fix(llama.cpp): fix eos without cache

Signed-off-by: GitHub <noreply@github.com>
Co-authored-by: mudler <mudler@users.noreply.github.com>

.dockerignore

@@ -3,3 +3,4 @@ models
 examples/chatbot-ui/models
 examples/rwkv/models
 examples/**/models
+Dockerfile

.env

@@ -18,7 +18,7 @@
 
 ## Default path for models
 #
-MODELS_PATH=/models
+# MODELS_PATH=/models
 
 ## Enable debug mode
 # DEBUG=true

.github/workflows/image-pr.yml

@@ -21,6 +21,7 @@ jobs:
       cuda-minor-version: ${{ matrix.cuda-minor-version }}
       platforms: ${{ matrix.platforms }}
       runs-on: ${{ matrix.runs-on }}
+      base-image: ${{ matrix.base-image }}
     secrets:
       dockerUsername: ${{ secrets.DOCKERHUB_USERNAME }}
       dockerPassword: ${{ secrets.DOCKERHUB_PASSWORD }}
@@ -39,6 +40,7 @@ jobs:
             ffmpeg: 'true'
             image-type: 'extras'
             runs-on: 'arc-runner-set'
+            base-image: "ubuntu:22.04"
           - build-type: 'cublas'
             cuda-major-version: "12"
             cuda-minor-version: "1"
@@ -48,6 +50,23 @@ jobs:
             ffmpeg: 'true'
             image-type: 'extras'
             runs-on: 'arc-runner-set'
+            base-image: "ubuntu:22.04"
+          - build-type: 'hipblas'
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-hipblas'
+            ffmpeg: 'false'
+            image-type: 'extras'
+            base-image: "rocm/dev-ubuntu-22.04:6.0-complete"
+            runs-on: 'arc-runner-set'
+          - build-type: 'sycl_f16'
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            base-image: "intel/oneapi-basekit:2024.0.1-devel-ubuntu22.04"
+            tag-suffix: 'sycl-f16-ffmpeg'
+            ffmpeg: 'true'
+            image-type: 'extras'
+            runs-on: 'arc-runner-set'
   core-image-build:
     uses: ./.github/workflows/image_build.yml
     with:
@@ -60,6 +79,7 @@ jobs:
       cuda-minor-version: ${{ matrix.cuda-minor-version }}
       platforms: ${{ matrix.platforms }}
       runs-on: ${{ matrix.runs-on }}
+      base-image: ${{ matrix.base-image }}
     secrets:
       dockerUsername: ${{ secrets.DOCKERHUB_USERNAME }}
       dockerPassword: ${{ secrets.DOCKERHUB_PASSWORD }}
@@ -75,6 +95,15 @@ jobs:
             ffmpeg: 'true'
             image-type: 'core'
             runs-on: 'ubuntu-latest'
+            base-image: "ubuntu:22.04"
+          - build-type: 'sycl_f16'
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            base-image: "intel/oneapi-basekit:2024.0.1-devel-ubuntu22.04"
+            tag-suffix: 'sycl-f16-ffmpeg-core'
+            ffmpeg: 'true'
+            image-type: 'core'
+            runs-on: 'arc-runner-set'
           - build-type: 'cublas'
             cuda-major-version: "12"
             cuda-minor-version: "1"
@@ -84,3 +113,4 @@ jobs:
             ffmpeg: 'true'
             image-type: 'core'
             runs-on: 'ubuntu-latest'
+            base-image: "ubuntu:22.04"

.github/workflows/image.yml

@@ -13,7 +13,7 @@ concurrency:
   cancel-in-progress: true
 
 jobs:
-  extras-image-build:
+  self-hosted-jobs:
     uses: ./.github/workflows/image_build.yml
     with:
       tag-latest: ${{ matrix.tag-latest }}
@@ -25,6 +25,7 @@ jobs:
       cuda-minor-version: ${{ matrix.cuda-minor-version }}
       platforms: ${{ matrix.platforms }}
       runs-on: ${{ matrix.runs-on }}
+      base-image: ${{ matrix.base-image }}
     secrets:
       dockerUsername: ${{ secrets.DOCKERHUB_USERNAME }}
       dockerPassword: ${{ secrets.DOCKERHUB_PASSWORD }}
@@ -36,6 +37,7 @@ jobs:
       max-parallel: ${{ github.event_name != 'pull_request' && 2 || 4 }}
       matrix:
         include:
+          # Extra images
           - build-type: ''
             #platforms: 'linux/amd64,linux/arm64'
             platforms: 'linux/amd64'
@@ -44,6 +46,7 @@ jobs:
             ffmpeg: ''
             image-type: 'extras'
             runs-on: 'arc-runner-set'
+            base-image: "ubuntu:22.04"
           - build-type: ''
             platforms: 'linux/amd64'
             tag-latest: 'false'
@@ -51,6 +54,7 @@ jobs:
             ffmpeg: 'true'
             image-type: 'extras'
             runs-on: 'arc-runner-set'
+            base-image: "ubuntu:22.04"
           - build-type: 'cublas'
             cuda-major-version: "11"
             cuda-minor-version: "7"
@@ -60,6 +64,7 @@ jobs:
             ffmpeg: ''
             image-type: 'extras'
             runs-on: 'arc-runner-set'
+            base-image: "ubuntu:22.04"
           - build-type: 'cublas'
             cuda-major-version: "12"
             cuda-minor-version: "1"
@@ -69,6 +74,7 @@ jobs:
             ffmpeg: ''
             image-type: 'extras'
             runs-on: 'arc-runner-set'
+            base-image: "ubuntu:22.04"
           - build-type: 'cublas'
             cuda-major-version: "11"
             cuda-minor-version: "7"
@@ -78,6 +84,7 @@ jobs:
             ffmpeg: 'true'
             image-type: 'extras'
             runs-on: 'arc-runner-set'
+            base-image: "ubuntu:22.04"
           - build-type: 'cublas'
             cuda-major-version: "12"
             cuda-minor-version: "1"
@@ -87,6 +94,7 @@ jobs:
             ffmpeg: 'true'
             image-type: 'extras'
             runs-on: 'arc-runner-set'
+            base-image: "ubuntu:22.04"
           - build-type: ''
             #platforms: 'linux/amd64,linux/arm64'
             platforms: 'linux/amd64'
@@ -94,7 +102,90 @@ jobs:
             tag-suffix: ''
             ffmpeg: ''
             image-type: 'extras'
+            base-image: "ubuntu:22.04"
             runs-on: 'arc-runner-set'
+          - build-type: 'hipblas'
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-hipblas-ffmpeg'
+            ffmpeg: 'true'
+            image-type: 'extras'
+            base-image: "rocm/dev-ubuntu-22.04:6.0-complete"
+            runs-on: 'arc-runner-set'
+          - build-type: 'hipblas'
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-hipblas'
+            ffmpeg: 'false'
+            image-type: 'extras'
+            base-image: "rocm/dev-ubuntu-22.04:6.0-complete"
+            runs-on: 'arc-runner-set'
+          - build-type: 'sycl_f16'
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            base-image: "intel/oneapi-basekit:2024.0.1-devel-ubuntu22.04"
+            tag-suffix: '-sycl-f16-ffmpeg'
+            ffmpeg: 'true'
+            image-type: 'extras'
+            runs-on: 'arc-runner-set'
+          - build-type: 'sycl_f32'
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            base-image: "intel/oneapi-basekit:2024.0.1-devel-ubuntu22.04"
+            tag-suffix: '-sycl-f32-ffmpeg'
+            ffmpeg: 'true'
+            image-type: 'extras'
+            runs-on: 'arc-runner-set'
+          # Core images
+          - build-type: 'sycl_f16'
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            base-image: "intel/oneapi-basekit:2024.0.1-devel-ubuntu22.04"
+            tag-suffix: '-sycl-f16-core'
+            ffmpeg: 'false'
+            image-type: 'core'
+            runs-on: 'arc-runner-set'
+          - build-type: 'sycl_f32'
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            base-image: "intel/oneapi-basekit:2024.0.1-devel-ubuntu22.04"
+            tag-suffix: '-sycl-f32-core'
+            ffmpeg: 'false'
+            image-type: 'core'
+            runs-on: 'arc-runner-set'
+          - build-type: 'sycl_f16'
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            base-image: "intel/oneapi-basekit:2024.0.1-devel-ubuntu22.04"
+            tag-suffix: '-sycl-f16-ffmpeg-core'
+            ffmpeg: 'true'
+            image-type: 'core'
+            runs-on: 'arc-runner-set'
+          - build-type: 'sycl_f32'
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            base-image: "intel/oneapi-basekit:2024.0.1-devel-ubuntu22.04"
+            tag-suffix: '-sycl-f32-ffmpeg-core'
+            ffmpeg: 'true'
+            image-type: 'core'
+            runs-on: 'arc-runner-set'
+          - build-type: 'hipblas'
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-hipblas-ffmpeg-core'
+            ffmpeg: 'true'
+            image-type: 'core'
+            base-image: "rocm/dev-ubuntu-22.04:6.0-complete"
+            runs-on: 'arc-runner-set'
+          - build-type: 'hipblas'
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-hipblas-core'
+            ffmpeg: 'false'
+            image-type: 'core'
+            base-image: "rocm/dev-ubuntu-22.04:6.0-complete"
+            runs-on: 'arc-runner-set'
+  
   core-image-build:
     uses: ./.github/workflows/image_build.yml
     with:
@@ -107,6 +198,7 @@ jobs:
       cuda-minor-version: ${{ matrix.cuda-minor-version }}
       platforms: ${{ matrix.platforms }}
       runs-on: ${{ matrix.runs-on }}
+      base-image: ${{ matrix.base-image }}
     secrets:
       dockerUsername: ${{ secrets.DOCKERHUB_USERNAME }}
       dockerPassword: ${{ secrets.DOCKERHUB_PASSWORD }}
@@ -121,6 +213,7 @@ jobs:
             tag-suffix: '-ffmpeg-core'
             ffmpeg: 'true'
             image-type: 'core'
+            base-image: "ubuntu:22.04"
             runs-on: 'ubuntu-latest'
           - build-type: 'cublas'
             cuda-major-version: "11"
@@ -130,6 +223,7 @@ jobs:
             tag-suffix: '-cublas-cuda11-core'
             ffmpeg: ''
             image-type: 'core'
+            base-image: "ubuntu:22.04"
             runs-on: 'ubuntu-latest'
           - build-type: 'cublas'
             cuda-major-version: "12"
@@ -139,6 +233,7 @@ jobs:
             tag-suffix: '-cublas-cuda12-core'
             ffmpeg: ''
             image-type: 'core'
+            base-image: "ubuntu:22.04"
             runs-on: 'ubuntu-latest'
           - build-type: 'cublas'
             cuda-major-version: "11"
@@ -149,6 +244,7 @@ jobs:
             ffmpeg: 'true'
             image-type: 'core'
             runs-on: 'ubuntu-latest'
+            base-image: "ubuntu:22.04"
           - build-type: 'cublas'
             cuda-major-version: "12"
             cuda-minor-version: "1"
@@ -158,3 +254,4 @@ jobs:
             ffmpeg: 'true'
             image-type: 'core'
             runs-on: 'ubuntu-latest'
+            base-image: "ubuntu:22.04"

.github/workflows/image_build.yml

@@ -4,6 +4,11 @@ name: 'build container images (reusable)'
 on:
   workflow_call:
     inputs:
+      base-image:
+        description: 'Base image'
+        required: false
+        default: ''
+        type: string
       build-type:
         description: 'Build type'
         default: ''
@@ -64,42 +69,47 @@ jobs:
           && sudo apt-get install -y git
       - name: Checkout
         uses: actions/checkout@v4
-      # - name: Release space from worker
-      #   run: |
-      #     echo "Listing top largest packages"
-      #     pkgs=$(dpkg-query -Wf '${Installed-Size}\t${Package}\t${Status}\n' | awk '$NF == "installed"{print $1 "\t" $2}' | sort -nr)
-      #     head -n 30 <<< "${pkgs}"
-      #     echo
-      #     df -h
-      #     echo
-      #     sudo apt-get remove -y '^llvm-.*|^libllvm.*' || true
-      #     sudo apt-get remove --auto-remove android-sdk-platform-tools || true
-      #     sudo apt-get purge --auto-remove android-sdk-platform-tools || true
-      #     sudo rm -rf /usr/local/lib/android
-      #     sudo apt-get remove -y '^dotnet-.*|^aspnetcore-.*' || true
-      #     sudo rm -rf /usr/share/dotnet
-      #     sudo apt-get remove -y '^mono-.*' || true
-      #     sudo apt-get remove -y '^ghc-.*' || true
-      #     sudo apt-get remove -y '.*jdk.*|.*jre.*' || true
-      #     sudo apt-get remove -y 'php.*' || true
-      #     sudo apt-get remove -y hhvm powershell firefox monodoc-manual msbuild || true
-      #     sudo apt-get remove -y '^google-.*' || true
-      #     sudo apt-get remove -y azure-cli || true
-      #     sudo apt-get remove -y '^mongo.*-.*|^postgresql-.*|^mysql-.*|^mssql-.*' || true
-      #     sudo apt-get remove -y '^gfortran-.*' || true
-      #     sudo apt-get remove -y microsoft-edge-stable || true
-      #     sudo apt-get remove -y firefox || true
-      #     sudo apt-get remove -y powershell || true
-      #     sudo apt-get remove -y r-base-core || true
-      #     sudo apt-get autoremove -y
-      #     sudo apt-get clean
-      #     echo
-      #     echo "Listing top largest packages"
-      #     pkgs=$(dpkg-query -Wf '${Installed-Size}\t${Package}\t${Status}\n' | awk '$NF == "installed"{print $1 "\t" $2}' | sort -nr)
-      #     head -n 30 <<< "${pkgs}"
-      #     echo
-      #     sudo rm -rfv build || true
-      #     df -h
+      - name: Release space from worker
+        if: inputs.runs-on == 'ubuntu-latest'
+        run: |
+          echo "Listing top largest packages"
+          pkgs=$(dpkg-query -Wf '${Installed-Size}\t${Package}\t${Status}\n' | awk '$NF == "installed"{print $1 "\t" $2}' | sort -nr)
+          head -n 30 <<< "${pkgs}"
+          echo
+          df -h
+          echo
+          sudo apt-get remove -y '^llvm-.*|^libllvm.*' || true
+          sudo apt-get remove --auto-remove android-sdk-platform-tools || true
+          sudo apt-get purge --auto-remove android-sdk-platform-tools || true
+          sudo rm -rf /usr/local/lib/android
+          sudo apt-get remove -y '^dotnet-.*|^aspnetcore-.*' || true
+          sudo rm -rf /usr/share/dotnet
+          sudo apt-get remove -y '^mono-.*' || true
+          sudo apt-get remove -y '^ghc-.*' || true
+          sudo apt-get remove -y '.*jdk.*|.*jre.*' || true
+          sudo apt-get remove -y 'php.*' || true
+          sudo apt-get remove -y hhvm powershell firefox monodoc-manual msbuild || true
+          sudo apt-get remove -y '^google-.*' || true
+          sudo apt-get remove -y azure-cli || true
+          sudo apt-get remove -y '^mongo.*-.*|^postgresql-.*|^mysql-.*|^mssql-.*' || true
+          sudo apt-get remove -y '^gfortran-.*' || true
+          sudo apt-get remove -y microsoft-edge-stable || true
+          sudo apt-get remove -y firefox || true
+          sudo apt-get remove -y powershell || true
+          sudo apt-get remove -y r-base-core || true
+          sudo apt-get autoremove -y
+          sudo apt-get clean
+          echo
+          echo "Listing top largest packages"
+          pkgs=$(dpkg-query -Wf '${Installed-Size}\t${Package}\t${Status}\n' | awk '$NF == "installed"{print $1 "\t" $2}' | sort -nr)
+          head -n 30 <<< "${pkgs}"
+          echo
+          sudo rm -rfv build || true
+          sudo rm -rf /usr/share/dotnet || true
+          sudo rm -rf /opt/ghc || true
+          sudo rm -rf "/usr/local/share/boost" || true
+          sudo rm -rf "$AGENT_TOOLSDIRECTORY" || true
+          df -h
       - name: Docker meta
         id: meta
         uses: docker/metadata-action@v5
@@ -149,6 +159,7 @@ jobs:
             CUDA_MINOR_VERSION=${{ inputs.cuda-minor-version }}
             FFMPEG=${{ inputs.ffmpeg }}
             IMAGE_TYPE=${{ inputs.image-type }}
+            BASE_IMAGE=${{ inputs.base-image }}
           context: .
           file: ./Dockerfile
           platforms: ${{ inputs.platforms }}

.github/workflows/release.yaml

@@ -20,6 +20,10 @@ jobs:
             defines: '-DLLAMA_AVX2=OFF'
           - build: 'avx512'
             defines: '-DLLAMA_AVX512=ON'
+          - build: 'cuda12'
+            defines: ''
+          - build: 'cuda11'
+            defines: ''
     runs-on: ubuntu-latest
     steps:
       - name: Clone
@@ -33,7 +37,18 @@ jobs:
         run: |
           sudo apt-get update
           sudo apt-get install build-essential ffmpeg
-
+      - name: Install CUDA Dependencies
+        if: ${{ matrix.build == 'cuda12' || matrix.build == 'cuda11' }}
+        run: |
+          if [ "${{ matrix.build }}" == "cuda12" ]; then
+            export CUDA_VERSION=12-3
+          else
+            export CUDA_VERSION=11-7
+          fi
+          curl -O https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/cuda-keyring_1.1-1_all.deb
+          sudo dpkg -i cuda-keyring_1.1-1_all.deb
+          sudo apt-get update
+          sudo apt-get install -y cuda-nvcc-${CUDA_VERSION} libcublas-dev-${CUDA_VERSION}
       - name: Cache grpc
         id: cache-grpc
         uses: actions/cache@v3
@@ -50,14 +65,19 @@ jobs:
       - name: Install gRPC
         run: |
           cd grpc && cd cmake/build && sudo make -j12 install
-
       - name: Build
         id: build
         env:
           CMAKE_ARGS: "${{ matrix.defines }}"
           BUILD_ID: "${{ matrix.build }}"
         run: |
-          STATIC=true make dist
+          if [ "${{ matrix.build }}" == "cuda12" ] || [ "${{ matrix.build }}" == "cuda11" ]; then
+            export BUILD_TYPE=cublas
+            export PATH=/usr/local/cuda/bin:$PATH
+            make dist
+          else
+            STATIC=true make dist
+          fi
       - uses: actions/upload-artifact@v3
         with:
           name: ${{ matrix.build }}
@@ -69,6 +89,35 @@ jobs:
           files: |
             release/*
 
+  build-stablediffusion:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Clone
+        uses: actions/checkout@v4
+        with:
+          submodules: true
+      - uses: actions/setup-go@v4
+        with:
+          go-version: '>=1.21.0'
+      - name: Dependencies
+        run: |
+          sudo apt-get install -y --no-install-recommends libopencv-dev
+          sudo ln -s /usr/include/opencv4/opencv2 /usr/include/opencv2
+      - name: Build stablediffusion
+        run: |
+          make backend-assets/grpc/stablediffusion
+          mkdir -p release && cp backend-assets/grpc/stablediffusion release
+      - uses: actions/upload-artifact@v3
+        with:
+          name: stablediffusion
+          path: release/
+      - name: Release
+        uses: softprops/action-gh-release@v1
+        if: startsWith(github.ref, 'refs/tags/')
+        with:
+          files: |
+            release/*
+
   build-macOS:
     strategy:
       matrix:
@@ -109,4 +158,4 @@ jobs:
         if: startsWith(github.ref, 'refs/tags/')
         with:
           files: |
-            release/*
+            release/*

.github/workflows/test-extra.yml

@@ -164,34 +164,74 @@ jobs:
 
            
 
-  tests-bark:
-    runs-on: ubuntu-latest
-    steps:
-      - name: Clone
-        uses: actions/checkout@v4
-        with: 
-          submodules: true
-      - name: Dependencies
-        run: |
-          sudo apt-get update
-          sudo apt-get install build-essential ffmpeg
-          curl https://repo.anaconda.com/pkgs/misc/gpgkeys/anaconda.asc | gpg --dearmor > conda.gpg && \
-             sudo install -o root -g root -m 644 conda.gpg /usr/share/keyrings/conda-archive-keyring.gpg && \
-              gpg --keyring /usr/share/keyrings/conda-archive-keyring.gpg --no-default-keyring --fingerprint 34161F5BF5EB1D4BFBBB8F0A8AEB4F8B29D82806 && \
-             sudo /bin/bash -c 'echo "deb [arch=amd64 signed-by=/usr/share/keyrings/conda-archive-keyring.gpg] https://repo.anaconda.com/pkgs/misc/debrepo/conda stable main" > /etc/apt/sources.list.d/conda.list' && \
-             sudo /bin/bash -c 'echo "deb [arch=amd64 signed-by=/usr/share/keyrings/conda-archive-keyring.gpg] https://repo.anaconda.com/pkgs/misc/debrepo/conda stable main" | tee -a /etc/apt/sources.list.d/conda.list' && \
-             sudo apt-get update && \
-             sudo apt-get install -y conda
-          sudo apt-get install -y ca-certificates cmake curl patch
-          sudo apt-get install -y libopencv-dev && sudo ln -s /usr/include/opencv4/opencv2 /usr/include/opencv2
+  # tests-bark:
+  #   runs-on: ubuntu-latest
+  #   steps:
+  #     - name: Release space from worker
+  #       run: |
+  #           echo "Listing top largest packages"
+  #           pkgs=$(dpkg-query -Wf '${Installed-Size}\t${Package}\t${Status}\n' | awk '$NF == "installed"{print $1 "\t" $2}' | sort -nr)
+  #           head -n 30 <<< "${pkgs}"
+  #           echo
+  #           df -h
+  #           echo
+  #           sudo apt-get remove -y '^llvm-.*|^libllvm.*' || true
+  #           sudo apt-get remove --auto-remove android-sdk-platform-tools || true
+  #           sudo apt-get purge --auto-remove android-sdk-platform-tools || true
+  #           sudo rm -rf /usr/local/lib/android
+  #           sudo apt-get remove -y '^dotnet-.*|^aspnetcore-.*' || true
+  #           sudo rm -rf /usr/share/dotnet
+  #           sudo apt-get remove -y '^mono-.*' || true
+  #           sudo apt-get remove -y '^ghc-.*' || true
+  #           sudo apt-get remove -y '.*jdk.*|.*jre.*' || true
+  #           sudo apt-get remove -y 'php.*' || true
+  #           sudo apt-get remove -y hhvm powershell firefox monodoc-manual msbuild || true
+  #           sudo apt-get remove -y '^google-.*' || true
+  #           sudo apt-get remove -y azure-cli || true
+  #           sudo apt-get remove -y '^mongo.*-.*|^postgresql-.*|^mysql-.*|^mssql-.*' || true
+  #           sudo apt-get remove -y '^gfortran-.*' || true
+  #           sudo apt-get remove -y microsoft-edge-stable || true
+  #           sudo apt-get remove -y firefox || true
+  #           sudo apt-get remove -y powershell || true
+  #           sudo apt-get remove -y r-base-core || true
+  #           sudo apt-get autoremove -y
+  #           sudo apt-get clean
+  #           echo
+  #           echo "Listing top largest packages"
+  #           pkgs=$(dpkg-query -Wf '${Installed-Size}\t${Package}\t${Status}\n' | awk '$NF == "installed"{print $1 "\t" $2}' | sort -nr)
+  #           head -n 30 <<< "${pkgs}"
+  #           echo
+  #           sudo rm -rfv build || true
+  #           sudo rm -rf /usr/share/dotnet || true
+  #           sudo rm -rf /opt/ghc || true
+  #           sudo rm -rf "/usr/local/share/boost" || true
+  #           sudo rm -rf "$AGENT_TOOLSDIRECTORY" || true
+  #           df -h
+  #     - name: Clone
+  #       uses: actions/checkout@v4
+  #       with: 
+  #         submodules: true
+  #     - name: Dependencies
+  #       run: |
+  #         sudo apt-get update
+  #         sudo apt-get install build-essential ffmpeg
+  #         curl https://repo.anaconda.com/pkgs/misc/gpgkeys/anaconda.asc | gpg --dearmor > conda.gpg && \
+  #            sudo install -o root -g root -m 644 conda.gpg /usr/share/keyrings/conda-archive-keyring.gpg && \
+  #             gpg --keyring /usr/share/keyrings/conda-archive-keyring.gpg --no-default-keyring --fingerprint 34161F5BF5EB1D4BFBBB8F0A8AEB4F8B29D82806 && \
+  #            sudo /bin/bash -c 'echo "deb [arch=amd64 signed-by=/usr/share/keyrings/conda-archive-keyring.gpg] https://repo.anaconda.com/pkgs/misc/debrepo/conda stable main" > /etc/apt/sources.list.d/conda.list' && \
+  #            sudo /bin/bash -c 'echo "deb [arch=amd64 signed-by=/usr/share/keyrings/conda-archive-keyring.gpg] https://repo.anaconda.com/pkgs/misc/debrepo/conda stable main" | tee -a /etc/apt/sources.list.d/conda.list' && \
+  #            sudo apt-get update && \
+  #            sudo apt-get install -y conda
+  #         sudo apt-get install -y ca-certificates cmake curl patch
+  #         sudo apt-get install -y libopencv-dev && sudo ln -s /usr/include/opencv4/opencv2 /usr/include/opencv2
           
-          sudo rm -rfv /usr/bin/conda || true
+  #         sudo rm -rfv /usr/bin/conda || true
 
-      - name: Test bark
-        run: |
-           export PATH=$PATH:/opt/conda/bin
-           make -C backend/python/bark
-           make -C backend/python/bark test
+  #     - name: Test bark
+  #       run: |
+  #          export PATH=$PATH:/opt/conda/bin
+  #          make -C backend/python/bark
+  #          make -C backend/python/bark test
 
            
   # Below tests needs GPU. Commented out for now
@@ -274,4 +314,4 @@ jobs:
         run: |
            export PATH=$PATH:/opt/conda/bin
            make -C backend/python/coqui
-           make -C backend/python/coqui test
+           make -C backend/python/coqui test

.gitignore

@@ -21,6 +21,7 @@ local-ai
 !charts/*
 # prevent above rules from omitting the api/localai folder
 !api/localai
+!core/**/localai
 
 # Ignore models
 models/*
@@ -34,6 +35,7 @@ release/
 .idea
 
 # Generated during build
-backend-assets/
+backend-assets/*
+!backend-assets/.keep
 prepare
 /ggml-metal.metal

Dockerfile

@@ -1,10 +1,12 @@
-ARG GO_VERSION=1.21-bullseye
 ARG IMAGE_TYPE=extras
+ARG BASE_IMAGE=ubuntu:22.04
+
 # extras or core
+FROM ${BASE_IMAGE} as requirements-core
 
+USER root
 
-FROM golang:$GO_VERSION as requirements-core
-
+ARG GO_VERSION=1.21.7
 ARG BUILD_TYPE
 ARG CUDA_MAJOR_VERSION=11
 ARG CUDA_MINOR_VERSION=7
@@ -12,14 +14,17 @@ ARG TARGETARCH
 ARG TARGETVARIANT
 
 ENV BUILD_TYPE=${BUILD_TYPE}
-
+ENV DEBIAN_FRONTEND=noninteractive
 ENV EXTERNAL_GRPC_BACKENDS="coqui:/build/backend/python/coqui/run.sh,huggingface-embeddings:/build/backend/python/sentencetransformers/run.sh,petals:/build/backend/python/petals/run.sh,transformers:/build/backend/python/transformers/run.sh,sentencetransformers:/build/backend/python/sentencetransformers/run.sh,autogptq:/build/backend/python/autogptq/run.sh,bark:/build/backend/python/bark/run.sh,diffusers:/build/backend/python/diffusers/run.sh,exllama:/build/backend/python/exllama/run.sh,vall-e-x:/build/backend/python/vall-e-x/run.sh,vllm:/build/backend/python/vllm/run.sh,mamba:/build/backend/python/mamba/run.sh,exllama2:/build/backend/python/exllama2/run.sh,transformers-musicgen:/build/backend/python/transformers-musicgen/run.sh"
 
 ARG GO_TAGS="stablediffusion tinydream tts"
 
 RUN apt-get update && \
-    apt-get install -y ca-certificates curl patch pip cmake && apt-get clean
+    apt-get install -y ca-certificates curl patch pip cmake git && apt-get clean
 
+# Install Go
+RUN curl -L -s https://go.dev/dl/go$GO_VERSION.linux-$TARGETARCH.tar.gz | tar -C /usr/local -xz
+ENV PATH $PATH:/usr/local/go/bin
 
 COPY --chmod=644 custom-ca-certs/* /usr/local/share/ca-certificates/
 RUN update-ca-certificates
@@ -31,15 +36,19 @@ RUN echo "Target Variant: $TARGETVARIANT"
 # CuBLAS requirements
 RUN if [ "${BUILD_TYPE}" = "cublas" ]; then \
     apt-get install -y software-properties-common && \
-    apt-add-repository contrib && \
-    curl -O https://developer.download.nvidia.com/compute/cuda/repos/debian11/x86_64/cuda-keyring_1.0-1_all.deb && \
-    dpkg -i cuda-keyring_1.0-1_all.deb && \
-    rm -f cuda-keyring_1.0-1_all.deb && \
+    curl -O https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/cuda-keyring_1.1-1_all.deb && \
+    dpkg -i cuda-keyring_1.1-1_all.deb && \
+    rm -f cuda-keyring_1.1-1_all.deb && \
     apt-get update && \
-    apt-get install -y cuda-nvcc-${CUDA_MAJOR_VERSION}-${CUDA_MINOR_VERSION} libcublas-dev-${CUDA_MAJOR_VERSION}-${CUDA_MINOR_VERSION} libcusparse-dev-${CUDA_MAJOR_VERSION}-${CUDA_MINOR_VERSION} libcusolver-dev-${CUDA_MAJOR_VERSION}-${CUDA_MINOR_VERSION}  && apt-get clean \
+    apt-get install -y cuda-nvcc-${CUDA_MAJOR_VERSION}-${CUDA_MINOR_VERSION} libcurand-dev-${CUDA_MAJOR_VERSION}-${CUDA_MINOR_VERSION} libcublas-dev-${CUDA_MAJOR_VERSION}-${CUDA_MINOR_VERSION} libcusparse-dev-${CUDA_MAJOR_VERSION}-${CUDA_MINOR_VERSION} libcusolver-dev-${CUDA_MAJOR_VERSION}-${CUDA_MINOR_VERSION}  && apt-get clean \
     ; fi
+
+# Cuda
 ENV PATH /usr/local/cuda/bin:${PATH}
 
+# HipBLAS requirements
+ENV PATH /opt/rocm/bin:${PATH}
+
 # OpenBLAS requirements and stable diffusion
 RUN apt-get install -y \
     libopenblas-dev \
@@ -66,10 +75,16 @@ RUN curl https://repo.anaconda.com/pkgs/misc/gpgkeys/anaconda.asc | gpg --dearmo
     apt-get install -y conda && apt-get clean
 
 ENV PATH="/root/.cargo/bin:${PATH}"
+RUN apt-get install -y python3-pip && apt-get clean
 RUN pip install --upgrade pip
+
 RUN curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y
 RUN apt-get install -y espeak-ng espeak && apt-get clean
 
+RUN if [ ! -e /usr/bin/python ]; then \
+	  ln -s /usr/bin/python3 /usr/bin/python \
+    ; fi
+
 ###################################
 ###################################
 
@@ -90,6 +105,13 @@ COPY . .
 COPY .git .
 RUN make prepare
 
+# If we are building with clblas support, we need the libraries for the builds
+RUN if [ "${BUILD_TYPE}" = "clblas" ]; then \
+    apt-get update && \
+    apt-get install -y libclblast-dev && \
+    apt-get clean \
+    ; fi
+
 # stablediffusion does not tolerate a newer version of abseil, build it first
 RUN GRPC_BACKENDS=backend-assets/grpc/stablediffusion make build
 
@@ -133,6 +155,13 @@ RUN if [ "${FFMPEG}" = "true" ]; then \
     apt-get install -y ffmpeg && apt-get clean \
     ; fi
 
+# Add OpenCL
+RUN if [ "${BUILD_TYPE}" = "clblas" ]; then \
+    apt-get update && \
+    apt-get install -y libclblast1 && \
+    apt-get clean \
+    ; fi
+
 WORKDIR /build
 
 # we start fresh & re-copy all assets because `make build` does not clean up nicely after itself
@@ -157,43 +186,43 @@ COPY --from=builder /build/backend-assets/grpc/stablediffusion ./backend-assets/
 
 ## Duplicated from Makefile to avoid having a big layer that's hard to push
 RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
-	PATH=$PATH:/opt/conda/bin make -C backend/python/autogptq \
+	 make -C backend/python/autogptq \
     ; fi
 RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
-	PATH=$PATH:/opt/conda/bin make -C backend/python/bark \
+	 make -C backend/python/bark \
     ; fi
 RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
-	PATH=$PATH:/opt/conda/bin make -C backend/python/diffusers \
+	 make -C backend/python/diffusers \
     ; fi
 RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
-	PATH=$PATH:/opt/conda/bin make -C backend/python/vllm \
+	 make -C backend/python/vllm \
     ; fi
 RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
-	PATH=$PATH:/opt/conda/bin make -C backend/python/mamba \
+	 make -C backend/python/mamba \
     ; fi
 RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
-	PATH=$PATH:/opt/conda/bin make -C backend/python/sentencetransformers \
+	 make -C backend/python/sentencetransformers \
     ; fi
 RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
-	PATH=$PATH:/opt/conda/bin make -C backend/python/transformers \
+	 make -C backend/python/transformers \
     ; fi
 RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
-	PATH=$PATH:/opt/conda/bin make -C backend/python/vall-e-x \
+	 make -C backend/python/vall-e-x \
     ; fi
 RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
-	PATH=$PATH:/opt/conda/bin make -C backend/python/exllama \
+	 make -C backend/python/exllama \
     ; fi
 RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
-    PATH=$PATH:/opt/conda/bin make -C backend/python/exllama2 \
+     make -C backend/python/exllama2 \
     ; fi
 RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
-	PATH=$PATH:/opt/conda/bin make -C backend/python/petals \
+	 make -C backend/python/petals \
     ; fi
 RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
-	PATH=$PATH:/opt/conda/bin make -C backend/python/transformers-musicgen \
+	 make -C backend/python/transformers-musicgen \
     ; fi
 RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
-	PATH=$PATH:/opt/conda/bin make -C backend/python/coqui \
+	 make -C backend/python/coqui \
     ; fi
 
 # Make sure the models directory exists

Makefile

@@ -4,34 +4,31 @@ GOVET=$(GOCMD) vet
 BINARY_NAME=local-ai
 
 # llama.cpp versions
-GOLLAMA_VERSION?=aeba71ee842819da681ea537e78846dc75949ac0
+GOLLAMA_VERSION?=6a8041ef6b46d4712afc3ae791d1c2d73da0ad1c
 
 GOLLAMA_STABLE_VERSION?=50cee7712066d9e38306eccadcfbb44ea87df4b7
 
-CPPLLAMA_VERSION?=381ee195721d8e747ee31a60c0751822b3072f02
+CPPLLAMA_VERSION?=4755afd1cbd40d93c017e5b98c39796f52345314
 
 # gpt4all version
 GPT4ALL_REPO?=https://github.com/nomic-ai/gpt4all
 GPT4ALL_VERSION?=27a8b020c36b0df8f8b82a252d261cda47cf44b8
 
-# go-ggml-transformers version
-GOGGMLTRANSFORMERS_VERSION?=ffb09d7dd71e2cbc6c5d7d05357d230eea6f369a
-
 # go-rwkv version
 RWKV_REPO?=https://github.com/donomii/go-rwkv.cpp
-RWKV_VERSION?=633c5a3485c403cb2520693dc0991a25dace9f0f
+RWKV_VERSION?=661e7ae26d442f5cfebd2a0881b44e8c55949ec6
 
 # whisper.cpp version
-WHISPER_CPP_VERSION?=37a709f6558c6d9783199e2b8cbb136e1c41d346
+WHISPER_CPP_VERSION?=a56f435fd475afd7edf02bfbf9f8c77f527198c2
 
 # bert.cpp version
 BERT_VERSION?=6abe312cded14042f6b7c3cd8edf082713334a4d
 
 # go-piper version
-PIPER_VERSION?=d6b6275ba037dabdba4a8b65dfdf6b2a73a67f07
+PIPER_VERSION?=9d0100873a7dbb0824dfea40e8cec70a1b110759
 
 # stablediffusion version
-STABLEDIFFUSION_VERSION?=902db5f066fd137697e3b69d0fa10d4782bd2c2f
+STABLEDIFFUSION_VERSION?=362df9da29f882dbf09ade61972d16a1f53c3485
 
 # tinydream version
 TINYDREAM_VERSION?=772a9c0d9aaf768290e63cca3c904fe69faf677a
@@ -47,6 +44,8 @@ BUILD_ID?=git
 
 TEST_DIR=/tmp/test
 
+TEST_FLAKES?=5
+
 RANDOM := $(shell bash -c 'echo $$RANDOM')
 
 VERSION?=$(shell git describe --always --tags || echo "dev" )
@@ -92,14 +91,19 @@ ifeq ($(BUILD_TYPE),openblas)
 	export WHISPER_OPENBLAS=1
 endif
 
+
 ifeq ($(BUILD_TYPE),cublas)
-	CGO_LDFLAGS+=-lcublas -lcudart -L$(CUDA_LIBPATH)
+	CGO_LDFLAGS+=-lcublas -lcudart -lculibos -lcublasLt -L$(CUDA_LIBPATH)
 	export LLAMA_CUBLAS=1
+# required by whisper.cpp
 	export WHISPER_CUBLAS=1
+	CGO_LDFLAGS+=-L$(CUDA_PATH)/stubs -lcuda
 endif
 
 ifeq ($(BUILD_TYPE),hipblas)
 	ROCM_HOME ?= /opt/rocm
+	ROCM_PATH ?= /opt/rocm
+	LD_LIBRARY_PATH ?= /opt/rocm/lib:/opt/rocm/llvm/lib
 	export CXX=$(ROCM_HOME)/llvm/bin/clang++
 	export CC=$(ROCM_HOME)/llvm/bin/clang
 	# llama-ggml has no hipblas support, so override it here.
@@ -108,7 +112,7 @@ ifeq ($(BUILD_TYPE),hipblas)
 	GPU_TARGETS ?= gfx900,gfx90a,gfx1030,gfx1031,gfx1100
 	AMDGPU_TARGETS ?= "$(GPU_TARGETS)"
 	CMAKE_ARGS+=-DLLAMA_HIPBLAS=ON -DAMDGPU_TARGETS="$(AMDGPU_TARGETS)" -DGPU_TARGETS="$(GPU_TARGETS)"
-	CGO_LDFLAGS += -O3 --rtlib=compiler-rt -unwindlib=libgcc -lhipblas -lrocblas --hip-link
+	CGO_LDFLAGS += -O3 --rtlib=compiler-rt -unwindlib=libgcc -lhipblas -lrocblas --hip-link -L${ROCM_HOME}/lib/llvm/lib
 endif
 
 ifeq ($(BUILD_TYPE),metal)
@@ -145,8 +149,18 @@ ifeq ($(findstring tts,$(GO_TAGS)),tts)
 	OPTIONAL_GRPC+=backend-assets/grpc/piper
 endif
 
-ALL_GRPC_BACKENDS=backend-assets/grpc/langchain-huggingface backend-assets/grpc/falcon-ggml backend-assets/grpc/bert-embeddings backend-assets/grpc/llama backend-assets/grpc/llama-cpp backend-assets/grpc/llama-ggml backend-assets/grpc/gpt4all backend-assets/grpc/dolly backend-assets/grpc/gpt2 backend-assets/grpc/gptj backend-assets/grpc/gptneox backend-assets/grpc/mpt backend-assets/grpc/replit backend-assets/grpc/starcoder backend-assets/grpc/rwkv backend-assets/grpc/whisper $(OPTIONAL_GRPC)
+ALL_GRPC_BACKENDS=backend-assets/grpc/langchain-huggingface
+ALL_GRPC_BACKENDS+=backend-assets/grpc/bert-embeddings
+ALL_GRPC_BACKENDS+=backend-assets/grpc/llama
+ALL_GRPC_BACKENDS+=backend-assets/grpc/llama-cpp
+ALL_GRPC_BACKENDS+=backend-assets/grpc/llama-ggml
+ALL_GRPC_BACKENDS+=backend-assets/grpc/gpt4all
+ALL_GRPC_BACKENDS+=backend-assets/grpc/rwkv
+ALL_GRPC_BACKENDS+=backend-assets/grpc/whisper
+ALL_GRPC_BACKENDS+=$(OPTIONAL_GRPC)
+
 GRPC_BACKENDS?=$(ALL_GRPC_BACKENDS) $(OPTIONAL_GRPC)
+TEST_PATHS?=./api/... ./pkg/... ./core/...
 
