wip

.devcontainer-scripts/poststart.sh

@@ -2,6 +2,9 @@
 
 cd /workspace
 
+# Grab the pre-stashed backend assets to avoid build issues
+cp -r /build/backend-assets /workspace/backend-assets
+
 # Ensures generated source files are present upon load
 make prepare
 

.devcontainer/docker-compose-devcontainer.yml

@@ -4,6 +4,10 @@ services:
       context: ..
       dockerfile: Dockerfile
       target: devcontainer
+      args:
+      - FFMPEG=true
+      - IMAGE_TYPE=extras
+      - GO_TAGS=p2p tts
     env_file:
       - ../.env
     ports:

.dockerignore

@@ -3,13 +3,7 @@
 .vscode
 .devcontainer
 models
-backends
 examples/chatbot-ui/models
-backend/go/image/stablediffusion-ggml/build/
-backend/go/*/build
-backend/go/*/.cache
-backend/go/*/sources
-backend/go/*/package
 examples/rwkv/models
 examples/**/models
 Dockerfile*
@@ -20,4 +14,4 @@ __pycache__
 
 # backend virtual environments
 **/venv
-backend/python/**/source
+backend/python/**/source

.env

@@ -41,6 +41,13 @@
 ## Uncomment and set to true to enable rebuilding from source
 # REBUILD=true
 
+## Enable go tags, available: p2p, tts
+## p2p: enable distributed inferencing
+## tts: enables text-to-speech with go-piper 
+## (requires REBUILD=true)
+#
+# GO_TAGS=p2p
+
 ## Path where to store generated images
 # LOCALAI_IMAGE_PATH=/tmp/generated/images
 

.github/bump_deps.sh

@@ -3,20 +3,15 @@ set -xe
 REPO=$1
 BRANCH=$2
 VAR=$3
-FILE=$4
-
-if [ -z "$FILE" ]; then
-    FILE="Makefile"
-fi
 
 LAST_COMMIT=$(curl -s -H "Accept: application/vnd.github.VERSION.sha" "https://api.github.com/repos/$REPO/commits/$BRANCH")
 
 # Read $VAR from Makefile (only first match)
 set +e
-CURRENT_COMMIT="$(grep -m1 "^$VAR?=" $FILE | cut -d'=' -f2)"
+CURRENT_COMMIT="$(grep -m1 "^$VAR?=" Makefile | cut -d'=' -f2)"
 set -e
 
-sed -i $FILE -e "s/$VAR?=.*/$VAR?=$LAST_COMMIT/"
+sed -i Makefile -e "s/$VAR?=.*/$VAR?=$LAST_COMMIT/"
 
 if [ -z "$CURRENT_COMMIT" ]; then
     echo "Could not find $VAR in Makefile."

.github/dependabot.yml

@@ -61,6 +61,10 @@ updates:
     directory: "/backend/python/openvoice"
     schedule:
       interval: "weekly"
+  - package-ecosystem: "pip"
+    directory: "/backend/python/parler-tts"
+    schedule:
+      interval: "weekly"
   - package-ecosystem: "pip"
     directory: "/backend/python/rerankers"
     schedule:

.github/workflows/backend_build.yml

@@ -1,243 +0,0 @@
----
-name: 'build python backend container images (reusable)'
-
-on:
-  workflow_call:
-    inputs:
-      base-image:
-        description: 'Base image'
-        required: true
-        type: string
-      build-type:
-        description: 'Build type'
-        default: ''
-        type: string
-      cuda-major-version:
-        description: 'CUDA major version'
-        default: "12"
-        type: string
-      cuda-minor-version:
-        description: 'CUDA minor version'
-        default: "1"
-        type: string
-      platforms:
-        description: 'Platforms'
-        default: ''
-        type: string
-      tag-latest:
-        description: 'Tag latest'
-        default: ''
-        type: string
-      tag-suffix:
-        description: 'Tag suffix'
-        default: ''
-        type: string
-      runs-on:
-        description: 'Runs on'
-        required: true
-        default: ''
-        type: string
-      backend:
-        description: 'Backend to build'
-        required: true
-        type: string
-      context:
-        description: 'Build context'
-        required: true
-        type: string
-      dockerfile:
-        description: 'Build Dockerfile'
-        required: true
-        type: string
-      skip-drivers:
-        description: 'Skip drivers'
-        default: 'false'
-        type: string
-    secrets:
-      dockerUsername:
-        required: false
-      dockerPassword:
-        required: false
-      quayUsername:
-        required: true
-      quayPassword:
-        required: true
-
-jobs:
-  backend-build:
-    runs-on: ${{ inputs.runs-on }}
-    env:
-        quay_username: ${{ secrets.quayUsername }}
-    steps:
-
-
-      - name: Free Disk Space (Ubuntu)
-        if: inputs.runs-on == 'ubuntu-latest'
-        uses: jlumbroso/free-disk-space@main
-        with:
-          # this might remove tools that are actually needed,
-          # if set to "true" but frees about 6 GB
-          tool-cache: true
-          # all of these default to true, but feel free to set to
-          # "false" if necessary for your workflow
-          android: true
-          dotnet: true
-          haskell: true
-          large-packages: true
-          docker-images: true
-          swap-storage: true
-
-      - name: Force Install GIT latest
-        run: |
-          sudo apt-get update \
-          && sudo apt-get install -y software-properties-common \
-          && sudo apt-get update \
-          && sudo add-apt-repository -y ppa:git-core/ppa \
-          && sudo apt-get update \
-          && sudo apt-get install -y git
-
-      - name: Checkout
-        uses: actions/checkout@v5
-
-      - name: Release space from worker
-        if: inputs.runs-on == 'ubuntu-latest'
-        run: |
-          echo "Listing top largest packages"
-          pkgs=$(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 snapd || true
-          sudo apt-get purge --auto-remove android-sdk-platform-tools snapd || 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 "$AGENT_TOOLSDIRECTORY" || true
-          df -h
-
-      - name: Docker meta
-        id: meta
-        if: github.event_name != 'pull_request'
-        uses: docker/metadata-action@v5
-        with:
-          images: |
-            quay.io/go-skynet/local-ai-backends
-            localai/localai-backends
-          tags: |
-            type=ref,event=branch
-            type=semver,pattern={{raw}}
-            type=sha
-          flavor: |
-            latest=${{ inputs.tag-latest }}
-            suffix=${{ inputs.tag-suffix }},onlatest=true
-
-      - name: Docker meta for PR
-        id: meta_pull_request
-        if: github.event_name == 'pull_request'
-        uses: docker/metadata-action@v5
-        with:
-          images: |
-            quay.io/go-skynet/ci-tests
-          tags: |
-            type=ref,event=branch,suffix=${{ github.event.number }}-${{ inputs.backend }}-${{ inputs.build-type }}-${{ inputs.cuda-major-version }}-${{ inputs.cuda-minor-version }}
-            type=semver,pattern={{raw}},suffix=${{ github.event.number }}-${{ inputs.backend }}-${{ inputs.build-type }}-${{ inputs.cuda-major-version }}-${{ inputs.cuda-minor-version }}
-            type=sha,suffix=${{ github.event.number }}-${{ inputs.backend }}-${{ inputs.build-type }}-${{ inputs.cuda-major-version }}-${{ inputs.cuda-minor-version }}
-          flavor: |
-            latest=${{ inputs.tag-latest }}
-            suffix=${{ inputs.tag-suffix }},onlatest=true
-## End testing image
-      - name: Set up QEMU
-        uses: docker/setup-qemu-action@master
-        with:
-          platforms: all
-
-      - name: Set up Docker Buildx
-        id: buildx
-        uses: docker/setup-buildx-action@master
-
-      - name: Login to DockerHub
-        if: github.event_name != 'pull_request'
-        uses: docker/login-action@v3
-        with:
-          username: ${{ secrets.dockerUsername }}
-          password: ${{ secrets.dockerPassword }}
-
-      - name: Login to Quay.io
-        if: ${{ env.quay_username != '' }}
-        uses: docker/login-action@v3
-        with:
-          registry: quay.io
-          username: ${{ secrets.quayUsername }}
-          password: ${{ secrets.quayPassword }}
-
-      - name: Build and push
-        uses: docker/build-push-action@v6
-        if: github.event_name != 'pull_request'
-        with:
-          builder: ${{ steps.buildx.outputs.name }}
-          build-args: |
-            BUILD_TYPE=${{ inputs.build-type }}
-            SKIP_DRIVERS=${{ inputs.skip-drivers }}
-            CUDA_MAJOR_VERSION=${{ inputs.cuda-major-version }}
-            CUDA_MINOR_VERSION=${{ inputs.cuda-minor-version }}
-            BASE_IMAGE=${{ inputs.base-image }}
-            BACKEND=${{ inputs.backend }}
-          context: ${{ inputs.context }}
-          file: ${{ inputs.dockerfile }}
-          cache-from: type=gha
-          platforms: ${{ inputs.platforms }}
-          push: ${{ github.event_name != 'pull_request' }}
-          tags: ${{ steps.meta.outputs.tags }}
-          labels: ${{ steps.meta.outputs.labels }}
-
-      - name: Build and push (PR)
-        uses: docker/build-push-action@v6
-        if: github.event_name == 'pull_request'
-        with:
-          builder: ${{ steps.buildx.outputs.name }}
-          build-args: |
-            BUILD_TYPE=${{ inputs.build-type }}
-            SKIP_DRIVERS=${{ inputs.skip-drivers }}
-            CUDA_MAJOR_VERSION=${{ inputs.cuda-major-version }}
-            CUDA_MINOR_VERSION=${{ inputs.cuda-minor-version }}
-            BASE_IMAGE=${{ inputs.base-image }}
-            BACKEND=${{ inputs.backend }}
-          context: ${{ inputs.context }}
-          file: ${{ inputs.dockerfile }}
-          cache-from: type=gha
-          platforms: ${{ inputs.platforms }}
-          push: ${{ env.quay_username != '' }}
-          tags: ${{ steps.meta_pull_request.outputs.tags }}
-          labels: ${{ steps.meta_pull_request.outputs.labels }}
-
-
-
-      - name: job summary
-        run: |
-          echo "Built image: ${{ steps.meta.outputs.labels }}" >> $GITHUB_STEP_SUMMARY

.github/workflows/backend_build_darwin.yml

@@ -1,144 +0,0 @@
----
-name: 'build darwin python backend container images (reusable)'
-
-on:
-  workflow_call:
-    inputs:
-      backend:
-        description: 'Backend to build'
-        required: true
-        type: string
-      build-type:
-        description: 'Build type (e.g., mps)'
-        default: ''
-        type: string
-      use-pip:
-        description: 'Use pip to install dependencies'
-        default: false
-        type: boolean
-      lang:
-        description: 'Programming language (e.g. go)'
-        default: 'python'
-        type: string
-      go-version:
-        description: 'Go version to use'
-        default: '1.24.x'
-        type: string
-      tag-suffix:
-        description: 'Tag suffix for the built image'
-        required: true
-        type: string
-      runs-on:
-        description: 'Runner to use'
-        default: 'macOS-14'
-        type: string
-    secrets:
-      dockerUsername:
-        required: false
-      dockerPassword:
-        required: false
-      quayUsername:
-        required: true
-      quayPassword:
-        required: true
-
-jobs:
-  darwin-backend-build:
-    runs-on: ${{ inputs.runs-on }}
-    strategy:
-      matrix:
-        go-version: ['${{ inputs.go-version }}']
-    steps:
-      - name: Clone
-        uses: actions/checkout@v5
-        with:
-          submodules: true
-
-      - name: Setup Go ${{ matrix.go-version }}
-        uses: actions/setup-go@v5
-        with:
-          go-version: ${{ matrix.go-version }}
-          cache: false
-
-      # You can test your matrix by printing the current Go version
-      - name: Display Go version
-        run: go version
-
-      - name: Dependencies
-        run: |
-          brew install protobuf grpc make protoc-gen-go protoc-gen-go-grpc libomp llvm
-
-      - name: Build ${{ inputs.backend }}-darwin
-        run: |
-          make protogen-go
-          BACKEND=${{ inputs.backend }} BUILD_TYPE=${{ inputs.build-type }} USE_PIP=${{ inputs.use-pip }} make build-darwin-${{ inputs.lang }}-backend
-
-      - name: Upload ${{ inputs.backend }}.tar
-        uses: actions/upload-artifact@v4
-        with:
-          name: ${{ inputs.backend }}-tar
-          path: backend-images/${{ inputs.backend }}.tar
-
-  darwin-backend-publish:
-    needs: darwin-backend-build
-    if: github.event_name != 'pull_request'
-    runs-on: ubuntu-latest
-    steps:
-      - name: Download ${{ inputs.backend }}.tar
-        uses: actions/download-artifact@v5
-        with:
-          name: ${{ inputs.backend }}-tar
-          path: .
-
-      - name: Install crane
-        run: |
-          curl -L https://github.com/google/go-containerregistry/releases/latest/download/go-containerregistry_Linux_x86_64.tar.gz | tar -xz
-          sudo mv crane /usr/local/bin/
-
-      - name: Log in to DockerHub
-        run: |
-          echo "${{ secrets.dockerPassword }}" | crane auth login docker.io -u "${{ secrets.dockerUsername }}" --password-stdin
-
-      - name: Log in to quay.io
-        run: |
-          echo "${{ secrets.quayPassword }}" | crane auth login quay.io -u "${{ secrets.quayUsername }}" --password-stdin
-
-      - name: Docker meta
-        id: meta
-        uses: docker/metadata-action@v5
-        with:
-          images: |
-            localai/localai-backends
-          tags: |
-            type=ref,event=branch
-            type=semver,pattern={{raw}}
-            type=sha
-          flavor: |
-            latest=auto
-            suffix=${{ inputs.tag-suffix }},onlatest=true
-
-      - name: Docker meta
-        id: quaymeta
-        uses: docker/metadata-action@v5
-        with:
-          images: |
-            quay.io/go-skynet/local-ai-backends
-          tags: |
-            type=ref,event=branch
-            type=semver,pattern={{raw}}
-            type=sha
-          flavor: |
-            latest=auto
-            suffix=${{ inputs.tag-suffix }},onlatest=true
-
-      - name: Push Docker image (DockerHub)
-        run: |
-          for tag in $(echo "${{ steps.meta.outputs.tags }}" | tr ',' '\n'); do
-            crane push ${{ inputs.backend }}.tar $tag
-          done
-
-      - name: Push Docker image (Quay)
-        run: |
-          for tag in $(echo "${{ steps.quaymeta.outputs.tags }}" | tr ',' '\n'); do
-            crane push ${{ inputs.backend }}.tar $tag
-          done

.github/workflows/backend_pr.yml

@@ -1,78 +0,0 @@
-name: 'build backend container images (PR-filtered)'
-
-on:
-  pull_request:
-
-concurrency:
-  group: ci-backends-pr-${{ github.head_ref || github.ref }}-${{ github.repository }}
-  cancel-in-progress: true
-
-jobs:
-  generate-matrix:
-    runs-on: ubuntu-latest
-    outputs:
-      matrix: ${{ steps.set-matrix.outputs.matrix }}
-      matrix-darwin: ${{ steps.set-matrix.outputs.matrix-darwin }}
-      has-backends: ${{ steps.set-matrix.outputs.has-backends }}
-      has-backends-darwin: ${{ steps.set-matrix.outputs.has-backends-darwin }}
-    steps:
-      - name: Checkout repository
-        uses: actions/checkout@v5
-
-      - name: Setup Bun
-        uses: oven-sh/setup-bun@v2
-
-      - name: Install dependencies
-        run: |
-          bun add js-yaml
-          bun add @octokit/core
-
-      # filters the matrix in backend.yml
-      - name: Filter matrix for changed backends
-        id: set-matrix
-        env:
-          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-          GITHUB_EVENT_PATH: ${{ github.event_path }}
-        run: bun run scripts/changed-backends.js
-
-  backend-jobs:
-    needs: generate-matrix
-    uses: ./.github/workflows/backend_build.yml
-    if: needs.generate-matrix.outputs.has-backends == 'true'
-    with:
-      tag-latest: ${{ matrix.tag-latest }}
-      tag-suffix: ${{ matrix.tag-suffix }}
-      build-type: ${{ matrix.build-type }}
-      cuda-major-version: ${{ matrix.cuda-major-version }}
-      cuda-minor-version: ${{ matrix.cuda-minor-version }}
-      platforms: ${{ matrix.platforms }}
-      runs-on: ${{ matrix.runs-on }}
-      base-image: ${{ matrix.base-image }}
-      backend: ${{ matrix.backend }}
-      dockerfile: ${{ matrix.dockerfile }}
-      skip-drivers: ${{ matrix.skip-drivers }}
-      context: ${{ matrix.context }}
-    secrets:
-      quayUsername: ${{ secrets.LOCALAI_REGISTRY_USERNAME }}
-      quayPassword: ${{ secrets.LOCALAI_REGISTRY_PASSWORD }}
-    strategy:
-      fail-fast: true
-      matrix: ${{ fromJson(needs.generate-matrix.outputs.matrix) }}
-  backend-jobs-darwin:
-    needs: generate-matrix
-    uses: ./.github/workflows/backend_build_darwin.yml
-    if: needs.generate-matrix.outputs.has-backends-darwin == 'true'
-    with:
-      backend: ${{ matrix.backend }}
-      build-type: ${{ matrix.build-type }}
-      go-version: "1.24.x"
-      tag-suffix: ${{ matrix.tag-suffix }}
-      lang: ${{ matrix.lang || 'python' }}
-      use-pip: ${{ matrix.backend == 'diffusers' }}
-      runs-on: "macOS-14"
-    secrets:
-      quayUsername: ${{ secrets.LOCALAI_REGISTRY_USERNAME }}
-      quayPassword: ${{ secrets.LOCALAI_REGISTRY_PASSWORD }}
-    strategy:
-      fail-fast: true
-      matrix: ${{ fromJson(needs.generate-matrix.outputs.matrix-darwin) }}

