst) {
+/**
+ * @brief Applies a unary operation to an input tensor using the CANN backend.
+ *
+ * This templated function applies a unary operator to the source tensor of `dst`
+ * and stores the result in the destination tensor.
+ *
+ * @tparam unary_op A callable with the signature:
+ *         void(ggml_backend_cann_context&, aclTensor*, aclTensor*)
+ *         where the first aclTensor is the source and the second is the destination.
+ *
+ * @param ctx The CANN backend context for managing resources and execution.
+ * @param dst The destination tensor. Its src[0] is treated as the input tensor.
+ */
+template <void unary_op(ggml_backend_cann_context&, aclTensor*, aclTensor*)>
+    void ggml_cann_unary_op(ggml_backend_cann_context& ctx, ggml_tensor* dst) {
    ggml_tensor* src = dst->src[0];

    aclTensor* acl_src = ggml_cann_create_tensor(src);
    aclTensor* acl_dst = ggml_cann_create_tensor(dst);

-    uint64_t workspaceSize = 0;
-    aclOpExecutor* executor;
-    void* workspaceAddr = nullptr;
-
-    ACL_CHECK(getWorkspaceSize(acl_src, acl_dst, &workspaceSize, &executor));
-    if (workspaceSize > 0) {
-        ggml_cann_pool_alloc workspace_allocator(ctx.pool(), workspaceSize);
-        workspaceAddr = workspace_allocator.get();
-    }
-
-    aclrtStream main_stream = ctx.stream();
-    ACL_CHECK(execute(workspaceAddr, workspaceSize, executor, main_stream));
-
+    unary_op(ctx, acl_src, acl_dst);
    ACL_CHECK(aclDestroyTensor(acl_src));
    ACL_CHECK(aclDestroyTensor(acl_dst));
 }

-// Activation functions template for const aclTensors.
-template <aclnnStatus getWorkspaceSize(const aclTensor*, const aclTensor*,
-                                       uint64_t*, aclOpExecutor**),
-          aclnnStatus execute(void*, uint64_t, aclOpExecutor*,
-                              const aclrtStream)>
-void ggml_cann_activation(ggml_backend_cann_context& ctx, ggml_tensor* dst) {
-    ggml_tensor* src = dst->src[0];
-
-    aclTensor* acl_src = ggml_cann_create_tensor(src);
-    aclTensor* acl_dst = ggml_cann_create_tensor(dst);
-
-    uint64_t workspaceSize = 0;
-    aclOpExecutor* executor;
-    void* workspaceAddr = nullptr;
-
-    ACL_CHECK(getWorkspaceSize(acl_src, acl_dst, &workspaceSize, &executor));
-    if (workspaceSize > 0) {
-        ggml_cann_pool_alloc workspace_allocator(ctx.pool(), workspaceSize);
-        workspaceAddr = workspace_allocator.get();
-    }
-
-    aclrtStream main_stream = ctx.stream();
-    ACL_CHECK(execute(workspaceAddr, workspaceSize, executor, main_stream));
-
-    ACL_CHECK(aclDestroyTensor(acl_src));
-    ACL_CHECK(aclDestroyTensor(acl_dst));
-}
+/**
+ * @brief Helper macro to invoke a unary ACL operation using ggml_cann_unary_op.
+ *
+ * This macro defines an inline lambda wrapping a specific ACL operation name,
+ * and passes it to the templated ggml_cann_unary_op function. It simplifies
+ * calling unary ops by hiding the lambda boilerplate.
+ *
+ * Internally, the lambda will call:
+ * @code
+ * GGML_CANN_CALL_ACLNN_OP(OP_NAME, acl_src, acl_dst);
+ * @endcode
+ *
+ * @param OP_NAME The name of the ACL unary operator to invoke via GGML_CANN_CALL_ACLNN_OP.
+ *
+ * @see ggml_cann_unary_op
+ * @see GGML_CANN_CALL_ACLNN_OP
+ */
+#define GGML_CANN_CALL_UNARY_OP(OP_NAME)                         \
+    do {                                                         \
+        auto lambda = [](auto ctx, auto acl_src, auto acl_dst) { \
+            GGML_CANN_CALL_ACLNN_OP(OP_NAME, acl_src, acl_dst);  \
+        };                                                       \
+        ggml_cann_unary_op<lambda>(ctx, dst);                    \
+    }                                                            \
+    while (0)