 # If empty, then we build all
 ifeq ($(GRPC_BACKENDS),)
@@ -217,14 +231,6 @@ backend-assets/espeak-ng-data: sources/go-piper
 sources/gpt4all/gpt4all-bindings/golang/libgpt4all.a: sources/gpt4all
 	$(MAKE) -C sources/gpt4all/gpt4all-bindings/golang/ libgpt4all.a
 
-## CEREBRAS GPT
-sources/go-ggml-transformers:
-	git clone --recurse-submodules https://github.com/go-skynet/go-ggml-transformers.cpp sources/go-ggml-transformers
-	cd sources/go-ggml-transformers && git checkout -b build $(GOGPT2_VERSION) && git submodule update --init --recursive --depth 1
-
-sources/go-ggml-transformers/libtransformers.a: sources/go-ggml-transformers
-	$(MAKE) -C sources/go-ggml-transformers BUILD_TYPE=$(BUILD_TYPE) libtransformers.a
-
 sources/whisper.cpp:
 	git clone https://github.com/ggerganov/whisper.cpp.git sources/whisper.cpp
 	cd sources/whisper.cpp && git checkout -b build $(WHISPER_CPP_VERSION) && git submodule update --init --recursive --depth 1
@@ -250,14 +256,13 @@ sources/go-piper/libpiper_binding.a: sources/go-piper
 	$(MAKE) -C sources/go-piper libpiper_binding.a example/main
 
 backend/cpp/llama/llama.cpp:
-	LLAMA_VERSION=$(CPPLLAMA_VERSION) $(MAKE) -C backend/cpp/llama llama.cpp	
+	LLAMA_VERSION=$(CPPLLAMA_VERSION) $(MAKE) -C backend/cpp/llama llama.cpp
 
-get-sources: backend/cpp/llama/llama.cpp sources/go-llama sources/go-llama-ggml sources/go-ggml-transformers sources/gpt4all sources/go-piper sources/go-rwkv sources/whisper.cpp sources/go-bert sources/go-stable-diffusion sources/go-tiny-dream
+get-sources: backend/cpp/llama/llama.cpp sources/go-llama sources/go-llama-ggml sources/gpt4all sources/go-piper sources/go-rwkv sources/whisper.cpp sources/go-bert sources/go-stable-diffusion sources/go-tiny-dream
 	touch $@
 
 replace:
 	$(GOCMD) mod edit -replace github.com/nomic-ai/gpt4all/gpt4all-bindings/golang=$(CURDIR)/sources/gpt4all/gpt4all-bindings/golang
-	$(GOCMD) mod edit -replace github.com/go-skynet/go-ggml-transformers.cpp=$(CURDIR)/sources/go-ggml-transformers
 	$(GOCMD) mod edit -replace github.com/donomii/go-rwkv.cpp=$(CURDIR)/sources/go-rwkv
 	$(GOCMD) mod edit -replace github.com/ggerganov/whisper.cpp=$(CURDIR)/sources/whisper.cpp
 	$(GOCMD) mod edit -replace github.com/ggerganov/whisper.cpp/bindings/go=$(CURDIR)/sources/whisper.cpp/bindings/go
@@ -276,7 +281,6 @@ rebuild: ## Rebuilds the project
 	$(MAKE) -C sources/go-llama clean
 	$(MAKE) -C sources/go-llama-ggml clean
 	$(MAKE) -C sources/gpt4all/gpt4all-bindings/golang/ clean
-	$(MAKE) -C sources/go-ggml-transformers clean
 	$(MAKE) -C sources/go-rwkv clean
 	$(MAKE) -C sources/whisper.cpp clean
 	$(MAKE) -C sources/go-stable-diffusion clean
@@ -321,7 +325,7 @@ run: prepare ## run local-ai
 test-models/testmodel:
 	mkdir test-models
 	mkdir test-dir
-	wget -q https://huggingface.co/nnakasato/ggml-model-test/resolve/main/ggml-model-q4.bin -O test-models/testmodel
+	wget -q https://huggingface.co/TheBloke/orca_mini_3B-GGML/resolve/main/orca-mini-3b.ggmlv3.q4_0.bin -O test-models/testmodel
 	wget -q https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-base.en.bin -O test-models/whisper-en
 	wget -q https://huggingface.co/mudler/all-MiniLM-L6-v2/resolve/main/ggml-model-q4_0.bin -O test-models/bert
 	wget -q https://cdn.openai.com/whisper/draft-20220913a/micro-machines.wav -O test-dir/audio.wav
@@ -330,7 +334,7 @@ test-models/testmodel:
 	cp tests/models_fixtures/* test-models
 
 prepare-test: grpcs
-	cp -rf backend-assets api
+	cp -rf backend-assets core/http
 	cp tests/models_fixtures/* test-models
 
 test: prepare test-models/testmodel grpcs
@@ -338,7 +342,7 @@ test: prepare test-models/testmodel grpcs
 	export GO_TAGS="tts stablediffusion"
 	$(MAKE) prepare-test
 	HUGGINGFACE_GRPC=$(abspath ./)/backend/python/sentencetransformers/run.sh TEST_DIR=$(abspath ./)/test-dir/ FIXTURES=$(abspath ./)/tests/fixtures CONFIG_FILE=$(abspath ./)/test-models/config.yaml MODELS_PATH=$(abspath ./)/test-models \
-	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="!gpt4all && !llama && !llama-gguf"  --flake-attempts 5 --fail-fast -v -r ./api ./pkg
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="!gpt4all && !llama && !llama-gguf"  --flake-attempts $(TEST_FLAKES) --fail-fast -v -r $(TEST_PATHS)
 	$(MAKE) test-gpt4all
 	$(MAKE) test-llama
 	$(MAKE) test-llama-gguf
@@ -367,23 +371,23 @@ teardown-e2e:
 
 test-gpt4all: prepare-test
 	TEST_DIR=$(abspath ./)/test-dir/ FIXTURES=$(abspath ./)/tests/fixtures CONFIG_FILE=$(abspath ./)/test-models/config.yaml MODELS_PATH=$(abspath ./)/test-models \
-	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="gpt4all" --flake-attempts 5 -v -r ./api ./pkg
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="gpt4all" --flake-attempts 5 -v -r $(TEST_PATHS)
 
 test-llama: prepare-test
 	TEST_DIR=$(abspath ./)/test-dir/ FIXTURES=$(abspath ./)/tests/fixtures CONFIG_FILE=$(abspath ./)/test-models/config.yaml MODELS_PATH=$(abspath ./)/test-models \
-	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="llama" --flake-attempts 5 -v -r ./api ./pkg
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="llama" --flake-attempts 5 -v -r $(TEST_PATHS)
 
 test-llama-gguf: prepare-test
 	TEST_DIR=$(abspath ./)/test-dir/ FIXTURES=$(abspath ./)/tests/fixtures CONFIG_FILE=$(abspath ./)/test-models/config.yaml MODELS_PATH=$(abspath ./)/test-models \
-	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="llama-gguf" --flake-attempts 5 -v -r ./api ./pkg
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="llama-gguf" --flake-attempts 5 -v -r $(TEST_PATHS)
 
 test-tts: prepare-test
 	TEST_DIR=$(abspath ./)/test-dir/ FIXTURES=$(abspath ./)/tests/fixtures CONFIG_FILE=$(abspath ./)/test-models/config.yaml MODELS_PATH=$(abspath ./)/test-models \
-	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="tts" --flake-attempts 1 -v -r ./api ./pkg
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="tts" --flake-attempts 1 -v -r $(TEST_PATHS)
 
 test-stablediffusion: prepare-test
 	TEST_DIR=$(abspath ./)/test-dir/ FIXTURES=$(abspath ./)/tests/fixtures CONFIG_FILE=$(abspath ./)/test-models/config.yaml MODELS_PATH=$(abspath ./)/test-models \
-	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="stablediffusion" --flake-attempts 1 -v -r ./api ./pkg
+	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --label-filter="stablediffusion" --flake-attempts 1 -v -r $(TEST_PATHS)
 
 test-container:
 	docker build --target requirements -t local-ai-test-container .
@@ -461,9 +465,6 @@ backend-assets/grpc/llama: backend-assets/grpc sources/go-llama/libbinding.a
 	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-llama LIBRARY_PATH=$(CURDIR)/sources/go-llama \
 	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/llama ./backend/go/llm/llama/
 # TODO: every binary should have its own folder instead, so can have different  implementations
-ifeq ($(BUILD_TYPE),metal)
-	cp backend/cpp/llama/llama.cpp/ggml-metal.metal backend-assets/grpc/
-endif
 
 ## BACKEND CPP LLAMA START
 # Sets the variables in case it has to build the gRPC locally.
@@ -484,7 +485,7 @@ ifdef BUILD_GRPC_FOR_BACKEND_LLAMA
 	CMAKE_ARGS="${CMAKE_ARGS} ${ADDED_CMAKE_ARGS}" LLAMA_VERSION=$(CPPLLAMA_VERSION) $(MAKE) -C backend/cpp/llama grpc-server
 else
 	echo "BUILD_GRPC_FOR_BACKEND_LLAMA is not defined."
-	LLAMA_VERSION=$(CPPLLAMA_VERSION) $(MAKE) -C backend/cpp/llama grpc-server			
+	LLAMA_VERSION=$(CPPLLAMA_VERSION) $(MAKE) -C backend/cpp/llama grpc-server
 endif
 ## BACKEND CPP LLAMA END
 
@@ -493,7 +494,7 @@ backend-assets/grpc/llama-cpp: backend-assets/grpc backend/cpp/llama/grpc-server
 	cp -rfv backend/cpp/llama/grpc-server backend-assets/grpc/llama-cpp
 # TODO: every binary should have its own folder instead, so can have different metal implementations
 ifeq ($(BUILD_TYPE),metal)
-	cp backend/cpp/llama/llama.cpp/build/bin/ggml-metal.metal backend-assets/grpc/
+	cp backend/cpp/llama/llama.cpp/build/bin/default.metallib backend-assets/grpc/
 endif
 
 backend-assets/grpc/llama-ggml: backend-assets/grpc sources/go-llama-ggml/libbinding.a
@@ -505,38 +506,6 @@ backend-assets/grpc/gpt4all: backend-assets/grpc backend-assets/gpt4all sources/
 	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/gpt4all/gpt4all-bindings/golang/ LIBRARY_PATH=$(CURDIR)/sources/gpt4all/gpt4all-bindings/golang/ \
 	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/gpt4all ./backend/go/llm/gpt4all/
 
-backend-assets/grpc/dolly: backend-assets/grpc sources/go-ggml-transformers/libtransformers.a
-	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-ggml-transformers LIBRARY_PATH=$(CURDIR)/sources/go-ggml-transformers \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/dolly ./backend/go/llm/dolly/
-
-backend-assets/grpc/gpt2: backend-assets/grpc sources/go-ggml-transformers/libtransformers.a
-	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-ggml-transformers LIBRARY_PATH=$(CURDIR)/sources/go-ggml-transformers \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/gpt2 ./backend/go/llm/gpt2/
-
-backend-assets/grpc/gptj: backend-assets/grpc sources/go-ggml-transformers/libtransformers.a
-	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-ggml-transformers LIBRARY_PATH=$(CURDIR)/sources/go-ggml-transformers \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/gptj ./backend/go/llm/gptj/
-
-backend-assets/grpc/gptneox: backend-assets/grpc sources/go-ggml-transformers/libtransformers.a
-	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-ggml-transformers LIBRARY_PATH=$(CURDIR)/sources/go-ggml-transformers \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/gptneox ./backend/go/llm/gptneox/
-
-backend-assets/grpc/mpt: backend-assets/grpc sources/go-ggml-transformers/libtransformers.a
-	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-ggml-transformers LIBRARY_PATH=$(CURDIR)/sources/go-ggml-transformers \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/mpt ./backend/go/llm/mpt/
-
-backend-assets/grpc/replit: backend-assets/grpc sources/go-ggml-transformers/libtransformers.a
-	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-ggml-transformers LIBRARY_PATH=$(CURDIR)/sources/go-ggml-transformers \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/replit ./backend/go/llm/replit/
-
-backend-assets/grpc/falcon-ggml: backend-assets/grpc sources/go-ggml-transformers/libtransformers.a
-	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-ggml-transformers LIBRARY_PATH=$(CURDIR)/sources/go-ggml-transformers \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/falcon-ggml ./backend/go/llm/falcon-ggml/
-
-backend-assets/grpc/starcoder: backend-assets/grpc sources/go-ggml-transformers/libtransformers.a
-	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-ggml-transformers LIBRARY_PATH=$(CURDIR)/sources/go-ggml-transformers \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/starcoder ./backend/go/llm/starcoder/
-
 backend-assets/grpc/rwkv: backend-assets/grpc sources/go-rwkv/librwkv.a
 	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-rwkv LIBRARY_PATH=$(CURDIR)/sources/go-rwkv \
 	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/rwkv ./backend/go/llm/rwkv
@@ -550,6 +519,7 @@ backend-assets/grpc/langchain-huggingface: backend-assets/grpc
 
 backend-assets/grpc/stablediffusion: backend-assets/grpc
 	if [ ! -f backend-assets/grpc/stablediffusion ]; then \
+		$(MAKE) sources/go-stable-diffusion; \
 		$(MAKE) sources/go-stable-diffusion/libstablediffusion.a; \
 		CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/sources/go-stable-diffusion/ LIBRARY_PATH=$(CURDIR)/sources/go-stable-diffusion/ \
 		$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/stablediffusion ./backend/go/image/stablediffusion; \
@@ -568,3 +538,29 @@ backend-assets/grpc/whisper: backend-assets/grpc sources/whisper.cpp/libwhisper.
 	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/whisper ./backend/go/transcribe/
 
 grpcs: prepare $(GRPC_BACKENDS)
+
+DOCKER_IMAGE?=local-ai
+IMAGE_TYPE?=core
+BASE_IMAGE?=ubuntu:22.04
+
+docker:
+	docker build \
+		--build-arg BASE_IMAGE=$(BASE_IMAGE) \
+		--build-arg IMAGE_TYPE=$(IMAGE_TYPE) \
+		--build-arg GO_TAGS=$(GO_TAGS) \
+		--build-arg BUILD_TYPE=$(BUILD_TYPE) \
+		-t $(DOCKER_IMAGE) .
+
+docker-image-intel:
+	docker build \
+		--build-arg BASE_IMAGE=intel/oneapi-basekit:2024.0.1-devel-ubuntu22.04 \
+		--build-arg IMAGE_TYPE=$(IMAGE_TYPE) \
+		--build-arg GO_TAGS="none" \
+		--build-arg BUILD_TYPE=sycl_f32 -t $(DOCKER_IMAGE) .
+
+docker-image-intel-xpu:
+	docker build \
+		--build-arg BASE_IMAGE=intel/oneapi-basekit:2024.0.1-devel-ubuntu22.04 \
+		--build-arg IMAGE_TYPE=$(IMAGE_TYPE) \
+		--build-arg GO_TAGS="none" \
+		--build-arg BUILD_TYPE=sycl_f32 -t $(DOCKER_IMAGE) .

README.md

@@ -43,17 +43,24 @@
 
 [Roadmap](https://github.com/mudler/LocalAI/issues?q=is%3Aissue+is%3Aopen+label%3Aroadmap)
 
+- Parallel function calling: https://github.com/mudler/LocalAI/pull/1726
+- Upload file API: https://github.com/mudler/LocalAI/pull/1703
+- Tools API support: https://github.com/mudler/LocalAI/pull/1715
+- LLaVa 1.6: https://github.com/mudler/LocalAI/pull/1714
+- ROCm container images: https://github.com/mudler/LocalAI/pull/1595
+- Intel GPU support (sycl, transformers, diffusers): https://github.com/mudler/LocalAI/issues/1653
+- Deprecation of old backends: https://github.com/mudler/LocalAI/issues/1651
 - Mamba support: https://github.com/mudler/LocalAI/pull/1589
 - Start and share models with config file: https://github.com/mudler/LocalAI/pull/1522
 - 🐸 Coqui: https://github.com/mudler/LocalAI/pull/1489
-- Inline templates: https://github.com/mudler/LocalAI/pull/1452
-- Mixtral: https://github.com/mudler/LocalAI/pull/1449
 - Img2vid https://github.com/mudler/LocalAI/pull/1442
-- Musicgen https://github.com/mudler/LocalAI/pull/1387
 
 Hot topics (looking for contributors):
 - Backends v2: https://github.com/mudler/LocalAI/issues/1126
 - Improving UX v2: https://github.com/mudler/LocalAI/issues/1373
+- Assistant API: https://github.com/mudler/LocalAI/issues/1273
+- Moderation endpoint: https://github.com/mudler/LocalAI/issues/999
+- Vulkan: https://github.com/mudler/LocalAI/issues/1647
 
 If you want to help and contribute, issues up for grabs: https://github.com/mudler/LocalAI/issues?q=is%3Aissue+is%3Aopen+label%3A%22up+for+grabs%22
 
@@ -62,7 +69,7 @@ If you want to help and contribute, issues up for grabs: https://github.com/mudl
 For a detailed step-by-step introduction, refer to the [Getting Started](https://localai.io/basics/getting_started/index.html) guide. For those in a hurry, here's a straightforward one-liner to launch a LocalAI instance with [phi-2](https://huggingface.co/microsoft/phi-2) using `docker`:
 
 ```
-docker run -ti -p 8080:8080 localai/localai:v2.5.1-ffmpeg-core phi-2
+docker run -ti -p 8080:8080 localai/localai:v2.9.0-ffmpeg-core phi-2
 ```
 
 ## 🚀 [Features](https://localai.io/features/)
@@ -92,10 +99,6 @@ WebUIs:
 
 Model galleries
 - https://github.com/go-skynet/model-gallery
-  
-Auto Docker / Model setup
-- https://io.midori-ai.xyz/howtos/easy-localai-installer/
-- https://io.midori-ai.xyz/howtos/easy-model-installer/
 
 Other:
 - Helm chart https://github.com/go-skynet/helm-charts
@@ -106,17 +109,20 @@ Other:
 - Slack bot https://github.com/mudler/LocalAGI/tree/main/examples/slack
 - Telegram bot https://github.com/mudler/LocalAI/tree/master/examples/telegram-bot
 - Examples: https://github.com/mudler/LocalAI/tree/master/examples/
+  
 
 ### 🔗 Resources
 
-- 🆕 New! [LLM finetuning guide](https://localai.io/advanced/fine-tuning/)
+- 🆕 New! [LLM finetuning guide](https://localai.io/docs/advanced/fine-tuning/)
 - [How to build locally](https://localai.io/basics/build/index.html)
 - [How to install in Kubernetes](https://localai.io/basics/getting_started/index.html#run-localai-in-kubernetes)
-- [Projects integrating LocalAI](https://localai.io/integrations/)
+- [Projects integrating LocalAI](https://localai.io/docs/integrations/)
 - [How tos section](https://io.midori-ai.xyz/howtos/) (curated by our community)
 
 ## :book: 🎥 [Media, Blogs, Social](https://localai.io/basics/news/#media-blogs-social)
 
+- [Run LocalAI on AWS EKS with Pulumi](https://www.pulumi.com/ai/answers/tiZMDoZzZV6TLxgDXNBnFE/deploying-helm-charts-on-aws-eks)
+- [Run LocalAI on AWS](https://staleks.hashnode.dev/installing-localai-on-aws-ec2-instance)
 - [Create a slackbot for teams and OSS projects that answer to documentation](https://mudler.pm/posts/smart-slackbot-for-teams/)
 - [LocalAI meets k8sgpt](https://www.youtube.com/watch?v=PKrDNuJ_dfE)
 - [Question Answering on Documents locally with LangChain, LocalAI, Chroma, and GPT4All](https://mudler.pm/posts/localai-question-answering/)
@@ -176,7 +182,6 @@ LocalAI couldn't have been built without the help of great software already avai
 - https://github.com/ggerganov/whisper.cpp
 - https://github.com/saharNooby/rwkv.cpp
 - https://github.com/rhasspy/piper
-- https://github.com/cmp-nct/ggllm.cpp
 
 ## 🤗 Contributors
 

SECURITY.md

@@ -0,0 +1,42 @@
+# Security Policy
+
+## Introduction
+
+At LocalAI, we take the security of our software seriously. We understand the importance of protecting our community from vulnerabilities and are committed to ensuring the safety and security of our users.
+
+## Supported Versions
+
+We provide support and updates for certain versions of our software. The following table outlines which versions are currently supported with security updates:
+
+| Version | Supported          |
+| ------- | ------------------ |
+| > 2.0   | :white_check_mark: |
+| < 2.0   | :x:                |
+
+Please ensure that you are using a supported version to receive the latest security updates.
+
+## Reporting a Vulnerability
+
+We encourage the responsible disclosure of any security vulnerabilities. If you believe you've found a security issue in our software, we kindly ask you to follow the steps below to report it to us:
+
+1. **Email Us:** Send an email to [security@localai.io](mailto:security@localai.io) with a detailed report. Please do not disclose the vulnerability publicly or to any third parties before it has been addressed by us.
+
+2. **Expect a Response:** We aim to acknowledge receipt of vulnerability reports within 48 hours. Our security team will review your report and work closely with you to understand the impact and ensure a thorough investigation.
+
+3. **Collaboration:** If the vulnerability is accepted, we will work with you and our community to address the issue promptly. We'll keep you informed throughout the resolution process and may request additional information or collaboration.
+
+4. **Disclosure:** Once the vulnerability has been resolved, we encourage a coordinated disclosure. We believe in transparency and will work with you to ensure that our community is informed in a responsible manner.
+
+## Use of Third-Party Platforms
+
+As a Free and Open Source Software (FOSS) organization, we do not offer monetary bounties. However, researchers who wish to report vulnerabilities can also do so via [Huntr](https://huntr.dev/bounties), a platform that recognizes contributions to open source security.
+
+## Contact
+
+For any security-related inquiries beyond vulnerability reporting, please contact us at [security@localai.io](mailto:security@localai.io).
+
+## Acknowledgments
+
+We appreciate the efforts of those who contribute to the security of our project. Your responsible disclosure is invaluable to the safety and integrity of LocalAI.
+
+Thank you for helping us keep LocalAI secure.

api/api.go

@@ -1,283 +0,0 @@
-package api
-
-import (
-	"encoding/json"
-	"errors"
-	"fmt"
-	"os"
-	"strings"
-
-	config "github.com/go-skynet/LocalAI/api/config"
-	"github.com/go-skynet/LocalAI/api/localai"
-	"github.com/go-skynet/LocalAI/api/openai"
-	"github.com/go-skynet/LocalAI/api/options"
-	"github.com/go-skynet/LocalAI/api/schema"
-	"github.com/go-skynet/LocalAI/internal"
-	"github.com/go-skynet/LocalAI/metrics"
-	"github.com/go-skynet/LocalAI/pkg/assets"
-	"github.com/go-skynet/LocalAI/pkg/model"
-	"github.com/go-skynet/LocalAI/pkg/startup"
-
-	"github.com/gofiber/fiber/v2"
-	"github.com/gofiber/fiber/v2/middleware/cors"
-	"github.com/gofiber/fiber/v2/middleware/logger"
-	"github.com/gofiber/fiber/v2/middleware/recover"
-	"github.com/rs/zerolog"
-	"github.com/rs/zerolog/log"
-)
-
-func Startup(opts ...options.AppOption) (*options.Option, *config.ConfigLoader, error) {
-	options := options.NewOptions(opts...)
-
-	zerolog.SetGlobalLevel(zerolog.InfoLevel)
-	if options.Debug {
-		zerolog.SetGlobalLevel(zerolog.DebugLevel)
-	}
-
-	log.Info().Msgf("Starting LocalAI using %d threads, with models path: %s", options.Threads, options.Loader.ModelPath)
-	log.Info().Msgf("LocalAI version: %s", internal.PrintableVersion())
-
-	startup.PreloadModelsConfigurations(options.Loader.ModelPath, options.ModelsURL...)
-
-	cl := config.NewConfigLoader()
-	if err := cl.LoadConfigs(options.Loader.ModelPath); err != nil {
-		log.Error().Msgf("error loading config files: %s", err.Error())
-	}
-
-	if options.ConfigFile != "" {
-		if err := cl.LoadConfigFile(options.ConfigFile); err != nil {
-			log.Error().Msgf("error loading config file: %s", err.Error())
-		}
-	}
-
-	if err := cl.Preload(options.Loader.ModelPath); err != nil {
-		log.Error().Msgf("error downloading models: %s", err.Error())
-	}
-
-	if options.PreloadJSONModels != "" {
-		if err := localai.ApplyGalleryFromString(options.Loader.ModelPath, options.PreloadJSONModels, cl, options.Galleries); err != nil {
-			return nil, nil, err
-		}
-	}
-
-	if options.PreloadModelsFromPath != "" {
-		if err := localai.ApplyGalleryFromFile(options.Loader.ModelPath, options.PreloadModelsFromPath, cl, options.Galleries); err != nil {
-			return nil, nil, err
-		}
-	}
-
-	if options.Debug {
-		for _, v := range cl.ListConfigs() {
-			cfg, _ := cl.GetConfig(v)
-			log.Debug().Msgf("Model: %s (config: %+v)", v, cfg)
-		}
-	}
-
-	if options.AssetsDestination != "" {
-		// Extract files from the embedded FS
-		err := assets.ExtractFiles(options.BackendAssets, options.AssetsDestination)
-		log.Debug().Msgf("Extracting backend assets files to %s", options.AssetsDestination)
-		if err != nil {
-			log.Warn().Msgf("Failed extracting backend assets files: %s (might be required for some backends to work properly, like gpt4all)", err)
-		}
-	}
-
-	// turn off any process that was started by GRPC if the context is canceled
-	go func() {
-		<-options.Context.Done()
-		log.Debug().Msgf("Context canceled, shutting down")
-		options.Loader.StopAllGRPC()
-	}()
-
-	if options.WatchDog {
-		wd := model.NewWatchDog(
-			options.Loader,
-			options.WatchDogBusyTimeout,
-			options.WatchDogIdleTimeout,
-			options.WatchDogBusy,
-			options.WatchDogIdle)
-		options.Loader.SetWatchDog(wd)
-		go wd.Run()
-		go func() {
-			<-options.Context.Done()
-			log.Debug().Msgf("Context canceled, shutting down")
-			wd.Shutdown()
-		}()
-	}
-
-	return options, cl, nil
-}
-
-func App(opts ...options.AppOption) (*fiber.App, error) {
-
-	options, cl, err := Startup(opts...)
-	if err != nil {
-		return nil, fmt.Errorf("failed basic startup tasks with error %s", err.Error())
-	}
-
-	// Return errors as JSON responses
-	app := fiber.New(fiber.Config{
-		BodyLimit:             options.UploadLimitMB * 1024 * 1024, // this is the default limit of 4MB
-		DisableStartupMessage: options.DisableMessage,
-		// Override default error handler
-		ErrorHandler: func(ctx *fiber.Ctx, err error) error {
-			// Status code defaults to 500
-			code := fiber.StatusInternalServerError
-
-			// Retrieve the custom status code if it's a *fiber.Error
-			var e *fiber.Error
-			if errors.As(err, &e) {
-				code = e.Code
-			}
-
-			// Send custom error page
-			return ctx.Status(code).JSON(
-				schema.ErrorResponse{
-					Error: &schema.APIError{Message: err.Error(), Code: code},
-				},
-			)
-		},
-	})
-
-	if options.Debug {
-		app.Use(logger.New(logger.Config{
-			Format: "[${ip}]:${port} ${status} - ${method} ${path}\n",
-		}))
-	}
-
-	// Default middleware config
-	app.Use(recover.New())
-	if options.Metrics != nil {
-		app.Use(metrics.APIMiddleware(options.Metrics))
-	}
-
-	// Auth middleware checking if API key is valid. If no API key is set, no auth is required.
-	auth := func(c *fiber.Ctx) error {
-		if len(options.ApiKeys) == 0 {
-			return c.Next()
-		}
-
-		// Check for api_keys.json file
-		fileContent, err := os.ReadFile("api_keys.json")
-		if err == nil {
-			// Parse JSON content from the file
-			var fileKeys []string
-			err := json.Unmarshal(fileContent, &fileKeys)
-			if err != nil {
-				return c.Status(fiber.StatusInternalServerError).JSON(fiber.Map{"message": "Error parsing api_keys.json"})
-			}
-
-			// Add file keys to options.ApiKeys
-			options.ApiKeys = append(options.ApiKeys, fileKeys...)
-		}
-
-		if len(options.ApiKeys) == 0 {
-			return c.Next()
-		}
-
-		authHeader := c.Get("Authorization")
-		if authHeader == "" {
-			return c.Status(fiber.StatusUnauthorized).JSON(fiber.Map{"message": "Authorization header missing"})
-		}
-		authHeaderParts := strings.Split(authHeader, " ")
-		if len(authHeaderParts) != 2 || authHeaderParts[0] != "Bearer" {
-			return c.Status(fiber.StatusUnauthorized).JSON(fiber.Map{"message": "Invalid Authorization header format"})
-		}
-
-		apiKey := authHeaderParts[1]
-		for _, key := range options.ApiKeys {
-			if apiKey == key {
-				return c.Next()
-			}
-		}
-
-		return c.Status(fiber.StatusUnauthorized).JSON(fiber.Map{"message": "Invalid API key"})
-
-	}
-
-	if options.CORS {
-		var c func(ctx *fiber.Ctx) error
-		if options.CORSAllowOrigins == "" {
-			c = cors.New()
-		} else {
-			c = cors.New(cors.Config{AllowOrigins: options.CORSAllowOrigins})
-		}
-
-		app.Use(c)
-	}
-
-	// LocalAI API endpoints
-	galleryService := localai.NewGalleryService(options.Loader.ModelPath)
-	galleryService.Start(options.Context, cl)
-
-	app.Get("/version", auth, func(c *fiber.Ctx) error {
-		return c.JSON(struct {
-			Version string `json:"version"`
-		}{Version: internal.PrintableVersion()})
-	})
-
-	modelGalleryService := localai.CreateModelGalleryService(options.Galleries, options.Loader.ModelPath, galleryService)
-	app.Post("/models/apply", auth, modelGalleryService.ApplyModelGalleryEndpoint())
-	app.Get("/models/available", auth, modelGalleryService.ListModelFromGalleryEndpoint())
-	app.Get("/models/galleries", auth, modelGalleryService.ListModelGalleriesEndpoint())
-	app.Post("/models/galleries", auth, modelGalleryService.AddModelGalleryEndpoint())
-	app.Delete("/models/galleries", auth, modelGalleryService.RemoveModelGalleryEndpoint())
-	app.Get("/models/jobs/:uuid", auth, modelGalleryService.GetOpStatusEndpoint())
-	app.Get("/models/jobs", auth, modelGalleryService.GetAllStatusEndpoint())
-
-	// openAI compatible API endpoint
-
-	// chat
-	app.Post("/v1/chat/completions", auth, openai.ChatEndpoint(cl, options))
-	app.Post("/chat/completions", auth, openai.ChatEndpoint(cl, options))
-
-	// edit
-	app.Post("/v1/edits", auth, openai.EditEndpoint(cl, options))
-	app.Post("/edits", auth, openai.EditEndpoint(cl, options))
-
-	// completion
-	app.Post("/v1/completions", auth, openai.CompletionEndpoint(cl, options))
-	app.Post("/completions", auth, openai.CompletionEndpoint(cl, options))
-	app.Post("/v1/engines/:model/completions", auth, openai.CompletionEndpoint(cl, options))
-
-	// embeddings
-	app.Post("/v1/embeddings", auth, openai.EmbeddingsEndpoint(cl, options))
-	app.Post("/embeddings", auth, openai.EmbeddingsEndpoint(cl, options))
-	app.Post("/v1/engines/:model/embeddings", auth, openai.EmbeddingsEndpoint(cl, options))
-
-	// audio
-	app.Post("/v1/audio/transcriptions", auth, openai.TranscriptEndpoint(cl, options))
-	app.Post("/tts", auth, localai.TTSEndpoint(cl, options))
-
-	// images
-	app.Post("/v1/images/generations", auth, openai.ImageEndpoint(cl, options))
-
-	if options.ImageDir != "" {
-		app.Static("/generated-images", options.ImageDir)
-	}
-
-	if options.AudioDir != "" {
-		app.Static("/generated-audio", options.AudioDir)
-	}
-
-	ok := func(c *fiber.Ctx) error {
-		return c.SendStatus(200)
-	}
-
-	// Kubernetes health checks
-	app.Get("/healthz", ok)
-	app.Get("/readyz", ok)
-
-	// Experimental Backend Statistics Module
-	backendMonitor := localai.NewBackendMonitor(cl, options) // Split out for now
-	app.Get("/backend/monitor", localai.BackendMonitorEndpoint(backendMonitor))
-	app.Post("/backend/shutdown", localai.BackendShutdownEndpoint(backendMonitor))
-
-	// models
-	app.Get("/v1/models", auth, openai.ListModelsEndpoint(options.Loader, cl))
-	app.Get("/models", auth, openai.ListModelsEndpoint(options.Loader, cl))
-
-	app.Get("/metrics", metrics.MetricsHandler())
-
-	return app, nil
-}

api/backend/image.go

@@ -1,61 +0,0 @@
-package backend
-
-import (
-	config "github.com/go-skynet/LocalAI/api/config"
-	"github.com/go-skynet/LocalAI/api/options"
-	"github.com/go-skynet/LocalAI/pkg/grpc/proto"
-	model "github.com/go-skynet/LocalAI/pkg/model"
-)
-
-func ImageGeneration(height, width, mode, step, seed int, positive_prompt, negative_prompt, src, dst string, loader *model.ModelLoader, c config.Config, o *options.Option) (func() error, error) {
-
-	opts := modelOpts(c, o, []model.Option{
-		model.WithBackendString(c.Backend),
-		model.WithAssetDir(o.AssetsDestination),
-		model.WithThreads(uint32(c.Threads)),
-		model.WithContext(o.Context),
-		model.WithModel(c.Model),
-		model.WithLoadGRPCLoadModelOpts(&proto.ModelOptions{
-			CUDA:          c.CUDA || c.Diffusers.CUDA,
-			SchedulerType: c.Diffusers.SchedulerType,
-			PipelineType:  c.Diffusers.PipelineType,
-			CFGScale:      c.Diffusers.CFGScale,
-			LoraAdapter:   c.LoraAdapter,
-			LoraScale:     c.LoraScale,
-			LoraBase:      c.LoraBase,
-			IMG2IMG:       c.Diffusers.IMG2IMG,
-			CLIPModel:     c.Diffusers.ClipModel,
-			CLIPSubfolder: c.Diffusers.ClipSubFolder,
-			CLIPSkip:      int32(c.Diffusers.ClipSkip),
-			ControlNet:    c.Diffusers.ControlNet,
-		}),
-	})
-
-	inferenceModel, err := loader.BackendLoader(
-		opts...,
-	)
-	if err != nil {
-		return nil, err
-	}
-
-	fn := func() error {
-		_, err := inferenceModel.GenerateImage(
-			o.Context,
-			&proto.GenerateImageRequest{
-				Height:           int32(height),
-				Width:            int32(width),
-				Mode:             int32(mode),
-				Step:             int32(step),
-				Seed:             int32(seed),
-				CLIPSkip:         int32(c.Diffusers.ClipSkip),
-				PositivePrompt:   positive_prompt,
-				NegativePrompt:   negative_prompt,
-				Dst:              dst,
-				Src:              src,
-				EnableParameters: c.Diffusers.EnableParameters,
-			})
-		return err
-	}
-
-	return fn, nil
-}