.github/workflows/build-test.yaml

@@ -1,67 +0,0 @@
-name: Build test
-
-on:
-  push:
-    branches:
-      - master
-  pull_request:
-
-jobs:
-  build-test:
-    runs-on: ubuntu-latest
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v5
-        with:
-          fetch-depth: 0
-      - name: Set up Go
-        uses: actions/setup-go@v5
-        with:
-          go-version: 1.23
-      - name: Run GoReleaser
-        run: |
-          make dev-dist
-  launcher-build-darwin:
-    runs-on: macos-latest
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v5
-        with:
-          fetch-depth: 0
-      - name: Set up Go
-        uses: actions/setup-go@v5
-        with:
-          go-version: 1.23
-      - name: Build launcher for macOS ARM64
-        run: |
-          make build-launcher-darwin
-          ls -liah dist
-      - name: Upload macOS launcher artifacts
-        uses: actions/upload-artifact@v4
-        with:
-          name: launcher-macos
-          path: dist/
-          retention-days: 30
-      
-  launcher-build-linux:
-    runs-on: ubuntu-latest
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v5
-        with:
-          fetch-depth: 0
-      - name: Set up Go
-        uses: actions/setup-go@v5
-        with:
-          go-version: 1.23
-      - name: Build launcher for Linux
-        run: |
-          sudo apt-get update
-          sudo apt-get install golang gcc libgl1-mesa-dev xorg-dev libxkbcommon-dev
-          make build-launcher-linux
-      - name: Upload Linux launcher artifacts
-        uses: actions/upload-artifact@v4
-        with:
-          name: launcher-linux
-          path: local-ai-launcher-linux.tar.xz
-          retention-days: 30

.github/workflows/bump_deps.yaml

@@ -10,32 +10,30 @@ jobs:
       matrix:
         include:
           - repository: "ggml-org/llama.cpp"
-            variable: "LLAMA_VERSION"
+            variable: "CPPLLAMA_VERSION"
             branch: "master"
-            file: "backend/cpp/llama-cpp/Makefile"
           - repository: "ggml-org/whisper.cpp"
             variable: "WHISPER_CPP_VERSION"
             branch: "master"
-            file: "backend/go/whisper/Makefile"
           - repository: "PABannier/bark.cpp"
             variable: "BARKCPP_VERSION"
             branch: "main"
-            file: "Makefile"
           - repository: "leejet/stable-diffusion.cpp"
             variable: "STABLEDIFFUSION_GGML_VERSION"
             branch: "master"
-            file: "backend/go/stablediffusion-ggml/Makefile"
+          - repository: "mudler/go-stable-diffusion"
+            variable: "STABLEDIFFUSION_VERSION"
+            branch: "master"
           - repository: "mudler/go-piper"
             variable: "PIPER_VERSION"
             branch: "master"
-            file: "backend/go/piper/Makefile"
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v5
+      - uses: actions/checkout@v4
       - name: Bump dependencies 🔧
         id: bump
         run: |
-          bash .github/bump_deps.sh ${{ matrix.repository }} ${{ matrix.branch }} ${{ matrix.variable }} ${{ matrix.file }}
+          bash .github/bump_deps.sh ${{ matrix.repository }} ${{ matrix.branch }} ${{ matrix.variable }}
           {
             echo 'message<<EOF'
             cat "${{ matrix.variable }}_message.txt"

.github/workflows/bump_docs.yaml

@@ -12,7 +12,7 @@ jobs:
           - repository: "mudler/LocalAI"
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v5
+      - uses: actions/checkout@v4
       - name: Bump dependencies 🔧
         run: |
           bash .github/bump_docs.sh ${{ matrix.repository }}

.github/workflows/checksum_checker.yaml

@@ -5,7 +5,7 @@ on:
   workflow_dispatch:
 jobs:
   checksum_check:
-    runs-on: ubuntu-latest
+    runs-on: arc-runner-set
     steps:
       - name: Force Install GIT latest
         run: |
@@ -15,11 +15,12 @@ jobs:
           && sudo add-apt-repository -y ppa:git-core/ppa \
           && sudo apt-get update \
           && sudo apt-get install -y git
-      - uses: actions/checkout@v5
+      - uses: actions/checkout@v4
       - name: Install dependencies
         run: |
           sudo apt-get update
           sudo apt-get install -y pip wget
+          sudo pip install --upgrade pip
           pip install huggingface_hub
       - name: 'Setup yq'
         uses: dcarbone/install-yq-action@v1.3.1

.github/workflows/dependabot_auto.yml

@@ -20,7 +20,7 @@ jobs:
           skip-commit-verification: true
 
       - name: Checkout repository
-        uses: actions/checkout@v5
+        uses: actions/checkout@v4
 
       - name: Approve a PR if not already approved
         run: |

.github/workflows/deploy-explorer.yaml

@@ -15,7 +15,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Clone
-        uses: actions/checkout@v5
+        uses: actions/checkout@v4
         with:
           submodules: true
       - uses: actions/setup-go@v5
@@ -31,7 +31,7 @@ jobs:
           make protogen-go
       - name: Build api
         run: |
-          CGO_ENABLED=0 make build
+          CGO_ENABLED=0 make build-api
       - name: rm
         uses: appleboy/ssh-action@v1.2.2
         with:

.github/workflows/generate_grpc_cache.yaml

@@ -17,7 +17,7 @@ jobs:
       matrix:
         include:
           - grpc-base-image: ubuntu:22.04
-            runs-on: 'ubuntu-latest'
+            runs-on: 'arc-runner-set'
             platforms: 'linux/amd64,linux/arm64'
     runs-on: ${{matrix.runs-on}}
     steps:
@@ -73,7 +73,7 @@ jobs:
         uses: docker/setup-buildx-action@master
 
       - name: Checkout
-        uses: actions/checkout@v5
+        uses: actions/checkout@v4
 
       - name: Cache GRPC
         uses: docker/build-push-action@v6

.github/workflows/generate_intel_image.yaml

@@ -15,7 +15,7 @@ jobs:
     strategy:
       matrix:
         include:
-          - base-image: intel/oneapi-basekit:2025.2.0-0-devel-ubuntu22.04
+          - base-image: intel/oneapi-basekit:2025.1.0-0-devel-ubuntu22.04
             runs-on: 'ubuntu-latest'
             platforms: 'linux/amd64'
     runs-on: ${{matrix.runs-on}}
@@ -43,7 +43,7 @@ jobs:
         uses: docker/setup-buildx-action@master
 
       - name: Checkout
-        uses: actions/checkout@v5
+        uses: actions/checkout@v4
 
       - name: Cache Intel images
         uses: docker/build-push-action@v6

.github/workflows/image-pr.yml

@@ -9,11 +9,13 @@ concurrency:
   cancel-in-progress: true
 
 jobs:
-  image-build:
+  extras-image-build:
     uses: ./.github/workflows/image_build.yml
     with:
       tag-latest: ${{ matrix.tag-latest }}
       tag-suffix: ${{ matrix.tag-suffix }}
+      ffmpeg: ${{ matrix.ffmpeg }}
+      image-type: ${{ matrix.image-type }}
       build-type: ${{ matrix.build-type }}
       cuda-major-version: ${{ matrix.cuda-major-version }}
       cuda-minor-version: ${{ matrix.cuda-minor-version }}
@@ -34,35 +36,115 @@ jobs:
       fail-fast: false
       matrix:
         include:
+          # This is basically covered by the AIO test
+          # - build-type: ''
+          #   platforms: 'linux/amd64'
+          #   tag-latest: 'false'
+          #   tag-suffix: '-ffmpeg'
+          #   ffmpeg: 'true'
+          #   image-type: 'extras'
+          #   runs-on: 'arc-runner-set'
+          #   base-image: "ubuntu:22.04"
+          #   makeflags: "--jobs=3 --output-sync=target"
           - build-type: 'cublas'
             cuda-major-version: "12"
             cuda-minor-version: "0"
             platforms: 'linux/amd64'
             tag-latest: 'false'
-            tag-suffix: '-gpu-nvidia-cuda-12'
-            runs-on: 'ubuntu-latest'
+            tag-suffix: '-cublas-cuda12-ffmpeg'
+            ffmpeg: 'true'
+            image-type: 'extras'
+            runs-on: 'arc-runner-set'
             base-image: "ubuntu:22.04"
             makeflags: "--jobs=3 --output-sync=target"
           - build-type: 'hipblas'
             platforms: 'linux/amd64'
             tag-latest: 'false'
             tag-suffix: '-hipblas'
-            base-image: "rocm/dev-ubuntu-22.04:6.4.3"
+            ffmpeg: 'false'
+            image-type: 'extras'
+            base-image: "rocm/dev-ubuntu-22.04:6.1"
             grpc-base-image: "ubuntu:22.04"
-            runs-on: 'ubuntu-latest'
+            runs-on: 'arc-runner-set'
             makeflags: "--jobs=3 --output-sync=target"
-          - build-type: 'sycl'
+          - build-type: 'sycl_f16'
             platforms: 'linux/amd64'
             tag-latest: 'false'
             base-image: "quay.io/go-skynet/intel-oneapi-base:latest"
             grpc-base-image: "ubuntu:22.04"
-            tag-suffix: 'sycl'
-            runs-on: 'ubuntu-latest'
+            tag-suffix: 'sycl-f16-ffmpeg'
+            ffmpeg: 'true'
+            image-type: 'extras'
+            runs-on: 'arc-runner-set'
             makeflags: "--jobs=3 --output-sync=target"
           - build-type: 'vulkan'
             platforms: 'linux/amd64'
             tag-latest: 'false'
-            tag-suffix: '-vulkan-core'
+            tag-suffix: '-vulkan-ffmpeg-core'
+            ffmpeg: 'true'
+            image-type: 'core'
             runs-on: 'ubuntu-latest'
             base-image: "ubuntu:22.04"
             makeflags: "--jobs=4 --output-sync=target"
+  # core-image-build:
+  #   uses: ./.github/workflows/image_build.yml
+  #   with:
+  #     tag-latest: ${{ matrix.tag-latest }}
+  #     tag-suffix: ${{ matrix.tag-suffix }}
+  #     ffmpeg: ${{ matrix.ffmpeg }}
+  #     image-type: ${{ matrix.image-type }}
+  #     build-type: ${{ matrix.build-type }}
+  #     cuda-major-version: ${{ matrix.cuda-major-version }}
+  #     cuda-minor-version: ${{ matrix.cuda-minor-version }}
+  #     platforms: ${{ matrix.platforms }}
+  #     runs-on: ${{ matrix.runs-on }}
+  #     base-image: ${{ matrix.base-image }}
+  #     grpc-base-image: ${{ matrix.grpc-base-image }}
+  #     makeflags: ${{ matrix.makeflags }}
+  #   secrets:
+  #     dockerUsername: ${{ secrets.DOCKERHUB_USERNAME }}
+  #     dockerPassword: ${{ secrets.DOCKERHUB_PASSWORD }}
+  #     quayUsername: ${{ secrets.LOCALAI_REGISTRY_USERNAME }}
+  #     quayPassword: ${{ secrets.LOCALAI_REGISTRY_PASSWORD }}
+  #   strategy:
+  #     matrix:
+  #       include:
+          # - build-type: ''
+          #   platforms: 'linux/amd64'
+          #   tag-latest: 'false'
+          #   tag-suffix: '-ffmpeg-core'
+          #   ffmpeg: 'true'
+          #   image-type: 'core'
+          #   runs-on: 'ubuntu-latest'
+          #   base-image: "ubuntu:22.04"
+          #   makeflags: "--jobs=4 --output-sync=target"
+          # - build-type: 'sycl_f16'
+          #   platforms: 'linux/amd64'
+          #   tag-latest: 'false'
+          #   base-image: "quay.io/go-skynet/intel-oneapi-base:latest"
+          #   grpc-base-image: "ubuntu:22.04"
+          #   tag-suffix: 'sycl-f16-ffmpeg-core'
+          #   ffmpeg: 'true'
+          #   image-type: 'core'
+          #   runs-on: 'arc-runner-set'
+          #   makeflags: "--jobs=3 --output-sync=target"
+          # - build-type: 'cublas'
+          #   cuda-major-version: "12"
+          #   cuda-minor-version: "0"
+          #   platforms: 'linux/amd64'
+          #   tag-latest: 'false'
+          #   tag-suffix: '-cublas-cuda12-ffmpeg-core'
+          #   ffmpeg: 'true'
+          #   image-type: 'core'
+          #   runs-on: 'ubuntu-latest'
+          #   base-image: "ubuntu:22.04"
+          #   makeflags: "--jobs=4 --output-sync=target"
+          # - build-type: 'vulkan'
+          #   platforms: 'linux/amd64'
+          #   tag-latest: 'false'
+          #   tag-suffix: '-vulkan-ffmpeg-core'
+          #   ffmpeg: 'true'
+          #   image-type: 'core'
+          #   runs-on: 'ubuntu-latest'
+          #   base-image: "ubuntu:22.04"
+          #   makeflags: "--jobs=4 --output-sync=target"

.github/workflows/image.yml

@@ -18,6 +18,8 @@ jobs:
     with:
       tag-latest: ${{ matrix.tag-latest }}
       tag-suffix: ${{ matrix.tag-suffix }}
+      ffmpeg: ${{ matrix.ffmpeg }}
+      image-type: ${{ matrix.image-type }}
       build-type: ${{ matrix.build-type }}
       cuda-major-version: ${{ matrix.cuda-major-version }}
       cuda-minor-version: ${{ matrix.cuda-minor-version }}
@@ -27,29 +29,157 @@ jobs:
       grpc-base-image: ${{ matrix.grpc-base-image }}
       aio: ${{ matrix.aio }}
       makeflags: ${{ matrix.makeflags }}
+      latest-image: ${{ matrix.latest-image }}
+      latest-image-aio: ${{ matrix.latest-image-aio }}
     secrets:
       dockerUsername: ${{ secrets.DOCKERHUB_USERNAME }}
       dockerPassword: ${{ secrets.DOCKERHUB_PASSWORD }}
       quayUsername: ${{ secrets.LOCALAI_REGISTRY_USERNAME }}
       quayPassword: ${{ secrets.LOCALAI_REGISTRY_PASSWORD }}
     strategy:
+      # Pushing with all jobs in parallel
+      # eats the bandwidth of all the nodes
+      max-parallel: 2
       matrix:
         include:
           - build-type: 'hipblas'
             platforms: 'linux/amd64'
             tag-latest: 'auto'
-            tag-suffix: '-gpu-hipblas'
-            base-image: "rocm/dev-ubuntu-22.04:6.4.3"
-            grpc-base-image: "ubuntu:22.04"
-            runs-on: 'ubuntu-latest'
-            makeflags: "--jobs=3 --output-sync=target"
+            tag-suffix: '-hipblas-extras'
+            ffmpeg: 'true'
+            image-type: 'extras'
             aio: "-aio-gpu-hipblas"
+            base-image: "rocm/dev-ubuntu-22.04:6.1"
+            grpc-base-image: "ubuntu:22.04"
+            latest-image: 'latest-gpu-hipblas-extras'
+            latest-image-aio: 'latest-aio-gpu-hipblas'
+            runs-on: 'arc-runner-set'
+            makeflags: "--jobs=3 --output-sync=target"
+          - build-type: 'hipblas'
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-hipblas'
+            ffmpeg: 'true'
+            image-type: 'core'
+            base-image: "rocm/dev-ubuntu-22.04:6.1"
+            grpc-base-image: "ubuntu:22.04"
+            runs-on: 'arc-runner-set'
+            makeflags: "--jobs=3 --output-sync=target"
+            latest-image: 'latest-gpu-hipblas'
+  self-hosted-jobs:
+    uses: ./.github/workflows/image_build.yml
+    with:
+      tag-latest: ${{ matrix.tag-latest }}
+      tag-suffix: ${{ matrix.tag-suffix }}
+      ffmpeg: ${{ matrix.ffmpeg }}
+      image-type: ${{ matrix.image-type }}
+      build-type: ${{ matrix.build-type }}
+      cuda-major-version: ${{ matrix.cuda-major-version }}
+      cuda-minor-version: ${{ matrix.cuda-minor-version }}
+      platforms: ${{ matrix.platforms }}
+      runs-on: ${{ matrix.runs-on }}
+      base-image: ${{ matrix.base-image }}
+      grpc-base-image: ${{ matrix.grpc-base-image }}
+      aio: ${{ matrix.aio }}
+      makeflags: ${{ matrix.makeflags }}
+      latest-image: ${{ matrix.latest-image }}
+      latest-image-aio: ${{ matrix.latest-image-aio }}
+    secrets:
+      dockerUsername: ${{ secrets.DOCKERHUB_USERNAME }}
+      dockerPassword: ${{ secrets.DOCKERHUB_PASSWORD }}
+      quayUsername: ${{ secrets.LOCALAI_REGISTRY_USERNAME }}
+      quayPassword: ${{ secrets.LOCALAI_REGISTRY_PASSWORD }}
+    strategy:
+      # Pushing with all jobs in parallel
+      # eats the bandwidth of all the nodes
+      max-parallel: ${{ github.event_name != 'pull_request' && 5 || 8 }}
+      matrix:
+        include:
+          - build-type: 'cublas'
+            cuda-major-version: "11"
+            cuda-minor-version: "7"
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-cublas-cuda11-extras'
+            ffmpeg: 'true'
+            image-type: 'extras'
+            runs-on: 'arc-runner-set'
+            base-image: "ubuntu:22.04"
+            aio: "-aio-gpu-nvidia-cuda-11"
+            latest-image: 'latest-gpu-nvidia-cuda-11-extras'
+            latest-image-aio: 'latest-aio-gpu-nvidia-cuda-11'
+            makeflags: "--jobs=3 --output-sync=target"
+          - build-type: 'cublas'
+            cuda-major-version: "12"
+            cuda-minor-version: "0"
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-cublas-cuda12-extras'
+            ffmpeg: 'true'
+            image-type: 'extras'
+            runs-on: 'arc-runner-set'
+            base-image: "ubuntu:22.04"
+            aio: "-aio-gpu-nvidia-cuda-12"
+            latest-image: 'latest-gpu-nvidia-cuda-12-extras'
+            latest-image-aio: 'latest-aio-gpu-nvidia-cuda-12'
+            makeflags: "--jobs=3 --output-sync=target"
+          - build-type: 'sycl_f16'
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            base-image: "quay.io/go-skynet/intel-oneapi-base:latest"
+            grpc-base-image: "ubuntu:22.04"
+            tag-suffix: '-sycl-f16-extras'
+            ffmpeg: 'true'
+            image-type: 'extras'
+            runs-on: 'arc-runner-set'
+            aio: "-aio-gpu-intel-f16"
+            latest-image: 'latest-gpu-intel-f16-extras'
+            latest-image-aio: 'latest-aio-gpu-intel-f16'
+            makeflags: "--jobs=3 --output-sync=target"
+          - build-type: 'sycl_f32'
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            base-image: "quay.io/go-skynet/intel-oneapi-base:latest"
+            grpc-base-image: "ubuntu:22.04"
+            tag-suffix: '-sycl-f32-extras'
+            ffmpeg: 'true'
+            image-type: 'extras'
+            runs-on: 'arc-runner-set'
+            aio: "-aio-gpu-intel-f32"
+            latest-image: 'latest-gpu-intel-f32-extras'
+            latest-image-aio: 'latest-aio-gpu-intel-f32'
+            makeflags: "--jobs=3 --output-sync=target"
+          # Core images
+          - build-type: 'sycl_f16'
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            base-image: "quay.io/go-skynet/intel-oneapi-base:latest"
+            grpc-base-image: "ubuntu:22.04"
+            tag-suffix: '-sycl-f16'
+            ffmpeg: 'true'
+            image-type: 'core'
+            runs-on: 'arc-runner-set'
+            makeflags: "--jobs=3 --output-sync=target"
+            latest-image: 'latest-gpu-intel-f16'
+          - build-type: 'sycl_f32'
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            base-image: "quay.io/go-skynet/intel-oneapi-base:latest"
+            grpc-base-image: "ubuntu:22.04"
+            tag-suffix: '-sycl-f32'
+            ffmpeg: 'true'
+            image-type: 'core'
+            runs-on: 'arc-runner-set'
+            makeflags: "--jobs=3 --output-sync=target"
+            latest-image: 'latest-gpu-intel-f32'
 