 #endif  // CANN_ACLNN_OPS
--- a/ggml/src/ggml-cann/ggml-cann.cpp
+++ b/ggml/src/ggml-cann/ggml-cann.cpp
@@ -803,7 +803,7 @@ static enum ggml_status ggml_backend_cann_buffer_init_tensor(
        return GGML_STATUS_SUCCESS;
    }

-    // TODO: can backend doesn't support quantized yet. Just leave the code
+    // TODO: cann backend doesn't support quantized yet. Just leave the code
    // here.
    if (ggml_is_quantized(tensor->type)) {
        // Initialize padding to 0 to avoid possible NaN values
@@ -1300,47 +1300,59 @@ static bool ggml_cann_compute_forward(ggml_backend_cann_context& ctx,
            ggml_cann_dup(ctx, dst);
            break;
        case GGML_OP_ADD:
-            ggml_cann_add(ctx, dst);
+        case GGML_OP_ADD1:
+            ggml_cann_binary_op<aclnn_add>(ctx, dst);
+            break;
+        case GGML_OP_SUB:
+            ggml_cann_binary_op<aclnn_sub>(ctx, dst);
            break;
        case GGML_OP_ACC:
            ggml_cann_acc(ctx, dst);
            break;
        case GGML_OP_MUL:
-            ggml_cann_mul_div<aclnnMulGetWorkspaceSize, aclnnMul>(ctx, dst);
+            ggml_cann_binary_op<aclnn_mul>(ctx, dst);
            break;
        case GGML_OP_DIV:
-            ggml_cann_mul_div<aclnnDivGetWorkspaceSize, aclnnDiv>(ctx, dst);
+            ggml_cann_binary_op<aclnn_div>(ctx, dst);
            break;
        case GGML_OP_UNARY:
            switch (ggml_get_unary_op(dst)) {
+                case GGML_UNARY_OP_ABS:
+                    GGML_CANN_CALL_UNARY_OP(Abs);
+                    break;
+                case GGML_UNARY_OP_NEG:
+                    GGML_CANN_CALL_UNARY_OP(Neg);
+                    break;
                case GGML_UNARY_OP_GELU:
-                    ggml_cann_activation<aclnnGeluGetWorkspaceSize, aclnnGelu>(
-                        ctx, dst);
+                    GGML_CANN_CALL_UNARY_OP(Gelu);
                    break;
                case GGML_UNARY_OP_SILU:
-                    ggml_cann_activation<aclnnSiluGetWorkspaceSize, aclnnSilu>(
-                        ctx, dst);
+                    GGML_CANN_CALL_UNARY_OP(Silu);
                    break;
-                // TODO: Use faster gelu??
-                case GGML_UNARY_OP_GELU_QUICK:
-                    ggml_cann_activation<aclnnGeluGetWorkspaceSize, aclnnGelu>(
-                        ctx, dst);
+                case GGML_UNARY_OP_GELU_QUICK: {
+                        auto lambda = [](auto ctx, auto acl_src, auto acl_dst) {
+                            GGML_CANN_CALL_ACLNN_OP(GeluV2, acl_src, 0, acl_dst);
+                        };
+                        ggml_cann_unary_op<lambda>(ctx, dst);
+                    }
                    break;
                case GGML_UNARY_OP_TANH:
-                    ggml_cann_activation<aclnnTanhGetWorkspaceSize, aclnnTanh>(
-                        ctx, dst);
+                    GGML_CANN_CALL_UNARY_OP(Tanh);
                    break;
                case GGML_UNARY_OP_RELU:
-                    ggml_cann_activation<aclnnReluGetWorkspaceSize, aclnnRelu>(
-                        ctx, dst);
+                    GGML_CANN_CALL_UNARY_OP(Relu);
+                    break;
+                case GGML_UNARY_OP_SIGMOID:
+                    GGML_CANN_CALL_UNARY_OP(Sigmoid);
                    break;
                case GGML_UNARY_OP_HARDSIGMOID:
-                    ggml_cann_activation<aclnnHardsigmoidGetWorkspaceSize,
-                                         aclnnHardsigmoid>(ctx, dst);
+                    GGML_CANN_CALL_UNARY_OP(Hardsigmoid);
                    break;
                case GGML_UNARY_OP_HARDSWISH:
-                    ggml_cann_activation<aclnnHardswishGetWorkspaceSize,
-                                         aclnnHardswish>(ctx, dst);