api/backend/options.go

@@ -1,127 +0,0 @@
-package backend
-
-import (
-	"os"
-	"path/filepath"
-
-	pb "github.com/go-skynet/LocalAI/pkg/grpc/proto"
-	model "github.com/go-skynet/LocalAI/pkg/model"
-
-	config "github.com/go-skynet/LocalAI/api/config"
-	"github.com/go-skynet/LocalAI/api/options"
-)
-
-func modelOpts(c config.Config, o *options.Option, opts []model.Option) []model.Option {
-	if o.SingleBackend {
-		opts = append(opts, model.WithSingleActiveBackend())
-	}
-
-	if o.ParallelBackendRequests {
-		opts = append(opts, model.EnableParallelRequests)
-	}
-
-	if c.GRPC.Attempts != 0 {
-		opts = append(opts, model.WithGRPCAttempts(c.GRPC.Attempts))
-	}
-
-	if c.GRPC.AttemptsSleepTime != 0 {
-		opts = append(opts, model.WithGRPCAttemptsDelay(c.GRPC.AttemptsSleepTime))
-	}
-
-	for k, v := range o.ExternalGRPCBackends {
-		opts = append(opts, model.WithExternalBackend(k, v))
-	}
-
-	return opts
-}
-
-func gRPCModelOpts(c config.Config) *pb.ModelOptions {
-	b := 512
-	if c.Batch != 0 {
-		b = c.Batch
-	}
-
-	return &pb.ModelOptions{
-		ContextSize:    int32(c.ContextSize),
-		Seed:           int32(c.Seed),
-		NBatch:         int32(b),
-		NoMulMatQ:      c.NoMulMatQ,
-		CUDA:           c.CUDA, // diffusers, transformers
-		DraftModel:     c.DraftModel,
-		AudioPath:      c.VallE.AudioPath,
-		Quantization:   c.Quantization,
-		MMProj:         c.MMProj,
-		YarnExtFactor:  c.YarnExtFactor,
-		YarnAttnFactor: c.YarnAttnFactor,
-		YarnBetaFast:   c.YarnBetaFast,
-		YarnBetaSlow:   c.YarnBetaSlow,
-		LoraAdapter:    c.LoraAdapter,
-		LoraBase:       c.LoraBase,
-		LoraScale:      c.LoraScale,
-		NGQA:           c.NGQA,
-		RMSNormEps:     c.RMSNormEps,
-		F16Memory:      c.F16,
-		MLock:          c.MMlock,
-		RopeFreqBase:   c.RopeFreqBase,
-		RopeFreqScale:  c.RopeFreqScale,
-		NUMA:           c.NUMA,
-		Embeddings:     c.Embeddings,
-		LowVRAM:        c.LowVRAM,
-		NGPULayers:     int32(c.NGPULayers),
-		MMap:           c.MMap,
-		MainGPU:        c.MainGPU,
-		Threads:        int32(c.Threads),
-		TensorSplit:    c.TensorSplit,
-		// AutoGPTQ
-		ModelBaseName:    c.AutoGPTQ.ModelBaseName,
-		Device:           c.AutoGPTQ.Device,
-		UseTriton:        c.AutoGPTQ.Triton,
-		UseFastTokenizer: c.AutoGPTQ.UseFastTokenizer,
-		// RWKV
-		Tokenizer: c.Tokenizer,
-	}
-}
-
-func gRPCPredictOpts(c config.Config, modelPath string) *pb.PredictOptions {
-	promptCachePath := ""
-	if c.PromptCachePath != "" {
-		p := filepath.Join(modelPath, c.PromptCachePath)
-		os.MkdirAll(filepath.Dir(p), 0755)
-		promptCachePath = p
-	}
-	return &pb.PredictOptions{
-		Temperature:         float32(c.Temperature),
-		TopP:                float32(c.TopP),
-		NDraft:              c.NDraft,
-		TopK:                int32(c.TopK),
-		Tokens:              int32(c.Maxtokens),
-		Threads:             int32(c.Threads),
-		PromptCacheAll:      c.PromptCacheAll,
-		PromptCacheRO:       c.PromptCacheRO,
-		PromptCachePath:     promptCachePath,
-		F16KV:               c.F16,
-		DebugMode:           c.Debug,
-		Grammar:             c.Grammar,
-		NegativePromptScale: c.NegativePromptScale,
-		RopeFreqBase:        c.RopeFreqBase,
-		RopeFreqScale:       c.RopeFreqScale,
-		NegativePrompt:      c.NegativePrompt,
-		Mirostat:            int32(c.LLMConfig.Mirostat),
-		MirostatETA:         float32(c.LLMConfig.MirostatETA),
-		MirostatTAU:         float32(c.LLMConfig.MirostatTAU),
-		Debug:               c.Debug,
-		StopPrompts:         c.StopWords,
-		Repeat:              int32(c.RepeatPenalty),
-		NKeep:               int32(c.Keep),
-		Batch:               int32(c.Batch),
-		IgnoreEOS:           c.IgnoreEOS,
-		Seed:                int32(c.Seed),
-		FrequencyPenalty:    float32(c.FrequencyPenalty),
-		MLock:               c.MMlock,
-		MMap:                c.MMap,
-		MainGPU:             c.MainGPU,
-		TensorSplit:         c.TensorSplit,
-		TailFreeSamplingZ:   float32(c.TFZ),
-		TypicalP:            float32(c.TypicalP),
-	}
-}

api/backend/transcript.go

@@ -1,39 +0,0 @@
-package backend
-
-import (
-	"context"
-	"fmt"
-
-	config "github.com/go-skynet/LocalAI/api/config"
-	"github.com/go-skynet/LocalAI/api/schema"
-
-	"github.com/go-skynet/LocalAI/api/options"
-	"github.com/go-skynet/LocalAI/pkg/grpc/proto"
-	model "github.com/go-skynet/LocalAI/pkg/model"
-)
-
-func ModelTranscription(audio, language string, loader *model.ModelLoader, c config.Config, o *options.Option) (*schema.Result, error) {
-
-	opts := modelOpts(c, o, []model.Option{
-		model.WithBackendString(model.WhisperBackend),
-		model.WithModel(c.Model),
-		model.WithContext(o.Context),
-		model.WithThreads(uint32(c.Threads)),
-		model.WithAssetDir(o.AssetsDestination),
-	})
-
-	whisperModel, err := o.Loader.BackendLoader(opts...)
-	if err != nil {
-		return nil, err
-	}
-
-	if whisperModel == nil {
-		return nil, fmt.Errorf("could not load whisper model")
-	}
-
-	return whisperModel.AudioTranscription(context.Background(), &proto.TranscriptRequest{
-		Dst:      audio,
-		Language: language,
-		Threads:  uint32(c.Threads),
-	})
-}

api/backend/tts.go

@@ -1,79 +0,0 @@
-package backend
-
-import (
-	"context"
-	"fmt"
-	"os"
-	"path/filepath"
-
-	api_config "github.com/go-skynet/LocalAI/api/config"
-	"github.com/go-skynet/LocalAI/api/options"
-	"github.com/go-skynet/LocalAI/pkg/grpc/proto"
-	model "github.com/go-skynet/LocalAI/pkg/model"
-	"github.com/go-skynet/LocalAI/pkg/utils"
-)
-
-func generateUniqueFileName(dir, baseName, ext string) string {
-	counter := 1
-	fileName := baseName + ext
-
-	for {
-		filePath := filepath.Join(dir, fileName)
-		_, err := os.Stat(filePath)
-		if os.IsNotExist(err) {
-			return fileName
-		}
-
-		counter++
-		fileName = fmt.Sprintf("%s_%d%s", baseName, counter, ext)
-	}
-}
-
-func ModelTTS(backend, text, modelFile string, loader *model.ModelLoader, o *options.Option) (string, *proto.Result, error) {
-	bb := backend
-	if bb == "" {
-		bb = model.PiperBackend
-	}
-	opts := modelOpts(api_config.Config{}, o, []model.Option{
-		model.WithBackendString(bb),
-		model.WithModel(modelFile),
-		model.WithContext(o.Context),
-		model.WithAssetDir(o.AssetsDestination),
-	})
-	piperModel, err := o.Loader.BackendLoader(opts...)
-	if err != nil {
-		return "", nil, err
-	}
-
-	if piperModel == nil {
-		return "", nil, fmt.Errorf("could not load piper model")
-	}
-
-	if err := os.MkdirAll(o.AudioDir, 0755); err != nil {
-		return "", nil, fmt.Errorf("failed creating audio directory: %s", err)
-	}
-
-	fileName := generateUniqueFileName(o.AudioDir, "piper", ".wav")
-	filePath := filepath.Join(o.AudioDir, fileName)
-
-	// If the model file is not empty, we pass it joined with the model path
-	modelPath := ""
-	if modelFile != "" {
-		if bb != model.TransformersMusicGen {
-			modelPath = filepath.Join(o.Loader.ModelPath, modelFile)
-			if err := utils.VerifyPath(modelPath, o.Loader.ModelPath); err != nil {
-				return "", nil, err
-			}
-		} else {
-			modelPath = modelFile
-		}
-	}
-
-	res, err := piperModel.TTS(context.Background(), &proto.TTSRequest{
-		Text:  text,
-		Model: modelPath,
-		Dst:   filePath,
-	})
-
-	return filePath, res, err
-}

api/config/config.go

@@ -1,372 +0,0 @@
-package api_config
-
-import (
-	"errors"
-	"fmt"
-	"io/fs"
-	"os"
-	"path/filepath"
-	"strings"
-	"sync"
-
-	"github.com/go-skynet/LocalAI/pkg/downloader"
-	"github.com/go-skynet/LocalAI/pkg/utils"
-	"github.com/rs/zerolog/log"
-	"gopkg.in/yaml.v3"
-)
-
-type Config struct {
-	PredictionOptions `yaml:"parameters"`
-	Name              string `yaml:"name"`
-
-	F16            bool              `yaml:"f16"`
-	Threads        int               `yaml:"threads"`
-	Debug          bool              `yaml:"debug"`
-	Roles          map[string]string `yaml:"roles"`
-	Embeddings     bool              `yaml:"embeddings"`
-	Backend        string            `yaml:"backend"`
-	TemplateConfig TemplateConfig    `yaml:"template"`
-
-	PromptStrings, InputStrings                []string `yaml:"-"`
-	InputToken                                 [][]int  `yaml:"-"`
-	functionCallString, functionCallNameString string   `yaml:"-"`
-
-	FunctionsConfig Functions `yaml:"function"`
-
-	FeatureFlag FeatureFlag `yaml:"feature_flags"` // Feature Flag registry. We move fast, and features may break on a per model/backend basis. Registry for (usually temporary) flags that indicate aborting something early.
-	// LLM configs (GPT4ALL, Llama.cpp, ...)
-	LLMConfig `yaml:",inline"`
-
-	// AutoGPTQ specifics
-	AutoGPTQ AutoGPTQ `yaml:"autogptq"`
-
-	// Diffusers
-	Diffusers Diffusers `yaml:"diffusers"`
-	Step      int       `yaml:"step"`
-
-	// GRPC Options
-	GRPC GRPC `yaml:"grpc"`
-
-	// Vall-e-x
-	VallE VallE `yaml:"vall-e"`
-
-	// CUDA
-	// Explicitly enable CUDA or not (some backends might need it)
-	CUDA bool `yaml:"cuda"`
-
-	DownloadFiles []File `yaml:"download_files"`
-
-	Description string `yaml:"description"`
-	Usage       string `yaml:"usage"`
-}
-
-type File struct {
-	Filename string `yaml:"filename" json:"filename"`
-	SHA256   string `yaml:"sha256" json:"sha256"`
-	URI      string `yaml:"uri" json:"uri"`
-}
-
-type VallE struct {
-	AudioPath string `yaml:"audio_path"`
-}
-
-type FeatureFlag map[string]*bool
-
-func (ff FeatureFlag) Enabled(s string) bool {
-	v, exist := ff[s]
-	return exist && v != nil && *v
-}
-
-type GRPC struct {
-	Attempts          int `yaml:"attempts"`
-	AttemptsSleepTime int `yaml:"attempts_sleep_time"`
-}
-
-type Diffusers struct {
-	CUDA             bool    `yaml:"cuda"`
-	PipelineType     string  `yaml:"pipeline_type"`
-	SchedulerType    string  `yaml:"scheduler_type"`
-	EnableParameters string  `yaml:"enable_parameters"` // A list of comma separated parameters to specify
-	CFGScale         float32 `yaml:"cfg_scale"`         // Classifier-Free Guidance Scale
-	IMG2IMG          bool    `yaml:"img2img"`           // Image to Image Diffuser
-	ClipSkip         int     `yaml:"clip_skip"`         // Skip every N frames
-	ClipModel        string  `yaml:"clip_model"`        // Clip model to use
-	ClipSubFolder    string  `yaml:"clip_subfolder"`    // Subfolder to use for clip model
-	ControlNet       string  `yaml:"control_net"`
-}
-
-type LLMConfig struct {
-	SystemPrompt    string   `yaml:"system_prompt"`
-	TensorSplit     string   `yaml:"tensor_split"`
-	MainGPU         string   `yaml:"main_gpu"`
-	RMSNormEps      float32  `yaml:"rms_norm_eps"`
-	NGQA            int32    `yaml:"ngqa"`
-	PromptCachePath string   `yaml:"prompt_cache_path"`
-	PromptCacheAll  bool     `yaml:"prompt_cache_all"`
-	PromptCacheRO   bool     `yaml:"prompt_cache_ro"`
-	MirostatETA     float64  `yaml:"mirostat_eta"`
-	MirostatTAU     float64  `yaml:"mirostat_tau"`
-	Mirostat        int      `yaml:"mirostat"`
-	NGPULayers      int      `yaml:"gpu_layers"`
-	MMap            bool     `yaml:"mmap"`
-	MMlock          bool     `yaml:"mmlock"`
-	LowVRAM         bool     `yaml:"low_vram"`
-	Grammar         string   `yaml:"grammar"`
-	StopWords       []string `yaml:"stopwords"`
-	Cutstrings      []string `yaml:"cutstrings"`
-	TrimSpace       []string `yaml:"trimspace"`
-	TrimSuffix      []string `yaml:"trimsuffix"`
-
-	ContextSize  int     `yaml:"context_size"`
-	NUMA         bool    `yaml:"numa"`
-	LoraAdapter  string  `yaml:"lora_adapter"`
-	LoraBase     string  `yaml:"lora_base"`
-	LoraScale    float32 `yaml:"lora_scale"`
-	NoMulMatQ    bool    `yaml:"no_mulmatq"`
-	DraftModel   string  `yaml:"draft_model"`
-	NDraft       int32   `yaml:"n_draft"`
-	Quantization string  `yaml:"quantization"`
-	MMProj       string  `yaml:"mmproj"`
-
-	RopeScaling    string  `yaml:"rope_scaling"`
-	YarnExtFactor  float32 `yaml:"yarn_ext_factor"`
-	YarnAttnFactor float32 `yaml:"yarn_attn_factor"`
-	YarnBetaFast   float32 `yaml:"yarn_beta_fast"`
-	YarnBetaSlow   float32 `yaml:"yarn_beta_slow"`
-}
-
-type AutoGPTQ struct {
-	ModelBaseName    string `yaml:"model_base_name"`
-	Device           string `yaml:"device"`
-	Triton           bool   `yaml:"triton"`
-	UseFastTokenizer bool   `yaml:"use_fast_tokenizer"`
-}
-
-type Functions struct {
-	DisableNoAction         bool   `yaml:"disable_no_action"`
-	NoActionFunctionName    string `yaml:"no_action_function_name"`
-	NoActionDescriptionName string `yaml:"no_action_description_name"`
-}
-
-type TemplateConfig struct {
-	Chat        string `yaml:"chat"`
-	ChatMessage string `yaml:"chat_message"`
-	Completion  string `yaml:"completion"`
-	Edit        string `yaml:"edit"`
-	Functions   string `yaml:"function"`
-}
-
-type ConfigLoader struct {
-	configs map[string]Config
-	sync.Mutex
-}
-
-func (c *Config) SetFunctionCallString(s string) {
-	c.functionCallString = s
-}
-
-func (c *Config) SetFunctionCallNameString(s string) {
-	c.functionCallNameString = s
-}
-
-func (c *Config) ShouldUseFunctions() bool {
-	return ((c.functionCallString != "none" || c.functionCallString == "") || c.ShouldCallSpecificFunction())
-}
-
-func (c *Config) ShouldCallSpecificFunction() bool {
-	return len(c.functionCallNameString) > 0
-}
-
-func (c *Config) FunctionToCall() string {
-	return c.functionCallNameString
-}
-
-func defaultPredictOptions(modelFile string) PredictionOptions {
-	return PredictionOptions{
-		TopP:        0.7,
-		TopK:        80,
-		Maxtokens:   512,
-		Temperature: 0.9,
-		Model:       modelFile,
-	}
-}
-
-func DefaultConfig(modelFile string) *Config {
-	return &Config{
-		PredictionOptions: defaultPredictOptions(modelFile),
-	}
-}
-
-func NewConfigLoader() *ConfigLoader {
-	return &ConfigLoader{
-		configs: make(map[string]Config),
-	}
-}
-func ReadConfigFile(file string) ([]*Config, error) {
-	c := &[]*Config{}
-	f, err := os.ReadFile(file)
-	if err != nil {
-		return nil, fmt.Errorf("cannot read config file: %w", err)
-	}
-	if err := yaml.Unmarshal(f, c); err != nil {
-		return nil, fmt.Errorf("cannot unmarshal config file: %w", err)
-	}
-
-	return *c, nil
-}
-
-func ReadConfig(file string) (*Config, error) {
-	c := &Config{}
-	f, err := os.ReadFile(file)
-	if err != nil {
-		return nil, fmt.Errorf("cannot read config file: %w", err)
-	}
-	if err := yaml.Unmarshal(f, c); err != nil {
-		return nil, fmt.Errorf("cannot unmarshal config file: %w", err)
-	}
-
-	return c, nil
-}
-
-func (cm *ConfigLoader) LoadConfigFile(file string) error {
-	cm.Lock()
-	defer cm.Unlock()
-	c, err := ReadConfigFile(file)
-	if err != nil {
-		return fmt.Errorf("cannot load config file: %w", err)
-	}
-
-	for _, cc := range c {
-		cm.configs[cc.Name] = *cc
-	}
-	return nil
-}
-
-func (cm *ConfigLoader) LoadConfig(file string) error {
-	cm.Lock()
-	defer cm.Unlock()
-	c, err := ReadConfig(file)
-	if err != nil {
-		return fmt.Errorf("cannot read config file: %w", err)
-	}
-
-	cm.configs[c.Name] = *c
-	return nil
-}
-
-func (cm *ConfigLoader) GetConfig(m string) (Config, bool) {
-	cm.Lock()
-	defer cm.Unlock()
-	v, exists := cm.configs[m]
-	return v, exists
-}
-
-func (cm *ConfigLoader) GetAllConfigs() []Config {
-	cm.Lock()
-	defer cm.Unlock()
-	var res []Config
-	for _, v := range cm.configs {
-		res = append(res, v)
-	}
-	return res
-}
-
-func (cm *ConfigLoader) ListConfigs() []string {
-	cm.Lock()
-	defer cm.Unlock()
-	var res []string
-	for k := range cm.configs {
-		res = append(res, k)
-	}
-	return res
-}
-
-// Preload prepare models if they are not local but url or huggingface repositories
-func (cm *ConfigLoader) Preload(modelPath string) error {
-	cm.Lock()
-	defer cm.Unlock()
-
-	status := func(fileName, current, total string, percent float64) {
-		utils.DisplayDownloadFunction(fileName, current, total, percent)
-	}
-
-	log.Info().Msgf("Preloading models from %s", modelPath)
-
-	for i, config := range cm.configs {
-
-		// Download files and verify their SHA
-		for _, file := range config.DownloadFiles {
-			log.Debug().Msgf("Checking %q exists and matches SHA", file.Filename)
-
-			if err := utils.VerifyPath(file.Filename, modelPath); err != nil {
-				return err
-			}
-			// Create file path
-			filePath := filepath.Join(modelPath, file.Filename)
-
-			if err := downloader.DownloadFile(file.URI, filePath, file.SHA256, status); err != nil {
-				return err
-			}
-		}
-
-		modelURL := config.PredictionOptions.Model
-		modelURL = downloader.ConvertURL(modelURL)
-
-		if downloader.LooksLikeURL(modelURL) {
-			// md5 of model name
-			md5Name := utils.MD5(modelURL)
-
-			// check if file exists
-			if _, err := os.Stat(filepath.Join(modelPath, md5Name)); errors.Is(err, os.ErrNotExist) {
-				err := downloader.DownloadFile(modelURL, filepath.Join(modelPath, md5Name), "", status)
-				if err != nil {
-					return err
-				}
-			}
-
-			cc := cm.configs[i]
-			c := &cc
-			c.PredictionOptions.Model = md5Name
-			cm.configs[i] = *c
-		}
-		if cm.configs[i].Name != "" {
-			log.Info().Msgf("Model name: %s", cm.configs[i].Name)
-		}
-		if cm.configs[i].Description != "" {
-			log.Info().Msgf("Model description: %s", cm.configs[i].Description)
-		}
-		if cm.configs[i].Usage != "" {
-			log.Info().Msgf("Model usage: \n%s", cm.configs[i].Usage)
-		}
-	}
-	return nil
-}
-
-func (cm *ConfigLoader) LoadConfigs(path string) error {
-	cm.Lock()
-	defer cm.Unlock()
-	entries, err := os.ReadDir(path)
-	if err != nil {
-		return err
-	}
-	files := make([]fs.FileInfo, 0, len(entries))
-	for _, entry := range entries {
-		info, err := entry.Info()
-		if err != nil {
-			return err
-		}
-		files = append(files, info)
-	}
-	for _, file := range files {
-		// Skip templates, YAML and .keep files
-		if !strings.Contains(file.Name(), ".yaml") && !strings.Contains(file.Name(), ".yml") {
-			continue
-		}
-		c, err := ReadConfig(filepath.Join(path, file.Name()))
-		if err == nil {
-			cm.configs[c.Name] = *c
-		}
-	}
-
-	return nil
-}

api/localai/backend_monitor.go

@@ -1,162 +0,0 @@
-package localai
-
-import (
-	"context"
-	"fmt"
-	"strings"
-
-	config "github.com/go-skynet/LocalAI/api/config"
-	"github.com/go-skynet/LocalAI/pkg/grpc/proto"
-
-	"github.com/go-skynet/LocalAI/api/options"
-	"github.com/gofiber/fiber/v2"
-	"github.com/rs/zerolog/log"
-
-	gopsutil "github.com/shirou/gopsutil/v3/process"
-)
-
-type BackendMonitorRequest struct {
-	Model string `json:"model" yaml:"model"`
-}
-
-type BackendMonitorResponse struct {
-	MemoryInfo    *gopsutil.MemoryInfoStat
-	MemoryPercent float32
-	CPUPercent    float64
-}
-
-type BackendMonitor struct {
-	configLoader *config.ConfigLoader
-	options      *options.Option // Taking options in case we need to inspect ExternalGRPCBackends, though that's out of scope for now, hence the name.
-}
-
-func NewBackendMonitor(configLoader *config.ConfigLoader, options *options.Option) BackendMonitor {
-	return BackendMonitor{
-		configLoader: configLoader,
-		options:      options,
-	}
-}
-
-func (bm *BackendMonitor) SampleLocalBackendProcess(model string) (*BackendMonitorResponse, error) {
-	config, exists := bm.configLoader.GetConfig(model)
-	var backend string
-	if exists {
-		backend = config.Model
-	} else {
-		// Last ditch effort: use it raw, see if a backend happens to match.
-		backend = model
-	}
-
-	if !strings.HasSuffix(backend, ".bin") {
-		backend = fmt.Sprintf("%s.bin", backend)
-	}
-
-	pid, err := bm.options.Loader.GetGRPCPID(backend)
-
-	if err != nil {
-		log.Error().Msgf("model %s : failed to find pid %+v", model, err)
-		return nil, err
-	}
-
-	// Name is slightly frightening but this does _not_ create a new process, rather it looks up an existing process by PID.
-	backendProcess, err := gopsutil.NewProcess(int32(pid))
-
-	if err != nil {
-		log.Error().Msgf("model %s [PID %d] : error getting process info %+v", model, pid, err)
-		return nil, err
-	}
-
-	memInfo, err := backendProcess.MemoryInfo()
-
-	if err != nil {
-		log.Error().Msgf("model %s [PID %d] : error getting memory info %+v", model, pid, err)
-		return nil, err
-	}
-
-	memPercent, err := backendProcess.MemoryPercent()
-	if err != nil {
-		log.Error().Msgf("model %s [PID %d] : error getting memory percent %+v", model, pid, err)
-		return nil, err
-	}
-
-	cpuPercent, err := backendProcess.CPUPercent()
-	if err != nil {
-		log.Error().Msgf("model %s [PID %d] : error getting cpu percent %+v", model, pid, err)
-		return nil, err
-	}
-
-	return &BackendMonitorResponse{
-		MemoryInfo:    memInfo,
-		MemoryPercent: memPercent,
-		CPUPercent:    cpuPercent,
-	}, nil
-}
-
-func (bm BackendMonitor) getModelLoaderIDFromCtx(c *fiber.Ctx) (string, error) {
-	input := new(BackendMonitorRequest)
-	// Get input data from the request body
-	if err := c.BodyParser(input); err != nil {
-		return "", err
-	}
-
-	config, exists := bm.configLoader.GetConfig(input.Model)
-	var backendId string
-	if exists {
-		backendId = config.Model
-	} else {
-		// Last ditch effort: use it raw, see if a backend happens to match.
-		backendId = input.Model
-	}
-
-	if !strings.HasSuffix(backendId, ".bin") {
-		backendId = fmt.Sprintf("%s.bin", backendId)
-	}
-
-	return backendId, nil
-}
-
-func BackendMonitorEndpoint(bm BackendMonitor) func(c *fiber.Ctx) error {
-	return func(c *fiber.Ctx) error {
-
-		backendId, err := bm.getModelLoaderIDFromCtx(c)
-		if err != nil {
-			return err
-		}
-
-		model := bm.options.Loader.CheckIsLoaded(backendId)
-		if model == "" {
-			return fmt.Errorf("backend %s is not currently loaded", backendId)
-		}
-
-		status, rpcErr := model.GRPC(false, nil).Status(context.TODO())
-		if rpcErr != nil {
-			log.Warn().Msgf("backend %s experienced an error retrieving status info: %s", backendId, rpcErr.Error())
-			val, slbErr := bm.SampleLocalBackendProcess(backendId)
-			if slbErr != nil {
-				return fmt.Errorf("backend %s experienced an error retrieving status info via rpc: %s, then failed local node process sample: %s", backendId, rpcErr.Error(), slbErr.Error())
-			}
-			return c.JSON(proto.StatusResponse{
-				State: proto.StatusResponse_ERROR,
-				Memory: &proto.MemoryUsageData{
-					Total: val.MemoryInfo.VMS,
-					Breakdown: map[string]uint64{
-						"gopsutil-RSS": val.MemoryInfo.RSS,
-					},
-				},
-			})
-		}
-
-		return c.JSON(status)
-	}
-}
-
-func BackendShutdownEndpoint(bm BackendMonitor) func(c *fiber.Ctx) error {
-	return func(c *fiber.Ctx) error {
-		backendId, err := bm.getModelLoaderIDFromCtx(c)
-		if err != nil {
-			return err
-		}
-
-		return bm.options.Loader.ShutdownModel(backendId)
-	}
-}

api/localai/gallery.go

@@ -1,326 +0,0 @@
-package localai
-
-import (
-	"context"
-	"fmt"
-	"os"
-	"slices"
-	"strings"
-	"sync"
-
-	json "github.com/json-iterator/go"
-	"gopkg.in/yaml.v3"
-
-	config "github.com/go-skynet/LocalAI/api/config"
-	"github.com/go-skynet/LocalAI/pkg/gallery"
-	"github.com/go-skynet/LocalAI/pkg/utils"
-
-	"github.com/gofiber/fiber/v2"
-	"github.com/google/uuid"
-	"github.com/rs/zerolog/log"
-)
-
-type galleryOp struct {
-	req         gallery.GalleryModel
-	id          string
-	galleries   []gallery.Gallery
-	galleryName string
-}
-
-type galleryOpStatus struct {
-	FileName           string  `json:"file_name"`
-	Error              error   `json:"error"`
-	Processed          bool    `json:"processed"`
-	Message            string  `json:"message"`
-	Progress           float64 `json:"progress"`
-	TotalFileSize      string  `json:"file_size"`
-	DownloadedFileSize string  `json:"downloaded_size"`
-}
-
-type galleryApplier struct {
-	modelPath string
-	sync.Mutex
-	C        chan galleryOp
-	statuses map[string]*galleryOpStatus
-}
-
-func NewGalleryService(modelPath string) *galleryApplier {
-	return &galleryApplier{
-		modelPath: modelPath,
-		C:         make(chan galleryOp),
-		statuses:  make(map[string]*galleryOpStatus),
-	}
-}
-
-func prepareModel(modelPath string, req gallery.GalleryModel, cm *config.ConfigLoader, downloadStatus func(string, string, string, float64)) error {
-
-	config, err := gallery.GetGalleryConfigFromURL(req.URL)
-	if err != nil {
-		return err
-	}
-
-	config.Files = append(config.Files, req.AdditionalFiles...)
-
-	return gallery.InstallModel(modelPath, req.Name, &config, req.Overrides, downloadStatus)
-}
-
-func (g *galleryApplier) updateStatus(s string, op *galleryOpStatus) {
-	g.Lock()
-	defer g.Unlock()
-	g.statuses[s] = op
-}
-
-func (g *galleryApplier) getStatus(s string) *galleryOpStatus {
-	g.Lock()
-	defer g.Unlock()
-
-	return g.statuses[s]
-}
-
-func (g *galleryApplier) getAllStatus() map[string]*galleryOpStatus {
-	g.Lock()
-	defer g.Unlock()
-
-	return g.statuses
-}
-
-func (g *galleryApplier) Start(c context.Context, cm *config.ConfigLoader) {
-	go func() {
-		for {
-			select {
-			case <-c.Done():
-				return
-			case op := <-g.C:
-				utils.ResetDownloadTimers()
-
-				g.updateStatus(op.id, &galleryOpStatus{Message: "processing", Progress: 0})
-
-				// updates the status with an error
-				updateError := func(e error) {
-					g.updateStatus(op.id, &galleryOpStatus{Error: e, Processed: true, Message: "error: " + e.Error()})
-				}
-
-				// displayDownload displays the download progress
-				progressCallback := func(fileName string, current string, total string, percentage float64) {
-					g.updateStatus(op.id, &galleryOpStatus{Message: "processing", FileName: fileName, Progress: percentage, TotalFileSize: total, DownloadedFileSize: current})
-					utils.DisplayDownloadFunction(fileName, current, total, percentage)
-				}
-
-				var err error
-				// if the request contains a gallery name, we apply the gallery from the gallery list
-				if op.galleryName != "" {
-					if strings.Contains(op.galleryName, "@") {
-						err = gallery.InstallModelFromGallery(op.galleries, op.galleryName, g.modelPath, op.req, progressCallback)
-					} else {
-						err = gallery.InstallModelFromGalleryByName(op.galleries, op.galleryName, g.modelPath, op.req, progressCallback)
-					}
-				} else {
-					err = prepareModel(g.modelPath, op.req, cm, progressCallback)
-				}
-
-				if err != nil {
-					updateError(err)
-					continue
-				}
-
-				// Reload models
-				err = cm.LoadConfigs(g.modelPath)
-				if err != nil {
-					updateError(err)
-					continue
-				}
-
-				err = cm.Preload(g.modelPath)
-				if err != nil {
-					updateError(err)
-					continue
-				}
-
-				g.updateStatus(op.id, &galleryOpStatus{Processed: true, Message: "completed", Progress: 100})
-			}
-		}
-	}()
-}
-
-type galleryModel struct {
-	gallery.GalleryModel `yaml:",inline"` // https://github.com/go-yaml/yaml/issues/63
-	ID                   string           `json:"id"`
-}
-
-func processRequests(modelPath, s string, cm *config.ConfigLoader, galleries []gallery.Gallery, requests []galleryModel) error {
-	var err error
-	for _, r := range requests {
-		utils.ResetDownloadTimers()
-		if r.ID == "" {
-			err = prepareModel(modelPath, r.GalleryModel, cm, utils.DisplayDownloadFunction)
-		} else {
-			if strings.Contains(r.ID, "@") {
-				err = gallery.InstallModelFromGallery(
-					galleries, r.ID, modelPath, r.GalleryModel, utils.DisplayDownloadFunction)
-			} else {
-				err = gallery.InstallModelFromGalleryByName(
-					galleries, r.ID, modelPath, r.GalleryModel, utils.DisplayDownloadFunction)
-			}
-		}
-	}
-	return err
-}
-
-func ApplyGalleryFromFile(modelPath, s string, cm *config.ConfigLoader, galleries []gallery.Gallery) error {
-	dat, err := os.ReadFile(s)
-	if err != nil {
-		return err
-	}
-	var requests []galleryModel
-
-	if err := yaml.Unmarshal(dat, &requests); err != nil {
-		return err
-	}
-
-	return processRequests(modelPath, s, cm, galleries, requests)
-}
-
-func ApplyGalleryFromString(modelPath, s string, cm *config.ConfigLoader, galleries []gallery.Gallery) error {
-	var requests []galleryModel
-	err := json.Unmarshal([]byte(s), &requests)
-	if err != nil {
-		return err
-	}
-
-	return processRequests(modelPath, s, cm, galleries, requests)
-}
-
-/// Endpoint Service
-
-type ModelGalleryService struct {
-	galleries      []gallery.Gallery
-	modelPath      string
-	galleryApplier *galleryApplier
-}
-
-type GalleryModel struct {
-	ID string `json:"id"`
-	gallery.GalleryModel
-}
-
-func CreateModelGalleryService(galleries []gallery.Gallery, modelPath string, galleryApplier *galleryApplier) ModelGalleryService {
-	return ModelGalleryService{
-		galleries:      galleries,
-		modelPath:      modelPath,
-		galleryApplier: galleryApplier,
-	}
-}
-
-func (mgs *ModelGalleryService) GetOpStatusEndpoint() func(c *fiber.Ctx) error {
-	return func(c *fiber.Ctx) error {
-		status := mgs.galleryApplier.getStatus(c.Params("uuid"))
-		if status == nil {
-			return fmt.Errorf("could not find any status for ID")
-		}
-		return c.JSON(status)
-	}
-}
-
-func (mgs *ModelGalleryService) GetAllStatusEndpoint() func(c *fiber.Ctx) error {
-	return func(c *fiber.Ctx) error {
-		return c.JSON(mgs.galleryApplier.getAllStatus())
-	}
-}
-
-func (mgs *ModelGalleryService) ApplyModelGalleryEndpoint() func(c *fiber.Ctx) error {
-	return func(c *fiber.Ctx) error {
-		input := new(GalleryModel)
-		// Get input data from the request body
-		if err := c.BodyParser(input); err != nil {
-			return err
-		}
-
-		uuid, err := uuid.NewUUID()
-		if err != nil {
-			return err
-		}
-		mgs.galleryApplier.C <- galleryOp{
-			req:         input.GalleryModel,
-			id:          uuid.String(),
-			galleryName: input.ID,
-			galleries:   mgs.galleries,
-		}
-		return c.JSON(struct {
-			ID        string `json:"uuid"`
-			StatusURL string `json:"status"`
-		}{ID: uuid.String(), StatusURL: c.BaseURL() + "/models/jobs/" + uuid.String()})
-	}
-}
-
-func (mgs *ModelGalleryService) ListModelFromGalleryEndpoint() func(c *fiber.Ctx) error {
-	return func(c *fiber.Ctx) error {
-		log.Debug().Msgf("Listing models from galleries: %+v", mgs.galleries)
-
-		models, err := gallery.AvailableGalleryModels(mgs.galleries, mgs.modelPath)
-		if err != nil {
-			return err
-		}
-		log.Debug().Msgf("Models found from galleries: %+v", models)
-		for _, m := range models {
-			log.Debug().Msgf("Model found from galleries: %+v", m)
-		}
-		dat, err := json.Marshal(models)
-		if err != nil {
-			return err
-		}
-		return c.Send(dat)
-	}
-}
-
-// NOTE: This is different (and much simpler!) than above! This JUST lists the model galleries that have been loaded, not their contents!
-func (mgs *ModelGalleryService) ListModelGalleriesEndpoint() func(c *fiber.Ctx) error {
-	return func(c *fiber.Ctx) error {
-		log.Debug().Msgf("Listing model galleries %+v", mgs.galleries)
-		dat, err := json.Marshal(mgs.galleries)
-		if err != nil {
-			return err
-		}
-		return c.Send(dat)
-	}
-}
-
-func (mgs *ModelGalleryService) AddModelGalleryEndpoint() func(c *fiber.Ctx) error {
-	return func(c *fiber.Ctx) error {
-		input := new(gallery.Gallery)
-		// Get input data from the request body
-		if err := c.BodyParser(input); err != nil {
-			return err
-		}
-		if slices.ContainsFunc(mgs.galleries, func(gallery gallery.Gallery) bool {
-			return gallery.Name == input.Name
-		}) {
-			return fmt.Errorf("%s already exists", input.Name)
-		}
-		dat, err := json.Marshal(mgs.galleries)
-		if err != nil {
-			return err
-		}
-		log.Debug().Msgf("Adding %+v to gallery list", *input)
-		mgs.galleries = append(mgs.galleries, *input)
-		return c.Send(dat)
-	}
-}
-
-func (mgs *ModelGalleryService) RemoveModelGalleryEndpoint() func(c *fiber.Ctx) error {
-	return func(c *fiber.Ctx) error {
-		input := new(gallery.Gallery)
-		// Get input data from the request body
-		if err := c.BodyParser(input); err != nil {
-			return err
-		}
-		if !slices.ContainsFunc(mgs.galleries, func(gallery gallery.Gallery) bool {
-			return gallery.Name == input.Name
-		}) {
-			return fmt.Errorf("%s is not currently registered", input.Name)
-		}
-		mgs.galleries = slices.DeleteFunc(mgs.galleries, func(gallery gallery.Gallery) bool {
-			return gallery.Name == input.Name
-		})
-		return c.Send(nil)
-	}
-}