   core-image-build:
     uses: ./.github/workflows/image_build.yml
     with:
       tag-latest: ${{ matrix.tag-latest }}
       tag-suffix: ${{ matrix.tag-suffix }}
+      ffmpeg: ${{ matrix.ffmpeg }}
+      image-type: ${{ matrix.image-type }}
       build-type: ${{ matrix.build-type }}
       cuda-major-version: ${{ matrix.cuda-major-version }}
       cuda-minor-version: ${{ matrix.cuda-minor-version }}
@@ -59,6 +189,8 @@ jobs:
       base-image: ${{ matrix.base-image }}
       grpc-base-image: ${{ matrix.grpc-base-image }}
       makeflags: ${{ matrix.makeflags }}
+      latest-image: ${{ matrix.latest-image }}
+      latest-image-aio: ${{ matrix.latest-image-aio }}
       skip-drivers: ${{ matrix.skip-drivers }}
     secrets:
       dockerUsername: ${{ secrets.DOCKERHUB_USERNAME }}
@@ -66,64 +198,66 @@ jobs:
       quayUsername: ${{ secrets.LOCALAI_REGISTRY_USERNAME }}
       quayPassword: ${{ secrets.LOCALAI_REGISTRY_PASSWORD }}
     strategy:
-      #max-parallel: ${{ github.event_name != 'pull_request' && 2 || 4 }}
+      max-parallel: ${{ github.event_name != 'pull_request' && 2 || 4 }}
       matrix:
         include:
           - build-type: ''
             platforms: 'linux/amd64,linux/arm64'
             tag-latest: 'auto'
             tag-suffix: ''
+            ffmpeg: 'true'
+            image-type: 'core'
             base-image: "ubuntu:22.04"
-            runs-on: 'ubuntu-latest'
+            runs-on: 'arc-runner-set'
             aio: "-aio-cpu"
+            latest-image: 'latest-cpu'
+            latest-image-aio: 'latest-aio-cpu'
             makeflags: "--jobs=4 --output-sync=target"
             skip-drivers: 'false'
           - build-type: 'cublas'
             cuda-major-version: "11"
             cuda-minor-version: "7"
             platforms: 'linux/amd64'
-            tag-latest: 'auto'
-            tag-suffix: '-gpu-nvidia-cuda-11'
-            runs-on: 'ubuntu-latest'
+            tag-latest: 'false'
+            tag-suffix: '-cublas-cuda11'
+            ffmpeg: 'true'
+            image-type: 'core'
+            runs-on: 'arc-runner-set'
             base-image: "ubuntu:22.04"
             makeflags: "--jobs=4 --output-sync=target"
             skip-drivers: 'false'
-            aio: "-aio-gpu-nvidia-cuda-11"
+            latest-image: 'latest-gpu-nvidia-cuda-12'
           - build-type: 'cublas'
             cuda-major-version: "12"
             cuda-minor-version: "0"
             platforms: 'linux/amd64'
-            tag-latest: 'auto'
-            tag-suffix: '-gpu-nvidia-cuda-12'
-            runs-on: 'ubuntu-latest'
+            tag-latest: 'false'
+            tag-suffix: '-cublas-cuda12'
+            ffmpeg: 'true'
+            image-type: 'core'
+            runs-on: 'arc-runner-set'
             base-image: "ubuntu:22.04"
             skip-drivers: 'false'
             makeflags: "--jobs=4 --output-sync=target"
-            aio: "-aio-gpu-nvidia-cuda-12"
+            latest-image: 'latest-gpu-nvidia-cuda-12'
           - build-type: 'vulkan'
             platforms: 'linux/amd64'
-            tag-latest: 'auto'
-            tag-suffix: '-gpu-vulkan'
-            runs-on: 'ubuntu-latest'
+            tag-latest: 'false'
+            tag-suffix: '-vulkan'
+            ffmpeg: 'true'
+            image-type: 'core'
+            runs-on: 'arc-runner-set'
             base-image: "ubuntu:22.04"
             skip-drivers: 'false'
             makeflags: "--jobs=4 --output-sync=target"
-            aio: "-aio-gpu-vulkan"
-          - build-type: 'intel'
-            platforms: 'linux/amd64'
-            tag-latest: 'auto'
-            base-image: "quay.io/go-skynet/intel-oneapi-base:latest"
-            grpc-base-image: "ubuntu:22.04"
-            tag-suffix: '-gpu-intel'
-            runs-on: 'ubuntu-latest'
-            makeflags: "--jobs=3 --output-sync=target"
-            aio: "-aio-gpu-intel"
-
+            latest-image: 'latest-gpu-vulkan'
   gh-runner:
     uses: ./.github/workflows/image_build.yml
     with:
       tag-latest: ${{ matrix.tag-latest }}
       tag-suffix: ${{ matrix.tag-suffix }}
+      ffmpeg: ${{ matrix.ffmpeg }}
+      image-type: ${{ matrix.image-type }}
       build-type: ${{ matrix.build-type }}
       cuda-major-version: ${{ matrix.cuda-major-version }}
       cuda-minor-version: ${{ matrix.cuda-minor-version }}
@@ -133,6 +267,8 @@ jobs:
       base-image: ${{ matrix.base-image }}
       grpc-base-image: ${{ matrix.grpc-base-image }}
       makeflags: ${{ matrix.makeflags }}
+      latest-image: ${{ matrix.latest-image }}
+      latest-image-aio: ${{ matrix.latest-image-aio }}
       skip-drivers: ${{ matrix.skip-drivers }}
     secrets:
       dockerUsername: ${{ secrets.DOCKERHUB_USERNAME }}
@@ -146,9 +282,12 @@ jobs:
             cuda-major-version: "12"
             cuda-minor-version: "0"
             platforms: 'linux/arm64'
-            tag-latest: 'auto'
+            tag-latest: 'false'
             tag-suffix: '-nvidia-l4t-arm64'
+            latest-image: 'latest-nvidia-l4t-arm64'
+            ffmpeg: 'true'
+            image-type: 'core'
             base-image: "nvcr.io/nvidia/l4t-jetpack:r36.4.0"
             runs-on: 'ubuntu-24.04-arm'
             makeflags: "--jobs=4 --output-sync=target"
-            skip-drivers: 'true'
+            skip-drivers: 'true'

.github/workflows/image_build.yml

@@ -33,14 +33,30 @@ on:
         description: 'Tag latest'
         default: ''
         type: string
+      latest-image:
+          description: 'Tag latest'
+          default: ''
+          type: string
+      latest-image-aio:
+          description: 'Tag latest'
+          default: ''
+          type: string
       tag-suffix:
         description: 'Tag suffix'
         default: ''
         type: string
+      ffmpeg:
+        description: 'FFMPEG'
+        default: ''
+        type: string
       skip-drivers:
         description: 'Skip drivers by default'
         default: 'false'
         type: string
+      image-type:
+        description: 'Image type'
+        default: ''
+        type: string
       runs-on:
         description: 'Runs on'
         required: true
@@ -69,22 +85,6 @@ jobs:
   reusable_image-build:
     runs-on: ${{ inputs.runs-on }}
     steps:
-
-      - name: Free Disk Space (Ubuntu)
-        if: inputs.runs-on == 'ubuntu-latest'
-        uses: jlumbroso/free-disk-space@main
-        with:
-          # this might remove tools that are actually needed,
-          # if set to "true" but frees about 6 GB
-          tool-cache: true
-          # all of these default to true, but feel free to set to
-          # "false" if necessary for your workflow
-          android: true
-          dotnet: true
-          haskell: true
-          large-packages: true
-          docker-images: true
-          swap-storage: true
       - name: Force Install GIT latest
         run: |
           sudo apt-get update \
@@ -94,7 +94,7 @@ jobs:
           && sudo apt-get update \
           && sudo apt-get install -y git
       - name: Checkout
-        uses: actions/checkout@v5
+        uses: actions/checkout@v4
 
       - name: Release space from worker
         if: inputs.runs-on == 'ubuntu-latest'
@@ -106,8 +106,8 @@ jobs:
           df -h
           echo
           sudo apt-get remove -y '^llvm-.*|^libllvm.*' || true
-          sudo apt-get remove --auto-remove android-sdk-platform-tools snapd || true
-          sudo apt-get purge --auto-remove android-sdk-platform-tools snapd || 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
@@ -152,18 +152,18 @@ jobs:
             type=sha
           flavor: |
             latest=${{ inputs.tag-latest }}
-            suffix=${{ inputs.tag-suffix }},onlatest=true
+            suffix=${{ inputs.tag-suffix }}
       - name: Docker meta for PR
         id: meta_pull_request
         if: github.event_name == 'pull_request'
         uses: docker/metadata-action@v5
         with:
           images: |
-            quay.io/go-skynet/ci-tests
+            ttl.sh/localai-ci-pr-${{ github.event.number }}
           tags: |
-            type=ref,event=branch,suffix=localai${{ github.event.number }}-${{ inputs.build-type }}-${{ inputs.cuda-major-version }}-${{ inputs.cuda-minor-version }}
-            type=semver,pattern={{raw}},suffix=localai${{ github.event.number }}-${{ inputs.build-type }}-${{ inputs.cuda-major-version }}-${{ inputs.cuda-minor-version }}
-            type=sha,suffix=localai${{ github.event.number }}-${{ inputs.build-type }}-${{ inputs.cuda-major-version }}-${{ inputs.cuda-minor-version }}
+            type=ref,event=branch
+            type=semver,pattern={{raw}}
+            type=sha
           flavor: |
             latest=${{ inputs.tag-latest }}
             suffix=${{ inputs.tag-suffix }}
@@ -179,7 +179,7 @@ jobs:
             type=semver,pattern={{raw}}
           flavor: |
             latest=${{ inputs.tag-latest }}
-            suffix=${{ inputs.aio }},onlatest=true
+            suffix=${{ inputs.aio }}
 
       - name: Docker meta AIO (dockerhub)
         if: inputs.aio != ''
@@ -192,8 +192,7 @@ jobs:
             type=ref,event=branch
             type=semver,pattern={{raw}}
           flavor: |
-            latest=${{ inputs.tag-latest }}
-            suffix=${{ inputs.aio }},onlatest=true
+            suffix=${{ inputs.aio }}
 
       - name: Set up QEMU
         uses: docker/setup-qemu-action@master
@@ -232,6 +231,8 @@ jobs:
             BUILD_TYPE=${{ inputs.build-type }}
             CUDA_MAJOR_VERSION=${{ inputs.cuda-major-version }}
             CUDA_MINOR_VERSION=${{ inputs.cuda-minor-version }}
+            FFMPEG=${{ inputs.ffmpeg }}
+            IMAGE_TYPE=${{ inputs.image-type }}
             BASE_IMAGE=${{ inputs.base-image }}
             GRPC_BASE_IMAGE=${{ inputs.grpc-base-image || inputs.base-image }}
             GRPC_MAKEFLAGS=--jobs=4 --output-sync=target
@@ -259,6 +260,8 @@ jobs:
             BUILD_TYPE=${{ inputs.build-type }}
             CUDA_MAJOR_VERSION=${{ inputs.cuda-major-version }}
             CUDA_MINOR_VERSION=${{ inputs.cuda-minor-version }}
+            FFMPEG=${{ inputs.ffmpeg }}
+            IMAGE_TYPE=${{ inputs.image-type }}
             BASE_IMAGE=${{ inputs.base-image }}
             GRPC_BASE_IMAGE=${{ inputs.grpc-base-image || inputs.base-image }}
             GRPC_MAKEFLAGS=--jobs=4 --output-sync=target
@@ -269,9 +272,13 @@ jobs:
           file: ./Dockerfile
           cache-from: type=gha
           platforms: ${{ inputs.platforms }}
-          #push: true
+          push: true
           tags: ${{ steps.meta_pull_request.outputs.tags }}
           labels: ${{ steps.meta_pull_request.outputs.labels }}
+      - name: Testing image
+        if: github.event_name == 'pull_request'
+        run: |
+          echo "Image is available at ttl.sh/localai-ci-pr-${{ github.event.number }}:${{ steps.meta_pull_request.outputs.version }}" >> $GITHUB_STEP_SUMMARY
 ## End testing image
       - name: Build and push AIO image
         if: inputs.aio != ''
@@ -303,6 +310,32 @@ jobs:
           tags: ${{ steps.meta_aio_dockerhub.outputs.tags }}
           labels: ${{ steps.meta_aio_dockerhub.outputs.labels }}
 
+      - name: Cleanup
+        run: |
+          docker builder prune -f
+          docker system prune --force --volumes --all
+
+      - name: Latest tag
+        # run this on branches, when it is a tag and there is a latest-image defined
+        if: github.event_name != 'pull_request' && inputs.latest-image != ''  && github.ref_type == 'tag'
+        run: |
+          docker pull localai/localai:${{ steps.meta.outputs.version }}
+          docker tag localai/localai:${{ steps.meta.outputs.version }} localai/localai:${{ inputs.latest-image }}
+          docker push localai/localai:${{ inputs.latest-image }}
+          docker pull quay.io/go-skynet/local-ai:${{ steps.meta.outputs.version }}
+          docker tag quay.io/go-skynet/local-ai:${{ steps.meta.outputs.version }} quay.io/go-skynet/local-ai:${{ inputs.latest-image }}
+          docker push quay.io/go-skynet/local-ai:${{ inputs.latest-image }}
+      - name: Latest AIO tag
+        # run this on branches, when it is a tag and there is a latest-image defined
+        if: github.event_name != 'pull_request' && inputs.latest-image-aio != ''  && github.ref_type == 'tag'
+        run: |
+          docker pull localai/localai:${{ steps.meta_aio_dockerhub.outputs.version }}
+          docker tag localai/localai:${{ steps.meta_aio_dockerhub.outputs.version }} localai/localai:${{ inputs.latest-image-aio }}
+          docker push localai/localai:${{ inputs.latest-image-aio }}
+          docker pull quay.io/go-skynet/local-ai:${{ steps.meta_aio.outputs.version }}
+          docker tag quay.io/go-skynet/local-ai:${{ steps.meta_aio.outputs.version }} quay.io/go-skynet/local-ai:${{ inputs.latest-image-aio }}
+          docker push quay.io/go-skynet/local-ai:${{ inputs.latest-image-aio }}
+
       - name: job summary
         run: |
           echo "Built image: ${{ steps.meta.outputs.labels }}" >> $GITHUB_STEP_SUMMARY