+                    GGML_CANN_CALL_UNARY_OP(Hardswish);
+                    break;
+                case GGML_UNARY_OP_EXP:
+                    GGML_CANN_CALL_UNARY_OP(Exp);
                    break;
                default:
                    return false;
@@ -1382,7 +1394,12 @@ static bool ggml_cann_compute_forward(ggml_backend_cann_context& ctx,
            ggml_cann_scale(ctx, dst);
            break;
        case GGML_OP_SQR:
-            ggml_cann_sqr(ctx, dst);
+            GGML_ASSERT(dst->src[1] == nullptr);
+            dst->src[1] = dst->src[0];
+            ggml_cann_binary_op<aclnn_mul>(ctx, dst);
+            break;
+        case GGML_OP_SQRT:
+            GGML_CANN_CALL_UNARY_OP(Sqrt);
            break;
        case GGML_OP_CLAMP:
            ggml_cann_clamp(ctx, dst);
@@ -1414,6 +1431,9 @@ static bool ggml_cann_compute_forward(ggml_backend_cann_context& ctx,
        case GGML_OP_POOL_2D:
            ggml_cann_pool2d(ctx, dst);
            break;
+        case GGML_OP_SUM:
+            ggml_cann_sum(ctx, dst);
+            break;
        case GGML_OP_SUM_ROWS:
            ggml_cann_sum_rows(ctx, dst);
            break;
@@ -1424,11 +1444,11 @@ static bool ggml_cann_compute_forward(ggml_backend_cann_context& ctx,
            ggml_cann_argmax(ctx, dst);
            break;
        case GGML_OP_COS:
-            ggml_cann_cos(ctx, dst);
+            ggml_cann_unary_op<aclnn_cos>(ctx, dst);
            break;
        case GGML_OP_SIN:
-            ggml_cann_sin(ctx, dst);
-            break;
+            ggml_cann_unary_op<aclnn_sin>(ctx, dst);
+        break;
        default:
            return false;
    }
@@ -1679,13 +1699,17 @@ static bool ggml_backend_cann_supports_op(ggml_backend_dev_t dev,
    switch (op->op) {
        case GGML_OP_UNARY:
            switch (ggml_get_unary_op(op)) {
+                case GGML_UNARY_OP_ABS:
+                case GGML_UNARY_OP_NEG:
                case GGML_UNARY_OP_GELU:
                case GGML_UNARY_OP_SILU:
                case GGML_UNARY_OP_RELU:
+                case GGML_UNARY_OP_SIGMOID:
                case GGML_UNARY_OP_HARDSIGMOID:
                case GGML_UNARY_OP_HARDSWISH:
                case GGML_UNARY_OP_GELU_QUICK:
                case GGML_UNARY_OP_TANH:
+                case GGML_UNARY_OP_EXP:
                    return true;
                default:
                    return false;
@@ -1784,6 +1808,7 @@ static bool ggml_backend_cann_supports_op(ggml_backend_dev_t dev,
            // value of paddingW should be at most half of kernelW
            return (p0 <= (k0 / 2)) && (p1 <= (k1 / 2));
        }
+        case GGML_OP_SUM:
        case GGML_OP_DUP:
        case GGML_OP_IM2COL:
        case GGML_OP_CONCAT:
@@ -1795,11 +1820,14 @@ static bool ggml_backend_cann_supports_op(ggml_backend_dev_t dev,
        case GGML_OP_TRANSPOSE:
        case GGML_OP_NORM:
        case GGML_OP_ADD:
+        case GGML_OP_ADD1:
+        case GGML_OP_SUB:
        case GGML_OP_MUL:
        case GGML_OP_DIV:
        case GGML_OP_RMS_NORM:
        case GGML_OP_SCALE:
        case GGML_OP_SQR:
+        case GGML_OP_SQRT:
        case GGML_OP_CLAMP:
        case GGML_OP_DIAG_MASK_INF:
        case GGML_OP_SOFT_MAX:
--- a/ggml/src/ggml-cuda/cpy.cu
+++ b/ggml/src/ggml-cuda/cpy.cu
@@ -360,6 +360,9 @@ void ggml_cuda_cpy_dest_ptrs_copy(ggml_cuda_graph * cuda_graph, char ** host_des
    // copy destination pointers to GPU
    CUDA_CHECK(cudaMemcpyAsync(cuda_graph->dest_ptrs_d, host_dest_ptrs, host_dest_ptrs_size*sizeof(char *), cudaMemcpyHostToDevice, stream));
    cuda_graph->graph_cpynode_index = 0; // reset index
+#else
+    GGML_UNUSED(cuda_graph); GGML_UNUSED(host_dest_ptrs);
+    GGML_UNUSED(host_dest_ptrs_size); GGML_UNUSED(stream);
 #endif
 }