api/localai/localai.go

@@ -1,32 +0,0 @@
-package localai
-
-import (
-	"github.com/go-skynet/LocalAI/api/backend"
-	config "github.com/go-skynet/LocalAI/api/config"
-
-	"github.com/go-skynet/LocalAI/api/options"
-	"github.com/gofiber/fiber/v2"
-)
-
-type TTSRequest struct {
-	Model   string `json:"model" yaml:"model"`
-	Input   string `json:"input" yaml:"input"`
-	Backend string `json:"backend" yaml:"backend"`
-}
-
-func TTSEndpoint(cm *config.ConfigLoader, o *options.Option) func(c *fiber.Ctx) error {
-	return func(c *fiber.Ctx) error {
-
-		input := new(TTSRequest)
-		// Get input data from the request body
-		if err := c.BodyParser(input); err != nil {
-			return err
-		}
-
-		filePath, _, err := backend.ModelTTS(input.Backend, input.Input, input.Model, o.Loader, o)
-		if err != nil {
-			return err
-		}
-		return c.Download(filePath)
-	}
-}

api/openai/chat.go

@@ -1,399 +0,0 @@
-package openai
-
-import (
-	"bufio"
-	"bytes"
-	"encoding/json"
-	"fmt"
-	"strings"
-	"time"
-
-	"github.com/go-skynet/LocalAI/api/backend"
-	config "github.com/go-skynet/LocalAI/api/config"
-	"github.com/go-skynet/LocalAI/api/options"
-	"github.com/go-skynet/LocalAI/api/schema"
-	"github.com/go-skynet/LocalAI/pkg/grammar"
-	model "github.com/go-skynet/LocalAI/pkg/model"
-	"github.com/go-skynet/LocalAI/pkg/utils"
-	"github.com/gofiber/fiber/v2"
-	"github.com/google/uuid"
-	"github.com/rs/zerolog/log"
-	"github.com/valyala/fasthttp"
-)
-
-func ChatEndpoint(cm *config.ConfigLoader, o *options.Option) func(c *fiber.Ctx) error {
-	emptyMessage := ""
-	id := uuid.New().String()
-	created := int(time.Now().Unix())
-
-	process := func(s string, req *schema.OpenAIRequest, config *config.Config, loader *model.ModelLoader, responses chan schema.OpenAIResponse) {
-		initialMessage := schema.OpenAIResponse{
-			ID:      id,
-			Created: created,
-			Model:   req.Model, // we have to return what the user sent here, due to OpenAI spec.
-			Choices: []schema.Choice{{Delta: &schema.Message{Role: "assistant", Content: &emptyMessage}}},
-			Object:  "chat.completion.chunk",
-		}
-		responses <- initialMessage
-
-		ComputeChoices(req, s, config, o, loader, func(s string, c *[]schema.Choice) {}, func(s string, usage backend.TokenUsage) bool {
-			resp := schema.OpenAIResponse{
-				ID:      id,
-				Created: created,
-				Model:   req.Model, // we have to return what the user sent here, due to OpenAI spec.
-				Choices: []schema.Choice{{Delta: &schema.Message{Content: &s}, Index: 0}},
-				Object:  "chat.completion.chunk",
-				Usage: schema.OpenAIUsage{
-					PromptTokens:     usage.Prompt,
-					CompletionTokens: usage.Completion,
-					TotalTokens:      usage.Prompt + usage.Completion,
-				},
-			}
-
-			responses <- resp
-			return true
-		})
-		close(responses)
-	}
-	return func(c *fiber.Ctx) error {
-		processFunctions := false
-		funcs := grammar.Functions{}
-		modelFile, input, err := readInput(c, o, true)
-		if err != nil {
-			return fmt.Errorf("failed reading parameters from request:%w", err)
-		}
-
-		config, input, err := readConfig(modelFile, input, cm, o.Loader, o.Debug, o.Threads, o.ContextSize, o.F16)
-		if err != nil {
-			return fmt.Errorf("failed reading parameters from request:%w", err)
-		}
-		log.Debug().Msgf("Configuration read: %+v", config)
-
-		// Allow the user to set custom actions via config file
-		// to be "embedded" in each model
-		noActionName := "answer"
-		noActionDescription := "use this action to answer without performing any action"
-
-		if config.FunctionsConfig.NoActionFunctionName != "" {
-			noActionName = config.FunctionsConfig.NoActionFunctionName
-		}
-		if config.FunctionsConfig.NoActionDescriptionName != "" {
-			noActionDescription = config.FunctionsConfig.NoActionDescriptionName
-		}
-
-		if input.ResponseFormat.Type == "json_object" {
-			input.Grammar = grammar.JSONBNF
-		}
-
-		// process functions if we have any defined or if we have a function call string
-		if len(input.Functions) > 0 && config.ShouldUseFunctions() {
-			log.Debug().Msgf("Response needs to process functions")
-
-			processFunctions = true
-
-			noActionGrammar := grammar.Function{
-				Name:        noActionName,
-				Description: noActionDescription,
-				Parameters: map[string]interface{}{
-					"properties": map[string]interface{}{
-						"message": map[string]interface{}{
-							"type":        "string",
-							"description": "The message to reply the user with",
-						}},
-				},
-			}
-
-			// Append the no action function
-			funcs = append(funcs, input.Functions...)
-			if !config.FunctionsConfig.DisableNoAction {
-				funcs = append(funcs, noActionGrammar)
-			}
-
-			// Force picking one of the functions by the request
-			if config.FunctionToCall() != "" {
-				funcs = funcs.Select(config.FunctionToCall())
-			}
-
-			// Update input grammar
-			jsStruct := funcs.ToJSONStructure()
-			config.Grammar = jsStruct.Grammar("")
-		} else if input.JSONFunctionGrammarObject != nil {
-			config.Grammar = input.JSONFunctionGrammarObject.Grammar("")
-		}
-
-		// functions are not supported in stream mode (yet?)
-		toStream := input.Stream && !processFunctions
-
-		log.Debug().Msgf("Parameters: %+v", config)
-
-		var predInput string
-
-		suppressConfigSystemPrompt := false
-		mess := []string{}
-		for messageIndex, i := range input.Messages {
-			var content string
-			role := i.Role
-
-			// if function call, we might want to customize the role so we can display better that the "assistant called a json action"
-			// if an "assistant_function_call" role is defined, we use it, otherwise we use the role that is passed by in the request
-			if i.FunctionCall != nil && i.Role == "assistant" {
-				roleFn := "assistant_function_call"
-				r := config.Roles[roleFn]
-				if r != "" {
-					role = roleFn
-				}
-			}
-			r := config.Roles[role]
-			contentExists := i.Content != nil && i.StringContent != ""
-			// First attempt to populate content via a chat message specific template
-			if config.TemplateConfig.ChatMessage != "" {
-				chatMessageData := model.ChatMessageTemplateData{
-					SystemPrompt: config.SystemPrompt,
-					Role:         r,
-					RoleName:     role,
-					Content:      i.StringContent,
-					MessageIndex: messageIndex,
-				}
-				templatedChatMessage, err := o.Loader.EvaluateTemplateForChatMessage(config.TemplateConfig.ChatMessage, chatMessageData)
-				if err != nil {
-					log.Error().Msgf("error processing message %+v using template \"%s\": %v. Skipping!", chatMessageData, config.TemplateConfig.ChatMessage, err)
-				} else {
-					if templatedChatMessage == "" {
-						log.Warn().Msgf("template \"%s\" produced blank output for %+v. Skipping!", config.TemplateConfig.ChatMessage, chatMessageData)
-						continue // TODO: This continue is here intentionally to skip over the line `mess = append(mess, content)` below, and to prevent the sprintf
-					}
-					log.Debug().Msgf("templated message for chat: %s", templatedChatMessage)
-					content = templatedChatMessage
-				}
-			}
-			// If this model doesn't have such a template, or if that template fails to return a value, template at the message level.
-			if content == "" {
-				if r != "" {
-					if contentExists {
-						content = fmt.Sprint(r, i.StringContent)
-					}
-					if i.FunctionCall != nil {
-						j, err := json.Marshal(i.FunctionCall)
-						if err == nil {
-							if contentExists {
-								content += "\n" + fmt.Sprint(r, " ", string(j))
-							} else {
-								content = fmt.Sprint(r, " ", string(j))
-							}
-						}
-					}
-				} else {
-					if contentExists {
-						content = fmt.Sprint(i.StringContent)
-					}
-					if i.FunctionCall != nil {
-						j, err := json.Marshal(i.FunctionCall)
-						if err == nil {
-							if contentExists {
-								content += "\n" + string(j)
-							} else {
-								content = string(j)
-							}
-						}
-					}
-				}
-				// Special Handling: System. We care if it was printed at all, not the r branch, so check seperately
-				if contentExists && role == "system" {
-					suppressConfigSystemPrompt = true
-				}
-			}
-
-			mess = append(mess, content)
-		}
-
-		predInput = strings.Join(mess, "\n")
-		log.Debug().Msgf("Prompt (before templating): %s", predInput)
-
-		if toStream {
-			log.Debug().Msgf("Stream request received")
-			c.Context().SetContentType("text/event-stream")
-			//c.Response().Header.SetContentType(fiber.MIMETextHTMLCharsetUTF8)
-			//	c.Set("Content-Type", "text/event-stream")
-			c.Set("Cache-Control", "no-cache")
-			c.Set("Connection", "keep-alive")
-			c.Set("Transfer-Encoding", "chunked")
-		}
-
-		templateFile := ""
-
-		// A model can have a "file.bin.tmpl" file associated with a prompt template prefix
-		if o.Loader.ExistsInModelPath(fmt.Sprintf("%s.tmpl", config.Model)) {
-			templateFile = config.Model
-		}
-
-		if config.TemplateConfig.Chat != "" && !processFunctions {
-			templateFile = config.TemplateConfig.Chat
-		}
-
-		if config.TemplateConfig.Functions != "" && processFunctions {
-			templateFile = config.TemplateConfig.Functions
-		}
-
-		if templateFile != "" {
-			templatedInput, err := o.Loader.EvaluateTemplateForPrompt(model.ChatPromptTemplate, templateFile, model.PromptTemplateData{
-				SystemPrompt:         config.SystemPrompt,
-				SuppressSystemPrompt: suppressConfigSystemPrompt,
-				Input:                predInput,
-				Functions:            funcs,
-			})
-			if err == nil {
-				predInput = templatedInput
-				log.Debug().Msgf("Template found, input modified to: %s", predInput)
-			} else {
-				log.Debug().Msgf("Template failed loading: %s", err.Error())
-			}
-		}
-
-		log.Debug().Msgf("Prompt (after templating): %s", predInput)
-		if processFunctions {
-			log.Debug().Msgf("Grammar: %+v", config.Grammar)
-		}
-
-		if toStream {
-			responses := make(chan schema.OpenAIResponse)
-
-			go process(predInput, input, config, o.Loader, responses)
-
-			c.Context().SetBodyStreamWriter(fasthttp.StreamWriter(func(w *bufio.Writer) {
-
-				usage := &schema.OpenAIUsage{}
-
-				for ev := range responses {
-					usage = &ev.Usage // Copy a pointer to the latest usage chunk so that the stop message can reference it
-					var buf bytes.Buffer
-					enc := json.NewEncoder(&buf)
-					enc.Encode(ev)
-					log.Debug().Msgf("Sending chunk: %s", buf.String())
-					_, err := fmt.Fprintf(w, "data: %v\n", buf.String())
-					if err != nil {
-						log.Debug().Msgf("Sending chunk failed: %v", err)
-						input.Cancel()
-						break
-					}
-					w.Flush()
-				}
-
-				resp := &schema.OpenAIResponse{
-					ID:      id,
-					Created: created,
-					Model:   input.Model, // we have to return what the user sent here, due to OpenAI spec.
-					Choices: []schema.Choice{
-						{
-							FinishReason: "stop",
-							Index:        0,
-							Delta:        &schema.Message{Content: &emptyMessage},
-						}},
-					Object: "chat.completion.chunk",
-					Usage:  *usage,
-				}
-				respData, _ := json.Marshal(resp)
-
-				w.WriteString(fmt.Sprintf("data: %s\n\n", respData))
-				w.WriteString("data: [DONE]\n\n")
-				w.Flush()
-			}))
-			return nil
-		}
-
-		result, tokenUsage, err := ComputeChoices(input, predInput, config, o, o.Loader, func(s string, c *[]schema.Choice) {
-			if processFunctions {
-				// As we have to change the result before processing, we can't stream the answer (yet?)
-				ss := map[string]interface{}{}
-				// This prevent newlines to break JSON parsing for clients
-				s = utils.EscapeNewLines(s)
-				json.Unmarshal([]byte(s), &ss)
-				log.Debug().Msgf("Function return: %s %+v", s, ss)
-
-				// The grammar defines the function name as "function", while OpenAI returns "name"
-				func_name := ss["function"]
-				// Similarly, while here arguments is a map[string]interface{}, OpenAI actually want a stringified object
-				args := ss["arguments"] // arguments needs to be a string, but we return an object from the grammar result (TODO: fix)
-				d, _ := json.Marshal(args)
-
-				ss["arguments"] = string(d)
-				ss["name"] = func_name
-
-				// if do nothing, reply with a message
-				if func_name == noActionName {
-					log.Debug().Msgf("nothing to do, computing a reply")
-
-					// If there is a message that the LLM already sends as part of the JSON reply, use it
-					arguments := map[string]interface{}{}
-					json.Unmarshal([]byte(d), &arguments)
-					m, exists := arguments["message"]
-					if exists {
-						switch message := m.(type) {
-						case string:
-							if message != "" {
-								log.Debug().Msgf("Reply received from LLM: %s", message)
-								message = backend.Finetune(*config, predInput, message)
-								log.Debug().Msgf("Reply received from LLM(finetuned): %s", message)
-
-								*c = append(*c, schema.Choice{Message: &schema.Message{Role: "assistant", Content: &message}})
-								return
-							}
-						}
-					}
-
-					log.Debug().Msgf("No action received from LLM, without a message, computing a reply")
-					// Otherwise ask the LLM to understand the JSON output and the context, and return a message
-					// Note: This costs (in term of CPU) another computation
-					config.Grammar = ""
-					images := []string{}
-					for _, m := range input.Messages {
-						images = append(images, m.StringImages...)
-					}
-					predFunc, err := backend.ModelInference(input.Context, predInput, images, o.Loader, *config, o, nil)
-					if err != nil {
-						log.Error().Msgf("inference error: %s", err.Error())
-						return
-					}
-
-					prediction, err := predFunc()
-					if err != nil {
-						log.Error().Msgf("inference error: %s", err.Error())
-						return
-					}
-
-					fineTunedResponse := backend.Finetune(*config, predInput, prediction.Response)
-					*c = append(*c, schema.Choice{Message: &schema.Message{Role: "assistant", Content: &fineTunedResponse}})
-				} else {
-					// otherwise reply with the function call
-					*c = append(*c, schema.Choice{
-						FinishReason: "function_call",
-						Message:      &schema.Message{Role: "assistant", FunctionCall: ss},
-					})
-				}
-
-				return
-			}
-			*c = append(*c, schema.Choice{FinishReason: "stop", Index: 0, Message: &schema.Message{Role: "assistant", Content: &s}})
-		}, nil)
-		if err != nil {
-			return err
-		}
-
-		resp := &schema.OpenAIResponse{
-			ID:      id,
-			Created: created,
-			Model:   input.Model, // we have to return what the user sent here, due to OpenAI spec.
-			Choices: result,
-			Object:  "chat.completion",
-			Usage: schema.OpenAIUsage{
-				PromptTokens:     tokenUsage.Prompt,
-				CompletionTokens: tokenUsage.Completion,
-				TotalTokens:      tokenUsage.Prompt + tokenUsage.Completion,
-			},
-		}
-		respData, _ := json.Marshal(resp)
-		log.Debug().Msgf("Response: %s", respData)
-
-		// Return the prediction in the response body
-		return c.JSON(resp)
-	}
-}

backend/backend.proto

@@ -126,6 +126,11 @@ message ModelOptions {
 
   // vllm
   string Quantization = 40;
+  float  GPUMemoryUtilization = 50;
+  bool   TrustRemoteCode = 51;
+  bool   EnforceEager = 52;
+  int32  SwapSpace = 53;
+  int32  MaxModelLen = 54;
 
   string MMProj = 41;
 
@@ -134,6 +139,8 @@ message ModelOptions {
   float YarnAttnFactor = 45;
   float YarnBetaFast = 46;
   float YarnBetaSlow = 47;
+
+  string Type = 49;
 }
 
 message Result {
@@ -184,6 +191,7 @@ message TTSRequest {
   string text = 1;
   string model = 2;
   string dst = 3;
+  string voice = 4;
 }
 
 message TokenizationResponse {

backend/backend_grpc.pb.go

@@ -0,0 +1,457 @@
+// Code generated by protoc-gen-go-grpc. DO NOT EDIT.
+// versions:
+// - protoc-gen-go-grpc v1.2.0
+// - protoc             v4.23.4
+// source: backend/backend.proto
+
+package proto
+
+import (
+	context "context"
+	grpc "google.golang.org/grpc"
+	codes "google.golang.org/grpc/codes"
+	status "google.golang.org/grpc/status"
+)
+
+// This is a compile-time assertion to ensure that this generated file
+// is compatible with the grpc package it is being compiled against.
+// Requires gRPC-Go v1.32.0 or later.
+const _ = grpc.SupportPackageIsVersion7
+
+// BackendClient is the client API for Backend service.
+//
+// For semantics around ctx use and closing/ending streaming RPCs, please refer to https://pkg.go.dev/google.golang.org/grpc/?tab=doc#ClientConn.NewStream.
+type BackendClient interface {
+	Health(ctx context.Context, in *HealthMessage, opts ...grpc.CallOption) (*Reply, error)
+	Predict(ctx context.Context, in *PredictOptions, opts ...grpc.CallOption) (*Reply, error)
+	LoadModel(ctx context.Context, in *ModelOptions, opts ...grpc.CallOption) (*Result, error)
+	PredictStream(ctx context.Context, in *PredictOptions, opts ...grpc.CallOption) (Backend_PredictStreamClient, error)
+	Embedding(ctx context.Context, in *PredictOptions, opts ...grpc.CallOption) (*EmbeddingResult, error)
+	GenerateImage(ctx context.Context, in *GenerateImageRequest, opts ...grpc.CallOption) (*Result, error)
+	AudioTranscription(ctx context.Context, in *TranscriptRequest, opts ...grpc.CallOption) (*TranscriptResult, error)
+	TTS(ctx context.Context, in *TTSRequest, opts ...grpc.CallOption) (*Result, error)
+	TokenizeString(ctx context.Context, in *PredictOptions, opts ...grpc.CallOption) (*TokenizationResponse, error)
+	Status(ctx context.Context, in *HealthMessage, opts ...grpc.CallOption) (*StatusResponse, error)
+}
+
+type backendClient struct {
+	cc grpc.ClientConnInterface
+}
+
+func NewBackendClient(cc grpc.ClientConnInterface) BackendClient {
+	return &backendClient{cc}
+}
+
+func (c *backendClient) Health(ctx context.Context, in *HealthMessage, opts ...grpc.CallOption) (*Reply, error) {
+	out := new(Reply)
+	err := c.cc.Invoke(ctx, "/backend.Backend/Health", in, out, opts...)
+	if err != nil {
+		return nil, err
+	}
+	return out, nil
+}
+
+func (c *backendClient) Predict(ctx context.Context, in *PredictOptions, opts ...grpc.CallOption) (*Reply, error) {
+	out := new(Reply)
+	err := c.cc.Invoke(ctx, "/backend.Backend/Predict", in, out, opts...)
+	if err != nil {
+		return nil, err
+	}
+	return out, nil
+}
+
+func (c *backendClient) LoadModel(ctx context.Context, in *ModelOptions, opts ...grpc.CallOption) (*Result, error) {
+	out := new(Result)
+	err := c.cc.Invoke(ctx, "/backend.Backend/LoadModel", in, out, opts...)
+	if err != nil {
+		return nil, err
+	}
+	return out, nil
+}
+
+func (c *backendClient) PredictStream(ctx context.Context, in *PredictOptions, opts ...grpc.CallOption) (Backend_PredictStreamClient, error) {
+	stream, err := c.cc.NewStream(ctx, &Backend_ServiceDesc.Streams[0], "/backend.Backend/PredictStream", opts...)
+	if err != nil {
+		return nil, err
+	}
+	x := &backendPredictStreamClient{stream}
+	if err := x.ClientStream.SendMsg(in); err != nil {
+		return nil, err
+	}
+	if err := x.ClientStream.CloseSend(); err != nil {
+		return nil, err
+	}
+	return x, nil
+}
+
+type Backend_PredictStreamClient interface {
+	Recv() (*Reply, error)
+	grpc.ClientStream
+}
+
+type backendPredictStreamClient struct {
+	grpc.ClientStream
+}
+
+func (x *backendPredictStreamClient) Recv() (*Reply, error) {
+	m := new(Reply)
+	if err := x.ClientStream.RecvMsg(m); err != nil {
+		return nil, err
+	}
+	return m, nil
+}
+
+func (c *backendClient) Embedding(ctx context.Context, in *PredictOptions, opts ...grpc.CallOption) (*EmbeddingResult, error) {
+	out := new(EmbeddingResult)
+	err := c.cc.Invoke(ctx, "/backend.Backend/Embedding", in, out, opts...)
+	if err != nil {
+		return nil, err
+	}
+	return out, nil
+}
+
+func (c *backendClient) GenerateImage(ctx context.Context, in *GenerateImageRequest, opts ...grpc.CallOption) (*Result, error) {
+	out := new(Result)
+	err := c.cc.Invoke(ctx, "/backend.Backend/GenerateImage", in, out, opts...)
+	if err != nil {
+		return nil, err
+	}
+	return out, nil
+}
+
+func (c *backendClient) AudioTranscription(ctx context.Context, in *TranscriptRequest, opts ...grpc.CallOption) (*TranscriptResult, error) {
+	out := new(TranscriptResult)
+	err := c.cc.Invoke(ctx, "/backend.Backend/AudioTranscription", in, out, opts...)
+	if err != nil {
+		return nil, err
+	}
+	return out, nil
+}
+
+func (c *backendClient) TTS(ctx context.Context, in *TTSRequest, opts ...grpc.CallOption) (*Result, error) {
+	out := new(Result)
+	err := c.cc.Invoke(ctx, "/backend.Backend/TTS", in, out, opts...)
+	if err != nil {
+		return nil, err
+	}
+	return out, nil
+}
+
+func (c *backendClient) TokenizeString(ctx context.Context, in *PredictOptions, opts ...grpc.CallOption) (*TokenizationResponse, error) {
+	out := new(TokenizationResponse)
+	err := c.cc.Invoke(ctx, "/backend.Backend/TokenizeString", in, out, opts...)
+	if err != nil {
+		return nil, err
+	}
+	return out, nil
+}
+
+func (c *backendClient) Status(ctx context.Context, in *HealthMessage, opts ...grpc.CallOption) (*StatusResponse, error) {
+	out := new(StatusResponse)
+	err := c.cc.Invoke(ctx, "/backend.Backend/Status", in, out, opts...)
+	if err != nil {
+		return nil, err
+	}
+	return out, nil
+}
+
+// BackendServer is the server API for Backend service.
+// All implementations must embed UnimplementedBackendServer
+// for forward compatibility
+type BackendServer interface {
+	Health(context.Context, *HealthMessage) (*Reply, error)
+	Predict(context.Context, *PredictOptions) (*Reply, error)
+	LoadModel(context.Context, *ModelOptions) (*Result, error)
+	PredictStream(*PredictOptions, Backend_PredictStreamServer) error
+	Embedding(context.Context, *PredictOptions) (*EmbeddingResult, error)
+	GenerateImage(context.Context, *GenerateImageRequest) (*Result, error)
+	AudioTranscription(context.Context, *TranscriptRequest) (*TranscriptResult, error)
+	TTS(context.Context, *TTSRequest) (*Result, error)
+	TokenizeString(context.Context, *PredictOptions) (*TokenizationResponse, error)
+	Status(context.Context, *HealthMessage) (*StatusResponse, error)
+	mustEmbedUnimplementedBackendServer()
+}
+
+// UnimplementedBackendServer must be embedded to have forward compatible implementations.
+type UnimplementedBackendServer struct {
+}
+
+func (UnimplementedBackendServer) Health(context.Context, *HealthMessage) (*Reply, error) {
+	return nil, status.Errorf(codes.Unimplemented, "method Health not implemented")
+}
+func (UnimplementedBackendServer) Predict(context.Context, *PredictOptions) (*Reply, error) {
+	return nil, status.Errorf(codes.Unimplemented, "method Predict not implemented")
+}
+func (UnimplementedBackendServer) LoadModel(context.Context, *ModelOptions) (*Result, error) {
+	return nil, status.Errorf(codes.Unimplemented, "method LoadModel not implemented")
+}
+func (UnimplementedBackendServer) PredictStream(*PredictOptions, Backend_PredictStreamServer) error {
+	return status.Errorf(codes.Unimplemented, "method PredictStream not implemented")
+}
+func (UnimplementedBackendServer) Embedding(context.Context, *PredictOptions) (*EmbeddingResult, error) {
+	return nil, status.Errorf(codes.Unimplemented, "method Embedding not implemented")
+}
+func (UnimplementedBackendServer) GenerateImage(context.Context, *GenerateImageRequest) (*Result, error) {
+	return nil, status.Errorf(codes.Unimplemented, "method GenerateImage not implemented")
+}
+func (UnimplementedBackendServer) AudioTranscription(context.Context, *TranscriptRequest) (*TranscriptResult, error) {
+	return nil, status.Errorf(codes.Unimplemented, "method AudioTranscription not implemented")
+}
+func (UnimplementedBackendServer) TTS(context.Context, *TTSRequest) (*Result, error) {
+	return nil, status.Errorf(codes.Unimplemented, "method TTS not implemented")
+}
+func (UnimplementedBackendServer) TokenizeString(context.Context, *PredictOptions) (*TokenizationResponse, error) {
+	return nil, status.Errorf(codes.Unimplemented, "method TokenizeString not implemented")
+}
+func (UnimplementedBackendServer) Status(context.Context, *HealthMessage) (*StatusResponse, error) {
+	return nil, status.Errorf(codes.Unimplemented, "method Status not implemented")
+}
+func (UnimplementedBackendServer) mustEmbedUnimplementedBackendServer() {}
+
+// UnsafeBackendServer may be embedded to opt out of forward compatibility for this service.
+// Use of this interface is not recommended, as added methods to BackendServer will
+// result in compilation errors.
+type UnsafeBackendServer interface {
+	mustEmbedUnimplementedBackendServer()
+}
+
+func RegisterBackendServer(s grpc.ServiceRegistrar, srv BackendServer) {
+	s.RegisterService(&Backend_ServiceDesc, srv)
+}
+
+func _Backend_Health_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
+	in := new(HealthMessage)
+	if err := dec(in); err != nil {
+		return nil, err
+	}
+	if interceptor == nil {
+		return srv.(BackendServer).Health(ctx, in)
+	}
+	info := &grpc.UnaryServerInfo{
+		Server:     srv,
+		FullMethod: "/backend.Backend/Health",
+	}
+	handler := func(ctx context.Context, req interface{}) (interface{}, error) {
+		return srv.(BackendServer).Health(ctx, req.(*HealthMessage))
+	}
+	return interceptor(ctx, in, info, handler)
+}
+
+func _Backend_Predict_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
+	in := new(PredictOptions)
+	if err := dec(in); err != nil {
+		return nil, err
+	}
+	if interceptor == nil {
+		return srv.(BackendServer).Predict(ctx, in)
+	}
+	info := &grpc.UnaryServerInfo{
+		Server:     srv,
+		FullMethod: "/backend.Backend/Predict",
+	}
+	handler := func(ctx context.Context, req interface{}) (interface{}, error) {
+		return srv.(BackendServer).Predict(ctx, req.(*PredictOptions))
+	}
+	return interceptor(ctx, in, info, handler)
+}
+
+func _Backend_LoadModel_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
+	in := new(ModelOptions)
+	if err := dec(in); err != nil {
+		return nil, err
+	}
+	if interceptor == nil {
+		return srv.(BackendServer).LoadModel(ctx, in)
+	}
+	info := &grpc.UnaryServerInfo{
+		Server:     srv,
+		FullMethod: "/backend.Backend/LoadModel",
+	}
+	handler := func(ctx context.Context, req interface{}) (interface{}, error) {
+		return srv.(BackendServer).LoadModel(ctx, req.(*ModelOptions))
+	}
+	return interceptor(ctx, in, info, handler)
+}
+
+func _Backend_PredictStream_Handler(srv interface{}, stream grpc.ServerStream) error {
+	m := new(PredictOptions)
+	if err := stream.RecvMsg(m); err != nil {
+		return err
+	}
+	return srv.(BackendServer).PredictStream(m, &backendPredictStreamServer{stream})
+}
+
+type Backend_PredictStreamServer interface {
+	Send(*Reply) error
+	grpc.ServerStream
+}
+
+type backendPredictStreamServer struct {
+	grpc.ServerStream
+}
+
+func (x *backendPredictStreamServer) Send(m *Reply) error {
+	return x.ServerStream.SendMsg(m)
+}
+
+func _Backend_Embedding_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
+	in := new(PredictOptions)
+	if err := dec(in); err != nil {
+		return nil, err
+	}
+	if interceptor == nil {
+		return srv.(BackendServer).Embedding(ctx, in)
+	}
+	info := &grpc.UnaryServerInfo{
+		Server:     srv,
+		FullMethod: "/backend.Backend/Embedding",
+	}
+	handler := func(ctx context.Context, req interface{}) (interface{}, error) {
+		return srv.(BackendServer).Embedding(ctx, req.(*PredictOptions))
+	}
+	return interceptor(ctx, in, info, handler)
+}
+
+func _Backend_GenerateImage_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
+	in := new(GenerateImageRequest)
+	if err := dec(in); err != nil {
+		return nil, err
+	}
+	if interceptor == nil {
+		return srv.(BackendServer).GenerateImage(ctx, in)
+	}
+	info := &grpc.UnaryServerInfo{
+		Server:     srv,
+		FullMethod: "/backend.Backend/GenerateImage",
+	}
+	handler := func(ctx context.Context, req interface{}) (interface{}, error) {
+		return srv.(BackendServer).GenerateImage(ctx, req.(*GenerateImageRequest))
+	}
+	return interceptor(ctx, in, info, handler)
+}
+
+func _Backend_AudioTranscription_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
+	in := new(TranscriptRequest)
+	if err := dec(in); err != nil {
+		return nil, err
+	}
+	if interceptor == nil {
+		return srv.(BackendServer).AudioTranscription(ctx, in)
+	}
+	info := &grpc.UnaryServerInfo{
+		Server:     srv,
+		FullMethod: "/backend.Backend/AudioTranscription",
+	}
+	handler := func(ctx context.Context, req interface{}) (interface{}, error) {
+		return srv.(BackendServer).AudioTranscription(ctx, req.(*TranscriptRequest))
+	}
+	return interceptor(ctx, in, info, handler)
+}
+
+func _Backend_TTS_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
+	in := new(TTSRequest)
+	if err := dec(in); err != nil {
+		return nil, err
+	}
+	if interceptor == nil {
+		return srv.(BackendServer).TTS(ctx, in)
+	}
+	info := &grpc.UnaryServerInfo{
+		Server:     srv,
+		FullMethod: "/backend.Backend/TTS",
+	}
+	handler := func(ctx context.Context, req interface{}) (interface{}, error) {
+		return srv.(BackendServer).TTS(ctx, req.(*TTSRequest))
+	}
+	return interceptor(ctx, in, info, handler)
+}
+
+func _Backend_TokenizeString_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
+	in := new(PredictOptions)
+	if err := dec(in); err != nil {
+		return nil, err
+	}
+	if interceptor == nil {
+		return srv.(BackendServer).TokenizeString(ctx, in)
+	}
+	info := &grpc.UnaryServerInfo{
+		Server:     srv,
+		FullMethod: "/backend.Backend/TokenizeString",
+	}
+	handler := func(ctx context.Context, req interface{}) (interface{}, error) {
+		return srv.(BackendServer).TokenizeString(ctx, req.(*PredictOptions))
+	}
+	return interceptor(ctx, in, info, handler)
+}
+
+func _Backend_Status_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
+	in := new(HealthMessage)
+	if err := dec(in); err != nil {
+		return nil, err
+	}
+	if interceptor == nil {
+		return srv.(BackendServer).Status(ctx, in)
+	}
+	info := &grpc.UnaryServerInfo{
+		Server:     srv,
+		FullMethod: "/backend.Backend/Status",
+	}
+	handler := func(ctx context.Context, req interface{}) (interface{}, error) {
+		return srv.(BackendServer).Status(ctx, req.(*HealthMessage))
+	}
+	return interceptor(ctx, in, info, handler)
+}
+
+// Backend_ServiceDesc is the grpc.ServiceDesc for Backend service.
+// It's only intended for direct use with grpc.RegisterService,
+// and not to be introspected or modified (even as a copy)
+var Backend_ServiceDesc = grpc.ServiceDesc{
+	ServiceName: "backend.Backend",
+	HandlerType: (*BackendServer)(nil),
+	Methods: []grpc.MethodDesc{
+		{
+			MethodName: "Health",
+			Handler:    _Backend_Health_Handler,
+		},
+		{
+			MethodName: "Predict",
+			Handler:    _Backend_Predict_Handler,
+		},
+		{
+			MethodName: "LoadModel",
+			Handler:    _Backend_LoadModel_Handler,
+		},
+		{
+			MethodName: "Embedding",
+			Handler:    _Backend_Embedding_Handler,
+		},
+		{
+			MethodName: "GenerateImage",
+			Handler:    _Backend_GenerateImage_Handler,
+		},
+		{
+			MethodName: "AudioTranscription",
+			Handler:    _Backend_AudioTranscription_Handler,
+		},
+		{
+			MethodName: "TTS",
+			Handler:    _Backend_TTS_Handler,
+		},
+		{
+			MethodName: "TokenizeString",
+			Handler:    _Backend_TokenizeString_Handler,
+		},
+		{
+			MethodName: "Status",
+			Handler:    _Backend_Status_Handler,
+		},
+	},
+	Streams: []grpc.StreamDesc{
+		{
+			StreamName:    "PredictStream",
+			Handler:       _Backend_PredictStream_Handler,
+			ServerStreams: true,
+		},
+	},
+	Metadata: "backend/backend.proto",
+}

backend/cpp/llama/CMakeLists.txt

@@ -2,16 +2,20 @@
 ## XXX: In some versions of CMake clip wasn't being built before llama.
 ## This is an hack for now, but it should be fixed in the future.
 set(TARGET myclip)
-add_library(${TARGET} clip.cpp clip.h)
+add_library(${TARGET} clip.cpp clip.h llava.cpp llava.h)
 install(TARGETS ${TARGET} LIBRARY)
-target_link_libraries(${TARGET} PRIVATE common ggml ${CMAKE_THREAD_LIBS_INIT})
+target_include_directories(myclip PUBLIC .)
+target_include_directories(myclip PUBLIC ../..)
+target_include_directories(myclip PUBLIC ../../common)
+target_link_libraries(${TARGET} PRIVATE common ggml llama ${CMAKE_THREAD_LIBS_INIT})
 target_compile_features(${TARGET} PRIVATE cxx_std_11)
 if (NOT MSVC)
     target_compile_options(${TARGET} PRIVATE -Wno-cast-qual) # stb_image.h
 endif()
+# END CLIP hack
+
 
 set(TARGET grpc-server)
-# END CLIP hack
 set(CMAKE_CXX_STANDARD 17)
 cmake_minimum_required(VERSION 3.15)
 set(TARGET grpc-server)
@@ -70,7 +74,7 @@ add_library(hw_grpc_proto
   ${hw_proto_srcs}
   ${hw_proto_hdrs} )
 