.github/workflows/labeler.yml

@@ -9,4 +9,4 @@ jobs:
       pull-requests: write
     runs-on: ubuntu-latest
     steps:
-    - uses: actions/labeler@v6
+    - uses: actions/labeler@v5

.github/workflows/localaibot_automerge.yml

@@ -6,15 +6,14 @@ permissions:
   contents: write
   pull-requests: write
   packages: read
-  issues: write # for Homebrew/actions/post-comment
-  actions: write # to dispatch publish workflow
+
 jobs:
   dependabot:
     runs-on: ubuntu-latest
     if: ${{ github.actor == 'localai-bot' }}
     steps:
       - name: Checkout repository
-        uses: actions/checkout@v5
+        uses: actions/checkout@v4
 
       - name: Approve a PR if not already approved
         run: |

.github/workflows/notify-models.yaml

@@ -11,14 +11,14 @@ jobs:
         MODEL_NAME: gemma-3-12b-it
     runs-on: ubuntu-latest
     steps:
-    - uses: actions/checkout@v5
+    - uses: actions/checkout@v4
       with:
         fetch-depth: 0 # needed to checkout all branches for this Action to work
     - uses: mudler/localai-github-action@v1
       with:
         model: 'gemma-3-12b-it' # Any from models.localai.io, or from huggingface.com with: "huggingface://<repository>/file"
         # Check the PR diff using the current branch and the base branch of the PR
-    - uses: GrantBirki/git-diff-action@v2.8.1
+    - uses: GrantBirki/git-diff-action@v2.8.0
       id: git-diff-action
       with:
             json_diff_file_output: diff.json
@@ -90,16 +90,16 @@ jobs:
         MODEL_NAME: gemma-3-12b-it
     runs-on: ubuntu-latest
     steps:
-    - uses: actions/checkout@v5
+    - uses: actions/checkout@v4
       with:
         fetch-depth: 0 # needed to checkout all branches for this Action to work
     - name: Start LocalAI
       run: |
         echo "Starting LocalAI..."
-        docker run -e -ti -d --name local-ai -p 8080:8080 localai/localai:master run --debug $MODEL_NAME
+        docker run -e -ti -d --name local-ai -p 8080:8080 localai/localai:master-ffmpeg-core run --debug $MODEL_NAME
         until [ "`docker inspect -f {{.State.Health.Status}} local-ai`" == "healthy" ]; do echo "Waiting for container to be ready";  docker logs --tail 10 local-ai; sleep 2; done
       # Check the PR diff using the current branch and the base branch of the PR
-    - uses: GrantBirki/git-diff-action@v2.8.1
+    - uses: GrantBirki/git-diff-action@v2.8.0
       id: git-diff-action
       with:
             json_diff_file_output: diff.json

.github/workflows/release.yaml

@@ -1,64 +1,324 @@
-name: goreleaser
+name: Build and Release
 
 on:
   push:
+    branches:
+      - master
     tags:
       - 'v*'
+  pull_request:
+
+env:
+  GRPC_VERSION: v1.65.0
+
+permissions:
+  contents: write
+
+concurrency:
+  group: ci-releases-${{ github.head_ref || github.ref }}-${{ github.repository }}
+  cancel-in-progress: true
 
 jobs:
-  goreleaser:
+
+  build-linux-arm:
     runs-on: ubuntu-latest
     steps:
-      - name: Checkout
-        uses: actions/checkout@v5
+      - name: Clone
+        uses: actions/checkout@v4
         with:
-          fetch-depth: 0
-      - name: Set up Go
-        uses: actions/setup-go@v5
+          submodules: true
+      - uses: actions/setup-go@v5
         with:
-          go-version: 1.23
-      - name: Run GoReleaser
-        uses: goreleaser/goreleaser-action@v6
-        with:
-          version: v2.11.0
-          args: release --clean
-        env:
-          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-  launcher-build-darwin:
-    runs-on: macos-latest
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v5
-        with:
-          fetch-depth: 0
-      - name: Set up Go
-        uses: actions/setup-go@v5
-        with:
-          go-version: 1.23
-      - name: Build launcher for macOS ARM64
-        run: |
-          make build-launcher-darwin
-      - name: Upload DMG to Release
-        uses: softprops/action-gh-release@v2
-        with:
-          files: ./dist/LocalAI.dmg
-  launcher-build-linux:
-    runs-on: ubuntu-latest
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v5
-        with:
-          fetch-depth: 0
-      - name: Set up Go
-        uses: actions/setup-go@v5
-        with:
-          go-version: 1.23
-      - name: Build launcher for Linux
+          go-version: '1.21.x'
+          cache: false
+      - name: Dependencies
         run: |
           sudo apt-get update
-          sudo apt-get install golang gcc libgl1-mesa-dev xorg-dev libxkbcommon-dev
-          make build-launcher-linux
-      - name: Upload Linux launcher artifacts
-        uses: softprops/action-gh-release@v2
+          sudo apt-get install build-essential ffmpeg protobuf-compiler ccache upx-ucl gawk
+          sudo apt-get install -qy binutils-aarch64-linux-gnu gcc-aarch64-linux-gnu g++-aarch64-linux-gnu libgmock-dev
+          make install-go-tools
+      - name: Install CUDA Dependencies
+        run: |
+          curl -O https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/cross-linux-aarch64/cuda-keyring_1.1-1_all.deb
+          sudo dpkg -i cuda-keyring_1.1-1_all.deb
+          sudo apt-get update
+          sudo apt-get install -y cuda-cross-aarch64 cuda-nvcc-cross-aarch64-${CUDA_VERSION} libcublas-cross-aarch64-${CUDA_VERSION}
+        env:
+          CUDA_VERSION: 12-4
+      - name: Cache grpc
+        id: cache-grpc
+        uses: actions/cache@v4
         with:
-          files: ./local-ai-launcher-linux.tar.xz
+          path: grpc
+          key: ${{ runner.os }}-arm-grpc-${{ env.GRPC_VERSION }}
+      - name: Build grpc
+        if: steps.cache-grpc.outputs.cache-hit != 'true'
+        run: |
+
+          git clone --recurse-submodules -b ${{ env.GRPC_VERSION }} --depth 1 --shallow-submodules https://github.com/grpc/grpc && \
+          cd grpc && sed -i "216i\  TESTONLY" "third_party/abseil-cpp/absl/container/CMakeLists.txt" && mkdir -p cmake/build && \
+          cd cmake/build && cmake -DgRPC_INSTALL=ON \
+            -DgRPC_BUILD_TESTS=OFF \
+            ../.. && sudo make --jobs 5 --output-sync=target
+      - name: Install gRPC
+        run: |
+          GNU_HOST=aarch64-linux-gnu
+          C_COMPILER_ARM_LINUX=$GNU_HOST-gcc
+          CXX_COMPILER_ARM_LINUX=$GNU_HOST-g++
+
+          CROSS_TOOLCHAIN=/usr/$GNU_HOST
+          CROSS_STAGING_PREFIX=$CROSS_TOOLCHAIN/stage
+          CMAKE_CROSS_TOOLCHAIN=/tmp/arm.toolchain.cmake
+
+          # https://cmake.org/cmake/help/v3.13/manual/cmake-toolchains.7.html#cross-compiling-for-linux
+          echo "set(CMAKE_SYSTEM_NAME Linux)" >> $CMAKE_CROSS_TOOLCHAIN && \
+            echo "set(CMAKE_SYSTEM_PROCESSOR arm)" >> $CMAKE_CROSS_TOOLCHAIN && \
+            echo "set(CMAKE_STAGING_PREFIX $CROSS_STAGING_PREFIX)" >> $CMAKE_CROSS_TOOLCHAIN && \
+            echo "set(CMAKE_SYSROOT ${CROSS_TOOLCHAIN}/sysroot)" >> $CMAKE_CROSS_TOOLCHAIN && \
+            echo "set(CMAKE_C_COMPILER /usr/bin/$C_COMPILER_ARM_LINUX)" >> $CMAKE_CROSS_TOOLCHAIN && \
+            echo "set(CMAKE_CXX_COMPILER /usr/bin/$CXX_COMPILER_ARM_LINUX)" >> $CMAKE_CROSS_TOOLCHAIN && \
+            echo "set(CMAKE_FIND_ROOT_PATH_MODE_PROGRAM NEVER)" >> $CMAKE_CROSS_TOOLCHAIN && \
+            echo "set(CMAKE_FIND_ROOT_PATH_MODE_LIBRARY ONLY)" >> $CMAKE_CROSS_TOOLCHAIN && \
+            echo "set(CMAKE_FIND_ROOT_PATH_MODE_INCLUDE ONLY)" >> $CMAKE_CROSS_TOOLCHAIN && \
+            echo "set(CMAKE_FIND_ROOT_PATH_MODE_PACKAGE ONLY)" >> $CMAKE_CROSS_TOOLCHAIN
+          GRPC_DIR=$PWD/grpc
+          cd grpc && cd cmake/build && sudo make --jobs 5 --output-sync=target install && \
+          GRPC_CROSS_BUILD_DIR=$GRPC_DIR/cmake/cross_build && \
+          mkdir -p $GRPC_CROSS_BUILD_DIR && \
+          cd $GRPC_CROSS_BUILD_DIR && \
+          cmake -DCMAKE_TOOLCHAIN_FILE=$CMAKE_CROSS_TOOLCHAIN \
+            -DCMAKE_BUILD_TYPE=Release \
+            -DCMAKE_INSTALL_PREFIX=$CROSS_TOOLCHAIN/grpc_install \
+            ../.. && \
+          sudo make -j`nproc` install
+      - name: Build
+        id: build
+        run: |
+          GNU_HOST=aarch64-linux-gnu
+          C_COMPILER_ARM_LINUX=$GNU_HOST-gcc
+          CXX_COMPILER_ARM_LINUX=$GNU_HOST-g++
+
+          CROSS_TOOLCHAIN=/usr/$GNU_HOST
+          CROSS_STAGING_PREFIX=$CROSS_TOOLCHAIN/stage
+          CMAKE_CROSS_TOOLCHAIN=/tmp/arm.toolchain.cmake
+          go install google.golang.org/grpc/cmd/protoc-gen-go-grpc@1958fcbe2ca8bd93af633f11e97d44e567e945af
+          go install google.golang.org/protobuf/cmd/protoc-gen-go@v1.34.2
+          export PATH=$PATH:$GOPATH/bin
+          export PATH=/usr/local/cuda/bin:$PATH
+          sudo rm -rf /usr/aarch64-linux-gnu/lib/libstdc++.so.6
+          sudo cp -rf /usr/aarch64-linux-gnu/lib/libstdc++.so* /usr/aarch64-linux-gnu/lib/libstdc++.so.6
+          sudo cp /usr/aarch64-linux-gnu/lib/ld-linux-aarch64.so.1 ld.so
+          BACKEND_LIBS="./grpc/cmake/cross_build/third_party/re2/libre2.a ./grpc/cmake/cross_build/libgrpc.a ./grpc/cmake/cross_build/libgrpc++.a ./grpc/cmake/cross_build/third_party/protobuf/libprotobuf.a /usr/aarch64-linux-gnu/lib/libc.so.6 /usr/aarch64-linux-gnu/lib/libstdc++.so.6 /usr/aarch64-linux-gnu/lib/libgomp.so.1 /usr/aarch64-linux-gnu/lib/libm.so.6 /usr/aarch64-linux-gnu/lib/libgcc_s.so.1 /usr/aarch64-linux-gnu/lib/libdl.so.2 /usr/aarch64-linux-gnu/lib/libpthread.so.0 ./ld.so" \
+          GOOS=linux \
+          GOARCH=arm64 \
+          CMAKE_ARGS="-DProtobuf_INCLUDE_DIRS=$CROSS_STAGING_PREFIX/include -DProtobuf_DIR=$CROSS_STAGING_PREFIX/lib/cmake/protobuf -DgRPC_DIR=$CROSS_STAGING_PREFIX/lib/cmake/grpc -DCMAKE_TOOLCHAIN_FILE=$CMAKE_CROSS_TOOLCHAIN -DCMAKE_C_COMPILER=aarch64-linux-gnu-gcc -DCMAKE_CXX_COMPILER=aarch64-linux-gnu-g++" make dist-cross-linux-arm64
+      - uses: actions/upload-artifact@v4
+        with:
+          name: LocalAI-linux-arm64
+          path: release/
+      - name: Release
+        uses: softprops/action-gh-release@v2
+        if: startsWith(github.ref, 'refs/tags/')
+        with:
+          files: |
+            release/*
+      - name: Setup tmate session if tests fail
+        if: ${{ failure() }}
+        uses: mxschmitt/action-tmate@v3.22
+        with:
+          detached: true
+          connect-timeout-seconds: 180
+          limit-access-to-actor: true
+  build-linux:
+    runs-on: arc-runner-set
+    steps:
+      - name: Force Install GIT latest
+        run: |
+          sudo apt-get update \
+          && sudo apt-get install -y software-properties-common \
+          && sudo apt-get update \
+          && sudo add-apt-repository -y ppa:git-core/ppa \
+          && sudo apt-get update \
+          && sudo apt-get install -y git
+      - name: Clone
+        uses: actions/checkout@v4
+        with:
+          submodules: true
+      - uses: actions/setup-go@v5
+        with:
+          go-version: '1.21.x'
+          cache: false
+      - name: Dependencies
+        run: |
+          sudo apt-get update
+          sudo apt-get install -y wget curl build-essential ffmpeg protobuf-compiler ccache upx-ucl gawk cmake libgmock-dev
+          make install-go-tools
+      - name: Intel Dependencies
+        run: |
+          wget -O- https://apt.repos.intel.com/intel-gpg-keys/GPG-PUB-KEY-INTEL-SW-PRODUCTS.PUB | gpg --dearmor | sudo tee /usr/share/keyrings/oneapi-archive-keyring.gpg > /dev/null
+          echo "deb [signed-by=/usr/share/keyrings/oneapi-archive-keyring.gpg] https://apt.repos.intel.com/oneapi all main" | sudo tee /etc/apt/sources.list.d/oneAPI.list
+          sudo apt update
+          sudo apt install -y intel-basekit
+      - name: Install CUDA Dependencies
+        run: |
+          curl -O https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/cuda-keyring_1.1-1_all.deb
+          sudo dpkg -i cuda-keyring_1.1-1_all.deb
+          sudo apt-get update
+          sudo apt-get install -y cuda-nvcc-${CUDA_VERSION} libcublas-dev-${CUDA_VERSION}
+        env:
+          CUDA_VERSION: 12-5
+      - name: "Install Hipblas"
+        env:
+          ROCM_VERSION: "6.1"
+          AMDGPU_VERSION: "6.1"
+        run: |
+            set -ex
+
+            sudo apt-get update
+            sudo DEBIAN_FRONTEND=noninteractive apt-get install -y --no-install-recommends ca-certificates curl libnuma-dev gnupg
+
+            curl -sL https://repo.radeon.com/rocm/rocm.gpg.key | sudo apt-key add -
+
+            printf "deb [arch=amd64] https://repo.radeon.com/rocm/apt/$ROCM_VERSION/ jammy main" | sudo tee /etc/apt/sources.list.d/rocm.list
+
+            printf "deb [arch=amd64] https://repo.radeon.com/amdgpu/$AMDGPU_VERSION/ubuntu jammy main" | sudo tee /etc/apt/sources.list.d/amdgpu.list
+            printf 'Package: *\nPin: release o=repo.radeon.com\nPin-Priority: 600' | sudo tee /etc/apt/preferences.d/rocm-pin-600
+            sudo apt-get update
+
+            sudo DEBIAN_FRONTEND=noninteractive apt-get install -y \
+                hipblas-dev rocm-dev \
+                rocblas-dev
+
+            sudo apt-get clean
+            sudo rm -rf /var/lib/apt/lists/*
+            sudo ldconfig
+      - name: Cache grpc
+        id: cache-grpc
+        uses: actions/cache@v4
+        with:
+          path: grpc
+          key: ${{ runner.os }}-grpc-${{ env.GRPC_VERSION }}
+      - name: Build grpc
+        if: steps.cache-grpc.outputs.cache-hit != 'true'
+        run: |
+          git clone --recurse-submodules -b ${{ env.GRPC_VERSION }} --depth 1 --shallow-submodules https://github.com/grpc/grpc && \
+          cd grpc && sed -i "216i\  TESTONLY" "third_party/abseil-cpp/absl/container/CMakeLists.txt" && mkdir -p cmake/build && \
+          cd cmake/build && cmake -DgRPC_INSTALL=ON \
+            -DgRPC_BUILD_TESTS=OFF \
+            ../.. && sudo make --jobs 5 --output-sync=target
+      - name: Install gRPC
+        run: |
+          cd grpc && cd cmake/build && sudo make --jobs 5 --output-sync=target install
+      # BACKEND_LIBS needed for gpu-workload: /opt/intel/oneapi/*/lib/libiomp5.so /opt/intel/oneapi/*/lib/libmkl_core.so /opt/intel/oneapi/*/lib/libmkl_core.so.2 /opt/intel/oneapi/*/lib/libmkl_intel_ilp64.so /opt/intel/oneapi/*/lib/libmkl_intel_ilp64.so.2 /opt/intel/oneapi/*/lib/libmkl_sycl_blas.so /opt/intel/oneapi/*/lib/libmkl_sycl_blas.so.4 /opt/intel/oneapi/*/lib/libmkl_tbb_thread.so /opt/intel/oneapi/*/lib/libmkl_tbb_thread.so.2 /opt/intel/oneapi/*/lib/libsycl.so /opt/intel/oneapi/*/lib/libsycl.so.7 /opt/intel/oneapi/*/lib/libsycl.so.7.1.0 /opt/rocm-*/lib/libamdhip64.so /opt/rocm-*/lib/libamdhip64.so.5 /opt/rocm-*/lib/libamdhip64.so.6 /opt/rocm-*/lib/libamdhip64.so.6.1.60100 /opt/rocm-*/lib/libhipblas.so /opt/rocm-*/lib/libhipblas.so.2 /opt/rocm-*/lib/libhipblas.so.2.1.60100 /opt/rocm-*/lib/librocblas.so /opt/rocm-*/lib/librocblas.so.4 /opt/rocm-*/lib/librocblas.so.4.1.60100 /usr/lib/x86_64-linux-gnu/libstdc++.so.6 /usr/lib/x86_64-linux-gnu/libOpenCL.so.1 /usr/lib/x86_64-linux-gnu/libOpenCL.so.1.0.0 /usr/lib/x86_64-linux-gnu/libm.so.6 /usr/lib/x86_64-linux-gnu/libgcc_s.so.1 /usr/lib/x86_64-linux-gnu/libc.so.6 /usr/lib/x86_64-linux-gnu/librt.so.1 /usr/local/cuda-*/targets/x86_64-linux/lib/libcublas.so /usr/local/cuda-*/targets/x86_64-linux/lib/libcublasLt.so /usr/local/cuda-*/targets/x86_64-linux/lib/libcudart.so /usr/local/cuda-*/targets/x86_64-linux/lib/stubs/libcuda.so
+      - name: Build
+        id: build
+        run: |
+          go install google.golang.org/grpc/cmd/protoc-gen-go-grpc@1958fcbe2ca8bd93af633f11e97d44e567e945af
+          go install google.golang.org/protobuf/cmd/protoc-gen-go@v1.34.2
+          export PATH=$PATH:$GOPATH/bin
+          export PATH=/usr/local/cuda/bin:$PATH
+          export PATH=/opt/rocm/bin:$PATH
+          source /opt/intel/oneapi/setvars.sh
+          sudo cp /lib64/ld-linux-x86-64.so.2 ld.so
+          BACKEND_LIBS="./ld.so ./sources/go-piper/piper/build/fi/lib/libfmt.a ./sources/go-piper/piper-phonemize/pi/lib/libonnxruntime.so.1.14.1 ./sources/go-piper/piper-phonemize/pi/src/libespeak-ng/libespeak-ng.so /usr/lib/x86_64-linux-gnu/libdl.so.2 /usr/lib/x86_64-linux-gnu/librt.so.1 /usr/lib/x86_64-linux-gnu/libpthread.so.0 ./sources/go-piper/piper-phonemize/pi/lib/libpiper_phonemize.so.1 ./sources/go-piper/piper/build/si/lib/libspdlog.a ./sources/go-piper/espeak/ei/lib/libucd.so" \
+          make -j4 dist
+      - uses: actions/upload-artifact@v4
+        with:
+          name: LocalAI-linux
+          path: release/
+      - name: Release
+        uses: softprops/action-gh-release@v2
+        if: startsWith(github.ref, 'refs/tags/')
+        with:
+          files: |
+            release/*
+      - name: Setup tmate session if tests fail
+        if: ${{ failure() }}
+        uses: mxschmitt/action-tmate@v3.22
+        with:
+          detached: true
+          connect-timeout-seconds: 180
+          limit-access-to-actor: true
+
+
+  build-macOS-x86_64:
+    runs-on: macos-13
+    steps:
+      - name: Clone
+        uses: actions/checkout@v4
+        with:
+          submodules: true
+      - uses: actions/setup-go@v5
+        with:
+          go-version: '1.21.x'
+          cache: false
+      - name: Dependencies
+        run: |
+          brew install protobuf grpc
+          make install-go-tools
+      - name: Build
+        id: build
+        run: |
+          export C_INCLUDE_PATH=/usr/local/include
+          export CPLUS_INCLUDE_PATH=/usr/local/include
+          export PATH=$PATH:$GOPATH/bin
+          export SKIP_GRPC_BACKEND=backend-assets/grpc/whisper
+          make dist
+      - uses: actions/upload-artifact@v4
+        with:
+          name: LocalAI-MacOS-x86_64
+          path: release/
+      - name: Release
+        uses: softprops/action-gh-release@v2
+        if: startsWith(github.ref, 'refs/tags/')
+        with:
+          files: |
+            release/*
+      - name: Setup tmate session if tests fail
+        if: ${{ failure() }}
+        uses: mxschmitt/action-tmate@v3.22
+        with:
+          detached: true
+          connect-timeout-seconds: 180
+          limit-access-to-actor: true
+
+  build-macOS-arm64:
+    runs-on: macos-14
+    steps:
+      - name: Clone
+        uses: actions/checkout@v4
+        with:
+          submodules: true
+      - uses: actions/setup-go@v5
+        with:
+          go-version: '1.21.x'
+          cache: false
+      - name: Dependencies
+        run: |
+          brew install protobuf grpc libomp llvm
+          make install-go-tools
+      - name: Build
+        id: build
+        run: |
+          export C_INCLUDE_PATH=/usr/local/include
+          export CPLUS_INCLUDE_PATH=/usr/local/include
+          export PATH=$PATH:$GOPATH/bin
+          export CC=/opt/homebrew/opt/llvm/bin/clang
+          make dist
+      - uses: actions/upload-artifact@v4
+        with:
+          name: LocalAI-MacOS-arm64
+          path: release/
+      - name: Release
+        uses: softprops/action-gh-release@v2
+        if: startsWith(github.ref, 'refs/tags/')
+        with:
+          files: |
+            release/*
+      - name: Setup tmate session if tests fail
+        if: ${{ failure() }}
+        uses: mxschmitt/action-tmate@v3.22
+        with:
+          detached: true
+          connect-timeout-seconds: 180
+          limit-access-to-actor: true