--- a/ggml/src/ggml-cuda/fattn-common.cuh
+++ b/ggml/src/ggml-cuda/fattn-common.cuh
@@ -62,7 +62,7 @@ static __device__ __forceinline__ T vec_dot_fattn_vec_KQ_q4_0(
    T sum = 0.0f;

 #pragma unroll
-    for (int k_KQ_0 = 0; k_KQ_0 < D/sizeof(int); k_KQ_0 += warp_size) {
+    for (int k_KQ_0 = 0; k_KQ_0 < int(D/sizeof(int)); k_KQ_0 += warp_size) {
        const int k_KQ = k_KQ_0 + threadIdx.x;

        const int ib    = k_KQ /  QI8_1;
@@ -102,7 +102,7 @@ static __device__ __forceinline__ T vec_dot_fattn_vec_KQ_q4_1(
    T sum = 0.0f;

 #pragma unroll
-    for (int k_KQ_0 = 0; k_KQ_0 < D/sizeof(int); k_KQ_0 += warp_size) {
+    for (int k_KQ_0 = 0; k_KQ_0 < int(D/sizeof(int)); k_KQ_0 += warp_size) {
        const int k_KQ = k_KQ_0 + threadIdx.x;

        const int ib    = k_KQ /  QI8_1;
@@ -146,7 +146,7 @@ static __device__ __forceinline__ T vec_dot_fattn_vec_KQ_q5_0(
    T sum = 0.0f;

 #pragma unroll
-    for (int k_KQ_0 = 0; k_KQ_0 < D/sizeof(int); k_KQ_0 += warp_size) {
+    for (int k_KQ_0 = 0; k_KQ_0 < int(D/sizeof(int)); k_KQ_0 += warp_size) {
        const int k_KQ = k_KQ_0 + threadIdx.x;

        const int ib    = k_KQ /  QI8_1;
@@ -193,7 +193,7 @@ static __device__ __forceinline__ T vec_dot_fattn_vec_KQ_q5_1(
    T sum = 0.0f;

 #pragma unroll
-    for (int k_KQ_0 = 0; k_KQ_0 < D/sizeof(int); k_KQ_0 += warp_size) {
+    for (int k_KQ_0 = 0; k_KQ_0 < int(D/sizeof(int)); k_KQ_0 += warp_size) {
        const int k_KQ = k_KQ_0 + threadIdx.x;

        const int ib    = k_KQ /  QI8_1;
@@ -244,7 +244,7 @@ static __device__ __forceinline__ T vec_dot_fattn_vec_KQ_q8_0(
    T sum = 0.0f;