-add_executable(${TARGET} grpc-server.cpp json.hpp )
+add_executable(${TARGET} grpc-server.cpp utils.hpp json.hpp)
 target_link_libraries(${TARGET} PRIVATE common llama myclip ${CMAKE_THREAD_LIBS_INIT} absl::flags hw_grpc_proto
   absl::flags_parse
   gRPC::${_REFLECTION}

backend/cpp/llama/Makefile

@@ -3,6 +3,7 @@ LLAMA_VERSION?=
 
 CMAKE_ARGS?=
 BUILD_TYPE?=
+ONEAPI_VARS?=/opt/intel/oneapi/setvars.sh
 
 # If build type is cublas, then we set -DLLAMA_CUBLAS=ON to CMAKE_ARGS automatically
 ifeq ($(BUILD_TYPE),cublas)
@@ -11,12 +12,21 @@ ifeq ($(BUILD_TYPE),cublas)
 # to CMAKE_ARGS automatically
 else ifeq ($(BUILD_TYPE),openblas)
 	CMAKE_ARGS+=-DLLAMA_BLAS=ON -DLLAMA_BLAS_VENDOR=OpenBLAS
-# If build type is clblast (openCL) we set -DLLAMA_CLBLAST=ON -DCLBlast_DIR=/some/path
-else ifeq ($(BUILD_TYPE),clblast)
+# If build type is clblas (openCL) we set -DLLAMA_CLBLAST=ON -DCLBlast_DIR=/some/path
+else ifeq ($(BUILD_TYPE),clblas)
 	CMAKE_ARGS+=-DLLAMA_CLBLAST=ON -DCLBlast_DIR=/some/path
 # If it's hipblas we do have also to set CC=/opt/rocm/llvm/bin/clang CXX=/opt/rocm/llvm/bin/clang++ 
 else ifeq ($(BUILD_TYPE),hipblas)
 	CMAKE_ARGS+=-DLLAMA_HIPBLAS=ON
+# If it's OSX, DO NOT embed the metal library - -DLLAMA_METAL_EMBED_LIBRARY=ON requires further investigation
+endif
+
+ifeq ($(BUILD_TYPE),sycl_f16)
+	CMAKE_ARGS+=-DLLAMA_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx -DLLAMA_SYCL_F16=ON
+endif
+
+ifeq ($(BUILD_TYPE),sycl_f32)
+	CMAKE_ARGS+=-DLLAMA_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx
 endif
 
 llama.cpp:
@@ -31,10 +41,14 @@ llama.cpp/examples/grpc-server:
 	cp -r $(abspath ./)/CMakeLists.txt llama.cpp/examples/grpc-server/
 	cp -r $(abspath ./)/grpc-server.cpp llama.cpp/examples/grpc-server/
 	cp -rfv $(abspath ./)/json.hpp llama.cpp/examples/grpc-server/
+	cp -rfv $(abspath ./)/utils.hpp llama.cpp/examples/grpc-server/
 	echo "add_subdirectory(grpc-server)" >> llama.cpp/examples/CMakeLists.txt
 ## XXX: In some versions of CMake clip wasn't being built before llama.
 ## This is an hack for now, but it should be fixed in the future.
 	cp -rfv llama.cpp/examples/llava/clip.h llama.cpp/examples/grpc-server/clip.h
+	cp -rfv llama.cpp/examples/llava/llava.cpp llama.cpp/examples/grpc-server/llava.cpp
+	echo '#include "llama.h"' > llama.cpp/examples/grpc-server/llava.h
+	cat llama.cpp/examples/llava/llava.h >> llama.cpp/examples/grpc-server/llava.h
 	cp -rfv llama.cpp/examples/llava/clip.cpp llama.cpp/examples/grpc-server/clip.cpp
 
 rebuild:
@@ -49,5 +63,10 @@ clean:
 	rm -rf grpc-server
 
 grpc-server: llama.cpp llama.cpp/examples/grpc-server
+ifneq (,$(findstring sycl,$(BUILD_TYPE)))
+	bash -c "source $(ONEAPI_VARS); \
+	cd llama.cpp && mkdir -p build && cd build && cmake .. $(CMAKE_ARGS) && cmake --build . --config Release"	
+else
 	cd llama.cpp && mkdir -p build && cd build && cmake .. $(CMAKE_ARGS) && cmake --build . --config Release
+endif
 	cp llama.cpp/build/bin/grpc-server .

backend/cpp/llama/utils.hpp

@@ -0,0 +1,510 @@
+// https://github.com/ggerganov/llama.cpp/blob/master/examples/server/utils.hpp
+
+#pragma once
+
+#include <string>
+#include <vector>
+#include <set>
+#include <mutex>
+#include <condition_variable>
+#include <unordered_map>
+
+#include "json.hpp"
+
+#include "../llava/clip.h"
+
+using json = nlohmann::json;
+
+extern bool server_verbose;
+
+#ifndef SERVER_VERBOSE
+#define SERVER_VERBOSE 1
+#endif
+
+#if SERVER_VERBOSE != 1
+#define LOG_VERBOSE(MSG, ...)
+#else
+#define LOG_VERBOSE(MSG, ...)                                            \
+    do                                                                   \
+    {                                                                    \
+        if (server_verbose)                                              \
+        {                                                                \
+            server_log("VERBOSE", __func__, __LINE__, MSG, __VA_ARGS__); \
+        }                                                                \
+    } while (0)
+#endif
+
+#define LOG_ERROR(  MSG, ...) server_log("ERROR",   __func__, __LINE__, MSG, __VA_ARGS__)
+#define LOG_WARNING(MSG, ...) server_log("WARNING", __func__, __LINE__, MSG, __VA_ARGS__)
+#define LOG_INFO(   MSG, ...) server_log("INFO",    __func__, __LINE__, MSG, __VA_ARGS__)
+
+//
+// parallel
+//
+
+enum server_state {
+    SERVER_STATE_LOADING_MODEL,  // Server is starting up, model not fully loaded yet
+    SERVER_STATE_READY,          // Server is ready and model is loaded
+    SERVER_STATE_ERROR           // An error occurred, load_model failed
+};
+
+enum task_type {
+    TASK_TYPE_COMPLETION,
+    TASK_TYPE_CANCEL,
+    TASK_TYPE_NEXT_RESPONSE
+};
+
+struct task_server {
+    int id = -1; // to be filled by llama_server_queue
+    int target_id;
+    task_type type;
+    json data;
+    bool infill_mode = false;
+    bool embedding_mode = false;
+    int multitask_id = -1;
+};
+
+struct task_result {
+    int id;
+    int multitask_id = -1;
+    bool stop;
+    bool error;
+    json result_json;
+};
+
+struct task_multi {
+    int id;
+    std::set<int> subtasks_remaining{};
+    std::vector<task_result> results{};
+};
+
+// TODO: can become bool if we can't find use of more states
+enum slot_state
+{
+    IDLE,
+    PROCESSING,
+};
+
+enum slot_command
+{
+    NONE,
+    LOAD_PROMPT,
+    RELEASE,
+};
+
+struct slot_params
+{
+    bool stream       = true;
+    bool cache_prompt = false; // remember the prompt to avoid reprocessing all prompt
+
+    uint32_t seed      = -1; // RNG seed
+    int32_t  n_keep    =  0; // number of tokens to keep from initial prompt
+    int32_t  n_predict = -1; // new tokens to predict
+
+    std::vector<std::string> antiprompt;
+
+    json input_prefix;
+    json input_suffix;
+};
+
+struct slot_image
+{
+    int32_t id;
+
+    bool request_encode_image = false;
+    float * image_embedding = nullptr;
+    int32_t image_tokens = 0;
+
+    clip_image_u8 * img_data;
+
+    std::string prefix_prompt; // before of this image
+};
+
+// completion token output with probabilities
+struct completion_token_output
+{
+    struct token_prob
+    {
+        llama_token tok;
+        float prob;
+    };
+
+    std::vector<token_prob> probs;
+    llama_token tok;
+    std::string text_to_send;
+};
+
+static inline void server_log(const char *level, const char *function, int line,
+                       const char *message, const nlohmann::ordered_json &extra)
+{
+    nlohmann::ordered_json log
+    {
+        {"timestamp", time(nullptr)},
+        {"level",     level},
+        {"function",  function},
+        {"line",      line},
+        {"message",   message},
+    };
+
+    if (!extra.empty())
+    {
+        log.merge_patch(extra);
+    }
+
+    const std::string str = log.dump(-1, ' ', false, json::error_handler_t::replace);
+    printf("%.*s\n", (int)str.size(), str.data());
+    fflush(stdout);
+}
+
+//
+// server utils
+//
+
+template <typename T>
+static T json_value(const json &body, const std::string &key, const T &default_value)
+{
+    // Fallback null to default value
+    return body.contains(key) && !body.at(key).is_null()
+        ? body.value(key, default_value)
+        : default_value;
+}
+
+inline std::string format_chatml(std::vector<json> messages)
+{
+    std::ostringstream chatml_msgs;
+
+    for (auto it = messages.begin(); it != messages.end(); ++it) {
+        chatml_msgs << "<|im_start|>"
+                    << json_value(*it, "role",    std::string("user")) << '\n';
+        chatml_msgs << json_value(*it, "content", std::string(""))
+                    << "<|im_end|>\n";
+    }
+
+    chatml_msgs << "<|im_start|>assistant" << '\n';
+
+    return chatml_msgs.str();
+}
+
+//
+// work queue utils
+//
+
+struct llama_server_queue {
+    int id = 0;
+    std::mutex mutex_tasks;
+    // queues
+    std::vector<task_server> queue_tasks;
+    std::vector<task_server> queue_tasks_deferred;
+    std::vector<task_multi> queue_multitasks;
+    std::condition_variable condition_tasks;
+    // callback functions
+    std::function<void(task_server&)> callback_new_task;
+    std::function<void(task_multi&)> callback_finish_multitask;
+    std::function<void(void)> callback_all_task_finished;
+
+    // Add a new task to the end of the queue
+    int post(task_server task) {
+        std::unique_lock<std::mutex> lock(mutex_tasks);
+        if (task.id == -1) {
+            task.id = id++;
+        }
+        queue_tasks.push_back(std::move(task));
+        condition_tasks.notify_one();
+        return task.id;
+    }
+
+    // Add a new task, but defer until one slot is available
+    void defer(task_server task) {
+        std::unique_lock<std::mutex> lock(mutex_tasks);
+        queue_tasks_deferred.push_back(std::move(task));
+    }
+
+    // Get the next id for creating anew task
+    int get_new_id() {
+        std::unique_lock<std::mutex> lock(mutex_tasks);
+        return id++;
+    }
+
+    // Register function to process a new task
+    void on_new_task(std::function<void(task_server&)> callback) {
+        callback_new_task = callback;
+    }
+
+    // Register function to process a multitask
+    void on_finish_multitask(std::function<void(task_multi&)> callback) {
+        callback_finish_multitask = callback;
+    }
+
+    // Register the function to be called when the batch of tasks is finished
+    void on_all_tasks_finished(std::function<void(void)> callback) {
+        callback_all_task_finished = callback;
+    }
+
+    // Call when the state of one slot is changed
+    void notify_slot_changed() {
+        // move deferred tasks back to main loop
+        std::unique_lock<std::mutex> lock(mutex_tasks);
+        for (auto & task : queue_tasks_deferred) {
+            queue_tasks.push_back(std::move(task));
+        }
+        queue_tasks_deferred.clear();
+    }
+
+    // Start the main loop. This call is blocking
+    [[noreturn]]
+    void start_loop() {
+        while (true) {
+            // new task arrived
+            LOG_VERBOSE("have new task", {});
+            {
+                while (true)
+                {
+                    std::unique_lock<std::mutex> lock(mutex_tasks);
+                    if (queue_tasks.empty()) {
+                        lock.unlock();
+                        break;
+                    }
+                    task_server task = queue_tasks.front();
+                    queue_tasks.erase(queue_tasks.begin());
+                    lock.unlock();
+                    LOG_VERBOSE("callback_new_task", {});
+                    callback_new_task(task);
+                }
+                LOG_VERBOSE("callback_all_task_finished", {});
+                // process and update all the multitasks
+                auto queue_iterator = queue_multitasks.begin();
+                while (queue_iterator != queue_multitasks.end())
+                {
+                    if (queue_iterator->subtasks_remaining.empty())
+                    {
+                        // all subtasks done == multitask is done
+                        task_multi current_multitask = *queue_iterator;
+                        callback_finish_multitask(current_multitask);
+                        // remove this multitask
+                        queue_iterator = queue_multitasks.erase(queue_iterator);
+                    }
+                    else
+                    {
+                        ++queue_iterator;
+                    }
+                }
+                // all tasks in the current loop is finished
+                callback_all_task_finished();
+            }
+            LOG_VERBOSE("wait for new task", {});
+            // wait for new task
+            {
+                std::unique_lock<std::mutex> lock(mutex_tasks);
+                if (queue_tasks.empty()) {
+                    condition_tasks.wait(lock, [&]{
+                        return !queue_tasks.empty();
+                    });
+                }
+            }
+        }
+    }
+
+    //
+    // functions to manage multitasks
+    //
+
+    // add a multitask by specifying the id of all subtask (subtask is a task_server)
+    void add_multitask(int multitask_id, std::vector<int>& sub_ids)
+    {
+        std::lock_guard<std::mutex> lock(mutex_tasks);
+        task_multi multi;
+        multi.id = multitask_id;
+        std::copy(sub_ids.begin(), sub_ids.end(), std::inserter(multi.subtasks_remaining, multi.subtasks_remaining.end()));
+        queue_multitasks.push_back(multi);
+    }
+
+    // updatethe remaining subtasks, while appending results to multitask
+    void update_multitask(int multitask_id, int subtask_id, task_result& result)
+    {
+        std::lock_guard<std::mutex> lock(mutex_tasks);
+        for (auto& multitask : queue_multitasks)
+        {
+            if (multitask.id == multitask_id)
+            {
+                multitask.subtasks_remaining.erase(subtask_id);
+                multitask.results.push_back(result);
+            }
+        }
+    }
+};
+
+struct llama_server_response {
+    typedef std::function<void(int, int, task_result&)> callback_multitask_t;
+    callback_multitask_t callback_update_multitask;
+    // for keeping track of all tasks waiting for the result
+    std::set<int> waiting_task_ids;
+    // the main result queue
+    std::vector<task_result> queue_results;
+    std::mutex mutex_results;
+    std::condition_variable condition_results;
+
+    void add_waiting_task_id(int task_id) {
+        std::unique_lock<std::mutex> lock(mutex_results);
+        waiting_task_ids.insert(task_id);
+    }
+
+    void remove_waiting_task_id(int task_id) {
+        std::unique_lock<std::mutex> lock(mutex_results);
+        waiting_task_ids.erase(task_id);
+    }
+
+    // This function blocks the thread until there is a response for this task_id
+    task_result recv(int task_id) {
+        while (true)
+        {
+            std::unique_lock<std::mutex> lock(mutex_results);
+            condition_results.wait(lock, [&]{
+                return !queue_results.empty();
+            });
+            LOG_VERBOSE("condition_results unblock", {});
+
+            for (int i = 0; i < (int) queue_results.size(); i++)
+            {
+                if (queue_results[i].id == task_id)
+                {
+                    assert(queue_results[i].multitask_id == -1);
+                    task_result res = queue_results[i];
+                    queue_results.erase(queue_results.begin() + i);
+                    return res;
+                }
+            }
+        }
+
+        // should never reach here
+    }
+
+    // Register the function to update multitask
+    void on_multitask_update(callback_multitask_t callback) {
+        callback_update_multitask = callback;
+    }
+
+    // Send a new result to a waiting task_id
+    void send(task_result result) {
+        std::unique_lock<std::mutex> lock(mutex_results);
+        LOG_VERBOSE("send new result", {});
+        for (auto& task_id : waiting_task_ids) {
+            // LOG_TEE("waiting task id %i \n", task_id);
+            // for now, tasks that have associated parent multitasks just get erased once multitask picks up the result
+            if (result.multitask_id == task_id)
+            {
+                LOG_VERBOSE("callback_update_multitask", {});
+                callback_update_multitask(task_id, result.id, result);
+                continue;
+            }
+
+            if (result.id == task_id)
+            {
+                LOG_VERBOSE("queue_results.push_back", {});
+                queue_results.push_back(result);
+                condition_results.notify_one();
+                return;
+            }
+        }
+    }
+};
+
+//
+// base64 utils (TODO: move to common in the future)
+//
+
+static const std::string base64_chars =
+             "ABCDEFGHIJKLMNOPQRSTUVWXYZ"
+             "abcdefghijklmnopqrstuvwxyz"
+             "0123456789+/";
+
+static inline bool is_base64(uint8_t c)
+{
+    return (isalnum(c) || (c == '+') || (c == '/'));
+}
+
+static inline std::vector<uint8_t> base64_decode(const std::string & encoded_string)
+{
+    int i = 0;
+    int j = 0;
+    int in_ = 0;
+
+    int in_len = encoded_string.size();
+
+    uint8_t char_array_4[4];
+    uint8_t char_array_3[3];
+
+    std::vector<uint8_t> ret;
+
+    while (in_len-- && (encoded_string[in_] != '=') && is_base64(encoded_string[in_]))
+    {
+        char_array_4[i++] = encoded_string[in_]; in_++;
+        if (i == 4)
+        {
+            for (i = 0; i <4; i++)
+            {
+                char_array_4[i] = base64_chars.find(char_array_4[i]);
+            }
+
+            char_array_3[0] = ((char_array_4[0]      ) << 2) + ((char_array_4[1] & 0x30) >> 4);
+            char_array_3[1] = ((char_array_4[1] & 0xf) << 4) + ((char_array_4[2] & 0x3c) >> 2);
+            char_array_3[2] = ((char_array_4[2] & 0x3) << 6) +   char_array_4[3];
+
+            for (i = 0; (i < 3); i++)
+            {
+                ret.push_back(char_array_3[i]);
+            }
+            i = 0;
+        }
+    }
+
+    if (i)
+    {
+        for (j = i; j <4; j++)
+        {
+            char_array_4[j] = 0;
+        }
+
+        for (j = 0; j <4; j++)
+        {
+            char_array_4[j] = base64_chars.find(char_array_4[j]);
+        }
+
+        char_array_3[0] = ((char_array_4[0]      ) << 2) + ((char_array_4[1] & 0x30) >> 4);
+        char_array_3[1] = ((char_array_4[1] & 0xf) << 4) + ((char_array_4[2] & 0x3c) >> 2);
+        char_array_3[2] = ((char_array_4[2] & 0x3) << 6) +   char_array_4[3];
+
+        for (j = 0; (j < i - 1); j++)
+        {
+            ret.push_back(char_array_3[j]);
+        }
+    }
+
+    return ret;
+}
+
+//
+// random string / id
+//
+
+static std::string random_string()
+{
+    static const std::string str("0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz");
+
+    std::random_device rd;
+    std::mt19937 generator(rd());
+
+    std::string result(32, ' ');
+
+    for (int i = 0; i < 32; ++i) {
+        result[i] = str[generator() % str.size()];
+    }
+
+    return result;
+}
+
+static std::string gen_chatcmplid()
+{
+    std::stringstream chatcmplid;
+    chatcmplid << "chatcmpl-" << random_string();
+    return chatcmplid.str();
+}

backend/go/llm/dolly/main.go

@@ -1,23 +0,0 @@
-package main
-
-// Note: this is started internally by LocalAI and a server is allocated for each model
-
-import (
-	"flag"
-
-	transformers "github.com/go-skynet/LocalAI/backend/go/llm/transformers"
-
-	grpc "github.com/go-skynet/LocalAI/pkg/grpc"
-)
-
-var (
-	addr = flag.String("addr", "localhost:50051", "the address to connect to")
-)
-
-func main() {
-	flag.Parse()
-
-	if err := grpc.StartServer(*addr, &transformers.Dolly{}); err != nil {
-		panic(err)
-	}
-}

backend/go/llm/falcon-ggml/main.go

@@ -1,23 +0,0 @@
-package main
-
-// Note: this is started internally by LocalAI and a server is allocated for each model
-
-import (
-	"flag"
-
-	transformers "github.com/go-skynet/LocalAI/backend/go/llm/transformers"
-
-	grpc "github.com/go-skynet/LocalAI/pkg/grpc"
-)
-
-var (
-	addr = flag.String("addr", "localhost:50051", "the address to connect to")
-)
-
-func main() {
-	flag.Parse()
-
-	if err := grpc.StartServer(*addr, &transformers.Falcon{}); err != nil {
-		panic(err)
-	}
-}

backend/go/llm/gpt2/main.go

@@ -1,23 +0,0 @@
-package main
-
-// Note: this is started internally by LocalAI and a server is allocated for each model
-
-import (
-	"flag"
-
-	transformers "github.com/go-skynet/LocalAI/backend/go/llm/transformers"
-
-	grpc "github.com/go-skynet/LocalAI/pkg/grpc"
-)
-
-var (
-	addr = flag.String("addr", "localhost:50051", "the address to connect to")
-)
-
-func main() {
-	flag.Parse()
-
-	if err := grpc.StartServer(*addr, &transformers.GPT2{}); err != nil {
-		panic(err)
-	}
-}

backend/go/llm/gptj/main.go

@@ -1,23 +0,0 @@
-package main
-
-// Note: this is started internally by LocalAI and a server is allocated for each model
-
-import (
-	"flag"
-
-	transformers "github.com/go-skynet/LocalAI/backend/go/llm/transformers"
-
-	grpc "github.com/go-skynet/LocalAI/pkg/grpc"
-)
-
-var (
-	addr = flag.String("addr", "localhost:50051", "the address to connect to")
-)
-
-func main() {
-	flag.Parse()
-
-	if err := grpc.StartServer(*addr, &transformers.GPTJ{}); err != nil {
-		panic(err)
-	}
-}

backend/go/llm/gptneox/main.go

@@ -1,23 +0,0 @@
-package main
-
-// Note: this is started internally by LocalAI and a server is allocated for each model
-
-import (
-	"flag"
-
-	transformers "github.com/go-skynet/LocalAI/backend/go/llm/transformers"
-
-	grpc "github.com/go-skynet/LocalAI/pkg/grpc"
-)
-
-var (
-	addr = flag.String("addr", "localhost:50051", "the address to connect to")
-)
-
-func main() {
-	flag.Parse()
-
-	if err := grpc.StartServer(*addr, &transformers.GPTNeoX{}); err != nil {
-		panic(err)
-	}
-}

backend/go/llm/mpt/main.go

@@ -1,23 +0,0 @@
-package main
-
-// Note: this is started internally by LocalAI and a server is allocated for each model
-
-import (
-	"flag"
-
-	transformers "github.com/go-skynet/LocalAI/backend/go/llm/transformers"
-
-	grpc "github.com/go-skynet/LocalAI/pkg/grpc"
-)
-
-var (
-	addr = flag.String("addr", "localhost:50051", "the address to connect to")
-)
-
-func main() {
-	flag.Parse()
-
-	if err := grpc.StartServer(*addr, &transformers.MPT{}); err != nil {
-		panic(err)
-	}
-}

backend/go/llm/replit/main.go

@@ -1,23 +0,0 @@
-package main
-
-// Note: this is started internally by LocalAI and a server is allocated for each model
-
-import (
-	"flag"
-
-	transformers "github.com/go-skynet/LocalAI/backend/go/llm/transformers"
-
-	grpc "github.com/go-skynet/LocalAI/pkg/grpc"
-)
-
-var (
-	addr = flag.String("addr", "localhost:50051", "the address to connect to")
-)
-
-func main() {
-	flag.Parse()
-
-	if err := grpc.StartServer(*addr, &transformers.Replit{}); err != nil {
-		panic(err)
-	}
-}

backend/go/llm/starcoder/main.go

@@ -1,23 +0,0 @@
-package main
-
-// Note: this is started internally by LocalAI and a server is allocated for each model
-
-import (
-	"flag"
-
-	transformers "github.com/go-skynet/LocalAI/backend/go/llm/transformers"
-
-	grpc "github.com/go-skynet/LocalAI/pkg/grpc"
-)
-
-var (
-	addr = flag.String("addr", "localhost:50051", "the address to connect to")
-)
-
-func main() {
-	flag.Parse()
-
-	if err := grpc.StartServer(*addr, &transformers.Starcoder{}); err != nil {
-		panic(err)
-	}
-}

backend/go/llm/transformers/dolly.go

@@ -1,44 +0,0 @@
-package transformers
-
-// This is a wrapper to statisfy the GRPC service interface
-// It is meant to be used by the main executable that is the server for the specific backend type (falcon, gpt3, etc)
-import (
-	"fmt"
-
-	"github.com/go-skynet/LocalAI/pkg/grpc/base"
-	pb "github.com/go-skynet/LocalAI/pkg/grpc/proto"
-
-	transformers "github.com/go-skynet/go-ggml-transformers.cpp"
-)
-
-type Dolly struct {
-	base.SingleThread
-
-	dolly *transformers.Dolly
-}
-
-func (llm *Dolly) Load(opts *pb.ModelOptions) error {
-	model, err := transformers.NewDolly(opts.ModelFile)
-	llm.dolly = model
-	return err
-}
-
-func (llm *Dolly) Predict(opts *pb.PredictOptions) (string, error) {
-	return llm.dolly.Predict(opts.Prompt, buildPredictOptions(opts)...)
-}
-
-// fallback to Predict
-func (llm *Dolly) PredictStream(opts *pb.PredictOptions, results chan string) error {
-
-	go func() {
-		res, err := llm.dolly.Predict(opts.Prompt, buildPredictOptions(opts)...)
-
-		if err != nil {
-			fmt.Println("err: ", err)
-		}
-		results <- res
-		close(results)
-	}()
-
-	return nil
-}

backend/go/llm/transformers/falcon.go

@@ -1,43 +0,0 @@
-package transformers
-
-// This is a wrapper to statisfy the GRPC service interface
-// It is meant to be used by the main executable that is the server for the specific backend type (falcon, gpt3, etc)
-import (
-	"fmt"
-
-	"github.com/go-skynet/LocalAI/pkg/grpc/base"
-	pb "github.com/go-skynet/LocalAI/pkg/grpc/proto"
-
-	transformers "github.com/go-skynet/go-ggml-transformers.cpp"
-)
-
-type Falcon struct {
-	base.SingleThread
-
-	falcon *transformers.Falcon
-}
-
-func (llm *Falcon) Load(opts *pb.ModelOptions) error {
-	model, err := transformers.NewFalcon(opts.ModelFile)
-	llm.falcon = model
-	return err
-}
-
-func (llm *Falcon) Predict(opts *pb.PredictOptions) (string, error) {
-	return llm.falcon.Predict(opts.Prompt, buildPredictOptions(opts)...)
-}
-
-// fallback to Predict
-func (llm *Falcon) PredictStream(opts *pb.PredictOptions, results chan string) error {
-	go func() {
-		res, err := llm.falcon.Predict(opts.Prompt, buildPredictOptions(opts)...)
-
-		if err != nil {
-			fmt.Println("err: ", err)
-		}
-		results <- res
-		close(results)
-	}()
-
-	return nil
-}

backend/go/llm/transformers/gpt2.go

@@ -1,42 +0,0 @@
-package transformers
-
-// This is a wrapper to statisfy the GRPC service interface
-// It is meant to be used by the main executable that is the server for the specific backend type (falcon, gpt3, etc)
-import (
-	"fmt"
-
-	"github.com/go-skynet/LocalAI/pkg/grpc/base"
-	pb "github.com/go-skynet/LocalAI/pkg/grpc/proto"
-
-	transformers "github.com/go-skynet/go-ggml-transformers.cpp"
-)
-
-type GPT2 struct {
-	base.SingleThread
-
-	gpt2 *transformers.GPT2
-}
-
-func (llm *GPT2) Load(opts *pb.ModelOptions) error {
-	model, err := transformers.New(opts.ModelFile)
-	llm.gpt2 = model
-	return err
-}
-
-func (llm *GPT2) Predict(opts *pb.PredictOptions) (string, error) {
-	return llm.gpt2.Predict(opts.Prompt, buildPredictOptions(opts)...)
-}
-
-// fallback to Predict
-func (llm *GPT2) PredictStream(opts *pb.PredictOptions, results chan string) error {
-	go func() {
-		res, err := llm.gpt2.Predict(opts.Prompt, buildPredictOptions(opts)...)
-
-		if err != nil {
-			fmt.Println("err: ", err)
-		}
-		results <- res
-		close(results)
-	}()
-	return nil
-}

backend/go/llm/transformers/gptj.go

@@ -1,42 +0,0 @@
-package transformers
-
-// This is a wrapper to statisfy the GRPC service interface
-// It is meant to be used by the main executable that is the server for the specific backend type (falcon, gpt3, etc)
-import (
-	"fmt"
-
-	"github.com/go-skynet/LocalAI/pkg/grpc/base"
-	pb "github.com/go-skynet/LocalAI/pkg/grpc/proto"
-
-	transformers "github.com/go-skynet/go-ggml-transformers.cpp"
-)
-
-type GPTJ struct {
-	base.SingleThread
-
-	gptj *transformers.GPTJ
-}
-
-func (llm *GPTJ) Load(opts *pb.ModelOptions) error {
-	model, err := transformers.NewGPTJ(opts.ModelFile)
-	llm.gptj = model
-	return err
-}
-
-func (llm *GPTJ) Predict(opts *pb.PredictOptions) (string, error) {
-	return llm.gptj.Predict(opts.Prompt, buildPredictOptions(opts)...)
-}
-
-// fallback to Predict
-func (llm *GPTJ) PredictStream(opts *pb.PredictOptions, results chan string) error {
-	go func() {
-		res, err := llm.gptj.Predict(opts.Prompt, buildPredictOptions(opts)...)
-
-		if err != nil {
-			fmt.Println("err: ", err)
-		}
-		results <- res
-		close(results)
-	}()
-	return nil
-}

backend/go/llm/transformers/gptneox.go

@@ -1,42 +0,0 @@
-package transformers
-
-// This is a wrapper to statisfy the GRPC service interface
-// It is meant to be used by the main executable that is the server for the specific backend type (falcon, gpt3, etc)
-import (
-	"fmt"
-
-	"github.com/go-skynet/LocalAI/pkg/grpc/base"
-	pb "github.com/go-skynet/LocalAI/pkg/grpc/proto"
-
-	transformers "github.com/go-skynet/go-ggml-transformers.cpp"
-)
-
-type GPTNeoX struct {
-	base.SingleThread
-
-	gptneox *transformers.GPTNeoX
-}
-
-func (llm *GPTNeoX) Load(opts *pb.ModelOptions) error {
-	model, err := transformers.NewGPTNeoX(opts.ModelFile)
-	llm.gptneox = model
-	return err
-}
-
-func (llm *GPTNeoX) Predict(opts *pb.PredictOptions) (string, error) {
-	return llm.gptneox.Predict(opts.Prompt, buildPredictOptions(opts)...)
-}
-
-// fallback to Predict
-func (llm *GPTNeoX) PredictStream(opts *pb.PredictOptions, results chan string) error {
-	go func() {
-		res, err := llm.gptneox.Predict(opts.Prompt, buildPredictOptions(opts)...)
-
-		if err != nil {
-			fmt.Println("err: ", err)
-		}
-		results <- res
-		close(results)
-	}()
-	return nil
-}

backend/go/llm/transformers/mpt.go

@@ -1,42 +0,0 @@
-package transformers
-
-// This is a wrapper to statisfy the GRPC service interface
-// It is meant to be used by the main executable that is the server for the specific backend type (falcon, gpt3, etc)
-import (
-	"fmt"
-
-	"github.com/go-skynet/LocalAI/pkg/grpc/base"
-	pb "github.com/go-skynet/LocalAI/pkg/grpc/proto"
-
-	transformers "github.com/go-skynet/go-ggml-transformers.cpp"
-)
-
-type MPT struct {
-	base.SingleThread
-
-	mpt *transformers.MPT
-}
-
-func (llm *MPT) Load(opts *pb.ModelOptions) error {
-	model, err := transformers.NewMPT(opts.ModelFile)
-	llm.mpt = model
-	return err
-}
-
-func (llm *MPT) Predict(opts *pb.PredictOptions) (string, error) {
-	return llm.mpt.Predict(opts.Prompt, buildPredictOptions(opts)...)
-}
-
-// fallback to Predict
-func (llm *MPT) PredictStream(opts *pb.PredictOptions, results chan string) error {
-	go func() {
-		res, err := llm.mpt.Predict(opts.Prompt, buildPredictOptions(opts)...)
-
-		if err != nil {
-			fmt.Println("err: ", err)
-		}
-		results <- res
-		close(results)
-	}()
-	return nil
-}

backend/go/llm/transformers/predict.go

@@ -1,26 +0,0 @@
-package transformers
-
-import (
-	pb "github.com/go-skynet/LocalAI/pkg/grpc/proto"
-	transformers "github.com/go-skynet/go-ggml-transformers.cpp"
-)
-
-func buildPredictOptions(opts *pb.PredictOptions) []transformers.PredictOption {
-	predictOptions := []transformers.PredictOption{
-		transformers.SetTemperature(float64(opts.Temperature)),
-		transformers.SetTopP(float64(opts.TopP)),
-		transformers.SetTopK(int(opts.TopK)),
-		transformers.SetTokens(int(opts.Tokens)),
-		transformers.SetThreads(int(opts.Threads)),
-	}
-
-	if opts.Batch != 0 {
-		predictOptions = append(predictOptions, transformers.SetBatch(int(opts.Batch)))
-	}
-
-	if opts.Seed != 0 {
-		predictOptions = append(predictOptions, transformers.SetSeed(int(opts.Seed)))
-	}
-
-	return predictOptions
-}

backend/go/llm/transformers/replit.go

@@ -1,42 +0,0 @@
-package transformers
-
-// This is a wrapper to statisfy the GRPC service interface
-// It is meant to be used by the main executable that is the server for the specific backend type (falcon, gpt3, etc)
-import (
-	"fmt"
-
-	"github.com/go-skynet/LocalAI/pkg/grpc/base"
-	pb "github.com/go-skynet/LocalAI/pkg/grpc/proto"
-
-	transformers "github.com/go-skynet/go-ggml-transformers.cpp"
-)
-
-type Replit struct {
-	base.SingleThread
-
-	replit *transformers.Replit
-}
-
-func (llm *Replit) Load(opts *pb.ModelOptions) error {
-	model, err := transformers.NewReplit(opts.ModelFile)
-	llm.replit = model
-	return err
-}
-
-func (llm *Replit) Predict(opts *pb.PredictOptions) (string, error) {
-	return llm.replit.Predict(opts.Prompt, buildPredictOptions(opts)...)
-}
-
-// fallback to Predict
-func (llm *Replit) PredictStream(opts *pb.PredictOptions, results chan string) error {
-	go func() {
-		res, err := llm.replit.Predict(opts.Prompt, buildPredictOptions(opts)...)
-
-		if err != nil {
-			fmt.Println("err: ", err)
-		}
-		results <- res
-		close(results)
-	}()
-	return nil
-}

backend/go/llm/transformers/starcoder.go

@@ -1,43 +0,0 @@
-package transformers
-
-// This is a wrapper to statisfy the GRPC service interface
-// It is meant to be used by the main executable that is the server for the specific backend type (falcon, gpt3, etc)
-import (
-	"fmt"
-
-	"github.com/go-skynet/LocalAI/pkg/grpc/base"
-	pb "github.com/go-skynet/LocalAI/pkg/grpc/proto"
-
-	transformers "github.com/go-skynet/go-ggml-transformers.cpp"
-)
-
-type Starcoder struct {
-	base.SingleThread
-
-	starcoder *transformers.Starcoder
-}
-
-func (llm *Starcoder) Load(opts *pb.ModelOptions) error {
-	model, err := transformers.NewStarcoder(opts.ModelFile)
-	llm.starcoder = model
-	return err
-}
-
-func (llm *Starcoder) Predict(opts *pb.PredictOptions) (string, error) {
-	return llm.starcoder.Predict(opts.Prompt, buildPredictOptions(opts)...)
-}
-
-// fallback to Predict
-func (llm *Starcoder) PredictStream(opts *pb.PredictOptions, results chan string) error {
-	go func() {
-		res, err := llm.starcoder.Predict(opts.Prompt, buildPredictOptions(opts)...)
-
-		if err != nil {
-			fmt.Println("err: ", err)
-		}
-		results <- res
-		close(results)
-	}()
-
-	return nil
-}

backend/go/transcribe/transcript.go

@@ -8,24 +8,24 @@ import (
 
 	"github.com/ggerganov/whisper.cpp/bindings/go/pkg/whisper"
 	"github.com/go-audio/wav"
-	"github.com/go-skynet/LocalAI/api/schema"
+	"github.com/go-skynet/LocalAI/core/schema"
 )
 
-func sh(c string) (string, error) {
-	cmd := exec.Command("/bin/sh", "-c", c)
+func runCommand(command []string) (string, error) {
+	cmd := exec.Command(command[0], command[1:]...)
 	cmd.Env = os.Environ()
-	o, err := cmd.CombinedOutput()
-	return string(o), err
+	out, err := cmd.CombinedOutput()
+	return string(out), err
 }
 
-// AudioToWav converts audio to wav for transcribe. It bashes out to ffmpeg
+// AudioToWav converts audio to wav for transcribe.
 // TODO: use https://github.com/mccoyst/ogg?
 func audioToWav(src, dst string) error {
-	out, err := sh(fmt.Sprintf("ffmpeg -i %s -format s16le -ar 16000 -ac 1 -acodec pcm_s16le %s", src, dst))
+    command := []string{"ffmpeg", "-i", src, "-format", "s16le", "-ar", "16000", "-ac", "1", "-acodec", "pcm_s16le", dst}
+	out, err := runCommand(command)
 	if err != nil {
 		return fmt.Errorf("error: %w out: %s", err, out)
 	}
-
 	return nil
 }
 

backend/go/transcribe/whisper.go

@@ -4,7 +4,7 @@ package main
 // It is meant to be used by the main executable that is the server for the specific backend type (falcon, gpt3, etc)
 import (
 	"github.com/ggerganov/whisper.cpp/bindings/go/pkg/whisper"
-	"github.com/go-skynet/LocalAI/api/schema"
+	"github.com/go-skynet/LocalAI/core/schema"
 	"github.com/go-skynet/LocalAI/pkg/grpc/base"
 	pb "github.com/go-skynet/LocalAI/pkg/grpc/proto"
 )

backend/python/autogptq/autogptq.py

@@ -33,7 +33,7 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
             model = AutoGPTQForCausalLM.from_quantized(request.Model,
                     model_basename=request.ModelBaseName,
                     use_safetensors=True,
-                    trust_remote_code=True,
+                    trust_remote_code=request.TrustRemoteCode,
                     device=device,
                     use_triton=request.UseTriton,
                     quantize_config=None)

backend/python/autogptq/autogptq.yml

@@ -71,7 +71,7 @@ dependencies:
       - regex==2023.10.3
       - requests==2.31.0
       - rouge==1.0.1
-      - safetensors==0.3.3
+      - safetensors>=0.3.3
       - six==1.16.0
       - sympy==1.12
       - tokenizers==0.14.0

backend/python/common-env/transformers/Makefile

@@ -4,6 +4,17 @@ ifeq ($(BUILD_TYPE), cublas)
 	CONDA_ENV_PATH = "transformers-nvidia.yml"
 endif
 
+ifeq ($(BUILD_TYPE), hipblas)
+	CONDA_ENV_PATH = "transformers-rocm.yml"
+endif
+
+# Intel GPU are supposed to have dependencies installed in the main python
+# environment, so we skip conda installation for SYCL builds.
+# https://github.com/intel/intel-extension-for-pytorch/issues/538
+ifneq (,$(findstring sycl,$(BUILD_TYPE)))
+export SKIP_CONDA=1
+endif
+
 .PHONY: transformers
 transformers:
 	@echo "Installing $(CONDA_ENV_PATH)..."

backend/python/common-env/transformers/install.sh

@@ -1,24 +1,38 @@
 #!/bin/bash
 set -ex
 
+SKIP_CONDA=${SKIP_CONDA:-0}
+
 # Check if environment exist
 conda_env_exists(){
     ! conda list --name "${@}" >/dev/null 2>/dev/null
 }
 