.github/workflows/secscan.yaml

@@ -14,11 +14,11 @@ jobs:
       GO111MODULE: on
     steps:
       - name: Checkout Source
-        uses: actions/checkout@v5
+        uses: actions/checkout@v4
         if: ${{ github.actor != 'dependabot[bot]' }}
       - name: Run Gosec Security Scanner
         if: ${{ github.actor != 'dependabot[bot]' }}
-        uses: securego/gosec@v2.22.9
+        uses: securego/gosec@v2.22.4
         with:
           # we let the report trigger content trigger a failure using the GitHub Security features.
           args: '-no-fail -fmt sarif -out results.sarif ./...'

.github/workflows/stalebot.yml

@@ -1,24 +0,0 @@
-name: 'Close stale issues and PRs'
-permissions:
-  issues: write
-  pull-requests: write
-on:
-  schedule:
-    - cron: '30 1 * * *'
-
-jobs:
-  stale:
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/stale@3a9db7e6a41a89f618792c92c0e97cc736e1b13f # v9
-        with:
-          stale-issue-message: 'This issue is stale because it has been open 90 days with no activity. Remove stale label or comment or this will be closed in 5 days.'
-          stale-pr-message: 'This PR is stale because it has been open 90 days with no activity. Remove stale label or comment or this will be closed in 10 days.'
-          close-issue-message: 'This issue was closed because it has been stalled for 5 days with no activity.'
-          close-pr-message: 'This PR was closed because it has been stalled for 10 days with no activity.'
-          days-before-issue-stale: 90
-          days-before-pr-stale: 90
-          days-before-issue-close: 5
-          days-before-pr-close: 10
-          exempt-issue-labels: 'roadmap'
-          exempt-pr-labels: 'roadmap'

.github/workflows/test-extra.yml

@@ -14,33 +14,11 @@ concurrency:
   cancel-in-progress: true
 
 jobs:
-  # Requires CUDA
-  # tests-chatterbox-tts:
-  #   runs-on: ubuntu-latest
-  #   steps:
-  #     - name: Clone
-  #       uses: actions/checkout@v5
-  #       with:
-  #         submodules: true
-  #     - name: Dependencies
-  #       run: |
-  #         sudo apt-get update
-  #         sudo apt-get install build-essential ffmpeg
-  #         # Install UV
-  #         curl -LsSf https://astral.sh/uv/install.sh | sh
-  #         sudo apt-get install -y ca-certificates cmake curl patch python3-pip
-  #         sudo apt-get install -y libopencv-dev
-  #         pip install --user --no-cache-dir grpcio-tools==1.64.1
-
-  #     - name: Test chatterbox-tts
-  #       run: |
-  #          make --jobs=5 --output-sync=target -C backend/python/chatterbox
-  #          make --jobs=5 --output-sync=target -C backend/python/chatterbox test
   tests-transformers:
     runs-on: ubuntu-latest
     steps:
       - name: Clone
-        uses: actions/checkout@v5
+        uses: actions/checkout@v4
         with:
           submodules: true
       - name: Dependencies
@@ -61,7 +39,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Clone
-        uses: actions/checkout@v5
+        uses: actions/checkout@v4
         with:
           submodules: true
       - name: Dependencies
@@ -83,7 +61,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Clone
-        uses: actions/checkout@v5
+        uses: actions/checkout@v4
         with:
           submodules: true
       - name: Dependencies
@@ -104,7 +82,7 @@ jobs:
   #  runs-on: ubuntu-latest
   #  steps:
   #    - name: Clone
-  #      uses: actions/checkout@v5
+  #      uses: actions/checkout@v4
   #      with:
   #        submodules: true
   #    - name: Dependencies
@@ -124,7 +102,7 @@ jobs:
   #   runs-on: ubuntu-latest
   #   steps:
   #     - name: Clone
-  #       uses: actions/checkout@v5
+  #       uses: actions/checkout@v4
   #       with:
   #         submodules: true
   #     - name: Dependencies
@@ -186,7 +164,7 @@ jobs:
   #           sudo rm -rf "$AGENT_TOOLSDIRECTORY" || true
   #           df -h
   #     - name: Clone
-  #       uses: actions/checkout@v5
+  #       uses: actions/checkout@v4
   #       with:
   #         submodules: true
   #     - name: Dependencies
@@ -211,7 +189,7 @@ jobs:
   #   runs-on: ubuntu-latest
   #   steps:
   #     - name: Clone
-  #       uses: actions/checkout@v5
+  #       uses: actions/checkout@v4
   #       with:
   #         submodules: true
   #     - name: Dependencies
@@ -232,7 +210,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Clone
-        uses: actions/checkout@v5
+        uses: actions/checkout@v4
         with:
           submodules: true
       - name: Dependencies

.github/workflows/test.yml

@@ -23,20 +23,6 @@ jobs:
       matrix:
         go-version: ['1.21.x']
     steps:
-      - name: Free Disk Space (Ubuntu)
-        uses: jlumbroso/free-disk-space@main
-        with:
-          # this might remove tools that are actually needed,
-          # if set to "true" but frees about 6 GB
-          tool-cache: true
-          # all of these default to true, but feel free to set to
-          # "false" if necessary for your workflow
-          android: true
-          dotnet: true
-          haskell: true
-          large-packages: true
-          docker-images: true
-          swap-storage: true
       - name: Release space from worker
         run: |
           echo "Listing top largest packages"
@@ -70,7 +56,7 @@ jobs:
           sudo rm -rfv build || true
           df -h
       - name: Clone
-        uses: actions/checkout@v5
+        uses: actions/checkout@v4
         with:
           submodules: true
       - name: Setup Go ${{ matrix.go-version }}
@@ -81,20 +67,18 @@ jobs:
       # You can test your matrix by printing the current Go version
       - name: Display Go version
         run: go version
-      - name: Proto Dependencies
-        run: |
-          # Install protoc
-          curl -L -s https://github.com/protocolbuffers/protobuf/releases/download/v26.1/protoc-26.1-linux-x86_64.zip -o protoc.zip && \
-          unzip -j -d /usr/local/bin protoc.zip bin/protoc && \
-          rm protoc.zip
-          go install google.golang.org/protobuf/cmd/protoc-gen-go@v1.34.2
-          go install google.golang.org/grpc/cmd/protoc-gen-go-grpc@1958fcbe2ca8bd93af633f11e97d44e567e945af
-          PATH="$PATH:$HOME/go/bin" make protogen-go
       - name: Dependencies
         run: |
           sudo apt-get update
           sudo apt-get install build-essential ccache upx-ucl curl ffmpeg
           sudo apt-get install -y libgmock-dev clang
+          curl https://repo.anaconda.com/pkgs/misc/gpgkeys/anaconda.asc | gpg --dearmor > conda.gpg && \
+             sudo install -o root -g root -m 644 conda.gpg /usr/share/keyrings/conda-archive-keyring.gpg && \
+             gpg --keyring /usr/share/keyrings/conda-archive-keyring.gpg --no-default-keyring --fingerprint 34161F5BF5EB1D4BFBBB8F0A8AEB4F8B29D82806 && \
+             sudo /bin/bash -c 'echo "deb [arch=amd64 signed-by=/usr/share/keyrings/conda-archive-keyring.gpg] https://repo.anaconda.com/pkgs/misc/debrepo/conda stable main" > /etc/apt/sources.list.d/conda.list' && \
+             sudo /bin/bash -c 'echo "deb [arch=amd64 signed-by=/usr/share/keyrings/conda-archive-keyring.gpg] https://repo.anaconda.com/pkgs/misc/debrepo/conda stable main" | tee -a /etc/apt/sources.list.d/conda.list' && \
+             sudo apt-get update && \
+             sudo apt-get install -y conda
           # Install UV
           curl -LsSf https://astral.sh/uv/install.sh | sh
           sudo apt-get install -y ca-certificates cmake patch python3-pip unzip
@@ -110,15 +94,38 @@ jobs:
           sudo apt-get install -y cuda-nvcc-${CUDA_VERSION} libcublas-dev-${CUDA_VERSION}
           export CUDACXX=/usr/local/cuda/bin/nvcc
 
+          go install google.golang.org/protobuf/cmd/protoc-gen-go@v1.34.2
+          go install google.golang.org/grpc/cmd/protoc-gen-go-grpc@1958fcbe2ca8bd93af633f11e97d44e567e945af
+          go install github.com/GeertJohan/go.rice/rice@latest
 