 #pragma unroll
-    for (int k_KQ_0 = 0; k_KQ_0 < D/sizeof(int); k_KQ_0 += warp_size) {
+    for (int k_KQ_0 = 0; k_KQ_0 < int(D/sizeof(int)); k_KQ_0 += warp_size) {
        const int k_KQ = k_KQ_0 + threadIdx.x;

        const int ib  = k_KQ / QI8_0;
--- a/ggml/src/ggml-cuda/fattn-tile-f32.cu
+++ b/ggml/src/ggml-cuda/fattn-tile-f32.cu
@@ -52,6 +52,18 @@ static __global__ void flash_attn_tile_ext_f32(
    return;
 #endif // FP16_MMA_AVAILABLE
    if (use_logit_softcap && !(D == 128 || D == 256)) {
+        GGML_UNUSED(Q); GGML_UNUSED(K); GGML_UNUSED(V); GGML_UNUSED(mask);
+        GGML_UNUSED(dst); GGML_UNUSED(dst_meta); GGML_UNUSED(scale);
+        GGML_UNUSED(max_bias); GGML_UNUSED(m0); GGML_UNUSED(m1);
+        GGML_UNUSED(n_head_log2); GGML_UNUSED(logit_softcap);
+        GGML_UNUSED(ne00); GGML_UNUSED(ne01); GGML_UNUSED(ne02);
+        GGML_UNUSED(ne03); GGML_UNUSED(ne10); GGML_UNUSED(ne11);
+        GGML_UNUSED(ne12); GGML_UNUSED(ne13); GGML_UNUSED(ne31);
+        GGML_UNUSED(nb31); GGML_UNUSED(nb01); GGML_UNUSED(nb02);
+        GGML_UNUSED(nb03); GGML_UNUSED(nb11); GGML_UNUSED(nb12);
+        GGML_UNUSED(nb13); GGML_UNUSED(nb21); GGML_UNUSED(nb22);
+        GGML_UNUSED(nb23); GGML_UNUSED(ne0); GGML_UNUSED(ne1);
+        GGML_UNUSED(ne2); GGML_UNUSED(ne3);
        NO_DEVICE_CODE;
        return;
    }
--- a/ggml/src/ggml-cuda/fattn-vec-f32.cuh
+++ b/ggml/src/ggml-cuda/fattn-vec-f32.cuh
@@ -45,6 +45,18 @@ static __global__ void flash_attn_vec_ext_f32(

    // Skip unused kernel variants for faster compilation:
    if (use_logit_softcap && !(D == 128 || D == 256)) {
+        GGML_UNUSED(Q); GGML_UNUSED(K); GGML_UNUSED(V); GGML_UNUSED(mask);
+        GGML_UNUSED(dst); GGML_UNUSED(dst_meta); GGML_UNUSED(scale);
+        GGML_UNUSED(max_bias); GGML_UNUSED(m0); GGML_UNUSED(m1);
+        GGML_UNUSED(n_head_log2); GGML_UNUSED(logit_softcap);
+        GGML_UNUSED(ne00); GGML_UNUSED(ne01); GGML_UNUSED(ne02);
+        GGML_UNUSED(ne03); GGML_UNUSED(ne10); GGML_UNUSED(ne11);
+        GGML_UNUSED(ne12); GGML_UNUSED(ne13); GGML_UNUSED(ne31);
+        GGML_UNUSED(nb31); GGML_UNUSED(nb01); GGML_UNUSED(nb02);
+        GGML_UNUSED(nb03); GGML_UNUSED(nb11); GGML_UNUSED(nb12);
+        GGML_UNUSED(nb13); GGML_UNUSED(nb21); GGML_UNUSED(nb22);
+        GGML_UNUSED(nb23); GGML_UNUSED(ne0); GGML_UNUSED(ne1);
+        GGML_UNUSED(ne2); GGML_UNUSED(ne3);
        NO_DEVICE_CODE;
        return;
    }
@@ -114,7 +126,7 @@ static __global__ void flash_attn_vec_ext_f32(
            // Set memory to zero if out of bounds:
            if (ncols > 2 && ic0 + j >= ne01) {
 #pragma unroll
-                for (int i0 = 0; i0 < D/sizeof(int); i0 += WARP_SIZE) {
+                for (int i0 = 0; i0 < int(D/sizeof(int)); i0 += WARP_SIZE) {
                    const int i = i0 + threadIdx.x;

                    tmp_q_i32[i] = 0;
@@ -127,7 +139,7 @@ static __global__ void flash_attn_vec_ext_f32(