-if conda_env_exists "transformers" ; then
-    echo "Creating virtual environment..."
-    conda env create --name transformers --file $1
-    echo "Virtual environment created."
-else 
-    echo "Virtual environment already exists."
+if [ $SKIP_CONDA -eq 1 ]; then
+    echo "Skipping conda environment installation"
+else
+    export PATH=$PATH:/opt/conda/bin
+    if conda_env_exists "transformers" ; then
+        echo "Creating virtual environment..."
+        conda env create --name transformers --file $1
+        echo "Virtual environment created."
+    else 
+        echo "Virtual environment already exists."
+    fi
+fi
+
+if [ -d "/opt/intel" ]; then
+    # Intel GPU: If the directory exists, we assume we are using the intel image
+    # (no conda env)
+    # https://github.com/intel/intel-extension-for-pytorch/issues/538
+    pip install intel-extension-for-transformers datasets sentencepiece tiktoken neural_speed
 fi
 
 if [ "$PIP_CACHE_PURGE" = true ] ; then
-    export PATH=$PATH:/opt/conda/bin
-
-    # Activate conda environment
-    source activate transformers
+    if [ $SKIP_CONDA -eq 0 ]; then
+        # Activate conda environment
+        source activate transformers
+    fi
 
     pip cache purge
 fi

backend/python/common-env/transformers/transformers-nvidia.yml

@@ -30,12 +30,14 @@ dependencies:
       - async-timeout==4.0.3
       - attrs==23.1.0
       - bark==0.1.5
+      - bitsandbytes==0.43.0
       - boto3==1.28.61
       - botocore==1.31.61
       - certifi==2023.7.22
+      - TTS==0.22.0
       - charset-normalizer==3.3.0
       - datasets==2.14.5
-      - sentence-transformers==2.2.2
+      - sentence-transformers==2.5.1 # Updated Version
       - sentencepiece==0.1.99
       - dill==0.3.7
       - einops==0.7.0
@@ -80,8 +82,8 @@ dependencies:
       - requests==2.31.0
       - rouge==1.0.1
       - s3transfer==0.7.0
-      - safetensors==0.3.3
-      - scipy==1.11.3
+      - safetensors>=0.4.1
+      - scipy==1.12.0 # Updated Version
       - six==1.16.0
       - sympy==1.12
       - tokenizers
@@ -112,7 +114,7 @@ dependencies:
       - sudachipy
       - sudachidict_core
       - vocos
-      - vllm==0.2.7
-      - transformers>=4.36.0  # Required for Mixtral.
+      - vllm==0.3.2
+      - transformers>=4.38.2  # Updated Version
       - xformers==0.0.23.post1  
 prefix: /opt/conda/envs/transformers

backend/python/common-env/transformers/transformers-rocm.yml

@@ -0,0 +1,109 @@
+name: transformers
+channels:
+  - defaults
+dependencies:
+  - _libgcc_mutex=0.1=main
+  - _openmp_mutex=5.1=1_gnu
+  - bzip2=1.0.8=h7b6447c_0
+  - ca-certificates=2023.08.22=h06a4308_0
+  - ld_impl_linux-64=2.38=h1181459_1
+  - libffi=3.4.4=h6a678d5_0
+  - libgcc-ng=11.2.0=h1234567_1
+  - libgomp=11.2.0=h1234567_1
+  - libstdcxx-ng=11.2.0=h1234567_1
+  - libuuid=1.41.5=h5eee18b_0
+  - ncurses=6.4=h6a678d5_0
+  - openssl=3.0.11=h7f8727e_2
+  - pip=23.2.1=py311h06a4308_0
+  - python=3.11.5=h955ad1f_0
+  - readline=8.2=h5eee18b_0
+  - setuptools=68.0.0=py311h06a4308_0
+  - sqlite=3.41.2=h5eee18b_0
+  - tk=8.6.12=h1ccaba5_0
+  - wheel=0.41.2=py311h06a4308_0
+  - xz=5.4.2=h5eee18b_0
+  - zlib=1.2.13=h5eee18b_0
+  - pip:
+      - --pre
+      - --extra-index-url https://download.pytorch.org/whl/nightly/
+      - accelerate==0.23.0
+      - aiohttp==3.8.5
+      - aiosignal==1.3.1
+      - async-timeout==4.0.3
+      - attrs==23.1.0
+      - bark==0.1.5
+      - boto3==1.28.61
+      - botocore==1.31.61
+      - certifi==2023.7.22
+      - TTS==0.22.0
+      - charset-normalizer==3.3.0
+      - datasets==2.14.5
+      - sentence-transformers==2.5.1 # Updated Version
+      - sentencepiece==0.1.99
+      - dill==0.3.7
+      - einops==0.7.0
+      - encodec==0.1.1
+      - filelock==3.12.4
+      - frozenlist==1.4.0
+      - fsspec==2023.6.0
+      - funcy==2.0
+      - grpcio==1.59.0
+      - huggingface-hub
+      - idna==3.4
+      - jinja2==3.1.2
+      - jmespath==1.0.1
+      - markupsafe==2.1.3
+      - mpmath==1.3.0
+      - multidict==6.0.4
+      - multiprocess==0.70.15
+      - networkx
+      - numpy==1.26.0
+      - packaging==23.2
+      - pandas
+      - peft==0.5.0
+      - protobuf==4.24.4
+      - psutil==5.9.5
+      - pyarrow==13.0.0
+      - python-dateutil==2.8.2
+      - pytz==2023.3.post1
+      - pyyaml==6.0.1
+      - regex==2023.10.3
+      - requests==2.31.0
+      - rouge==1.0.1
+      - s3transfer==0.7.0
+      - safetensors>=0.4.1
+      - scipy==1.12.0 # Updated Version
+      - six==1.16.0
+      - sympy==1.12
+      - tokenizers
+      - torch
+      - torchaudio
+      - tqdm==4.66.1
+      - triton==2.1.0
+      - typing-extensions==4.8.0
+      - tzdata==2023.3
+      - auto-gptq==0.6.0
+      - urllib3==1.26.17
+      - xxhash==3.4.1
+      - yarl==1.9.2
+      - soundfile
+      - langid
+      - wget
+      - unidecode
+      - pyopenjtalk-prebuilt
+      - pypinyin
+      - inflect
+      - cn2an
+      - jieba
+      - eng_to_ipa
+      - openai-whisper
+      - matplotlib
+      - gradio==3.41.2
+      - nltk
+      - sudachipy
+      - sudachidict_core
+      - vocos
+      - vllm==0.3.2
+      - transformers>=4.38.2  # Updated Version
+      - xformers==0.0.23.post1
+prefix: /opt/conda/envs/transformers

backend/python/common-env/transformers/transformers.yml

@@ -36,7 +36,7 @@ dependencies:
       - TTS==0.22.0
       - charset-normalizer==3.3.0
       - datasets==2.14.5
-      - sentence-transformers==2.2.2
+      - sentence-transformers==2.5.1 # Updated Version
       - sentencepiece==0.1.99
       - dill==0.3.7
       - einops==0.7.0
@@ -69,8 +69,8 @@ dependencies:
       - requests==2.31.0
       - rouge==1.0.1
       - s3transfer==0.7.0
-      - safetensors==0.3.3
-      - scipy==1.11.3
+      - safetensors>=0.4.1
+      - scipy==1.12.0 # Updated Version
       - six==1.16.0
       - sympy==1.12
       - tokenizers
@@ -101,7 +101,7 @@ dependencies:
       - sudachipy
       - sudachidict_core
       - vocos
-      - vllm==0.2.7
-      - transformers>=4.36.0  # Required for Mixtral.
+      - vllm==0.3.2
+      - transformers>=4.38.2  # Updated Version
       - xformers==0.0.23.post1  
-prefix: /opt/conda/envs/transformers
+prefix: /opt/conda/envs/transformers

backend/python/diffusers/Makefile

@@ -1,8 +1,20 @@
+export CONDA_ENV_PATH = "diffusers.yml"
+
+ifeq ($(BUILD_TYPE), hipblas)
+export CONDA_ENV_PATH = "diffusers-rocm.yml"
+endif
+
+# Intel GPU are supposed to have dependencies installed in the main python
+# environment, so we skip conda installation for SYCL builds.
+# https://github.com/intel/intel-extension-for-pytorch/issues/538
+ifneq (,$(findstring sycl,$(BUILD_TYPE)))
+export SKIP_CONDA=1
+endif
+
 .PHONY: diffusers
 diffusers:
-	@echo "Creating virtual environment..."
-	@conda env create --name diffusers --file diffusers.yml
-	@echo "Virtual environment created."
+	@echo "Installing $(CONDA_ENV_PATH)..."
+	bash install.sh $(CONDA_ENV_PATH)
 
 .PHONY: run
 run:
@@ -11,4 +23,4 @@ run:
 	@echo "Diffusers run."
 
 test:
-	bash test.sh
+	bash test.sh

backend/python/diffusers/backend_diffusers.py

@@ -21,14 +21,15 @@ from diffusers import StableDiffusionXLPipeline, StableDiffusionDepth2ImgPipelin
 from diffusers import StableDiffusionImg2ImgPipeline, AutoPipelineForText2Image, ControlNetModel, StableVideoDiffusionPipeline
 from diffusers.pipelines.stable_diffusion import safety_checker
 from diffusers.utils import load_image,export_to_video
-from compel import Compel
+from compel import Compel, ReturnedEmbeddingsType
 
 from transformers import CLIPTextModel
 from safetensors.torch import load_file
 
 
 _ONE_DAY_IN_SECONDS = 60 * 60 * 24
-COMPEL=os.environ.get("COMPEL", "1") == "1"
+COMPEL=os.environ.get("COMPEL", "0") == "1"
+XPU=os.environ.get("XPU", "0") == "1"
 CLIPSKIP=os.environ.get("CLIPSKIP", "1") == "1"
 SAFETENSORS=os.environ.get("SAFETENSORS", "1") == "1"
 CHUNK_SIZE=os.environ.get("CHUNK_SIZE", "8")
@@ -36,6 +37,10 @@ FPS=os.environ.get("FPS", "7")
 DISABLE_CPU_OFFLOAD=os.environ.get("DISABLE_CPU_OFFLOAD", "0") == "1"
 FRAMES=os.environ.get("FRAMES", "64")
 
+if XPU:
+    import intel_extension_for_pytorch as ipex
+    print(ipex.xpu.get_device_name(0))
+
 # If MAX_WORKERS are specified in the environment use it, otherwise default to 1
 MAX_WORKERS = int(os.environ.get('PYTHON_GRPC_MAX_WORKERS', '1'))
 
@@ -231,8 +236,13 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
             if request.SchedulerType != "":
                 self.pipe.scheduler = get_scheduler(request.SchedulerType, self.pipe.scheduler.config)
                 
-            if not self.img2vid:
-                self.compel = Compel(tokenizer=self.pipe.tokenizer, text_encoder=self.pipe.text_encoder)
+            if COMPEL:
+                self.compel = Compel(
+                    tokenizer=[self.pipe.tokenizer, self.pipe.tokenizer_2 ], 
+                    text_encoder=[self.pipe.text_encoder, self.pipe.text_encoder_2],
+                    returned_embeddings_type=ReturnedEmbeddingsType.PENULTIMATE_HIDDEN_STATES_NON_NORMALIZED,
+                    requires_pooled=[False, True]
+                    )
 
 
             if request.ControlNet:
@@ -247,6 +257,8 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
                 self.pipe.to('cuda')
                 if self.controlnet:
                     self.controlnet.to('cuda')
+            if XPU:
+                self.pipe = self.pipe.to("xpu")
             # Assume directory from request.ModelFile.
             # Only if request.LoraAdapter it's not an absolute path
             if request.LoraAdapter and request.ModelFile != "" and not os.path.isabs(request.LoraAdapter) and request.LoraAdapter:
@@ -386,8 +398,9 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
 
         image = {}
         if COMPEL:
-            conditioning = self.compel.build_conditioning_tensor(prompt)
-            kwargs["prompt_embeds"]= conditioning
+            conditioning, pooled = self.compel.build_conditioning_tensor(prompt)
+            kwargs["prompt_embeds"] = conditioning
+            kwargs["pooled_prompt_embeds"] = pooled
             # pass the kwargs dictionary to the self.pipe method
             image = self.pipe(
                 guidance_scale=self.cfg_scale,

backend/python/diffusers/diffusers-rocm.yml

@@ -0,0 +1,64 @@
+name: diffusers
+channels:
+  - defaults
+dependencies:
+  - _libgcc_mutex=0.1=main
+  - _openmp_mutex=5.1=1_gnu
+  - bzip2=1.0.8=h7b6447c_0
+  - ca-certificates=2023.08.22=h06a4308_0
+  - ld_impl_linux-64=2.38=h1181459_1
+  - libffi=3.4.4=h6a678d5_0
+  - libgcc-ng=11.2.0=h1234567_1
+  - libgomp=11.2.0=h1234567_1
+  - libstdcxx-ng=11.2.0=h1234567_1
+  - libuuid=1.41.5=h5eee18b_0
+  - ncurses=6.4=h6a678d5_0
+  - openssl=3.0.11=h7f8727e_2
+  - pip=23.2.1=py311h06a4308_0
+  - python=3.11.5=h955ad1f_0
+  - readline=8.2=h5eee18b_0
+  - setuptools=68.0.0=py311h06a4308_0
+  - sqlite=3.41.2=h5eee18b_0
+  - tk=8.6.12=h1ccaba5_0
+  - tzdata=2023c=h04d1e81_0
+  - wheel=0.41.2=py311h06a4308_0
+  - xz=5.4.2=h5eee18b_0
+  - zlib=1.2.13=h5eee18b_0
+  - pip:
+      - --pre
+      - --extra-index-url https://download.pytorch.org/whl/nightly/
+      - accelerate>=0.11.0
+      - certifi==2023.7.22
+      - charset-normalizer==3.3.0
+      - compel==2.0.2
+      - diffusers==0.24.0
+      - filelock==3.12.4
+      - fsspec==2023.9.2
+      - grpcio==1.59.0
+      - huggingface-hub>=0.19.4
+      - idna==3.4
+      - importlib-metadata==6.8.0
+      - jinja2==3.1.2
+      - markupsafe==2.1.3
+      - mpmath==1.3.0
+      - networkx==3.1
+      - numpy==1.26.0
+      - omegaconf
+      - packaging==23.2
+      - pillow==10.0.1
+      - protobuf==4.24.4
+      - psutil==5.9.5
+      - pyparsing==3.1.1
+      - pyyaml==6.0.1
+      - regex==2023.10.3
+      - requests==2.31.0
+      - safetensors==0.4.0
+      - sympy==1.12
+      - tqdm==4.66.1
+      - transformers>=4.25.1
+      - triton==2.1.0
+      - typing-extensions==4.8.0
+      - urllib3==2.0.6
+      - zipp==3.17.0
+      - torch
+prefix: /opt/conda/envs/diffusers

backend/python/diffusers/diffusers.yml

@@ -71,4 +71,4 @@ dependencies:
       - typing-extensions==4.8.0
       - urllib3==2.0.6
       - zipp==3.17.0
-prefix: /opt/conda/envs/diffusers
+prefix: /opt/conda/envs/diffusers

backend/python/diffusers/install.sh

@@ -0,0 +1,50 @@
+#!/bin/bash
+set -ex
+
+SKIP_CONDA=${SKIP_CONDA:-0}
+
+# Check if environment exist
+conda_env_exists(){
+    ! conda list --name "${@}" >/dev/null 2>/dev/null
+}
+
+if [ $SKIP_CONDA -eq 1 ]; then
+    echo "Skipping conda environment installation"
+else
+    export PATH=$PATH:/opt/conda/bin
+    if conda_env_exists "diffusers" ; then
+        echo "Creating virtual environment..."
+        conda env create --name diffusers --file $1
+        echo "Virtual environment created."
+    else 
+        echo "Virtual environment already exists."
+    fi
+fi
+
+if [ -d "/opt/intel" ]; then
+    # Intel GPU: If the directory exists, we assume we are using the Intel image
+    # https://github.com/intel/intel-extension-for-pytorch/issues/538
+    pip install torch==2.1.0a0 \
+                torchvision==0.16.0a0 \
+                torchaudio==2.1.0a0 \
+                intel-extension-for-pytorch==2.1.10+xpu \
+                --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
+    
+    pip install google-api-python-client \
+                grpcio \
+                grpcio-tools \
+                diffusers==0.24.0 \
+                transformers>=4.25.1 \
+                accelerate \
+                compel==2.0.2 \
+                Pillow
+fi
+
+if [ "$PIP_CACHE_PURGE" = true ] ; then
+    if [ $SKIP_CONDA -ne 1 ]; then
+        # Activate conda environment
+        source activate diffusers
+    fi
+
+    pip cache purge
+fi

backend/python/diffusers/run.sh

@@ -3,10 +3,15 @@
 ##
 ## A bash script wrapper that runs the diffusers server with conda
 
-export PATH=$PATH:/opt/conda/bin
-
-# Activate conda environment
-source activate diffusers
+if [ -d "/opt/intel" ]; then
+    # Assumes we are using the Intel oneAPI container image
+    # https://github.com/intel/intel-extension-for-pytorch/issues/538
+    export XPU=1
+else
+    export PATH=$PATH:/opt/conda/bin
+    # Activate conda environment
+    source activate diffusers
+fi
 
 # get the directory where the bash script is located
 DIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )" >/dev/null 2>&1 && pwd )"

backend/python/exllama/Makefile

@@ -1,7 +1,8 @@
+export CONDA_ENV_PATH = "exllama.yml"
+
 .PHONY: exllama
 exllama:
-	$(MAKE) -C ../common-env/transformers
-	bash install.sh
+	bash install.sh ${CONDA_ENV_PATH}
 
 .PHONY: run
 run:

backend/python/exllama/install.sh

@@ -1,14 +1,27 @@
 #!/bin/bash
+set -ex
 
-##
-## A bash script installs the required dependencies of VALL-E-X and prepares the environment
 export PATH=$PATH:/opt/conda/bin
 
-# Activate conda environment
-source activate transformers
+if [ "$BUILD_TYPE" != "cublas" ]; then
+    echo "[exllama] Attention!!! Nvidia GPU is required - skipping installation"
+    exit 0
+fi
 
-echo $CONDA_PREFIX
+# Check if environment exist
+conda_env_exists(){
+    ! conda list --name "${@}" >/dev/null 2>/dev/null
+}
 
+if conda_env_exists "exllama" ; then
+    echo "Creating virtual environment..."
+    conda env create --name exllama --file $1
+    echo "Virtual environment created."
+else
+    echo "Virtual environment already exists."
+fi
+
+source activate exllama
 
 git clone https://github.com/turboderp/exllama $CONDA_PREFIX/exllama && pushd $CONDA_PREFIX/exllama && pip install -r requirements.txt && popd
 

backend/python/exllama/run.sh

@@ -2,11 +2,10 @@
 
 ##
 ## A bash script wrapper that runs the exllama server with conda
-
 export PATH=$PATH:/opt/conda/bin
 
 # Activate conda environment
-source activate transformers
+source activate exllama
 
 # get the directory where the bash script is located
 DIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )" >/dev/null 2>&1 && pwd )"

backend/python/exllama2/install.sh

@@ -1,15 +1,29 @@
 #!/bin/bash
-
+set -e
 ##
 ## A bash script installs the required dependencies of VALL-E-X and prepares the environment
-export PATH=$PATH:/opt/conda/bin
+export SHA=c0ddebaaaf8ffd1b3529c2bb654e650bce2f790f
 
-# Activate conda environment
+if [ "$BUILD_TYPE" != "cublas" ]; then
+    echo "[exllamav2] Attention!!! Nvidia GPU is required - skipping installation"
+    exit 0
+fi
+
+export PATH=$PATH:/opt/conda/bin
 source activate transformers
 
 echo $CONDA_PREFIX
 
-git clone https://github.com/turboderp/exllamav2 $CONDA_PREFIX/exllamav2 && pushd $CONDA_PREFIX/exllamav2 && pip install -r requirements.txt && popd
+git clone https://github.com/turboderp/exllamav2 $CONDA_PREFIX/exllamav2
+
+pushd $CONDA_PREFIX/exllamav2
+
+git checkout -b build $SHA
+
+# TODO: this needs to be pinned within the conda environments
+pip install -r requirements.txt
+
+popd
 
 cp -rfv $CONDA_PREFIX/exllamav2/* ./  
 

backend/python/mamba/install.sh

@@ -1,14 +1,15 @@
 #!/bin/bash
-
+set -e
 ##
 ## A bash script installs the required dependencies of VALL-E-X and prepares the environment
-export PATH=$PATH:/opt/conda/bin
 
 if [ "$BUILD_TYPE" != "cublas" ]; then
     echo "[mamba] Attention!!! nvcc is required - skipping installation"
     exit 0
 fi
 
+export PATH=$PATH:/opt/conda/bin
+
 # Activate conda environment
 source activate transformers
 

backend/python/petals/Makefile

@@ -1,7 +1,7 @@
 .PHONY: petals
 petals:
 	@echo "Creating virtual environment..."
-	@conda env create --name petals --file petals.yml
+	bash install.sh "petals.yml"
 	@echo "Virtual environment created."
 
 .PHONY: run

backend/python/petals/install.sh

@@ -0,0 +1,5 @@
+#!/bin/bash
+
+export PATH=$PATH:/opt/conda/bin
+
+conda env create --name petals --file $1

backend/python/transformers/run.sh

@@ -3,10 +3,16 @@
 ##
 ## A bash script wrapper that runs the transformers server with conda
 
-export PATH=$PATH:/opt/conda/bin
 
-# Activate conda environment
-source activate transformers
+if [ -d "/opt/intel" ]; then
+    # Assumes we are using the Intel oneAPI container image
+    # https://github.com/intel/intel-extension-for-pytorch/issues/538
+    export XPU=1
+else
+    export PATH=$PATH:/opt/conda/bin
+    # Activate conda environment
+    source activate transformers
+fi
 
 # get the directory where the bash script is located
 DIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )" >/dev/null 2>&1 && pwd )"

backend/python/transformers/transformers_server.py

@@ -15,8 +15,16 @@ import backend_pb2_grpc
 
 import grpc
 import torch
+import torch.cuda
+
+XPU=os.environ.get("XPU", "0") == "1"
+if XPU:
+    import intel_extension_for_pytorch as ipex
+    from intel_extension_for_transformers.transformers.modeling import AutoModelForCausalLM
+    from transformers import AutoTokenizer, AutoModel, set_seed
+else:
+    from transformers import AutoTokenizer, AutoModel, AutoModelForCausalLM, set_seed, BitsAndBytesConfig
 
-from transformers import AutoTokenizer, AutoModel
 
 _ONE_DAY_IN_SECONDS = 60 * 60 * 24
 
@@ -67,19 +75,60 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
             A Result object that contains the result of the LoadModel operation.
         """
         model_name = request.Model
-        try:
-            self.model = AutoModel.from_pretrained(model_name, trust_remote_code=True) # trust_remote_code is needed to use the encode method with embeddings models like jinai-v2
-            self.tokenizer = AutoTokenizer.from_pretrained(model_name)
 
-            if request.CUDA:
+        compute = "auto"
+        if request.F16Memory == True:
+            compute=torch.bfloat16
+
+        self.CUDA = request.CUDA
+
+        device_map="cpu"
+
+        quantization = None
+
+        if self.CUDA:
+            if request.Device:
+                device_map=request.Device
+            else:
+                device_map="cuda:0"
+            if request.Quantization == "bnb_4bit":
+                quantization = BitsAndBytesConfig(
+                    load_in_4bit = True,
+                    bnb_4bit_compute_dtype = compute,
+                    bnb_4bit_quant_type = "nf4",
+                    bnb_4bit_use_double_quant = True,
+                    load_in_8bit = False,
+                )
+            elif request.Quantization == "bnb_8bit":
+                quantization = BitsAndBytesConfig(
+                    load_in_4bit=False,
+                    bnb_4bit_compute_dtype = None,
+                    load_in_8bit=True,                                   
+                )
+                                                   
+    
+        try:
+            if request.Type == "AutoModelForCausalLM":
+                if XPU:
+                    if quantization == "xpu_4bit":
+                        xpu_4bit = True
+                    self.model = AutoModelForCausalLM.from_pretrained(model_name, trust_remote_code=request.TrustRemoteCode,
+                                              device_map="xpu", load_in_4bit=xpu_4bit)
+                else:
+                    self.model = AutoModelForCausalLM.from_pretrained(model_name, trust_remote_code=request.TrustRemoteCode, use_safetensors=True, quantization_config=quantization, device_map=device_map, torch_dtype=compute)
+            else:
+                self.model = AutoModel.from_pretrained(model_name, trust_remote_code=request.TrustRemoteCode,  use_safetensors=True,  quantization_config=quantization, device_map=device_map, torch_dtype=compute)
+            self.tokenizer = AutoTokenizer.from_pretrained(model_name, use_safetensors=True)
+            self.XPU = False
+
+            if XPU:
+                self.XPU = True
                 try:
-                    # TODO: also tensorflow, make configurable
-                    import torch.cuda
-                    if torch.cuda.is_available():
-                        print("Loading model", model_name, "to CUDA.", file=sys.stderr)
-                        self.model = self.model.to("cuda")
+                    print("Optimizing model", model_name, "to XPU.", file=sys.stderr)
+                    self.model = ipex.optimize_transformers(self.model, inplace=True, dtype=torch.float16, device="xpu")
                 except Exception as err:
-                    print("Not using CUDA:", err, file=sys.stderr)
+                    print("Not using XPU:", err, file=sys.stderr)
+
         except Exception as err:
             return backend_pb2.Result(success=False, message=f"Unexpected {err=}, {type(err)=}")
         # Implement your logic here for the LoadModel service
@@ -98,6 +147,7 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
             An EmbeddingResult object that contains the calculated embeddings.
         """
 
+        set_seed(request.Seed)
         # Tokenize input
         max_length = 512
         if request.Tokens != 0:
@@ -105,13 +155,60 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
         encoded_input = self.tokenizer(request.Embeddings, padding=True, truncation=True, max_length=max_length, return_tensors="pt")    
 
         # Create word embeddings
-        model_output = self.model(**encoded_input)
+        if self.CUDA:
+            encoded_input = encoded_input.to("cuda")
+
+        with torch.no_grad():    
+            model_output = self.model(**encoded_input)
 
         # Pool to get sentence embeddings; i.e. generate one 1024 vector for the entire sentence
-        sentence_embeddings = mean_pooling(model_output, encoded_input['attention_mask']).detach().numpy()
+        sentence_embeddings = mean_pooling(model_output, encoded_input['attention_mask'])
         print("Calculated embeddings for: " + request.Embeddings, file=sys.stderr)
         print("Embeddings:", sentence_embeddings, file=sys.stderr)
-        return backend_pb2.EmbeddingResult(embeddings=sentence_embeddings)
+        return backend_pb2.EmbeddingResult(embeddings=sentence_embeddings[0])
+
+    def Predict(self, request, context):
+        """
+        Generates text based on the given prompt and sampling parameters.
+
+        Args:
+            request: The predict request.
+            context: The gRPC context.
+
+        Returns:
+            backend_pb2.Reply: The predict result.
+        """
+        set_seed(request.Seed)
+        if request.TopP == 0:
+            request.TopP = 0.9
+
+        max_tokens = 200
+        if request.Tokens > 0:
+            max_tokens = request.Tokens
+
+        inputs = self.tokenizer(request.Prompt, return_tensors="pt").input_ids
+        if self.CUDA:
+            inputs = inputs.to("cuda")
+        if XPU:
+            inputs = inputs.to("xpu")
+
+        outputs = self.model.generate(inputs,max_new_tokens=max_tokens, temperature=request.Temperature, top_p=request.TopP, do_sample=True, pad_token_id=self.tokenizer.eos_token_id)
+        generated_text = self.tokenizer.batch_decode(outputs[:, inputs.shape[1]:], skip_special_tokens=True)[0]
+
+        return backend_pb2.Reply(message=bytes(generated_text, encoding='utf-8'))
+
+    def PredictStream(self, request, context):
+        """
+        Generates text based on the given prompt and sampling parameters, and streams the results.
+
+        Args:
+            request: The predict stream request.
+            context: The gRPC context.
+
+        Returns:
+            backend_pb2.Result: The predict stream result.
+        """
+        yield self.Predict(request, context)
 
 
 def serve(address):

backend/python/vall-e-x/Makefile

@@ -1,3 +1,7 @@
+ifneq (,$(findstring sycl,$(BUILD_TYPE)))
+export SKIP_CONDA=1
+endif
+
 .PHONY: ttsvalle
 ttsvalle:
 	$(MAKE) -C ../common-env/transformers

backend/python/vall-e-x/install.sh

@@ -2,15 +2,18 @@
 
 ##
 ## A bash script installs the required dependencies of VALL-E-X and prepares the environment
-export PATH=$PATH:/opt/conda/bin
 export SHA=3faaf8ccadb154d63b38070caf518ce9309ea0f4
 
-# Activate conda environment
-source activate transformers
+SKIP_CONDA=${SKIP_CONDA:-0}
 
-echo $CONDA_PREFIX
+if [ $SKIP_CONDA -ne 1 ]; then
+    source activate transformers
+else
+    export PATH=$PATH:/opt/conda/bin
+    CONDA_PREFIX=$PWD
+fi
 
-git clone https://github.com/Plachtaa/VALL-E-X.git $CONDA_PREFIX/vall-e-x && pushd $CONDA_PREFIX/vall-e-x && git checkout -b build $SHA && pip install -r requirements.txt && popd
+git clone https://github.com/Plachtaa/VALL-E-X.git $CONDA_PREFIX/vall-e-x && pushd $CONDA_PREFIX/vall-e-x && git checkout -b build $SHA && popd
 
 cp -rfv $CONDA_PREFIX/vall-e-x/* ./
 

backend/python/vall-e-x/ttsvalle.py

@@ -55,6 +55,7 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
             print("Preparing models, please wait", file=sys.stderr)
             # download and load all models
             preload_models()
+            self.clonedVoice = False
             # Assume directory from request.ModelFile.
             # Only if request.LoraAdapter it's not an absolute path
             if request.AudioPath and request.ModelFile != "" and not os.path.isabs(request.AudioPath):
@@ -65,6 +66,7 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
             if request.AudioPath != "":
                 print("Generating model", file=sys.stderr)
                 make_prompt(name=model_name, audio_prompt_path=request.AudioPath)
+                self.clonedVoice = True
                 ### Use given transcript
                 ##make_prompt(name=model_name, audio_prompt_path="paimon_prompt.wav",
                 ##                transcript="Just, what was that? Paimon thought we were gonna get eaten.")
@@ -91,6 +93,8 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
         try:
             audio_array = None
             if model != "":
+                if self.clonedVoice:
+                    model = os.path.basename(request.model)
                 audio_array = generate_audio(request.text, prompt=model)
             else:
                 audio_array = generate_audio(request.text)

backend/python/vall-e-x/ttsvalle.yml

@@ -79,7 +79,7 @@ dependencies:
       - pypinyin==0.49.0
       - python-multipart==0.0.6
       - regex==2023.10.3
-      - safetensors==0.4.0
+      - safetensors>=0.4.0
       - semantic-version==2.10.0
       - soundfile==0.12.1
       - starlette==0.27.0

backend/python/vllm/backend_vllm.py

@@ -1,6 +1,6 @@
 #!/usr/bin/env python3
+import asyncio
 from concurrent import futures
-import time
 import argparse
 import signal
 import sys
@@ -10,7 +10,10 @@ import backend_pb2
 import backend_pb2_grpc
 
 import grpc
-from vllm import LLM, SamplingParams
+from vllm.engine.arg_utils import AsyncEngineArgs
+from vllm.engine.async_llm_engine import AsyncLLMEngine
+from vllm.sampling_params import SamplingParams
+from vllm.utils import random_uuid
 
 _ONE_DAY_IN_SECONDS = 60 * 60 * 24
 
@@ -79,16 +82,30 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
         Returns:
             backend_pb2.Result: The load model result.
         """
+        engine_args = AsyncEngineArgs(
+            model=request.Model,
+        )
+
+        if request.Quantization != "":
+            engine_args.quantization = request.Quantization
+        if request.GPUMemoryUtilization != 0:
+            engine_args.gpu_memory_utilization = request.GPUMemoryUtilization
+        if request.TrustRemoteCode:
+            engine_args.trust_remote_code = request.TrustRemoteCode
+        if request.EnforceEager:
+            engine_args.enforce_eager = request.EnforceEager
+        if request.SwapSpace != 0:
+            engine_args.swap_space = request.SwapSpace
+        if request.MaxModelLen != 0:
+            engine_args.max_model_len = request.MaxModelLen
+
         try:
-            if request.Quantization != "":
-                self.llm = LLM(model=request.Model, quantization=request.Quantization)
-            else:
-                self.llm = LLM(model=request.Model)
+            self.llm = AsyncLLMEngine.from_engine_args(engine_args)
         except Exception as err:
             return backend_pb2.Result(success=False, message=f"Unexpected {err=}, {type(err)=}")
         return backend_pb2.Result(message="Model loaded successfully", success=True)
 
-    def Predict(self, request, context):
+    async def Predict(self, request, context):
         """
         Generates text based on the given prompt and sampling parameters.
 
@@ -99,24 +116,11 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
         Returns:
             backend_pb2.Reply: The predict result.
         """
-        if request.TopP == 0:
-            request.TopP = 0.9
+        gen = self._predict(request, context, streaming=False)
+        res = await gen.__anext__()
+        return res
 
-        max_tokens = 200
-        if request.Tokens > 0:
-            max_tokens = request.Tokens
-
-        sampling_params = SamplingParams(max_tokens=max_tokens, temperature=request.Temperature, top_p=request.TopP)
-        outputs = self.llm.generate([request.Prompt], sampling_params)
-
-        generated_text = outputs[0].outputs[0].text
-        # Remove prompt from response if present
-        if request.Prompt in generated_text:
-            generated_text = generated_text.replace(request.Prompt, "")
-
-        return backend_pb2.Reply(message=bytes(generated_text, encoding='utf-8'))
-
-    def PredictStream(self, request, context):
+    async def PredictStream(self, request, context):
         """
         Generates text based on the given prompt and sampling parameters, and streams the results.
 