           # The python3-grpc-tools package in 22.04 is too old
-          pip install --user grpcio-tools==1.71.0 grpcio==1.71.0
+          pip install --user grpcio-tools
 
           make -C backend/python/transformers
 
-          make backends/huggingface backends/llama-cpp backends/local-store backends/silero-vad backends/piper backends/whisper backends/stablediffusion-ggml
+          # Pre-build piper before we start tests in order to have shared libraries in place
+          make sources/go-piper && \
+          GO_TAGS="tts" make -C sources/go-piper piper.o && \
+          sudo cp -rfv sources/go-piper/piper-phonemize/pi/lib/. /usr/lib/
         env:
           CUDA_VERSION: 12-4
+      - name: Cache grpc
+        id: cache-grpc
+        uses: actions/cache@v4
+        with:
+          path: grpc
+          key: ${{ runner.os }}-grpc-${{ env.GRPC_VERSION }}
+      - name: Build grpc
+        if: steps.cache-grpc.outputs.cache-hit != 'true'
+        run: |
+          git clone --recurse-submodules -b ${{ env.GRPC_VERSION }} --depth 1 --jobs 5 --shallow-submodules https://github.com/grpc/grpc && \
+          cd grpc && sed -i "216i\  TESTONLY" "third_party/abseil-cpp/absl/container/CMakeLists.txt" && mkdir -p cmake/build && cd cmake/build && \
+          cmake -DgRPC_INSTALL=ON \
+            -DgRPC_BUILD_TESTS=OFF \
+            ../.. && sudo make --jobs 5
+      - name: Install gRPC
+        run: |
+          cd grpc && cd cmake/build && sudo make --jobs 5 install
       - name: Test
         run: |
           PATH="$PATH:/root/go/bin" GO_TAGS="tts" make --jobs 5 --output-sync=target test
@@ -166,7 +173,7 @@ jobs:
           sudo rm -rfv build || true
           df -h
       - name: Clone
-        uses: actions/checkout@v5
+        uses: actions/checkout@v4
         with:
           submodules: true
       - name: Dependencies
@@ -177,10 +184,16 @@ jobs:
           rm protoc.zip
           go install google.golang.org/protobuf/cmd/protoc-gen-go@v1.34.2
           go install google.golang.org/grpc/cmd/protoc-gen-go-grpc@1958fcbe2ca8bd93af633f11e97d44e567e945af
+          go install github.com/GeertJohan/go.rice/rice@latest
           PATH="$PATH:$HOME/go/bin" make protogen-go
+      - name: Build images
+        run: |
+          docker build --build-arg FFMPEG=true --build-arg IMAGE_TYPE=extras --build-arg EXTRA_BACKENDS=rerankers --build-arg MAKEFLAGS="--jobs=5 --output-sync=target" -t local-ai:tests -f Dockerfile .
+          BASE_IMAGE=local-ai:tests DOCKER_AIO_IMAGE=local-ai-aio:test make docker-aio
       - name: Test
         run: |
-            PATH="$PATH:$HOME/go/bin" make backends/local-store backends/silero-vad backends/llama-cpp backends/whisper backends/piper backends/stablediffusion-ggml docker-build-aio e2e-aio
+            PATH="$PATH:$HOME/go/bin" LOCALAI_MODELS_DIR=$PWD/models LOCALAI_IMAGE_TAG=test LOCALAI_IMAGE=local-ai-aio \
+            make run-e2e-aio
       - name: Setup tmate session if tests fail
         if: ${{ failure() }}
         uses: mxschmitt/action-tmate@v3.22
@@ -196,7 +209,7 @@ jobs:
         go-version: ['1.21.x']
     steps:
       - name: Clone
-        uses: actions/checkout@v5
+        uses: actions/checkout@v4
         with:
           submodules: true
       - name: Setup Go ${{ matrix.go-version }}
@@ -210,11 +223,8 @@ jobs:
       - name: Dependencies
         run: |
           brew install protobuf grpc make protoc-gen-go protoc-gen-go-grpc libomp llvm
-          pip install --user --no-cache-dir grpcio-tools==1.71.0 grpcio==1.71.0
-      - name: Build llama-cpp-darwin
-        run: |
-          make protogen-go
-          make backends/llama-cpp-darwin
+          pip install --user --no-cache-dir grpcio-tools
+          go install github.com/GeertJohan/go.rice/rice@latest
       - name: Test
         run: |
           export C_INCLUDE_PATH=/usr/local/include
@@ -222,8 +232,7 @@ jobs:
           export CC=/opt/homebrew/opt/llvm/bin/clang
           # Used to run the newer GNUMake version from brew that supports --output-sync
           export PATH="/opt/homebrew/opt/make/libexec/gnubin:$PATH"
-          PATH="$PATH:$HOME/go/bin" make protogen-go
-          PATH="$PATH:$HOME/go/bin" BUILD_TYPE="GITHUB_CI_HAS_BROKEN_METAL" CMAKE_ARGS="-DGGML_F16C=OFF -DGGML_AVX512=OFF -DGGML_AVX2=OFF -DGGML_FMA=OFF" make --jobs 4 --output-sync=target test
+          BUILD_TYPE="GITHUB_CI_HAS_BROKEN_METAL" CMAKE_ARGS="-DGGML_F16C=OFF -DGGML_AVX512=OFF -DGGML_AVX2=OFF -DGGML_FMA=OFF" make --jobs 4 --output-sync=target test
       - name: Setup tmate session if tests fail
         if: ${{ failure() }}
         uses: mxschmitt/action-tmate@v3.22

.github/workflows/update_swagger.yaml

@@ -9,7 +9,7 @@ jobs:
       fail-fast: false
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v5
+      - uses: actions/checkout@v4
       - uses: actions/setup-go@v5
         with:
           go-version: 'stable'

.github/workflows/yaml-check.yml

@@ -8,7 +8,7 @@ jobs:
     steps:
       - name: 'Checkout'
         uses: actions/checkout@master
-      - name: 'Yamllint model gallery'
+      - name: 'Yamllint'
         uses: karancode/yamllint-github-action@master
         with:
           yamllint_file_or_dir: 'gallery'
@@ -16,11 +16,3 @@ jobs:
           yamllint_comment: true
         env:
           GITHUB_ACCESS_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-      - name: 'Yamllint Backend gallery'
-        uses: karancode/yamllint-github-action@master
-        with:
-          yamllint_file_or_dir: 'backend'
-          yamllint_strict: false
-          yamllint_comment: true
-        env:
-          GITHUB_ACCESS_TOKEN: ${{ secrets.GITHUB_TOKEN }}

.gitignore

@@ -5,14 +5,9 @@ __pycache__/
 *.o
 get-sources
 prepare-sources
-/backend/cpp/llama-cpp/grpc-server
-/backend/cpp/llama-cpp/llama.cpp
+/backend/cpp/llama/grpc-server
+/backend/cpp/llama/llama.cpp
 /backend/cpp/llama-*
-!backend/cpp/llama-cpp
-/backends
-/backend-images
-/result.yaml
-protoc
 
 *.log
 
@@ -24,7 +19,7 @@ go-bert
 
 # LocalAI build binary
 LocalAI
-/local-ai
+local-ai
 # prevent above rules from omitting the helm chart
 !charts/*
 # prevent above rules from omitting the api/localai folder
@@ -61,4 +56,4 @@ docs/static/gallery.html
 **/venv
 
 # per-developer customization files for the development container
-.devcontainer/customization/*
+.devcontainer/customization/*

.goreleaser.yaml

@@ -1,33 +0,0 @@
-version: 2
-before:
-  hooks:
-    - make protogen-go
-    - go mod tidy
-dist: release
-source:
-  enabled: true
-  name_template: '{{ .ProjectName }}-{{ .Tag }}-source'
-builds:
-  - main: ./cmd/local-ai
-    env:
-      - CGO_ENABLED=0
-    ldflags:
-      - -s -w
-      - -X "github.com/mudler/LocalAI/internal.Version={{ .Tag }}"
-      - -X "github.com/mudler/LocalAI/internal.Commit={{ .FullCommit }}"
-    goos:
-      - linux
-      - darwin
-      #- windows
-    goarch:
-      - amd64
-      - arm64
-archives:
-  - formats: [ 'binary' ] # this removes the tar of the archives, leaving the binaries alone
-    name_template: local-ai-{{ .Tag }}-{{ .Os }}-{{ .Arch }}{{ if .Arm }}v{{ .Arm }}{{ end }}
-checksum:
-  name_template: '{{ .ProjectName }}-{{ .Tag }}-checksums.txt'
-snapshot:
-  version_template: "{{ .Tag }}-next"
-changelog:
-  use: github-native

.vscode/launch.json

@@ -26,7 +26,7 @@
                 "LOCALAI_P2P": "true",
                 "LOCALAI_FEDERATED": "true"
             },
-            "buildFlags": ["-tags", "", "-v"],
+            "buildFlags": ["-tags", "p2p tts", "-v"],
             "envFile": "${workspaceFolder}/.env",
             "cwd": "${workspaceRoot}"
         }

Dockerfile

@@ -1,31 +1,120 @@
+ARG IMAGE_TYPE=extras
 ARG BASE_IMAGE=ubuntu:22.04
 ARG GRPC_BASE_IMAGE=${BASE_IMAGE}
 ARG INTEL_BASE_IMAGE=${BASE_IMAGE}
 
-FROM ${BASE_IMAGE} AS requirements
+# The requirements-core target is common to all images.  It should not be placed in requirements-core unless every single build will use it.
+FROM ${BASE_IMAGE} AS requirements-core
+
+USER root
+
+ARG GO_VERSION=1.22.6
+ARG CMAKE_VERSION=3.26.4
+ARG CMAKE_FROM_SOURCE=false
+ARG TARGETARCH
+ARG TARGETVARIANT
 
 ENV DEBIAN_FRONTEND=noninteractive
+ENV EXTERNAL_GRPC_BACKENDS="coqui:/build/backend/python/coqui/run.sh,transformers:/build/backend/python/transformers/run.sh,rerankers:/build/backend/python/rerankers/run.sh,bark:/build/backend/python/bark/run.sh,diffusers:/build/backend/python/diffusers/run.sh,faster-whisper:/build/backend/python/faster-whisper/run.sh,kokoro:/build/backend/python/kokoro/run.sh,vllm:/build/backend/python/vllm/run.sh,exllama2:/build/backend/python/exllama2/run.sh"
 
 RUN apt-get update && \
     apt-get install -y --no-install-recommends \
-        ca-certificates curl wget espeak-ng libgomp1 \
-        ffmpeg libopenblas-base libopenblas-dev && \
+        build-essential \
+        ccache \
+        ca-certificates \
+        curl libssl-dev \
+        git \
+        git-lfs \
+        unzip upx-ucl && \
     apt-get clean && \
     rm -rf /var/lib/apt/lists/*
 
+# Install CMake (the version in 22.04 is too old)
+RUN <<EOT bash
+    if [ "${CMAKE_FROM_SOURCE}}" = "true" ]; then
+        curl -L -s https://github.com/Kitware/CMake/releases/download/v${CMAKE_VERSION}/cmake-${CMAKE_VERSION}.tar.gz -o cmake.tar.gz && tar xvf cmake.tar.gz && cd cmake-${CMAKE_VERSION} && ./configure && make && make install
+    else
+        apt-get update && \
+        apt-get install -y \
+            cmake && \
+        apt-get clean && \
+        rm -rf /var/lib/apt/lists/*
+    fi
+EOT
+
+# Install Go
+RUN curl -L -s https://go.dev/dl/go${GO_VERSION}.linux-${TARGETARCH}.tar.gz | tar -C /usr/local -xz
+ENV PATH=$PATH:/root/go/bin:/usr/local/go/bin
+
+# Install grpc compilers and rice
+RUN go install google.golang.org/protobuf/cmd/protoc-gen-go@v1.34.2 && \
+    go install google.golang.org/grpc/cmd/protoc-gen-go-grpc@1958fcbe2ca8bd93af633f11e97d44e567e945af && \
+    go install github.com/GeertJohan/go.rice/rice@latest
+
+COPY --chmod=644 custom-ca-certs/* /usr/local/share/ca-certificates/
+RUN update-ca-certificates
+
+RUN test -n "$TARGETARCH" \
+    || (echo 'warn: missing $TARGETARCH, either set this `ARG` manually, or run using `docker buildkit`')
+
+# Use the variables in subsequent instructions
+RUN echo "Target Architecture: $TARGETARCH"
+RUN echo "Target Variant: $TARGETVARIANT"
+
+# Cuda
+ENV PATH=/usr/local/cuda/bin:${PATH}
+
+# HipBLAS requirements
+ENV PATH=/opt/rocm/bin:${PATH}
+
+# OpenBLAS requirements and stable diffusion
+RUN apt-get update && \
+    apt-get install -y --no-install-recommends \
+        libopenblas-dev && \
+    apt-get clean && \
+    rm -rf /var/lib/apt/lists/*
+
+WORKDIR /build
+
+###################################
+###################################
+
+# The requirements-extras target is for any builds with IMAGE_TYPE=extras. It should not be placed in this target unless every IMAGE_TYPE=extras build will use it
+FROM requirements-core AS requirements-extras
+
+# Install uv as a system package
+RUN curl -LsSf https://astral.sh/uv/install.sh | UV_INSTALL_DIR=/usr/bin sh
+ENV PATH="/root/.cargo/bin:${PATH}"
+
+RUN curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y
+RUN apt-get update && \
+    apt-get install -y --no-install-recommends \
+        espeak-ng \
+        espeak \
+        python3-pip \
+        python-is-python3 \
+        python3-dev llvm \
+        python3-venv && \
+    apt-get clean && \
+    rm -rf /var/lib/apt/lists/* && \
+    pip install --upgrade pip
+
+# Install grpcio-tools (the version in 22.04 is too old)
+RUN pip install --user grpcio-tools
+
+###################################
+###################################
+
 # The requirements-drivers target is for BUILD_TYPE specific items.  If you need to install something specific to CUDA, or specific to ROCM, it goes here.
-FROM requirements AS requirements-drivers
+# This target will be built on top of requirements-core or requirements-extras as retermined by the IMAGE_TYPE build-arg
+FROM requirements-${IMAGE_TYPE} AS requirements-drivers
 
 ARG BUILD_TYPE
 ARG CUDA_MAJOR_VERSION=12
 ARG CUDA_MINOR_VERSION=0
 ARG SKIP_DRIVERS=false
-ARG TARGETARCH
-ARG TARGETVARIANT
-ENV BUILD_TYPE=${BUILD_TYPE}
 
-RUN mkdir -p /run/localai
-RUN echo "default" > /run/localai/capability
+ENV BUILD_TYPE=${BUILD_TYPE}
 
 # Vulkan requirements
 RUN <<EOT bash
@@ -39,8 +128,7 @@ RUN <<EOT bash
         apt-get install -y \
             vulkan-sdk && \
         apt-get clean && \
-        rm -rf /var/lib/apt/lists/* && \
-        echo "vulkan" > /run/localai/capability
+        rm -rf /var/lib/apt/lists/*
     fi
 EOT
 
@@ -67,24 +155,7 @@ RUN <<EOT bash
             libcusparse-dev-${CUDA_MAJOR_VERSION}-${CUDA_MINOR_VERSION} \
             libcusolver-dev-${CUDA_MAJOR_VERSION}-${CUDA_MINOR_VERSION} && \
         apt-get clean && \
-        rm -rf /var/lib/apt/lists/* && \
-        echo "nvidia" > /run/localai/capability
-    fi
-EOT
-
-RUN <<EOT bash
-    if [ "${BUILD_TYPE}" = "cublas" ] && [ "${TARGETARCH}" = "arm64" ]; then
-        echo "nvidia-l4t" > /run/localai/capability
-    fi
-EOT
-
-# https://github.com/NVIDIA/Isaac-GR00T/issues/343
-RUN <<EOT bash
-    if [ "${BUILD_TYPE}" = "cublas" ] && [ "${TARGETARCH}" = "arm64" ]; then
-        wget https://developer.download.nvidia.com/compute/cudss/0.6.0/local_installers/cudss-local-tegra-repo-ubuntu2204-0.6.0_0.6.0-1_arm64.deb && \
-        dpkg -i cudss-local-tegra-repo-ubuntu2204-0.6.0_0.6.0-1_arm64.deb && \
-        cp /var/cudss-local-tegra-repo-ubuntu2204-0.6.0/cudss-*-keyring.gpg /usr/share/keyrings/ && \
-        apt-get update && apt-get -y install cudss
+        rm -rf /var/lib/apt/lists/*
     fi
 EOT
 
@@ -104,94 +175,11 @@ RUN if [ "${BUILD_TYPE}" = "hipblas" ] && [ "${SKIP_DRIVERS}" = "false" ]; then
             rocblas-dev && \
         apt-get clean && \
         rm -rf /var/lib/apt/lists/* && \
-        echo "amd" > /run/localai/capability && \
         # I have no idea why, but the ROCM lib packages don't trigger ldconfig after they install, which results in local-ai and others not being able
         # to locate the libraries. We run ldconfig ourselves to work around this packaging deficiency
         ldconfig \
     ; fi
 
-RUN if [ "${BUILD_TYPE}" = "hipblas" ]; then \
-    ln -s /opt/rocm-**/lib/llvm/lib/libomp.so /usr/lib/libomp.so \
-    ; fi
-
-RUN expr "${BUILD_TYPE}" = intel && echo "intel" > /run/localai/capability || echo "not intel"
-
-# Cuda
-ENV PATH=/usr/local/cuda/bin:${PATH}
-
-# HipBLAS requirements
-ENV PATH=/opt/rocm/bin:${PATH}
-
-###################################
-###################################
-
-# The requirements-core target is common to all images.  It should not be placed in requirements-core unless every single build will use it.
-FROM requirements-drivers AS build-requirements
-
-ARG GO_VERSION=1.22.6
-ARG CMAKE_VERSION=3.26.4
-ARG CMAKE_FROM_SOURCE=false
-ARG TARGETARCH
-ARG TARGETVARIANT
-
-
-RUN apt-get update && \
-    apt-get install -y --no-install-recommends \
-        build-essential \
-        ccache \
-        ca-certificates espeak-ng \
-        curl libssl-dev \
-        git \
-        git-lfs \
-        unzip upx-ucl python3 python-is-python3 && \
-    apt-get clean && \
-    rm -rf /var/lib/apt/lists/*
-
-# Install CMake (the version in 22.04 is too old)
-RUN <<EOT bash
-    if [ "${CMAKE_FROM_SOURCE}" = "true" ]; then
-        curl -L -s https://github.com/Kitware/CMake/releases/download/v${CMAKE_VERSION}/cmake-${CMAKE_VERSION}.tar.gz -o cmake.tar.gz && tar xvf cmake.tar.gz && cd cmake-${CMAKE_VERSION} && ./configure && make && make install
-    else
-        apt-get update && \
-        apt-get install -y \
-            cmake && \
-        apt-get clean && \
-        rm -rf /var/lib/apt/lists/*
-    fi
-EOT
-
-# Install Go
-RUN curl -L -s https://go.dev/dl/go${GO_VERSION}.linux-${TARGETARCH}.tar.gz | tar -C /usr/local -xz
-ENV PATH=$PATH:/root/go/bin:/usr/local/go/bin
-
-# Install grpc compilers
-RUN go install google.golang.org/protobuf/cmd/protoc-gen-go@v1.34.2 && \
-    go install google.golang.org/grpc/cmd/protoc-gen-go-grpc@1958fcbe2ca8bd93af633f11e97d44e567e945af
-
-COPY --chmod=644 custom-ca-certs/* /usr/local/share/ca-certificates/
-RUN update-ca-certificates
-
-
-# OpenBLAS requirements and stable diffusion
-RUN apt-get update && \
-    apt-get install -y --no-install-recommends \
-        libopenblas-dev && \
-    apt-get clean && \
-    rm -rf /var/lib/apt/lists/*
-
-RUN test -n "$TARGETARCH" \
-    || (echo 'warn: missing $TARGETARCH, either set this `ARG` manually, or run using `docker buildkit`')
-
-# Use the variables in subsequent instructions
-RUN echo "Target Architecture: $TARGETARCH"
-RUN echo "Target Variant: $TARGETVARIANT"
-
-
-
-
-WORKDIR /build
-
-
 ###################################
 ###################################
 