            const float * Q_f = (const float *) (Q + j*nb01);
 #pragma unroll
-            for (int i0 = 0; i0 < D/sizeof(int); i0 += WARP_SIZE) {
+            for (int i0 = 0; i0 < int(D/sizeof(int)); i0 += WARP_SIZE) {
                quantize_q8_1_to_shared<float2>(Q_f + 4*i0, scale, tmp_q_i32, tmp_q_ds);
            }
        }
@@ -140,7 +152,7 @@ static __global__ void flash_attn_vec_ext_f32(
            float2 * tmp_q_ds  = (float2 *) (tmp_q_i32 + D/sizeof(int));

 #pragma unroll
-            for (int i0 = 0; i0 < D/sizeof(int); i0 += WARP_SIZE) {
+            for (int i0 = 0; i0 < int(D/sizeof(int)); i0 += WARP_SIZE) {
                const int i = i0 + threadIdx.x;

                Q_i32[j][i0/WARP_SIZE] = tmp_q_i32[i];
--- a/licenses/LICENSE-curl
+++ b/licenses/LICENSE-curl
@@ -0,0 +1,9 @@
+Copyright (c) 1996 - 2025, Daniel Stenberg, daniel@haxx.se, and many contributors, see the THANKS file.
+
+All rights reserved.
+
+Permission to use, copy, modify, and distribute this software for any purpose with or without fee is hereby granted, provided that the above copyright notice and this permission notice appear in all copies.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT OF THIRD PARTY RIGHTS. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+Except as contained in this notice, the name of a copyright holder shall not be used in advertising or otherwise to promote the sale, use or other dealings in this Software without prior written authorization of the copyright holder.
--- a/licenses/LICENSE-httplib
+++ b/licenses/LICENSE-httplib
@@ -0,0 +1,21 @@
+The MIT License (MIT)
+
+Copyright (c) 2017 yhirose
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
--- a/licenses/LICENSE-jsonhpp
+++ b/licenses/LICENSE-jsonhpp
@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2013-2025 Niels Lohmann
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
--- a/examples/run/linenoise.cpp/LICENSE
+++ b/examples/run/linenoise.cpp/LICENSE
--- a/tests/CMakeLists.txt
+++ b/tests/CMakeLists.txt
@@ -131,9 +131,13 @@ if (NOT WIN32)
 endif()

 llama_target_and_test(test-log.cpp)
-llama_target_and_test(test-arg-parser.cpp)
 llama_target_and_test(test-chat-template.cpp)

+# this fails on windows (github hosted runner) due to curl DLL not found (exit code 0xc0000135)
+if (NOT WIN32)
+    llama_target_and_test(test-arg-parser.cpp)
+endif()
+
 # llama_target_and_test(test-opt.cpp) # SLOW
 llama_target_and_test(test-gguf.cpp)
 llama_target_and_test(test-backend-ops.cpp)
Author	SHA1	Message	Date
Xuan-Son Nguyen	bd3f59f812	cmake : enable curl by default (#12761 ) * cmake : enable curl by default * no curl if no examples * fix build * fix build-linux-cross * add windows-setup-curl * fix * shell * fix path * fix windows-latest-cmake* * run: include_directories * LLAMA_RUN_EXTRA_LIBS * sycl: no llama_curl * no test-arg-parser on windows * clarification * try riscv64 / arm64 * windows: include libcurl inside release binary * add msg * fix mac / ios / android build * will this fix xcode? * try clearing the cache * add bunch of licenses * revert clear cache * fix xcode * fix xcode (2) * fix typo	2025-04-07 13:35:19 +02:00
zhouwg	52b3d71f12	CANN: fix typo in ggml-cann (#12733 )	2025-04-07 19:34:14 +08:00
hipudding	d0d5b2232b	CANN: Refactor to reduce duplicate code (#12731 ) * CANN: Refactor to reduce duplicate code * CANN: fix review comment	2025-04-07 17:10:36 +08:00
R0CKSTAR	916c83bfe7	musa: fix compilation warnings in mp_22/31 (#12780 ) Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com>	2025-04-06 15:23:54 +02:00