@@ -127,30 +131,84 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
         Returns:
             backend_pb2.Result: The predict stream result.
         """
-        yield self.Predict(request, context)
+        iterations = self._predict(request, context, streaming=True)
+        try:
+            async for iteration in iterations:
+                yield iteration
+        finally:
+            await iterations.aclose()
 
-def serve(address):
-    server = grpc.server(futures.ThreadPoolExecutor(max_workers=MAX_WORKERS))
+    async def _predict(self, request, context, streaming=False):
+
+        # Build sampling parameters
+        sampling_params = SamplingParams(top_p=0.9, max_tokens=200)
+        if request.TopP != 0:
+            sampling_params.top_p = request.TopP
+        if request.Tokens > 0:
+            sampling_params.max_tokens = request.Tokens
+        if request.Temperature != 0:
+            sampling_params.temperature = request.Temperature
+        if request.TopK != 0:
+            sampling_params.top_k = request.TopK
+        if request.PresencePenalty != 0:
+            sampling_params.presence_penalty = request.PresencePenalty
+        if request.FrequencyPenalty != 0:
+            sampling_params.frequency_penalty = request.FrequencyPenalty
+        if request.StopPrompts:
+            sampling_params.stop = request.StopPrompts
+        if request.IgnoreEOS:
+            sampling_params.ignore_eos = request.IgnoreEOS
+        if request.Seed != 0:
+            sampling_params.seed = request.Seed
+
+        # Generate text
+        request_id = random_uuid()
+        outputs = self.llm.generate(request.Prompt, sampling_params, request_id)
+
+        # Stream the results
+        generated_text = ""
+        try:
+            async for request_output in outputs:
+                iteration_text = request_output.outputs[0].text
+
+                if streaming:
+                    # Remove text already sent as vllm concatenates the text from previous yields
+                    delta_iteration_text = iteration_text.removeprefix(generated_text)
+                    # Send the partial result
+                    yield backend_pb2.Reply(message=bytes(delta_iteration_text, encoding='utf-8'))
+
+                # Keep track of text generated
+                generated_text = iteration_text
+        finally:
+            await outputs.aclose()
+
+        # If streaming, we already sent everything
+        if streaming:
+            return
+
+        # Sending the final generated text
+        yield backend_pb2.Reply(message=bytes(generated_text, encoding='utf-8'))
+
+async def serve(address):
+    # Start asyncio gRPC server
+    server = grpc.aio.server(migration_thread_pool=futures.ThreadPoolExecutor(max_workers=MAX_WORKERS))
+    # Add the servicer to the server
     backend_pb2_grpc.add_BackendServicer_to_server(BackendServicer(), server)
+    # Bind the server to the address
     server.add_insecure_port(address)
-    server.start()
+
+    # Gracefully shutdown the server on SIGTERM or SIGINT
+    loop = asyncio.get_event_loop()
+    for sig in (signal.SIGINT, signal.SIGTERM):
+        loop.add_signal_handler(
+            sig, lambda: asyncio.ensure_future(server.stop(5))
+        )
+
+    # Start the server
+    await server.start()
     print("Server started. Listening on: " + address, file=sys.stderr)
-
-    # Define the signal handler function
-    def signal_handler(sig, frame):
-        print("Received termination signal. Shutting down...")
-        server.stop(0)
-        sys.exit(0)
-
-    # Set the signal handlers for SIGINT and SIGTERM
-    signal.signal(signal.SIGINT, signal_handler)
-    signal.signal(signal.SIGTERM, signal_handler)
-
-    try:
-        while True:
-            time.sleep(_ONE_DAY_IN_SECONDS)
-    except KeyboardInterrupt:
-        server.stop(0)
+    # Wait for the server to be terminated
+    await server.wait_for_termination()
 
 if __name__ == "__main__":
     parser = argparse.ArgumentParser(description="Run the gRPC server.")
@@ -159,4 +217,4 @@ if __name__ == "__main__":
     )
     args = parser.parse_args()
 
-    serve(args.addr)
+    asyncio.run(serve(args.addr))

core/backend/embeddings.go

@@ -3,36 +3,32 @@ package backend
 import (
 	"fmt"
 
-	config "github.com/go-skynet/LocalAI/api/config"
-	"github.com/go-skynet/LocalAI/api/options"
+	"github.com/go-skynet/LocalAI/core/config"
+
 	"github.com/go-skynet/LocalAI/pkg/grpc"
 	model "github.com/go-skynet/LocalAI/pkg/model"
 )
 
-func ModelEmbedding(s string, tokens []int, loader *model.ModelLoader, c config.Config, o *options.Option) (func() ([]float32, error), error) {
-	if !c.Embeddings {
-		return nil, fmt.Errorf("endpoint disabled for this model by API configuration")
-	}
+func ModelEmbedding(s string, tokens []int, loader *model.ModelLoader, backendConfig config.BackendConfig, appConfig *config.ApplicationConfig) (func() ([]float32, error), error) {
+	modelFile := backendConfig.Model
 
-	modelFile := c.Model
-
-	grpcOpts := gRPCModelOpts(c)
+	grpcOpts := gRPCModelOpts(backendConfig)
 
 	var inferenceModel interface{}
 	var err error
 
-	opts := modelOpts(c, o, []model.Option{
+	opts := modelOpts(backendConfig, appConfig, []model.Option{
 		model.WithLoadGRPCLoadModelOpts(grpcOpts),
-		model.WithThreads(uint32(c.Threads)),
-		model.WithAssetDir(o.AssetsDestination),
+		model.WithThreads(uint32(*backendConfig.Threads)),
+		model.WithAssetDir(appConfig.AssetsDestination),
 		model.WithModel(modelFile),
-		model.WithContext(o.Context),
+		model.WithContext(appConfig.Context),
 	})
 
-	if c.Backend == "" {
+	if backendConfig.Backend == "" {
 		inferenceModel, err = loader.GreedyLoader(opts...)
 	} else {
-		opts = append(opts, model.WithBackendString(c.Backend))
+		opts = append(opts, model.WithBackendString(backendConfig.Backend))
 		inferenceModel, err = loader.BackendLoader(opts...)
 	}
 	if err != nil {
@@ -41,9 +37,9 @@ func ModelEmbedding(s string, tokens []int, loader *model.ModelLoader, c config.
 
 	var fn func() ([]float32, error)
 	switch model := inferenceModel.(type) {
-	case *grpc.Client:
+	case grpc.Backend:
 		fn = func() ([]float32, error) {
-			predictOptions := gRPCPredictOpts(c, loader.ModelPath)
+			predictOptions := gRPCPredictOpts(backendConfig, loader.ModelPath)
 			if len(tokens) > 0 {
 				embeds := []int32{}
 
@@ -52,7 +48,7 @@ func ModelEmbedding(s string, tokens []int, loader *model.ModelLoader, c config.
 				}
 				predictOptions.EmbeddingTokens = embeds
 
-				res, err := model.Embeddings(o.Context, predictOptions)
+				res, err := model.Embeddings(appConfig.Context, predictOptions)
 				if err != nil {
 					return nil, err
 				}
@@ -61,7 +57,7 @@ func ModelEmbedding(s string, tokens []int, loader *model.ModelLoader, c config.
 			}
 			predictOptions.Embeddings = s
 
-			res, err := model.Embeddings(o.Context, predictOptions)
+			res, err := model.Embeddings(appConfig.Context, predictOptions)
 			if err != nil {
 				return nil, err
 			}

core/backend/image.go

@@ -0,0 +1,52 @@
+package backend
+
+import (
+	"github.com/go-skynet/LocalAI/core/config"
+
+	"github.com/go-skynet/LocalAI/pkg/grpc/proto"
+	model "github.com/go-skynet/LocalAI/pkg/model"
+)
+
+func ImageGeneration(height, width, mode, step, seed int, positive_prompt, negative_prompt, src, dst string, loader *model.ModelLoader, backendConfig config.BackendConfig, appConfig *config.ApplicationConfig) (func() error, error) {
+	threads := backendConfig.Threads
+	if *threads == 0 && appConfig.Threads != 0 {
+		threads = &appConfig.Threads
+	}
+	gRPCOpts := gRPCModelOpts(backendConfig)
+	opts := modelOpts(backendConfig, appConfig, []model.Option{
+		model.WithBackendString(backendConfig.Backend),
+		model.WithAssetDir(appConfig.AssetsDestination),
+		model.WithThreads(uint32(*threads)),
+		model.WithContext(appConfig.Context),
+		model.WithModel(backendConfig.Model),
+		model.WithLoadGRPCLoadModelOpts(gRPCOpts),
+	})
+
+	inferenceModel, err := loader.BackendLoader(
+		opts...,
+	)
+	if err != nil {
+		return nil, err
+	}
+
+	fn := func() error {
+		_, err := inferenceModel.GenerateImage(
+			appConfig.Context,
+			&proto.GenerateImageRequest{
+				Height:           int32(height),
+				Width:            int32(width),
+				Mode:             int32(mode),
+				Step:             int32(step),
+				Seed:             int32(seed),
+				CLIPSkip:         int32(backendConfig.Diffusers.ClipSkip),
+				PositivePrompt:   positive_prompt,
+				NegativePrompt:   negative_prompt,
+				Dst:              dst,
+				Src:              src,
+				EnableParameters: backendConfig.Diffusers.EnableParameters,
+			})
+		return err
+	}
+
+	return fn, nil
+}

core/backend/llm.go

@@ -8,8 +8,8 @@ import (
 	"sync"
 	"unicode/utf8"
 
-	config "github.com/go-skynet/LocalAI/api/config"
-	"github.com/go-skynet/LocalAI/api/options"
+	"github.com/go-skynet/LocalAI/core/config"
+
 	"github.com/go-skynet/LocalAI/pkg/gallery"
 	"github.com/go-skynet/LocalAI/pkg/grpc"
 	model "github.com/go-skynet/LocalAI/pkg/model"
@@ -26,17 +26,20 @@ type TokenUsage struct {
 	Completion int
 }
 
-func ModelInference(ctx context.Context, s string, images []string, loader *model.ModelLoader, c config.Config, o *options.Option, tokenCallback func(string, TokenUsage) bool) (func() (LLMResponse, error), error) {
+func ModelInference(ctx context.Context, s string, images []string, loader *model.ModelLoader, c config.BackendConfig, o *config.ApplicationConfig, tokenCallback func(string, TokenUsage) bool) (func() (LLMResponse, error), error) {
 	modelFile := c.Model
-
+	threads := c.Threads
+	if *threads == 0 && o.Threads != 0 {
+		threads = &o.Threads
+	}
 	grpcOpts := gRPCModelOpts(c)
 
-	var inferenceModel *grpc.Client
+	var inferenceModel grpc.Backend
 	var err error
 
 	opts := modelOpts(c, o, []model.Option{
 		model.WithLoadGRPCLoadModelOpts(grpcOpts),
-		model.WithThreads(uint32(c.Threads)), // some models uses this to allocate threads during startup
+		model.WithThreads(uint32(*threads)), // some models uses this to allocate threads during startup
 		model.WithAssetDir(o.AssetsDestination),
 		model.WithModel(modelFile),
 		model.WithContext(o.Context),
@@ -140,7 +143,7 @@ func ModelInference(ctx context.Context, s string, images []string, loader *mode
 var cutstrings map[string]*regexp.Regexp = make(map[string]*regexp.Regexp)
 var mu sync.Mutex = sync.Mutex{}
 
-func Finetune(config config.Config, input, prediction string) string {
+func Finetune(config config.BackendConfig, input, prediction string) string {
 	if config.Echo {
 		prediction = input + prediction
 	}

core/backend/options.go

@@ -0,0 +1,141 @@
+package backend
+
+import (
+	"os"
+	"path/filepath"
+
+	"github.com/go-skynet/LocalAI/core/config"
+	pb "github.com/go-skynet/LocalAI/pkg/grpc/proto"
+	model "github.com/go-skynet/LocalAI/pkg/model"
+)
+
+func modelOpts(c config.BackendConfig, so *config.ApplicationConfig, opts []model.Option) []model.Option {
+	if so.SingleBackend {
+		opts = append(opts, model.WithSingleActiveBackend())
+	}
+
+	if so.ParallelBackendRequests {
+		opts = append(opts, model.EnableParallelRequests)
+	}
+
+	if c.GRPC.Attempts != 0 {
+		opts = append(opts, model.WithGRPCAttempts(c.GRPC.Attempts))
+	}
+
+	if c.GRPC.AttemptsSleepTime != 0 {
+		opts = append(opts, model.WithGRPCAttemptsDelay(c.GRPC.AttemptsSleepTime))
+	}
+
+	for k, v := range so.ExternalGRPCBackends {
+		opts = append(opts, model.WithExternalBackend(k, v))
+	}
+
+	return opts
+}
+
+func gRPCModelOpts(c config.BackendConfig) *pb.ModelOptions {
+	b := 512
+	if c.Batch != 0 {
+		b = c.Batch
+	}
+
+	return &pb.ModelOptions{
+		CUDA:                 c.CUDA || c.Diffusers.CUDA,
+		SchedulerType:        c.Diffusers.SchedulerType,
+		PipelineType:         c.Diffusers.PipelineType,
+		CFGScale:             c.Diffusers.CFGScale,
+		LoraAdapter:          c.LoraAdapter,
+		LoraScale:            c.LoraScale,
+		F16Memory:            *c.F16,
+		LoraBase:             c.LoraBase,
+		IMG2IMG:              c.Diffusers.IMG2IMG,
+		CLIPModel:            c.Diffusers.ClipModel,
+		CLIPSubfolder:        c.Diffusers.ClipSubFolder,
+		CLIPSkip:             int32(c.Diffusers.ClipSkip),
+		ControlNet:           c.Diffusers.ControlNet,
+		ContextSize:          int32(*c.ContextSize),
+		Seed:                 int32(*c.Seed),
+		NBatch:               int32(b),
+		NoMulMatQ:            c.NoMulMatQ,
+		DraftModel:           c.DraftModel,
+		AudioPath:            c.VallE.AudioPath,
+		Quantization:         c.Quantization,
+		GPUMemoryUtilization: c.GPUMemoryUtilization,
+		TrustRemoteCode:      c.TrustRemoteCode,
+		EnforceEager:         c.EnforceEager,
+		SwapSpace:            int32(c.SwapSpace),
+		MaxModelLen:          int32(c.MaxModelLen),
+		MMProj:               c.MMProj,
+		YarnExtFactor:        c.YarnExtFactor,
+		YarnAttnFactor:       c.YarnAttnFactor,
+		YarnBetaFast:         c.YarnBetaFast,
+		YarnBetaSlow:         c.YarnBetaSlow,
+		NGQA:                 c.NGQA,
+		RMSNormEps:           c.RMSNormEps,
+		MLock:                *c.MMlock,
+		RopeFreqBase:         c.RopeFreqBase,
+		RopeScaling:          c.RopeScaling,
+		Type:                 c.ModelType,
+		RopeFreqScale:        c.RopeFreqScale,
+		NUMA:                 c.NUMA,
+		Embeddings:           c.Embeddings,
+		LowVRAM:              *c.LowVRAM,
+		NGPULayers:           int32(*c.NGPULayers),
+		MMap:                 *c.MMap,
+		MainGPU:              c.MainGPU,
+		Threads:              int32(*c.Threads),
+		TensorSplit:          c.TensorSplit,
+		// AutoGPTQ
+		ModelBaseName:    c.AutoGPTQ.ModelBaseName,
+		Device:           c.AutoGPTQ.Device,
+		UseTriton:        c.AutoGPTQ.Triton,
+		UseFastTokenizer: c.AutoGPTQ.UseFastTokenizer,
+		// RWKV
+		Tokenizer: c.Tokenizer,
+	}
+}
+
+func gRPCPredictOpts(c config.BackendConfig, modelPath string) *pb.PredictOptions {
+	promptCachePath := ""
+	if c.PromptCachePath != "" {
+		p := filepath.Join(modelPath, c.PromptCachePath)
+		os.MkdirAll(filepath.Dir(p), 0755)
+		promptCachePath = p
+	}
+
+	return &pb.PredictOptions{
+		Temperature:         float32(*c.Temperature),
+		TopP:                float32(*c.TopP),
+		NDraft:              c.NDraft,
+		TopK:                int32(*c.TopK),
+		Tokens:              int32(*c.Maxtokens),
+		Threads:             int32(*c.Threads),
+		PromptCacheAll:      c.PromptCacheAll,
+		PromptCacheRO:       c.PromptCacheRO,
+		PromptCachePath:     promptCachePath,
+		F16KV:               *c.F16,
+		DebugMode:           *c.Debug,
+		Grammar:             c.Grammar,
+		NegativePromptScale: c.NegativePromptScale,
+		RopeFreqBase:        c.RopeFreqBase,
+		RopeFreqScale:       c.RopeFreqScale,
+		NegativePrompt:      c.NegativePrompt,
+		Mirostat:            int32(*c.LLMConfig.Mirostat),
+		MirostatETA:         float32(*c.LLMConfig.MirostatETA),
+		MirostatTAU:         float32(*c.LLMConfig.MirostatTAU),
+		Debug:               *c.Debug,
+		StopPrompts:         c.StopWords,
+		Repeat:              int32(c.RepeatPenalty),
+		NKeep:               int32(c.Keep),
+		Batch:               int32(c.Batch),
+		IgnoreEOS:           c.IgnoreEOS,
+		Seed:                int32(*c.Seed),
+		FrequencyPenalty:    float32(c.FrequencyPenalty),
+		MLock:               *c.MMlock,
+		MMap:                *c.MMap,
+		MainGPU:             c.MainGPU,
+		TensorSplit:         c.TensorSplit,
+		TailFreeSamplingZ:   float32(c.TFZ),
+		TypicalP:            float32(c.TypicalP),
+	}
+}

core/backend/transcript.go

@@ -0,0 +1,38 @@
+package backend
+
+import (
+	"context"
+	"fmt"
+
+	"github.com/go-skynet/LocalAI/core/config"
+	"github.com/go-skynet/LocalAI/core/schema"
+
+	"github.com/go-skynet/LocalAI/pkg/grpc/proto"
+	model "github.com/go-skynet/LocalAI/pkg/model"
+)
+
+func ModelTranscription(audio, language string, ml *model.ModelLoader, backendConfig config.BackendConfig, appConfig *config.ApplicationConfig) (*schema.Result, error) {
+
+	opts := modelOpts(backendConfig, appConfig, []model.Option{
+		model.WithBackendString(model.WhisperBackend),
+		model.WithModel(backendConfig.Model),
+		model.WithContext(appConfig.Context),
+		model.WithThreads(uint32(*backendConfig.Threads)),
+		model.WithAssetDir(appConfig.AssetsDestination),
+	})
+
+	whisperModel, err := ml.BackendLoader(opts...)
+	if err != nil {
+		return nil, err
+	}
+
+	if whisperModel == nil {
+		return nil, fmt.Errorf("could not load whisper model")
+	}
+
+	return whisperModel.AudioTranscription(context.Background(), &proto.TranscriptRequest{
+		Dst:      audio,
+		Language: language,
+		Threads:  uint32(*backendConfig.Threads),
+	})
+}

core/backend/tts.go

@@ -0,0 +1,89 @@
+package backend
+
+import (
+	"context"
+	"fmt"
+	"os"
+	"path/filepath"
+
+	"github.com/go-skynet/LocalAI/core/config"
+
+	"github.com/go-skynet/LocalAI/pkg/grpc/proto"
+	model "github.com/go-skynet/LocalAI/pkg/model"
+	"github.com/go-skynet/LocalAI/pkg/utils"
+)
+
+func generateUniqueFileName(dir, baseName, ext string) string {
+	counter := 1
+	fileName := baseName + ext
+
+	for {
+		filePath := filepath.Join(dir, fileName)
+		_, err := os.Stat(filePath)
+		if os.IsNotExist(err) {
+			return fileName
+		}
+
+		counter++
+		fileName = fmt.Sprintf("%s_%d%s", baseName, counter, ext)
+	}
+}
+
+func ModelTTS(backend, text, modelFile, voice string, loader *model.ModelLoader, appConfig *config.ApplicationConfig, backendConfig config.BackendConfig) (string, *proto.Result, error) {
+	bb := backend
+	if bb == "" {
+		bb = model.PiperBackend
+	}
+
+	grpcOpts := gRPCModelOpts(backendConfig)
+
+	opts := modelOpts(config.BackendConfig{}, appConfig, []model.Option{
+		model.WithBackendString(bb),
+		model.WithModel(modelFile),
+		model.WithContext(appConfig.Context),
+		model.WithAssetDir(appConfig.AssetsDestination),
+		model.WithLoadGRPCLoadModelOpts(grpcOpts),
+	})
+	ttsModel, err := loader.BackendLoader(opts...)
+	if err != nil {
+		return "", nil, err
+	}
+
+	if ttsModel == nil {
+		return "", nil, fmt.Errorf("could not load piper model")
+	}
+
+	if err := os.MkdirAll(appConfig.AudioDir, 0755); err != nil {
+		return "", nil, fmt.Errorf("failed creating audio directory: %s", err)
+	}
+
+	fileName := generateUniqueFileName(appConfig.AudioDir, "tts", ".wav")
+	filePath := filepath.Join(appConfig.AudioDir, fileName)
+
+	// If the model file is not empty, we pass it joined with the model path
+	modelPath := ""
+	if modelFile != "" {
+		// If the model file is not empty, we pass it joined with the model path
+		// Checking first that it exists and is not outside ModelPath
+		// TODO: we should actually first check if the modelFile is looking like
+		// a FS path
+		mp := filepath.Join(loader.ModelPath, modelFile)
+		if _, err := os.Stat(mp); err == nil {
+			if err := utils.VerifyPath(mp, appConfig.ModelPath); err != nil {
+				return "", nil, err
+			}
+			modelPath = mp
+		} else {
+			modelPath = modelFile
+		}
+	}
+
+	res, err := ttsModel.TTS(context.Background(), &proto.TTSRequest{
+		Text:  text,
+		Model: modelPath,
+		Voice: voice,
+		Dst:   filePath,
+	})
+
+	return filePath, res, err
+}

core/config/application_config.go

@@ -1,4 +1,4 @@
-package options
+package config
 
 import (
 	"context"
@@ -6,27 +6,27 @@ import (
 	"encoding/json"
 	"time"
 
-	"github.com/go-skynet/LocalAI/metrics"
 	"github.com/go-skynet/LocalAI/pkg/gallery"
-	model "github.com/go-skynet/LocalAI/pkg/model"
 	"github.com/rs/zerolog/log"
 )
 
-type Option struct {
+type ApplicationConfig struct {
 	Context                             context.Context
 	ConfigFile                          string
-	Loader                              *model.ModelLoader
+	ModelPath                           string
 	UploadLimitMB, Threads, ContextSize int
 	F16                                 bool
 	Debug, DisableMessage               bool
 	ImageDir                            string
 	AudioDir                            string
+	UploadDir                           string
 	CORS                                bool
 	PreloadJSONModels                   string
 	PreloadModelsFromPath               string
 	CORSAllowOrigins                    string
 	ApiKeys                             []string
-	Metrics                             *metrics.Metrics
+
+	ModelLibraryURL string
 
 	Galleries []gallery.Gallery
 
@@ -49,10 +49,10 @@ type Option struct {
 	WatchDogBusyTimeout, WatchDogIdleTimeout time.Duration
 }
 
-type AppOption func(*Option)
+type AppOption func(*ApplicationConfig)
 
-func NewOptions(o ...AppOption) *Option {
-	opt := &Option{
+func NewApplicationConfig(o ...AppOption) *ApplicationConfig {
+	opt := &ApplicationConfig{
 		Context:        context.Background(),
 		UploadLimitMB:  15,
 		Threads:        1,
@@ -67,57 +67,69 @@ func NewOptions(o ...AppOption) *Option {
 }
 
 func WithModelsURL(urls ...string) AppOption {
-	return func(o *Option) {
+	return func(o *ApplicationConfig) {
 		o.ModelsURL = urls
 	}
 }
 
+func WithModelPath(path string) AppOption {
+	return func(o *ApplicationConfig) {
+		o.ModelPath = path
+	}
+}
+
 func WithCors(b bool) AppOption {
-	return func(o *Option) {
+	return func(o *ApplicationConfig) {
 		o.CORS = b
 	}
 }
 
-var EnableWatchDog = func(o *Option) {
+func WithModelLibraryURL(url string) AppOption {
+	return func(o *ApplicationConfig) {
+		o.ModelLibraryURL = url
+	}
+}
+
+var EnableWatchDog = func(o *ApplicationConfig) {
 	o.WatchDog = true
 }
 
-var EnableWatchDogIdleCheck = func(o *Option) {
+var EnableWatchDogIdleCheck = func(o *ApplicationConfig) {
 	o.WatchDog = true
 	o.WatchDogIdle = true
 }
 
-var EnableWatchDogBusyCheck = func(o *Option) {
+var EnableWatchDogBusyCheck = func(o *ApplicationConfig) {
 	o.WatchDog = true
 	o.WatchDogBusy = true
 }
 
 func SetWatchDogBusyTimeout(t time.Duration) AppOption {
-	return func(o *Option) {
+	return func(o *ApplicationConfig) {
 		o.WatchDogBusyTimeout = t
 	}
 }
 
 func SetWatchDogIdleTimeout(t time.Duration) AppOption {
-	return func(o *Option) {
+	return func(o *ApplicationConfig) {
 		o.WatchDogIdleTimeout = t
 	}
 }
 
-var EnableSingleBackend = func(o *Option) {
+var EnableSingleBackend = func(o *ApplicationConfig) {
 	o.SingleBackend = true
 }
 
-var EnableParallelBackendRequests = func(o *Option) {
+var EnableParallelBackendRequests = func(o *ApplicationConfig) {
 	o.ParallelBackendRequests = true
 }
 
-var EnableGalleriesAutoload = func(o *Option) {
+var EnableGalleriesAutoload = func(o *ApplicationConfig) {
 	o.AutoloadGalleries = true
 }
 
 func WithExternalBackend(name string, uri string) AppOption {
-	return func(o *Option) {
+	return func(o *ApplicationConfig) {
 		if o.ExternalGRPCBackends == nil {
 			o.ExternalGRPCBackends = make(map[string]string)
 		}
@@ -126,27 +138,26 @@ func WithExternalBackend(name string, uri string) AppOption {
 }
 
 func WithCorsAllowOrigins(b string) AppOption {
-	return func(o *Option) {
+	return func(o *ApplicationConfig) {
 		o.CORSAllowOrigins = b
 	}
 }
 
 func WithBackendAssetsOutput(out string) AppOption {
-	return func(o *Option) {
+	return func(o *ApplicationConfig) {
 		o.AssetsDestination = out
 	}
 }
 
 func WithBackendAssets(f embed.FS) AppOption {
-	return func(o *Option) {
+	return func(o *ApplicationConfig) {
 		o.BackendAssets = f
 	}
 }
 
 func WithStringGalleries(galls string) AppOption {
-	return func(o *Option) {
+	return func(o *ApplicationConfig) {
 		if galls == "" {
-			log.Debug().Msgf("no galleries to load")
 			o.Galleries = []gallery.Gallery{}
 			return
 		}
@@ -159,96 +170,96 @@ func WithStringGalleries(galls string) AppOption {
 }
 
 func WithGalleries(galleries []gallery.Gallery) AppOption {
-	return func(o *Option) {
+	return func(o *ApplicationConfig) {
 		o.Galleries = append(o.Galleries, galleries...)
 	}
 }
 
 func WithContext(ctx context.Context) AppOption {
-	return func(o *Option) {
+	return func(o *ApplicationConfig) {
 		o.Context = ctx
 	}
 }
 
 func WithYAMLConfigPreload(configFile string) AppOption {
-	return func(o *Option) {
+	return func(o *ApplicationConfig) {
 		o.PreloadModelsFromPath = configFile
 	}
 }
 
 func WithJSONStringPreload(configFile string) AppOption {
-	return func(o *Option) {
+	return func(o *ApplicationConfig) {
 		o.PreloadJSONModels = configFile
 	}
 }
 func WithConfigFile(configFile string) AppOption {
-	return func(o *Option) {
+	return func(o *ApplicationConfig) {
 		o.ConfigFile = configFile
 	}
 }
 
-func WithModelLoader(loader *model.ModelLoader) AppOption {
-	return func(o *Option) {
-		o.Loader = loader
-	}
-}
-
 func WithUploadLimitMB(limit int) AppOption {
-	return func(o *Option) {
+	return func(o *ApplicationConfig) {
 		o.UploadLimitMB = limit
 	}
 }
 
 func WithThreads(threads int) AppOption {
-	return func(o *Option) {
+	return func(o *ApplicationConfig) {
 		o.Threads = threads
 	}
 }
 
 func WithContextSize(ctxSize int) AppOption {
-	return func(o *Option) {
+	return func(o *ApplicationConfig) {
 		o.ContextSize = ctxSize
 	}
 }
 
 func WithF16(f16 bool) AppOption {
-	return func(o *Option) {
+	return func(o *ApplicationConfig) {
 		o.F16 = f16
 	}
 }
 
 func WithDebug(debug bool) AppOption {
-	return func(o *Option) {
+	return func(o *ApplicationConfig) {
 		o.Debug = debug
 	}
 }
 
 func WithDisableMessage(disableMessage bool) AppOption {
-	return func(o *Option) {
+	return func(o *ApplicationConfig) {
 		o.DisableMessage = disableMessage
 	}
 }
 
 func WithAudioDir(audioDir string) AppOption {
-	return func(o *Option) {
+	return func(o *ApplicationConfig) {
 		o.AudioDir = audioDir
 	}
 }
 
 func WithImageDir(imageDir string) AppOption {
-	return func(o *Option) {
+	return func(o *ApplicationConfig) {
 		o.ImageDir = imageDir
 	}
 }
 
+func WithUploadDir(uploadDir string) AppOption {
+	return func(o *ApplicationConfig) {
+		o.UploadDir = uploadDir
+	}
+}
+
 func WithApiKeys(apiKeys []string) AppOption {
-	return func(o *Option) {
+	return func(o *ApplicationConfig) {
 		o.ApiKeys = apiKeys
 	}
 }
 
-func WithMetrics(meter *metrics.Metrics) AppOption {
-	return func(o *Option) {
-		o.Metrics = meter
-	}
-}
+// func WithMetrics(meter *metrics.Metrics) AppOption {
+// 	return func(o *StartupOptions) {
+// 		o.Metrics = meter
+// 	}
+// }

core/config/backend_config.go

@@ -0,0 +1,565 @@
+package config
+
+import (
+	"errors"
+	"fmt"
+	"io/fs"
+	"math/rand"
+	"os"
+	"path/filepath"
+	"strings"
+	"sync"
+
+	"github.com/go-skynet/LocalAI/core/schema"
+	"github.com/go-skynet/LocalAI/pkg/downloader"
+	"github.com/go-skynet/LocalAI/pkg/utils"
+	"github.com/rs/zerolog/log"
+	"gopkg.in/yaml.v3"
+
+	"github.com/charmbracelet/glamour"
+)
+
+type BackendConfig struct {
+	schema.PredictionOptions `yaml:"parameters"`
+	Name                     string `yaml:"name"`
+
+	F16            *bool             `yaml:"f16"`
+	Threads        *int              `yaml:"threads"`
+	Debug          *bool             `yaml:"debug"`
+	Roles          map[string]string `yaml:"roles"`
+	Embeddings     bool              `yaml:"embeddings"`
+	Backend        string            `yaml:"backend"`
+	TemplateConfig TemplateConfig    `yaml:"template"`
+
+	PromptStrings, InputStrings                []string `yaml:"-"`
+	InputToken                                 [][]int  `yaml:"-"`
+	functionCallString, functionCallNameString string   `yaml:"-"`
+
+	FunctionsConfig Functions `yaml:"function"`
+
+	FeatureFlag FeatureFlag `yaml:"feature_flags"` // Feature Flag registry. We move fast, and features may break on a per model/backend basis. Registry for (usually temporary) flags that indicate aborting something early.
+	// LLM configs (GPT4ALL, Llama.cpp, ...)
+	LLMConfig `yaml:",inline"`
+
+	// AutoGPTQ specifics
+	AutoGPTQ AutoGPTQ `yaml:"autogptq"`
+
+	// Diffusers
+	Diffusers Diffusers `yaml:"diffusers"`
+	Step      int       `yaml:"step"`
+
+	// GRPC Options
+	GRPC GRPC `yaml:"grpc"`
+
+	// Vall-e-x
+	VallE VallE `yaml:"vall-e"`
+
+	// CUDA
+	// Explicitly enable CUDA or not (some backends might need it)
+	CUDA bool `yaml:"cuda"`
+
+	DownloadFiles []File `yaml:"download_files"`
+
+	Description string `yaml:"description"`
+	Usage       string `yaml:"usage"`
+}
+
+type File struct {
+	Filename string `yaml:"filename" json:"filename"`
+	SHA256   string `yaml:"sha256" json:"sha256"`
+	URI      string `yaml:"uri" json:"uri"`
+}
+
+type VallE struct {
+	AudioPath string `yaml:"audio_path"`
+}
+
+type FeatureFlag map[string]*bool
+
+func (ff FeatureFlag) Enabled(s string) bool {
+	v, exist := ff[s]
+	return exist && v != nil && *v
+}
+
+type GRPC struct {
+	Attempts          int `yaml:"attempts"`
+	AttemptsSleepTime int `yaml:"attempts_sleep_time"`
+}
+
+type Diffusers struct {
+	CUDA             bool    `yaml:"cuda"`
+	PipelineType     string  `yaml:"pipeline_type"`
+	SchedulerType    string  `yaml:"scheduler_type"`
+	EnableParameters string  `yaml:"enable_parameters"` // A list of comma separated parameters to specify
+	CFGScale         float32 `yaml:"cfg_scale"`         // Classifier-Free Guidance Scale
+	IMG2IMG          bool    `yaml:"img2img"`           // Image to Image Diffuser
+	ClipSkip         int     `yaml:"clip_skip"`         // Skip every N frames
+	ClipModel        string  `yaml:"clip_model"`        // Clip model to use
+	ClipSubFolder    string  `yaml:"clip_subfolder"`    // Subfolder to use for clip model
+	ControlNet       string  `yaml:"control_net"`
+}
+
+type LLMConfig struct {
+	SystemPrompt    string   `yaml:"system_prompt"`
+	TensorSplit     string   `yaml:"tensor_split"`
+	MainGPU         string   `yaml:"main_gpu"`
+	RMSNormEps      float32  `yaml:"rms_norm_eps"`
+	NGQA            int32    `yaml:"ngqa"`
+	PromptCachePath string   `yaml:"prompt_cache_path"`
+	PromptCacheAll  bool     `yaml:"prompt_cache_all"`
+	PromptCacheRO   bool     `yaml:"prompt_cache_ro"`
+	MirostatETA     *float64 `yaml:"mirostat_eta"`
+	MirostatTAU     *float64 `yaml:"mirostat_tau"`
+	Mirostat        *int     `yaml:"mirostat"`
+	NGPULayers      *int     `yaml:"gpu_layers"`
+	MMap            *bool    `yaml:"mmap"`
+	MMlock          *bool    `yaml:"mmlock"`
+	LowVRAM         *bool    `yaml:"low_vram"`
+	Grammar         string   `yaml:"grammar"`
+	StopWords       []string `yaml:"stopwords"`
+	Cutstrings      []string `yaml:"cutstrings"`
+	TrimSpace       []string `yaml:"trimspace"`
+	TrimSuffix      []string `yaml:"trimsuffix"`
+
+	ContextSize          *int    `yaml:"context_size"`
+	NUMA                 bool    `yaml:"numa"`
+	LoraAdapter          string  `yaml:"lora_adapter"`
+	LoraBase             string  `yaml:"lora_base"`
+	LoraScale            float32 `yaml:"lora_scale"`
+	NoMulMatQ            bool    `yaml:"no_mulmatq"`
+	DraftModel           string  `yaml:"draft_model"`
+	NDraft               int32   `yaml:"n_draft"`
+	Quantization         string  `yaml:"quantization"`
+	GPUMemoryUtilization float32 `yaml:"gpu_memory_utilization"` // vLLM
+	TrustRemoteCode      bool    `yaml:"trust_remote_code"`      // vLLM
+	EnforceEager         bool    `yaml:"enforce_eager"`          // vLLM
+	SwapSpace            int     `yaml:"swap_space"`             // vLLM
+	MaxModelLen          int     `yaml:"max_model_len"`          // vLLM
+	MMProj               string  `yaml:"mmproj"`
+
+	RopeScaling string `yaml:"rope_scaling"`
+	ModelType   string `yaml:"type"`
+
+	YarnExtFactor  float32 `yaml:"yarn_ext_factor"`
+	YarnAttnFactor float32 `yaml:"yarn_attn_factor"`
+	YarnBetaFast   float32 `yaml:"yarn_beta_fast"`
+	YarnBetaSlow   float32 `yaml:"yarn_beta_slow"`
+}
+
+type AutoGPTQ struct {
+	ModelBaseName    string `yaml:"model_base_name"`
+	Device           string `yaml:"device"`
+	Triton           bool   `yaml:"triton"`
+	UseFastTokenizer bool   `yaml:"use_fast_tokenizer"`
+}
+
+type Functions struct {
+	DisableNoAction         bool   `yaml:"disable_no_action"`
+	NoActionFunctionName    string `yaml:"no_action_function_name"`
+	NoActionDescriptionName string `yaml:"no_action_description_name"`
+	ParallelCalls           bool   `yaml:"parallel_calls"`
+}
+
+type TemplateConfig struct {
+	Chat        string `yaml:"chat"`
+	ChatMessage string `yaml:"chat_message"`
+	Completion  string `yaml:"completion"`
+	Edit        string `yaml:"edit"`
+	Functions   string `yaml:"function"`
+}
+
+func (c *BackendConfig) SetFunctionCallString(s string) {
+	c.functionCallString = s
+}
+
+func (c *BackendConfig) SetFunctionCallNameString(s string) {
+	c.functionCallNameString = s
+}
+
+func (c *BackendConfig) ShouldUseFunctions() bool {
+	return ((c.functionCallString != "none" || c.functionCallString == "") || c.ShouldCallSpecificFunction())
+}
+
+func (c *BackendConfig) ShouldCallSpecificFunction() bool {
+	return len(c.functionCallNameString) > 0
+}
+
+func (c *BackendConfig) FunctionToCall() string {
+	return c.functionCallNameString
+}
+
+func (cfg *BackendConfig) SetDefaults(debug bool, threads, ctx int, f16 bool) {
+	defaultTopP := 0.7
+	defaultTopK := 80
+	defaultTemp := 0.9
+	defaultMaxTokens := 2048
+	defaultMirostat := 2
+	defaultMirostatTAU := 5.0
+	defaultMirostatETA := 0.1
+
+	// Try to offload all GPU layers (if GPU is found)
+	defaultNGPULayers := 99999999
+
+	trueV := true
+	falseV := false
+
+	if cfg.Seed == nil {
+		//  random number generator seed
+		defaultSeed := int(rand.Int31())
+		cfg.Seed = &defaultSeed
+	}
+
+	if cfg.TopK == nil {
+		cfg.TopK = &defaultTopK
+	}
+
+	if cfg.MMap == nil {
+		// MMap is enabled by default
+		cfg.MMap = &trueV
+	}
+
+	if cfg.MMlock == nil {
+		// MMlock is disabled by default
+		cfg.MMlock = &falseV
+	}
+
+	if cfg.TopP == nil {
+		cfg.TopP = &defaultTopP
+	}
+	if cfg.Temperature == nil {
+		cfg.Temperature = &defaultTemp
+	}
+
+	if cfg.Maxtokens == nil {
+		cfg.Maxtokens = &defaultMaxTokens
+	}
+
+	if cfg.Mirostat == nil {
+		cfg.Mirostat = &defaultMirostat
+	}
+
+	if cfg.MirostatETA == nil {
+		cfg.MirostatETA = &defaultMirostatETA
+	}
+
+	if cfg.MirostatTAU == nil {
+		cfg.MirostatTAU = &defaultMirostatTAU
+	}
+	if cfg.NGPULayers == nil {
+		cfg.NGPULayers = &defaultNGPULayers
+	}
+
+	if cfg.LowVRAM == nil {
+		cfg.LowVRAM = &falseV
+	}
+
+	// Value passed by the top level are treated as default (no implicit defaults)
+	// defaults are set by the user
+	if ctx == 0 {
+		ctx = 1024
+	}
+
+	if cfg.ContextSize == nil {
+		cfg.ContextSize = &ctx
+	}
+
+	if threads == 0 {
+		// Threads can't be 0
+		threads = 4
+	}
+
+	if cfg.Threads == nil {
+		cfg.Threads = &threads
+	}
+
+	if cfg.F16 == nil {
+		cfg.F16 = &f16
+	}
+
+	if debug {
+		cfg.Debug = &debug
+	}
+}
+
+////// Config Loader ////////
+
+type BackendConfigLoader struct {
+	configs map[string]BackendConfig
+	sync.Mutex
+}
+
+type LoadOptions struct {
+	debug            bool
+	threads, ctxSize int
+	f16              bool
+}
+
+func LoadOptionDebug(debug bool) ConfigLoaderOption {
+	return func(o *LoadOptions) {
+		o.debug = debug
+	}
+}
+
+func LoadOptionThreads(threads int) ConfigLoaderOption {
+	return func(o *LoadOptions) {
+		o.threads = threads
+	}
+}
+
+func LoadOptionContextSize(ctxSize int) ConfigLoaderOption {
+	return func(o *LoadOptions) {
+		o.ctxSize = ctxSize
+	}
+}
+
+func LoadOptionF16(f16 bool) ConfigLoaderOption {
+	return func(o *LoadOptions) {
+		o.f16 = f16
+	}
+}
+
+type ConfigLoaderOption func(*LoadOptions)
+
+func (lo *LoadOptions) Apply(options ...ConfigLoaderOption) {
+	for _, l := range options {
+		l(lo)
+	}
+}
+
+// Load a config file for a model
+func (cl *BackendConfigLoader) LoadBackendConfigFileByName(modelName, modelPath string, opts ...ConfigLoaderOption) (*BackendConfig, error) {
+
+	lo := &LoadOptions{}
+	lo.Apply(opts...)
+
+	// Load a config file if present after the model name
+	cfg := &BackendConfig{
+		PredictionOptions: schema.PredictionOptions{
+			Model: modelName,
+		},
+	}
+
+	cfgExisting, exists := cl.GetBackendConfig(modelName)
+	if exists {
+		cfg = &cfgExisting
+	} else {
+		// Try loading a model config file
+		modelConfig := filepath.Join(modelPath, modelName+".yaml")
+		if _, err := os.Stat(modelConfig); err == nil {
+			if err := cl.LoadBackendConfig(modelConfig); err != nil {
+				return nil, fmt.Errorf("failed loading model config (%s) %s", modelConfig, err.Error())
+			}
+			cfgExisting, exists = cl.GetBackendConfig(modelName)
+			if exists {
+				cfg = &cfgExisting
+			}
+		}
+	}
+
+	cfg.SetDefaults(lo.debug, lo.threads, lo.ctxSize, lo.f16)
+
+	return cfg, nil
+}
+
+func NewBackendConfigLoader() *BackendConfigLoader {
+	return &BackendConfigLoader{
+		configs: make(map[string]BackendConfig),
+	}
+}
+func ReadBackendConfigFile(file string, opts ...ConfigLoaderOption) ([]*BackendConfig, error) {
+	lo := &LoadOptions{}
+	lo.Apply(opts...)
+
+	c := &[]*BackendConfig{}
+	f, err := os.ReadFile(file)
+	if err != nil {
+		return nil, fmt.Errorf("cannot read config file: %w", err)
+	}
+	if err := yaml.Unmarshal(f, c); err != nil {
+		return nil, fmt.Errorf("cannot unmarshal config file: %w", err)
+	}
+
+	for _, cc := range *c {
+		cc.SetDefaults(lo.debug, lo.threads, lo.ctxSize, lo.f16)
+	}
+
+	return *c, nil
+}
+
+func ReadBackendConfig(file string, opts ...ConfigLoaderOption) (*BackendConfig, error) {
+	lo := &LoadOptions{}
+	lo.Apply(opts...)
+
+	c := &BackendConfig{}
+	f, err := os.ReadFile(file)
+	if err != nil {
+		return nil, fmt.Errorf("cannot read config file: %w", err)
+	}
+	if err := yaml.Unmarshal(f, c); err != nil {
+		return nil, fmt.Errorf("cannot unmarshal config file: %w", err)
+	}
+
+	c.SetDefaults(lo.debug, lo.threads, lo.ctxSize, lo.f16)
+	return c, nil
+}
+
+func (cm *BackendConfigLoader) LoadBackendConfigFile(file string, opts ...ConfigLoaderOption) error {
+	cm.Lock()
+	defer cm.Unlock()
+	c, err := ReadBackendConfigFile(file, opts...)
+	if err != nil {
+		return fmt.Errorf("cannot load config file: %w", err)
+	}
+
+	for _, cc := range c {
+		cm.configs[cc.Name] = *cc
+	}
+	return nil
+}
+
+func (cl *BackendConfigLoader) LoadBackendConfig(file string, opts ...ConfigLoaderOption) error {
+	cl.Lock()
+	defer cl.Unlock()
+	c, err := ReadBackendConfig(file, opts...)
+	if err != nil {
+		return fmt.Errorf("cannot read config file: %w", err)
+	}
+
+	cl.configs[c.Name] = *c
+	return nil
+}
+
+func (cl *BackendConfigLoader) GetBackendConfig(m string) (BackendConfig, bool) {
+	cl.Lock()
+	defer cl.Unlock()
+	v, exists := cl.configs[m]
+	return v, exists
+}
+
+func (cl *BackendConfigLoader) GetAllBackendConfigs() []BackendConfig {
+	cl.Lock()
+	defer cl.Unlock()
+	var res []BackendConfig
+	for _, v := range cl.configs {
+		res = append(res, v)
+	}
+	return res
+}
+
+func (cl *BackendConfigLoader) ListBackendConfigs() []string {
+	cl.Lock()
+	defer cl.Unlock()
+	var res []string
+	for k := range cl.configs {
+		res = append(res, k)
+	}
+	return res
+}
+
+// Preload prepare models if they are not local but url or huggingface repositories
+func (cl *BackendConfigLoader) Preload(modelPath string) error {
+	cl.Lock()
+	defer cl.Unlock()
+
+	status := func(fileName, current, total string, percent float64) {
+		utils.DisplayDownloadFunction(fileName, current, total, percent)
+	}
+
+	log.Info().Msgf("Preloading models from %s", modelPath)
+
+	renderMode := "dark"
+	if os.Getenv("COLOR") != "" {
+		renderMode = os.Getenv("COLOR")
+	}
+
+	glamText := func(t string) {
+		out, err := glamour.Render(t, renderMode)
+		if err == nil && os.Getenv("NO_COLOR") == "" {
+			fmt.Println(out)
+		} else {
+			fmt.Println(t)
+		}
+	}
+
+	for i, config := range cl.configs {
+
+		// Download files and verify their SHA
+		for _, file := range config.DownloadFiles {
+			log.Debug().Msgf("Checking %q exists and matches SHA", file.Filename)
+
+			if err := utils.VerifyPath(file.Filename, modelPath); err != nil {
+				return err
+			}
+			// Create file path
+			filePath := filepath.Join(modelPath, file.Filename)
+
+			if err := downloader.DownloadFile(file.URI, filePath, file.SHA256, status); err != nil {
+				return err
+			}
+		}
+
+		modelURL := config.PredictionOptions.Model
+		modelURL = downloader.ConvertURL(modelURL)
+
+		if downloader.LooksLikeURL(modelURL) {
+			// md5 of model name
+			md5Name := utils.MD5(modelURL)
+
+			// check if file exists
+			if _, err := os.Stat(filepath.Join(modelPath, md5Name)); errors.Is(err, os.ErrNotExist) {
+				err := downloader.DownloadFile(modelURL, filepath.Join(modelPath, md5Name), "", status)
+				if err != nil {
+					return err
+				}
+			}
+
+			cc := cl.configs[i]
+			c := &cc
+			c.PredictionOptions.Model = md5Name
+			cl.configs[i] = *c
+		}
+		if cl.configs[i].Name != "" {
+			glamText(fmt.Sprintf("**Model name**: _%s_", cl.configs[i].Name))
+		}
+		if cl.configs[i].Description != "" {
+			//glamText("**Description**")
+			glamText(cl.configs[i].Description)
+		}
+		if cl.configs[i].Usage != "" {
+			//glamText("**Usage**")
+			glamText(cl.configs[i].Usage)
+		}
+	}
+	return nil
+}
+
+// LoadBackendConfigsFromPath reads all the configurations of the models from a path
+// (non-recursive)
+func (cm *BackendConfigLoader) LoadBackendConfigsFromPath(path string, opts ...ConfigLoaderOption) error {
+	cm.Lock()
+	defer cm.Unlock()
+	entries, err := os.ReadDir(path)
+	if err != nil {
+		return err
+	}
+	files := make([]fs.FileInfo, 0, len(entries))
+	for _, entry := range entries {
+		info, err := entry.Info()
+		if err != nil {
+			return err
+		}
+		files = append(files, info)
+	}
+	for _, file := range files {
+		// Skip templates, YAML and .keep files
+		if !strings.Contains(file.Name(), ".yaml") && !strings.Contains(file.Name(), ".yml") {
+			continue
+		}
+		c, err := ReadBackendConfig(filepath.Join(path, file.Name()), opts...)
+		if err == nil {
+			cm.configs[c.Name] = *c
+		}
+	}
+
+	return nil
+}