@@ -202,25 +190,69 @@ FROM ${INTEL_BASE_IMAGE} AS intel
 RUN wget -qO - https://repositories.intel.com/gpu/intel-graphics.key | \
 gpg --yes --dearmor --output /usr/share/keyrings/intel-graphics.gpg
 RUN echo "deb [arch=amd64 signed-by=/usr/share/keyrings/intel-graphics.gpg] https://repositories.intel.com/gpu/ubuntu jammy/lts/2350 unified" > /etc/apt/sources.list.d/intel-graphics.list
+
+###################################
+###################################
+
+# The grpc target does one thing, it builds and installs GRPC.  This is in it's own layer so that it can be effectively cached by CI.
+# You probably don't need to change anything here, and if you do, make sure that CI is adjusted so that the cache continues to work.
+FROM ${GRPC_BASE_IMAGE} AS grpc
+
+# This is a bit of a hack, but it's required in order to be able to effectively cache this layer in CI
+ARG GRPC_MAKEFLAGS="-j4 -Otarget"
+ARG GRPC_VERSION=v1.65.0
+ARG CMAKE_FROM_SOURCE=false
+ARG CMAKE_VERSION=3.26.4
+
+ENV MAKEFLAGS=${GRPC_MAKEFLAGS}
+
+WORKDIR /build
+
 RUN apt-get update && \
     apt-get install -y --no-install-recommends \
-        intel-oneapi-runtime-libs && \
+        ca-certificates \
+        build-essential curl libssl-dev \
+        git && \
     apt-get clean && \
     rm -rf /var/lib/apt/lists/*
 
+# Install CMake (the version in 22.04 is too old)
+RUN <<EOT bash
+    if [ "${CMAKE_FROM_SOURCE}}" = "true" ]; then
+        curl -L -s https://github.com/Kitware/CMake/releases/download/v${CMAKE_VERSION}/cmake-${CMAKE_VERSION}.tar.gz -o cmake.tar.gz && tar xvf cmake.tar.gz && cd cmake-${CMAKE_VERSION} && ./configure && make && make install
+    else
+        apt-get update && \
+        apt-get install -y \
+            cmake && \
+        apt-get clean && \
+        rm -rf /var/lib/apt/lists/*
+    fi
+EOT
+
+# We install GRPC to a different prefix here so that we can copy in only the build artifacts later
+# saves several hundred MB on the final docker image size vs copying in the entire GRPC source tree
+# and running make install in the target container
+RUN git clone --recurse-submodules --jobs 4 -b ${GRPC_VERSION} --depth 1 --shallow-submodules https://github.com/grpc/grpc && \
+    mkdir -p /build/grpc/cmake/build && \
+    cd /build/grpc/cmake/build && \
+    sed -i "216i\  TESTONLY" "../../third_party/abseil-cpp/absl/container/CMakeLists.txt" && \
+    cmake -DgRPC_INSTALL=ON -DgRPC_BUILD_TESTS=OFF -DCMAKE_INSTALL_PREFIX:PATH=/opt/grpc ../.. && \
+    make && \
+    make install && \
+    rm -rf /build
+
 ###################################
 ###################################
 
 # The builder-base target has the arguments, variables, and copies shared between full builder images and the uncompiled devcontainer
 
-FROM build-requirements AS builder-base
+FROM requirements-drivers AS builder-base
 
-ARG GO_TAGS=""
+ARG GO_TAGS="tts p2p"
 ARG GRPC_BACKENDS
 ARG MAKEFLAGS
 ARG LD_FLAGS="-s -w"
-ARG TARGETARCH
-ARG TARGETVARIANT
+
 ENV GRPC_BACKENDS=${GRPC_BACKENDS}
 ENV GO_TAGS=${GO_TAGS}
 ENV MAKEFLAGS=${MAKEFLAGS}
@@ -234,7 +266,9 @@ RUN echo "GO_TAGS: $GO_TAGS" && echo "TARGETARCH: $TARGETARCH"
 WORKDIR /build
 
 
-# We need protoc installed, and the version in 22.04 is too old.
+# We need protoc installed, and the version in 22.04 is too old.  We will create one as part installing the GRPC build below
+# but that will also being in a newer version of absl which stablediffusion cannot compile with.  This version of protoc is only
+# here so that we can generate the grpc code for the stablediffusion build
 RUN <<EOT bash
     if [ "amd64" = "$TARGETARCH" ]; then
         curl -L -s https://github.com/protocolbuffers/protobuf/releases/download/v27.1/protoc-27.1-linux-x86_64.zip -o protoc.zip && \
@@ -251,39 +285,34 @@ EOT
 ###################################
 ###################################
 
-# Compile backends first in a separate stage
-FROM builder-base AS builder-backends
-ARG TARGETARCH
-ARG TARGETVARIANT
-
-WORKDIR /build
-
-COPY ./Makefile .
-COPY ./backend ./backend
-COPY ./go.mod .
-COPY ./go.sum .
-COPY ./.git ./.git
-
-# Some of the Go backends use libs from the main src, we could further optimize the caching by building the CPP backends before here
-COPY ./pkg/grpc ./pkg/grpc
-COPY ./pkg/utils ./pkg/utils
-COPY ./pkg/langchain ./pkg/langchain
-
-RUN ls -l ./
-RUN make protogen-go
-
 # The builder target compiles LocalAI. This target is not the target that will be uploaded to the registry.
 # Adjustments to the build process should likely be made here.
-FROM builder-backends AS builder
+FROM builder-base AS builder
 
+# Install the pre-built GRPC
+COPY --from=grpc /opt/grpc /usr/local
+
+# Rebuild with defaults backends
 WORKDIR /build
 
 COPY . .
+COPY .git .
+
+RUN make prepare
 
 ## Build the binary
 ## If we're on arm64 AND using cublas/hipblas, skip some of the llama-compat backends to save space
 ## Otherwise just run the normal build
-RUN make build
+RUN if [ "${TARGETARCH}" = "arm64" ] || [ "${BUILD_TYPE}" = "hipblas" ]; then \
+        SKIP_GRPC_BACKEND="backend-assets/grpc/llama-cpp-avx512 backend-assets/grpc/llama-cpp-avx backend-assets/grpc/llama-cpp-avx2" make build; \
+    else \
+        make build; \
+    fi
+
+RUN if [ ! -d "/build/sources/go-piper/piper-phonemize/pi/lib/" ]; then \
+        mkdir -p /build/sources/go-piper/piper-phonemize/pi/lib/ \
+        touch /build/sources/go-piper/piper-phonemize/pi/lib/keep \
+    ; fi
 
 ###################################
 ###################################
@@ -293,11 +322,24 @@ RUN make build
 
 FROM builder-base AS devcontainer
 
+ARG FFMPEG
+
+COPY --from=grpc /opt/grpc /usr/local
+
 COPY .devcontainer-scripts /.devcontainer-scripts
 
+# Add FFmpeg
+RUN if [ "${FFMPEG}" = "true" ]; then \
+        apt-get update && \
+        apt-get install -y --no-install-recommends \
+            ffmpeg && \
+        apt-get clean && \
+        rm -rf /var/lib/apt/lists/* \
+    ; fi
+
 RUN apt-get update && \
     apt-get install -y --no-install-recommends \
-        ssh less
+        ssh less wget
 # For the devcontainer, leave apt functional in case additional devtools are needed at runtime.
 
 RUN go install github.com/go-delve/delve/cmd/dlv@latest
@@ -311,27 +353,98 @@ RUN go install github.com/mikefarah/yq/v4@latest
 # If you cannot find a more suitable place for an addition, this layer is a suitable place for it.
 FROM requirements-drivers
 
+ARG FFMPEG
+ARG BUILD_TYPE
+ARG TARGETARCH
+ARG IMAGE_TYPE=extras
+ARG EXTRA_BACKENDS
+ARG MAKEFLAGS
+
+ENV BUILD_TYPE=${BUILD_TYPE}
+ENV REBUILD=false
 ENV HEALTHCHECK_ENDPOINT=http://localhost:8080/readyz
+ENV MAKEFLAGS=${MAKEFLAGS}
 
 ARG CUDA_MAJOR_VERSION=12
 ENV NVIDIA_DRIVER_CAPABILITIES=compute,utility
 ENV NVIDIA_REQUIRE_CUDA="cuda>=${CUDA_MAJOR_VERSION}.0"
 ENV NVIDIA_VISIBLE_DEVICES=all
 
-WORKDIR /
+# Add FFmpeg
+RUN if [ "${FFMPEG}" = "true" ]; then \
+        apt-get update && \
+        apt-get install -y --no-install-recommends \
+            ffmpeg && \
+        apt-get clean && \
+        rm -rf /var/lib/apt/lists/* \
+    ; fi
 
-COPY ./entrypoint.sh .
+WORKDIR /build
+
+# we start fresh & re-copy all assets because `make build` does not clean up nicely after itself
+# so when `entrypoint.sh` runs `make build` again (which it does by default), the build would fail
+# see https://github.com/go-skynet/LocalAI/pull/658#discussion_r1241971626 and
+# https://github.com/go-skynet/LocalAI/pull/434
+COPY . .
+
+COPY --from=builder /build/sources ./sources/
+COPY --from=grpc /opt/grpc /usr/local
+
+RUN make prepare-sources
 
 # Copy the binary
 COPY --from=builder /build/local-ai ./
 
+# Copy shared libraries for piper
+COPY --from=builder /build/sources/go-piper/piper-phonemize/pi/lib/* /usr/lib/
+
+# Change the shell to bash so we can use [[ tests below
+SHELL ["/bin/bash", "-c"]
+# We try to strike a balance between individual layer size (as that affects total push time) and total image size
+# Splitting the backends into more groups with fewer items results in a larger image, but a smaller size for the largest layer
+# Splitting the backends into fewer groups with more items results in a smaller image, but a larger size for the largest layer
+
+RUN if [[ ( "${IMAGE_TYPE}" == "extras ")]]; then \
+        apt-get -qq -y install espeak-ng \
+    ; fi
+
+RUN if [[ ( "${EXTRA_BACKENDS}" =~ "coqui" || -z "${EXTRA_BACKENDS}" ) && "$IMAGE_TYPE" == "extras" ]]; then \
+        make -C backend/python/coqui \
+    ; fi && \
+    if [[ ( "${EXTRA_BACKENDS}" =~ "faster-whisper" || -z "${EXTRA_BACKENDS}" ) && "$IMAGE_TYPE" == "extras" ]]; then \
+        make -C backend/python/faster-whisper \
+    ; fi && \
+    if [[ ( "${EXTRA_BACKENDS}" =~ "diffusers" || -z "${EXTRA_BACKENDS}" ) && "$IMAGE_TYPE" == "extras" ]]; then \
+        make -C backend/python/diffusers \
+    ; fi
+
+RUN if [[ ( "${EXTRA_BACKENDS}" =~ "kokoro" || -z "${EXTRA_BACKENDS}" ) && "$IMAGE_TYPE" == "extras" ]]; then \
+        make -C backend/python/kokoro \
+    ; fi && \
+    if [[ ( "${EXTRA_BACKENDS}" =~ "exllama2" || -z "${EXTRA_BACKENDS}" ) && "$IMAGE_TYPE" == "extras" ]]; then \
+        make -C backend/python/exllama2 \
+    ; fi && \
+    if [[ ( "${EXTRA_BACKENDS}" =~ "transformers" || -z "${EXTRA_BACKENDS}" ) && "$IMAGE_TYPE" == "extras" ]]; then \
+        make -C backend/python/transformers \
+    ; fi
+
+RUN if [[ ( "${EXTRA_BACKENDS}" =~ "vllm" || -z "${EXTRA_BACKENDS}" ) && "$IMAGE_TYPE" == "extras" ]]; then \
+        make -C backend/python/vllm \
+    ; fi && \
+    if [[ ( "${EXTRA_BACKENDS}" =~ "bark" || -z "${EXTRA_BACKENDS}" ) && "$IMAGE_TYPE" == "extras" ]]; then \
+        make -C backend/python/bark \
+    ; fi && \
+    if [[ ( "${EXTRA_BACKENDS}" =~ "rerankers" || -z "${EXTRA_BACKENDS}" ) && "$IMAGE_TYPE" == "extras" ]]; then \
+        make -C backend/python/rerankers \
+    ; fi
+
 # Make sure the models directory exists
-RUN mkdir -p /models /backends
+RUN mkdir -p /build/models
 
 # Define the health check command
 HEALTHCHECK --interval=1m --timeout=10m --retries=10 \
   CMD curl -f ${HEALTHCHECK_ENDPOINT} || exit 1
 
-VOLUME /models /backends
+VOLUME /build/models
 EXPOSE 8080
-ENTRYPOINT [ "/entrypoint.sh" ]
+ENTRYPOINT [ "/build/entrypoint.sh" ]

Earthfile

@@ -0,0 +1,5 @@
+VERSION 0.7
+
+build:
+    FROM DOCKERFILE -f Dockerfile .
+    SAVE ARTIFACT /usr/bin/local-ai AS LOCAL local-ai

README.md

@@ -1,6 +1,6 @@
 <h1 align="center">
   <br>
-  <img width="300" src="./core/http/static/logo.png"> <br>
+  <img height="300" src="./core/http/static/logo.png"> <br>
 <br>
 </h1>
 
@@ -43,7 +43,7 @@
 
 > :bulb: Get help - [❓FAQ](https://localai.io/faq/) [💭Discussions](https://github.com/go-skynet/LocalAI/discussions) [:speech_balloon: Discord](https://discord.gg/uJAeKSAGDy) [:book: Documentation website](https://localai.io/)
 >
-> [💻 Quickstart](https://localai.io/basics/getting_started/) [🖼️ Models](https://models.localai.io/) [🚀 Roadmap](https://github.com/mudler/LocalAI/issues?q=is%3Aissue+is%3Aopen+label%3Aroadmap) [🌍 Explorer](https://explorer.localai.io) [🛫 Examples](https://github.com/mudler/LocalAI-examples) Try on 
+> [💻 Quickstart](https://localai.io/basics/getting_started/) [🖼️ Models](https://models.localai.io/) [🚀 Roadmap](https://github.com/mudler/LocalAI/issues?q=is%3Aissue+is%3Aopen+label%3Aroadmap) [🥽 Demo](https://demo.localai.io) [🌍 Explorer](https://explorer.localai.io) [🛫 Examples](https://github.com/mudler/LocalAI-examples) Try on 
 [![Telegram](https://img.shields.io/badge/Telegram-2CA5E0?style=for-the-badge&logo=telegram&logoColor=white)](https://t.me/localaiofficial_bot)
 
 [![tests](https://github.com/go-skynet/LocalAI/actions/workflows/test.yml/badge.svg)](https://github.com/go-skynet/LocalAI/actions/workflows/test.yml)[![Build and Release](https://github.com/go-skynet/LocalAI/actions/workflows/release.yaml/badge.svg)](https://github.com/go-skynet/LocalAI/actions/workflows/release.yaml)[![build container images](https://github.com/go-skynet/LocalAI/actions/workflows/image.yml/badge.svg)](https://github.com/go-skynet/LocalAI/actions/workflows/image.yml)[![Bump dependencies](https://github.com/go-skynet/LocalAI/actions/workflows/bump_deps.yaml/badge.svg)](https://github.com/go-skynet/LocalAI/actions/workflows/bump_deps.yaml)[![Artifact Hub](https://img.shields.io/endpoint?url=https://artifacthub.io/badge/repository/localai)](https://artifacthub.io/packages/search?repo=localai)
@@ -110,72 +110,25 @@ curl https://localai.io/install.sh | sh
 
 For more installation options, see [Installer Options](https://localai.io/docs/advanced/installer/).
 