core/config/config_test.go

@@ -1,11 +1,10 @@
-package api_config_test
+package config_test
 
 import (
 	"os"
 
-	. "github.com/go-skynet/LocalAI/api/config"
-	"github.com/go-skynet/LocalAI/api/options"
-	"github.com/go-skynet/LocalAI/pkg/model"
+	. "github.com/go-skynet/LocalAI/core/config"
+
 	. "github.com/onsi/ginkgo/v2"
 	. "github.com/onsi/gomega"
 )
@@ -19,7 +18,7 @@ var _ = Describe("Test cases for config related functions", func() {
 	Context("Test Read configuration functions", func() {
 		configFile = os.Getenv("CONFIG_FILE")
 		It("Test ReadConfigFile", func() {
-			config, err := ReadConfigFile(configFile)
+			config, err := ReadBackendConfigFile(configFile)
 			Expect(err).To(BeNil())
 			Expect(config).ToNot(BeNil())
 			// two configs in config.yaml
@@ -28,29 +27,26 @@ var _ = Describe("Test cases for config related functions", func() {
 		})
 
 		It("Test LoadConfigs", func() {
-			cm := NewConfigLoader()
-			opts := options.NewOptions()
-			modelLoader := model.NewModelLoader(os.Getenv("MODELS_PATH"))
-			options.WithModelLoader(modelLoader)(opts)
-
-			err := cm.LoadConfigs(opts.Loader.ModelPath)
+			cm := NewBackendConfigLoader()
+			opts := NewApplicationConfig()
+			err := cm.LoadBackendConfigsFromPath(opts.ModelPath)
 			Expect(err).To(BeNil())
-			Expect(cm.ListConfigs()).ToNot(BeNil())
+			Expect(cm.ListBackendConfigs()).ToNot(BeNil())
 
 			// config should includes gpt4all models's api.config
-			Expect(cm.ListConfigs()).To(ContainElements("gpt4all"))
+			Expect(cm.ListBackendConfigs()).To(ContainElements("gpt4all"))
 
 			// config should includes gpt2 models's api.config
-			Expect(cm.ListConfigs()).To(ContainElements("gpt4all-2"))
+			Expect(cm.ListBackendConfigs()).To(ContainElements("gpt4all-2"))
 
 			// config should includes text-embedding-ada-002 models's api.config
-			Expect(cm.ListConfigs()).To(ContainElements("text-embedding-ada-002"))
+			Expect(cm.ListBackendConfigs()).To(ContainElements("text-embedding-ada-002"))
 
 			// config should includes rwkv_test models's api.config
-			Expect(cm.ListConfigs()).To(ContainElements("rwkv_test"))
+			Expect(cm.ListBackendConfigs()).To(ContainElements("rwkv_test"))
 
 			// config should includes whisper-1 models's api.config
-			Expect(cm.ListConfigs()).To(ContainElements("whisper-1"))
+			Expect(cm.ListBackendConfigs()).To(ContainElements("whisper-1"))
 		})
 	})
 })

core/http/api.go

@@ -0,0 +1,242 @@
+package http
+
+import (
+	"encoding/json"
+	"errors"
+	"os"
+	"strings"
+
+	"github.com/go-skynet/LocalAI/core/http/endpoints/elevenlabs"
+	"github.com/go-skynet/LocalAI/core/http/endpoints/localai"
+	"github.com/go-skynet/LocalAI/core/http/endpoints/openai"
+
+	"github.com/go-skynet/LocalAI/core/config"
+	"github.com/go-skynet/LocalAI/core/schema"
+	"github.com/go-skynet/LocalAI/core/services"
+	"github.com/go-skynet/LocalAI/internal"
+	"github.com/go-skynet/LocalAI/pkg/model"
+
+	"github.com/gofiber/fiber/v2"
+	"github.com/gofiber/fiber/v2/middleware/cors"
+	"github.com/gofiber/fiber/v2/middleware/logger"
+	"github.com/gofiber/fiber/v2/middleware/recover"
+)
+
+func readAuthHeader(c *fiber.Ctx) string {
+	authHeader := c.Get("Authorization")
+
+	// elevenlabs
+	xApiKey := c.Get("xi-api-key")
+	if xApiKey != "" {
+		authHeader = "Bearer " + xApiKey
+	}
+
+	// anthropic
+	xApiKey = c.Get("x-api-key")
+	if xApiKey != "" {
+		authHeader = "Bearer " + xApiKey
+	}
+
+	return authHeader
+}
+
+func App(cl *config.BackendConfigLoader, ml *model.ModelLoader, appConfig *config.ApplicationConfig) (*fiber.App, error) {
+	// Return errors as JSON responses
+	app := fiber.New(fiber.Config{
+		BodyLimit:             appConfig.UploadLimitMB * 1024 * 1024, // this is the default limit of 4MB
+		DisableStartupMessage: appConfig.DisableMessage,
+		// Override default error handler
+		ErrorHandler: func(ctx *fiber.Ctx, err error) error {
+			// Status code defaults to 500
+			code := fiber.StatusInternalServerError
+
+			// Retrieve the custom status code if it's a *fiber.Error
+			var e *fiber.Error
+			if errors.As(err, &e) {
+				code = e.Code
+			}
+
+			// Send custom error page
+			return ctx.Status(code).JSON(
+				schema.ErrorResponse{
+					Error: &schema.APIError{Message: err.Error(), Code: code},
+				},
+			)
+		},
+	})
+
+	if appConfig.Debug {
+		app.Use(logger.New(logger.Config{
+			Format: "[${ip}]:${port} ${status} - ${method} ${path}\n",
+		}))
+	}
+
+	// Default middleware config
+
+	if !appConfig.Debug {
+		app.Use(recover.New())
+	}
+
+	metricsService, err := services.NewLocalAIMetricsService()
+	if err != nil {
+		return nil, err
+	}
+
+	if metricsService != nil {
+		app.Use(localai.LocalAIMetricsAPIMiddleware(metricsService))
+		app.Hooks().OnShutdown(func() error {
+			return metricsService.Shutdown()
+		})
+	}
+
+	// Auth middleware checking if API key is valid. If no API key is set, no auth is required.
+	auth := func(c *fiber.Ctx) error {
+		if len(appConfig.ApiKeys) == 0 {
+			return c.Next()
+		}
+
+		// Check for api_keys.json file
+		fileContent, err := os.ReadFile("api_keys.json")
+		if err == nil {
+			// Parse JSON content from the file
+			var fileKeys []string
+			err := json.Unmarshal(fileContent, &fileKeys)
+			if err != nil {
+				return c.Status(fiber.StatusInternalServerError).JSON(fiber.Map{"message": "Error parsing api_keys.json"})
+			}
+
+			// Add file keys to options.ApiKeys
+			appConfig.ApiKeys = append(appConfig.ApiKeys, fileKeys...)
+		}
+
+		if len(appConfig.ApiKeys) == 0 {
+			return c.Next()
+		}
+
+		authHeader := readAuthHeader(c)
+		if authHeader == "" {
+			return c.Status(fiber.StatusUnauthorized).JSON(fiber.Map{"message": "Authorization header missing"})
+		}
+
+		// If it's a bearer token
+		authHeaderParts := strings.Split(authHeader, " ")
+		if len(authHeaderParts) != 2 || authHeaderParts[0] != "Bearer" {
+			return c.Status(fiber.StatusUnauthorized).JSON(fiber.Map{"message": "Invalid Authorization header format"})
+		}
+
+		apiKey := authHeaderParts[1]
+		for _, key := range appConfig.ApiKeys {
+			if apiKey == key {
+				return c.Next()
+			}
+		}
+
+		return c.Status(fiber.StatusUnauthorized).JSON(fiber.Map{"message": "Invalid API key"})
+	}
+
+	if appConfig.CORS {
+		var c func(ctx *fiber.Ctx) error
+		if appConfig.CORSAllowOrigins == "" {
+			c = cors.New()
+		} else {
+			c = cors.New(cors.Config{AllowOrigins: appConfig.CORSAllowOrigins})
+		}
+
+		app.Use(c)
+	}
+
+	// LocalAI API endpoints
+	galleryService := services.NewGalleryService(appConfig.ModelPath)
+	galleryService.Start(appConfig.Context, cl)
+
+	app.Get("/version", auth, func(c *fiber.Ctx) error {
+		return c.JSON(struct {
+			Version string `json:"version"`
+		}{Version: internal.PrintableVersion()})
+	})
+
+	// Load upload json
+	openai.LoadUploadConfig(appConfig.UploadDir)
+
+	modelGalleryEndpointService := localai.CreateModelGalleryEndpointService(appConfig.Galleries, appConfig.ModelPath, galleryService)
+	app.Post("/models/apply", auth, modelGalleryEndpointService.ApplyModelGalleryEndpoint())
+	app.Get("/models/available", auth, modelGalleryEndpointService.ListModelFromGalleryEndpoint())
+	app.Get("/models/galleries", auth, modelGalleryEndpointService.ListModelGalleriesEndpoint())
+	app.Post("/models/galleries", auth, modelGalleryEndpointService.AddModelGalleryEndpoint())
+	app.Delete("/models/galleries", auth, modelGalleryEndpointService.RemoveModelGalleryEndpoint())
+	app.Get("/models/jobs/:uuid", auth, modelGalleryEndpointService.GetOpStatusEndpoint())
+	app.Get("/models/jobs", auth, modelGalleryEndpointService.GetAllStatusEndpoint())
+
+	app.Post("/tts", auth, localai.TTSEndpoint(cl, ml, appConfig))
+
+	// Elevenlabs
+	app.Post("/v1/text-to-speech/:voice-id", auth, elevenlabs.TTSEndpoint(cl, ml, appConfig))
+
+	// openAI compatible API endpoint
+
+	// chat
+	app.Post("/v1/chat/completions", auth, openai.ChatEndpoint(cl, ml, appConfig))
+	app.Post("/chat/completions", auth, openai.ChatEndpoint(cl, ml, appConfig))
+
+	// edit
+	app.Post("/v1/edits", auth, openai.EditEndpoint(cl, ml, appConfig))
+	app.Post("/edits", auth, openai.EditEndpoint(cl, ml, appConfig))
+
+	// files
+	app.Post("/v1/files", auth, openai.UploadFilesEndpoint(cl, appConfig))
+	app.Post("/files", auth, openai.UploadFilesEndpoint(cl, appConfig))
+	app.Get("/v1/files", auth, openai.ListFilesEndpoint(cl, appConfig))
+	app.Get("/files", auth, openai.ListFilesEndpoint(cl, appConfig))
+	app.Get("/v1/files/:file_id", auth, openai.GetFilesEndpoint(cl, appConfig))
+	app.Get("/files/:file_id", auth, openai.GetFilesEndpoint(cl, appConfig))
+	app.Delete("/v1/files/:file_id", auth, openai.DeleteFilesEndpoint(cl, appConfig))
+	app.Delete("/files/:file_id", auth, openai.DeleteFilesEndpoint(cl, appConfig))
+	app.Get("/v1/files/:file_id/content", auth, openai.GetFilesContentsEndpoint(cl, appConfig))
+	app.Get("/files/:file_id/content", auth, openai.GetFilesContentsEndpoint(cl, appConfig))
+
+	// completion
+	app.Post("/v1/completions", auth, openai.CompletionEndpoint(cl, ml, appConfig))
+	app.Post("/completions", auth, openai.CompletionEndpoint(cl, ml, appConfig))
+	app.Post("/v1/engines/:model/completions", auth, openai.CompletionEndpoint(cl, ml, appConfig))
+
+	// embeddings
+	app.Post("/v1/embeddings", auth, openai.EmbeddingsEndpoint(cl, ml, appConfig))
+	app.Post("/embeddings", auth, openai.EmbeddingsEndpoint(cl, ml, appConfig))
+	app.Post("/v1/engines/:model/embeddings", auth, openai.EmbeddingsEndpoint(cl, ml, appConfig))
+
+	// audio
+	app.Post("/v1/audio/transcriptions", auth, openai.TranscriptEndpoint(cl, ml, appConfig))
+	app.Post("/v1/audio/speech", auth, localai.TTSEndpoint(cl, ml, appConfig))
+
+	// images
+	app.Post("/v1/images/generations", auth, openai.ImageEndpoint(cl, ml, appConfig))
+
+	if appConfig.ImageDir != "" {
+		app.Static("/generated-images", appConfig.ImageDir)
+	}
+
+	if appConfig.AudioDir != "" {
+		app.Static("/generated-audio", appConfig.AudioDir)
+	}
+
+	ok := func(c *fiber.Ctx) error {
+		return c.SendStatus(200)
+	}
+
+	// Kubernetes health checks
+	app.Get("/healthz", ok)
+	app.Get("/readyz", ok)
+
+	// Experimental Backend Statistics Module
+	backendMonitor := services.NewBackendMonitor(cl, ml, appConfig) // Split out for now
+	app.Get("/backend/monitor", localai.BackendMonitorEndpoint(backendMonitor))
+	app.Post("/backend/shutdown", localai.BackendShutdownEndpoint(backendMonitor))
+
+	// models
+	app.Get("/v1/models", auth, openai.ListModelsEndpoint(cl, ml))
+	app.Get("/models", auth, openai.ListModelsEndpoint(cl, ml))
+
+	app.Get("/metrics", localai.LocalAIMetricsEndpoint())
+
+	return app, nil
+}

core/http/api_test.go

@@ -1,4 +1,4 @@
-package api_test
+package http_test
 
 import (
 	"bytes"
@@ -13,9 +13,10 @@ import (
 	"path/filepath"
 	"runtime"
 
-	. "github.com/go-skynet/LocalAI/api"
-	"github.com/go-skynet/LocalAI/api/options"
-	"github.com/go-skynet/LocalAI/metrics"
+	"github.com/go-skynet/LocalAI/core/config"
+	. "github.com/go-skynet/LocalAI/core/http"
+	"github.com/go-skynet/LocalAI/core/startup"
+
 	"github.com/go-skynet/LocalAI/pkg/downloader"
 	"github.com/go-skynet/LocalAI/pkg/gallery"
 	"github.com/go-skynet/LocalAI/pkg/model"
@@ -29,6 +30,15 @@ import (
 	"github.com/sashabaranov/go-openai/jsonschema"
 )
 
+const testPrompt = `### System:
+You are an AI assistant that follows instruction extremely well. Help as much as you can.
+
+### User:
+
+Can you help rephrasing sentences?
+
+### Response:`
+
 type modelApplyRequest struct {
 	ID        string                 `json:"id"`
 	URL       string                 `json:"url"`
@@ -118,25 +128,33 @@ var backendAssets embed.FS
 var _ = Describe("API test", func() {
 
 	var app *fiber.App
-	var modelLoader *model.ModelLoader
 	var client *openai.Client
 	var client2 *openaigo.Client
 	var c context.Context
 	var cancel context.CancelFunc
 	var tmpdir string
+	var modelDir string
+	var bcl *config.BackendConfigLoader
+	var ml *model.ModelLoader
+	var applicationConfig *config.ApplicationConfig
 
-	commonOpts := []options.AppOption{
-		options.WithDebug(true),
-		options.WithDisableMessage(true),
+	commonOpts := []config.AppOption{
+		config.WithDebug(true),
+		config.WithDisableMessage(true),
 	}
 
 	Context("API with ephemeral models", func() {
-		BeforeEach(func() {
+
+		BeforeEach(func(sc SpecContext) {
 			var err error
 			tmpdir, err = os.MkdirTemp("", "")
 			Expect(err).ToNot(HaveOccurred())
 
-			modelLoader = model.NewModelLoader(tmpdir)
+			modelDir = filepath.Join(tmpdir, "models")
+			backendAssetsDir := filepath.Join(tmpdir, "backend-assets")
+			err = os.Mkdir(backendAssetsDir, 0755)
+			Expect(err).ToNot(HaveOccurred())
+
 			c, cancel = context.WithCancel(context.Background())
 
 			g := []gallery.GalleryModel{
@@ -163,16 +181,18 @@ var _ = Describe("API test", func() {
 				},
 			}
 
-			metricsService, err := metrics.SetupMetrics()
+			bcl, ml, applicationConfig, err = startup.Startup(
+				append(commonOpts,
+					config.WithContext(c),
+					config.WithGalleries(galleries),
+					config.WithModelPath(modelDir),
+					config.WithBackendAssets(backendAssets),
+					config.WithBackendAssetsOutput(backendAssetsDir))...)
 			Expect(err).ToNot(HaveOccurred())
 
-			app, err = App(
-				append(commonOpts,
-					options.WithMetrics(metricsService),
-					options.WithContext(c),
-					options.WithGalleries(galleries),
-					options.WithModelLoader(modelLoader), options.WithBackendAssets(backendAssets), options.WithBackendAssetsOutput(tmpdir))...)
+			app, err = App(bcl, ml, applicationConfig)
 			Expect(err).ToNot(HaveOccurred())
+
 			go app.Listen("127.0.0.1:9090")
 
 			defaultConfig := openai.DefaultConfig("")
@@ -189,15 +209,21 @@ var _ = Describe("API test", func() {
 			}, "2m").ShouldNot(HaveOccurred())
 		})
 
-		AfterEach(func() {
+		AfterEach(func(sc SpecContext) {
 			cancel()
-			app.Shutdown()
-			os.RemoveAll(tmpdir)
+			if app != nil {
+				err := app.Shutdown()
+				Expect(err).ToNot(HaveOccurred())
+			}
+			err := os.RemoveAll(tmpdir)
+			Expect(err).ToNot(HaveOccurred())
+			_, err = os.ReadDir(tmpdir)
+			Expect(err).To(HaveOccurred())
 		})
 
 		Context("Applying models", func() {
-			It("applies models from a gallery", func() {
 
+			It("applies models from a gallery", func() {
 				models := getModels("http://127.0.0.1:9090/models/available")
 				Expect(len(models)).To(Equal(2), fmt.Sprint(models))
 				Expect(models[0].Installed).To(BeFalse(), fmt.Sprint(models))
@@ -219,10 +245,10 @@ var _ = Describe("API test", func() {
 				}, "360s", "10s").Should(Equal(true))
 				Expect(resp["message"]).ToNot(ContainSubstring("error"))
 
-				dat, err := os.ReadFile(filepath.Join(tmpdir, "bert2.yaml"))
+				dat, err := os.ReadFile(filepath.Join(modelDir, "bert2.yaml"))
 				Expect(err).ToNot(HaveOccurred())
 
-				_, err = os.ReadFile(filepath.Join(tmpdir, "foo.yaml"))
+				_, err = os.ReadFile(filepath.Join(modelDir, "foo.yaml"))
 				Expect(err).ToNot(HaveOccurred())
 
 				content := map[string]interface{}{}
@@ -244,6 +270,7 @@ var _ = Describe("API test", func() {
 				}
 			})
 			It("overrides models", func() {
+
 				response := postModelApplyRequest("http://127.0.0.1:9090/models/apply", modelApplyRequest{
 					URL:  "https://raw.githubusercontent.com/go-skynet/model-gallery/main/bert-embeddings.yaml",
 					Name: "bert",
@@ -261,7 +288,7 @@ var _ = Describe("API test", func() {
 					return response["processed"].(bool)
 				}, "360s", "10s").Should(Equal(true))
 
-				dat, err := os.ReadFile(filepath.Join(tmpdir, "bert.yaml"))
+				dat, err := os.ReadFile(filepath.Join(modelDir, "bert.yaml"))
 				Expect(err).ToNot(HaveOccurred())
 
 				content := map[string]interface{}{}
@@ -285,7 +312,7 @@ var _ = Describe("API test", func() {
 					return response["processed"].(bool)
 				}, "360s", "10s").Should(Equal(true))
 
-				dat, err := os.ReadFile(filepath.Join(tmpdir, "bert.yaml"))
+				dat, err := os.ReadFile(filepath.Join(modelDir, "bert.yaml"))
 				Expect(err).ToNot(HaveOccurred())
 
 				content := map[string]interface{}{}
@@ -359,7 +386,7 @@ var _ = Describe("API test", func() {
 				var res map[string]string
 				err = json.Unmarshal([]byte(resp2.Choices[0].Message.FunctionCall.Arguments), &res)
 				Expect(err).ToNot(HaveOccurred())
-				Expect(res["location"]).To(Equal("San Francisco, California, United States"), fmt.Sprint(res))
+				Expect(res["location"]).To(Equal("San Francisco"), fmt.Sprint(res))
 				Expect(res["unit"]).To(Equal("celcius"), fmt.Sprint(res))
 				Expect(string(resp2.Choices[0].FinishReason)).To(Equal("function_call"), fmt.Sprint(resp2.Choices[0].FinishReason))
 
@@ -474,8 +501,11 @@ var _ = Describe("API test", func() {
 			var err error
 			tmpdir, err = os.MkdirTemp("", "")
 			Expect(err).ToNot(HaveOccurred())
+			modelDir = filepath.Join(tmpdir, "models")
+			backendAssetsDir := filepath.Join(tmpdir, "backend-assets")
+			err = os.Mkdir(backendAssetsDir, 0755)
+			Expect(err).ToNot(HaveOccurred())
 
-			modelLoader = model.NewModelLoader(tmpdir)
 			c, cancel = context.WithCancel(context.Background())
 
 			galleries := []gallery.Gallery{
@@ -485,21 +515,20 @@ var _ = Describe("API test", func() {
 				},
 			}
 
-			metricsService, err := metrics.SetupMetrics()
-			Expect(err).ToNot(HaveOccurred())
-
-			app, err = App(
+			bcl, ml, applicationConfig, err = startup.Startup(
 				append(commonOpts,
-					options.WithContext(c),
-					options.WithMetrics(metricsService),
-					options.WithAudioDir(tmpdir),
-					options.WithImageDir(tmpdir),
-					options.WithGalleries(galleries),
-					options.WithModelLoader(modelLoader),
-					options.WithBackendAssets(backendAssets),
-					options.WithBackendAssetsOutput(tmpdir))...,
+					config.WithContext(c),
+					config.WithAudioDir(tmpdir),
+					config.WithImageDir(tmpdir),
+					config.WithGalleries(galleries),
+					config.WithModelPath(modelDir),
+					config.WithBackendAssets(backendAssets),
+					config.WithBackendAssetsOutput(tmpdir))...,
 			)
 			Expect(err).ToNot(HaveOccurred())
+			app, err = App(bcl, ml, applicationConfig)
+			Expect(err).ToNot(HaveOccurred())
+
 			go app.Listen("127.0.0.1:9090")
 
 			defaultConfig := openai.DefaultConfig("")
@@ -518,8 +547,14 @@ var _ = Describe("API test", func() {
 
 		AfterEach(func() {
 			cancel()
-			app.Shutdown()
-			os.RemoveAll(tmpdir)
+			if app != nil {
+				err := app.Shutdown()
+				Expect(err).ToNot(HaveOccurred())
+			}
+			err := os.RemoveAll(tmpdir)
+			Expect(err).ToNot(HaveOccurred())
+			_, err = os.ReadDir(tmpdir)
+			Expect(err).To(HaveOccurred())
 		})
 		It("installs and is capable to run tts", Label("tts"), func() {
 			if runtime.GOOS != "linux" {
@@ -590,20 +625,20 @@ var _ = Describe("API test", func() {
 
 	Context("API query", func() {
 		BeforeEach(func() {
-			modelLoader = model.NewModelLoader(os.Getenv("MODELS_PATH"))
+			modelPath := os.Getenv("MODELS_PATH")
 			c, cancel = context.WithCancel(context.Background())
 
-			metricsService, err := metrics.SetupMetrics()
-			Expect(err).ToNot(HaveOccurred())
+			var err error
 
-			app, err = App(
+			bcl, ml, applicationConfig, err = startup.Startup(
 				append(commonOpts,
-					options.WithExternalBackend("huggingface", os.Getenv("HUGGINGFACE_GRPC")),
-					options.WithContext(c),
-					options.WithModelLoader(modelLoader),
-					options.WithMetrics(metricsService),
+					config.WithExternalBackend("huggingface", os.Getenv("HUGGINGFACE_GRPC")),
+					config.WithContext(c),
+					config.WithModelPath(modelPath),
 				)...)
 			Expect(err).ToNot(HaveOccurred())
+			app, err = App(bcl, ml, applicationConfig)
+			Expect(err).ToNot(HaveOccurred())
 			go app.Listen("127.0.0.1:9090")
 
 			defaultConfig := openai.DefaultConfig("")
@@ -621,7 +656,10 @@ var _ = Describe("API test", func() {
 		})
 		AfterEach(func() {
 			cancel()
-			app.Shutdown()
+			if app != nil {
+				err := app.Shutdown()
+				Expect(err).ToNot(HaveOccurred())
+			}
 		})
 		It("returns the models list", func() {
 			models, err := client.ListModels(context.TODO())
@@ -629,28 +667,28 @@ var _ = Describe("API test", func() {
 			Expect(len(models.Models)).To(Equal(6)) // If "config.yaml" should be included, this should be 8?
 		})
 		It("can generate completions", func() {
-			resp, err := client.CreateCompletion(context.TODO(), openai.CompletionRequest{Model: "testmodel", Prompt: "abcdedfghikl"})
+			resp, err := client.CreateCompletion(context.TODO(), openai.CompletionRequest{Model: "testmodel", Prompt: testPrompt})
 			Expect(err).ToNot(HaveOccurred())
 			Expect(len(resp.Choices)).To(Equal(1))
 			Expect(resp.Choices[0].Text).ToNot(BeEmpty())
 		})
 
 		It("can generate chat completions ", func() {
-			resp, err := client.CreateChatCompletion(context.TODO(), openai.ChatCompletionRequest{Model: "testmodel", Messages: []openai.ChatCompletionMessage{openai.ChatCompletionMessage{Role: "user", Content: "abcdedfghikl"}}})
+			resp, err := client.CreateChatCompletion(context.TODO(), openai.ChatCompletionRequest{Model: "testmodel", Messages: []openai.ChatCompletionMessage{openai.ChatCompletionMessage{Role: "user", Content: testPrompt}}})
 			Expect(err).ToNot(HaveOccurred())
 			Expect(len(resp.Choices)).To(Equal(1))
 			Expect(resp.Choices[0].Message.Content).ToNot(BeEmpty())
 		})
 
 		It("can generate completions from model configs", func() {
-			resp, err := client.CreateCompletion(context.TODO(), openai.CompletionRequest{Model: "gpt4all", Prompt: "abcdedfghikl"})
+			resp, err := client.CreateCompletion(context.TODO(), openai.CompletionRequest{Model: "gpt4all", Prompt: testPrompt})
 			Expect(err).ToNot(HaveOccurred())
 			Expect(len(resp.Choices)).To(Equal(1))
 			Expect(resp.Choices[0].Text).ToNot(BeEmpty())
 		})
 
 		It("can generate chat completions from model configs", func() {
-			resp, err := client.CreateChatCompletion(context.TODO(), openai.ChatCompletionRequest{Model: "gpt4all-2", Messages: []openai.ChatCompletionMessage{openai.ChatCompletionMessage{Role: "user", Content: "abcdedfghikl"}}})
+			resp, err := client.CreateChatCompletion(context.TODO(), openai.ChatCompletionRequest{Model: "gpt4all-2", Messages: []openai.ChatCompletionMessage{openai.ChatCompletionMessage{Role: "user", Content: testPrompt}}})
 			Expect(err).ToNot(HaveOccurred())
 			Expect(len(resp.Choices)).To(Equal(1))
 			Expect(resp.Choices[0].Message.Content).ToNot(BeEmpty())
@@ -658,7 +696,7 @@ var _ = Describe("API test", func() {
 
 		It("returns errors", func() {
 			backends := len(model.AutoLoadBackends) + 1 // +1 for huggingface
-			_, err := client.CreateCompletion(context.TODO(), openai.CompletionRequest{Model: "foomodel", Prompt: "abcdedfghikl"})
+			_, err := client.CreateCompletion(context.TODO(), openai.CompletionRequest{Model: "foomodel", Prompt: testPrompt})
 			Expect(err).To(HaveOccurred())
 			Expect(err.Error()).To(ContainSubstring(fmt.Sprintf("error, status code: 500, message: could not load model - all backends returned error: %d errors occurred:", backends)))
 		})
@@ -802,20 +840,20 @@ var _ = Describe("API test", func() {
 
 	Context("Config file", func() {
 		BeforeEach(func() {
-			modelLoader = model.NewModelLoader(os.Getenv("MODELS_PATH"))
+			modelPath := os.Getenv("MODELS_PATH")
 			c, cancel = context.WithCancel(context.Background())
 
-			metricsService, err := metrics.SetupMetrics()
-			Expect(err).ToNot(HaveOccurred())
-
-			app, err = App(
+			var err error
+			bcl, ml, applicationConfig, err = startup.Startup(
 				append(commonOpts,
-					options.WithContext(c),
-					options.WithMetrics(metricsService),
-					options.WithModelLoader(modelLoader),
-					options.WithConfigFile(os.Getenv("CONFIG_FILE")))...,
+					config.WithContext(c),
+					config.WithModelPath(modelPath),
+					config.WithConfigFile(os.Getenv("CONFIG_FILE")))...,
 			)
 			Expect(err).ToNot(HaveOccurred())
+			app, err = App(bcl, ml, applicationConfig)
+			Expect(err).ToNot(HaveOccurred())
+
 			go app.Listen("127.0.0.1:9090")
 
 			defaultConfig := openai.DefaultConfig("")
@@ -831,16 +869,19 @@ var _ = Describe("API test", func() {
 		})
 		AfterEach(func() {
 			cancel()
-			app.Shutdown()
+			if app != nil {
+				err := app.Shutdown()
+				Expect(err).ToNot(HaveOccurred())
+			}
 		})
 		It("can generate chat completions from config file (list1)", func() {
-			resp, err := client.CreateChatCompletion(context.TODO(), openai.ChatCompletionRequest{Model: "list1", Messages: []openai.ChatCompletionMessage{{Role: "user", Content: "abcdedfghikl"}}})
+			resp, err := client.CreateChatCompletion(context.TODO(), openai.ChatCompletionRequest{Model: "list1", Messages: []openai.ChatCompletionMessage{{Role: "user", Content: testPrompt}}})
 			Expect(err).ToNot(HaveOccurred())
 			Expect(len(resp.Choices)).To(Equal(1))
 			Expect(resp.Choices[0].Message.Content).ToNot(BeEmpty())
 		})
 		It("can generate chat completions from config file (list2)", func() {
-			resp, err := client.CreateChatCompletion(context.TODO(), openai.ChatCompletionRequest{Model: "list2", Messages: []openai.ChatCompletionMessage{{Role: "user", Content: "abcdedfghikl"}}})
+			resp, err := client.CreateChatCompletion(context.TODO(), openai.ChatCompletionRequest{Model: "list2", Messages: []openai.ChatCompletionMessage{{Role: "user", Content: testPrompt}}})
 			Expect(err).ToNot(HaveOccurred())
 			Expect(len(resp.Choices)).To(Equal(1))
 			Expect(resp.Choices[0].Message.Content).ToNot(BeEmpty())

core/http/apt_suite_test.go

@@ -1,4 +1,4 @@
-package api_test
+package http_test
 
 import (
 	"testing"