-### macOS Download:
-
-<a href="https://github.com/mudler/LocalAI/releases/latest/download/LocalAI.dmg">
-  <img src="https://img.shields.io/badge/Download-macOS-blue?style=for-the-badge&logo=apple&logoColor=white" alt="Download LocalAI for macOS"/>
-</a>
-
 Or run with docker:
 
 ### CPU only image:
-
+```bash
+docker run -ti --name local-ai -p 8080:8080 localai/localai:latest-cpu
+```
+### Nvidia GPU:
+```bash
+docker run -ti --name local-ai -p 8080:8080 --gpus all localai/localai:latest-gpu-nvidia-cuda-12
+```
+### CPU and GPU image (bigger size):
 ```bash
 docker run -ti --name local-ai -p 8080:8080 localai/localai:latest
 ```
-
-### NVIDIA GPU Images:
-
+### AIO images (it will pre-download a set of models ready for use, see https://localai.io/basics/container/)
 ```bash
-# CUDA 12.0
-docker run -ti --name local-ai -p 8080:8080 --gpus all localai/localai:latest-gpu-nvidia-cuda-12
-
-# CUDA 11.7
-docker run -ti --name local-ai -p 8080:8080 --gpus all localai/localai:latest-gpu-nvidia-cuda-11
-
-# NVIDIA Jetson (L4T) ARM64
-docker run -ti --name local-ai -p 8080:8080 --gpus all localai/localai:latest-nvidia-l4t-arm64
-```
-
-### AMD GPU Images (ROCm):
-
-```bash
-docker run -ti --name local-ai -p 8080:8080 --device=/dev/kfd --device=/dev/dri --group-add=video localai/localai:latest-gpu-hipblas
-```
-
-### Intel GPU Images (oneAPI):
-
-```bash
-docker run -ti --name local-ai -p 8080:8080 --device=/dev/dri/card1 --device=/dev/dri/renderD128 localai/localai:latest-gpu-intel
-```
-
-### Vulkan GPU Images:
-
-```bash
-docker run -ti --name local-ai -p 8080:8080 localai/localai:latest-gpu-vulkan
-```
-
-### AIO Images (pre-downloaded models):
-
-```bash
-# CPU version
 docker run -ti --name local-ai -p 8080:8080 localai/localai:latest-aio-cpu
-
-# NVIDIA CUDA 12 version
-docker run -ti --name local-ai -p 8080:8080 --gpus all localai/localai:latest-aio-gpu-nvidia-cuda-12
-
-# NVIDIA CUDA 11 version
-docker run -ti --name local-ai -p 8080:8080 --gpus all localai/localai:latest-aio-gpu-nvidia-cuda-11
-
-# Intel GPU version
-docker run -ti --name local-ai -p 8080:8080 localai/localai:latest-aio-gpu-intel
-
-# AMD GPU version
-docker run -ti --name local-ai -p 8080:8080 --device=/dev/kfd --device=/dev/dri --group-add=video localai/localai:latest-aio-gpu-hipblas
 ```
 
-For more information about the AIO images and pre-downloaded models, see [Container Documentation](https://localai.io/basics/container/).
-
 To load models:
 
 ```bash
@@ -191,19 +144,10 @@ local-ai run https://gist.githubusercontent.com/.../phi-2.yaml
 local-ai run oci://localai/phi-2:latest
 ```
 
-> ⚡ **Automatic Backend Detection**: When you install models from the gallery or YAML files, LocalAI automatically detects your system's GPU capabilities (NVIDIA, AMD, Intel) and downloads the appropriate backend. For advanced configuration options, see [GPU Acceleration](https://localai.io/features/gpu-acceleration/#automatic-backend-detection).
-
 For more information, see [💻 Getting started](https://localai.io/basics/getting_started/index.html)
 
 ## 📰 Latest project news
 
-- August 2025: MLX, MLX-VLM, Diffusers and llama.cpp are now supported on Mac M1/M2/M3+ chips ( with `development` suffix in the gallery ): https://github.com/mudler/LocalAI/pull/6049 https://github.com/mudler/LocalAI/pull/6119 https://github.com/mudler/LocalAI/pull/6121 https://github.com/mudler/LocalAI/pull/6060
-- July/August 2025: 🔍 [Object Detection](https://localai.io/features/object-detection/) added to the API featuring [rf-detr](https://github.com/roboflow/rf-detr)
-- July 2025: All backends migrated outside of the main binary. LocalAI is now more lightweight, small, and automatically downloads the required backend to run the model. [Read the release notes](https://github.com/mudler/LocalAI/releases/tag/v3.2.0)
-- June 2025: [Backend management](https://github.com/mudler/LocalAI/pull/5607) has been added. Attention: extras images are going to be deprecated from the next release! Read [the backend management PR](https://github.com/mudler/LocalAI/pull/5607).
-- May 2025: [Audio input](https://github.com/mudler/LocalAI/pull/5466) and [Reranking](https://github.com/mudler/LocalAI/pull/5396) in llama.cpp backend, [Realtime API](https://github.com/mudler/LocalAI/pull/5392),  Support to Gemma, SmollVLM, and more multimodal models (available in the gallery).
-- May 2025: Important: image name changes [See release](https://github.com/mudler/LocalAI/releases/tag/v2.29.0)
-- Apr 2025: Rebrand, WebUI enhancements
 - Apr 2025: [LocalAGI](https://github.com/mudler/LocalAGI) and [LocalRecall](https://github.com/mudler/LocalRecall) join the LocalAI family stack.
 - Apr 2025: WebUI overhaul, AIO images updates
 - Feb 2025: Backend cleanup, Breaking changes, new backends (kokoro, OutelTTS, faster-whisper), Nvidia L4T images
@@ -222,7 +166,6 @@ Roadmap items: [List of issues](https://github.com/mudler/LocalAI/issues?q=is%3A
 
 ## 🚀 [Features](https://localai.io/features/)
 
-- 🧩 [Backend Gallery](https://localai.io/backends/): Install/remove backends on the fly, powered by OCI images — fully customizable and API-driven.
 - 📖 [Text generation with GPTs](https://localai.io/features/text-generation/) (`llama.cpp`, `transformers`, `vllm` ... [:book: and more](https://localai.io/model-compatibility/index.html#model-compatibility-table))
 - 🗣 [Text to Audio](https://localai.io/features/text-to-audio/)
 - 🔈 [Audio to Text](https://localai.io/features/audio-to-text/) (Audio transcription with `whisper.cpp`)
@@ -232,67 +175,12 @@ Roadmap items: [List of issues](https://github.com/mudler/LocalAI/issues?q=is%3A
 - ✍️ [Constrained grammars](https://localai.io/features/constrained_grammars/)
 - 🖼️ [Download Models directly from Huggingface ](https://localai.io/models/)
 - 🥽 [Vision API](https://localai.io/features/gpt-vision/)
-- 🔍 [Object Detection](https://localai.io/features/object-detection/)
 - 📈 [Reranker API](https://localai.io/features/reranker/)
 - 🆕🖧 [P2P Inferencing](https://localai.io/features/distribute/)
 - [Agentic capabilities](https://github.com/mudler/LocalAGI)
 - 🔊 Voice activity detection (Silero-VAD support)
 - 🌍 Integrated WebUI!
 
-## 🧩 Supported Backends & Acceleration
-
-LocalAI supports a comprehensive range of AI backends with multiple acceleration options:
-
-### Text Generation & Language Models
-| Backend | Description | Acceleration Support |
-|---------|-------------|---------------------|
-| **llama.cpp** | LLM inference in C/C++ | CUDA 11/12, ROCm, Intel SYCL, Vulkan, Metal, CPU |
-| **vLLM** | Fast LLM inference with PagedAttention | CUDA 12, ROCm, Intel |
-| **transformers** | HuggingFace transformers framework | CUDA 11/12, ROCm, Intel, CPU |
-| **exllama2** | GPTQ inference library | CUDA 12 |
-| **MLX** | Apple Silicon LLM inference | Metal (M1/M2/M3+) |
-| **MLX-VLM** | Apple Silicon Vision-Language Models | Metal (M1/M2/M3+) |
-
-### Audio & Speech Processing
-| Backend | Description | Acceleration Support |
-|---------|-------------|---------------------|
-| **whisper.cpp** | OpenAI Whisper in C/C++ | CUDA 12, ROCm, Intel SYCL, Vulkan, CPU |
-| **faster-whisper** | Fast Whisper with CTranslate2 | CUDA 12, ROCm, Intel, CPU |
-| **bark** | Text-to-audio generation | CUDA 12, ROCm, Intel |
-| **bark-cpp** | C++ implementation of Bark | CUDA, Metal, CPU |
-| **coqui** | Advanced TTS with 1100+ languages | CUDA 12, ROCm, Intel, CPU |
-| **kokoro** | Lightweight TTS model | CUDA 12, ROCm, Intel, CPU |
-| **chatterbox** | Production-grade TTS | CUDA 11/12, CPU |
-| **piper** | Fast neural TTS system | CPU |
-| **kitten-tts** | Kitten TTS models | CPU |
-| **silero-vad** | Voice Activity Detection | CPU |
-
-### Image & Video Generation
-| Backend | Description | Acceleration Support |
-|---------|-------------|---------------------|
-| **stablediffusion.cpp** | Stable Diffusion in C/C++ | CUDA 12, Intel SYCL, Vulkan, CPU |
-| **diffusers** | HuggingFace diffusion models | CUDA 11/12, ROCm, Intel, Metal, CPU |
-
-### Specialized AI Tasks
-| Backend | Description | Acceleration Support |
-|---------|-------------|---------------------|
-| **rfdetr** | Real-time object detection | CUDA 12, Intel, CPU |
-| **rerankers** | Document reranking API | CUDA 11/12, ROCm, Intel, CPU |
-| **local-store** | Vector database | CPU |
-| **huggingface** | HuggingFace API integration | API-based |
-
-### Hardware Acceleration Matrix
-
-| Acceleration Type | Supported Backends | Hardware Support |
-|-------------------|-------------------|------------------|
-| **NVIDIA CUDA 11** | llama.cpp, whisper, stablediffusion, diffusers, rerankers, bark, chatterbox | Nvidia hardware |
-| **NVIDIA CUDA 12** | All CUDA-compatible backends | Nvidia hardware |
-| **AMD ROCm** | llama.cpp, whisper, vllm, transformers, diffusers, rerankers, coqui, kokoro, bark | AMD Graphics |
-| **Intel oneAPI** | llama.cpp, whisper, stablediffusion, vllm, transformers, diffusers, rfdetr, rerankers, exllama2, coqui, kokoro, bark | Intel Arc, Intel iGPUs |
-| **Apple Metal** | llama.cpp, whisper, diffusers, MLX, MLX-VLM, bark-cpp | Apple M1/M2/M3+ |
-| **Vulkan** | llama.cpp, whisper, stablediffusion | Cross-platform GPUs |
-| **NVIDIA Jetson** | llama.cpp, whisper, stablediffusion, diffusers, rfdetr | ARM64 embedded AI |
-| **CPU Optimized** | All backends | AVX/AVX2/AVX512, quantization support |
 
 ### 🔗 Community and integrations
 
@@ -307,9 +195,6 @@ WebUIs:
 Model galleries
 - https://github.com/go-skynet/model-gallery
 
-Voice:
-- https://github.com/richiejp/VoxInput
-
 Other:
 - Helm chart https://github.com/go-skynet/helm-charts
 - VSCode extension https://github.com/badgooooor/localai-vscode-plugin

aio/cpu/embeddings.yaml

@@ -1,6 +1,5 @@
 embeddings: true
 name: text-embedding-ada-002
-backend: llama-cpp
 parameters:
   model: huggingface://bartowski/granite-embedding-107m-multilingual-GGUF/granite-embedding-107m-multilingual-f16.gguf
 

aio/cpu/rerank.yaml

@@ -1,13 +1,7 @@
 name: jina-reranker-v1-base-en
-reranking: true
-f16: true
+backend: rerankers
 parameters:
-  model: jina-reranker-v1-tiny-en.f16.gguf
-backend: llama-cpp
-download_files:
-  - filename: jina-reranker-v1-tiny-en.f16.gguf
-    sha256: 5f696cf0d0f3d347c4a279eee8270e5918554cdac0ed1f632f2619e4e8341407
-    uri: huggingface://mradermacher/jina-reranker-v1-tiny-en-GGUF/jina-reranker-v1-tiny-en.f16.gguf 
+  model: cross-encoder
 
 usage: |
     You can test this model with curl like this:

aio/cpu/text-to-speech.yaml

@@ -2,7 +2,7 @@ name: tts-1
 download_files:
   - filename: voice-en-us-amy-low.tar.gz
     uri: https://github.com/rhasspy/piper/releases/download/v0.0.2/voice-en-us-amy-low.tar.gz
-backend: piper
+
 parameters:
   model: en-us-amy-low.onnx
 

aio/cpu/text-to-text.yaml

@@ -1,6 +1,5 @@
 context_size: 8192
 f16: true
-backend: llama-cpp
 function:
   grammar:
     no_mixed_free_string: true

aio/cpu/vision.yaml

@@ -1,11 +1,10 @@
 context_size: 4096
 f16: true
-backend: llama-cpp
 mmap: true
-mmproj: minicpm-v-4_5-mmproj-f16.gguf
+mmproj: minicpm-v-2_6-mmproj-f16.gguf
 name: gpt-4o
 parameters:
-  model: minicpm-v-4_5-Q4_K_M.gguf
+  model: minicpm-v-2_6-Q4_K_M.gguf
 stopwords:
 - <|im_end|>
 - <dummy32000>
@@ -42,9 +41,9 @@ template:
     <|im_start|>assistant
 
 download_files:
-- filename: minicpm-v-4_5-Q4_K_M.gguf
-  sha256: c1c3c33100b15b4caf7319acce4e23c0eb0ce1cbd12f70e8d24f05aa67b7512f
-  uri: huggingface://openbmb/MiniCPM-V-4_5-gguf/ggml-model-Q4_K_M.gguf
-- filename: minicpm-v-4_5-mmproj-f16.gguf
-  uri: huggingface://openbmb/MiniCPM-V-4_5-gguf/mmproj-model-f16.gguf
-  sha256: 7a7225a32e8d453aaa3d22d8c579b5bf833c253f784cdb05c99c9a76fd616df8
+- filename: minicpm-v-2_6-Q4_K_M.gguf
+  sha256: 3a4078d53b46f22989adbf998ce5a3fd090b6541f112d7e936eb4204a04100b1
+  uri: huggingface://openbmb/MiniCPM-V-2_6-gguf/ggml-model-Q4_K_M.gguf
+- filename: minicpm-v-2_6-mmproj-f16.gguf
+  uri: huggingface://openbmb/MiniCPM-V-2_6-gguf/mmproj-model-f16.gguf
+  sha256: 4485f68a0f1aa404c391e788ea88ea653c100d8e98fe572698f701e5809711fd

aio/entrypoint.sh

@@ -135,4 +135,4 @@ check_vars
 
 echo "===> Starting LocalAI[$PROFILE] with the following models: $MODELS"
 
-exec /entrypoint.sh "$@"
+exec /build/entrypoint.sh "$@"

aio/gpu-8g/embeddings.yaml

@@ -1,6 +1,5 @@
 embeddings: true
 name: text-embedding-ada-002
-backend: llama-cpp
 parameters:
   model: huggingface://bartowski/granite-embedding-107m-multilingual-GGUF/granite-embedding-107m-multilingual-f16.gguf
 

aio/gpu-8g/rerank.yaml

@@ -1,13 +1,7 @@
 name: jina-reranker-v1-base-en
-reranking: true
-f16: true
+backend: rerankers
 parameters:
-  model: jina-reranker-v1-tiny-en.f16.gguf
-backend: llama-cpp
-download_files:
-  - filename: jina-reranker-v1-tiny-en.f16.gguf
-    sha256: 5f696cf0d0f3d347c4a279eee8270e5918554cdac0ed1f632f2619e4e8341407
-    uri: huggingface://mradermacher/jina-reranker-v1-tiny-en-GGUF/jina-reranker-v1-tiny-en.f16.gguf 
+  model: cross-encoder
 
 usage: |
     You can test this model with curl like this:

aio/gpu-8g/text-to-speech.yaml

@@ -2,7 +2,7 @@ name: tts-1
 download_files:
   - filename: voice-en-us-amy-low.tar.gz
     uri: https://github.com/rhasspy/piper/releases/download/v0.0.2/voice-en-us-amy-low.tar.gz
-backend: piper
+
 parameters:
   model: en-us-amy-low.onnx
 

aio/gpu-8g/text-to-text.yaml

@@ -1,6 +1,5 @@
 context_size: 4096
 f16: true
-backend: llama-cpp
 function:
   capture_llm_results:
   - (?s)<Thought>(.*?)</Thought>

aio/gpu-8g/vision.yaml

@@ -1,11 +1,10 @@
 context_size: 4096
-backend: llama-cpp
 f16: true
 mmap: true
-mmproj: minicpm-v-4_5-mmproj-f16.gguf
+mmproj: minicpm-v-2_6-mmproj-f16.gguf
 name: gpt-4o
 parameters:
-  model: minicpm-v-4_5-Q4_K_M.gguf
+  model: minicpm-v-2_6-Q4_K_M.gguf
 stopwords:
 - <|im_end|>
 - <dummy32000>
@@ -42,9 +41,9 @@ template:
     <|im_start|>assistant
 
 download_files:
-- filename: minicpm-v-4_5-Q4_K_M.gguf
-  sha256: c1c3c33100b15b4caf7319acce4e23c0eb0ce1cbd12f70e8d24f05aa67b7512f
-  uri: huggingface://openbmb/MiniCPM-V-4_5-gguf/ggml-model-Q4_K_M.gguf
-- filename: minicpm-v-4_5-mmproj-f16.gguf
-  uri: huggingface://openbmb/MiniCPM-V-4_5-gguf/mmproj-model-f16.gguf
-  sha256: 7a7225a32e8d453aaa3d22d8c579b5bf833c253f784cdb05c99c9a76fd616df8
+- filename: minicpm-v-2_6-Q4_K_M.gguf
+  sha256: 3a4078d53b46f22989adbf998ce5a3fd090b6541f112d7e936eb4204a04100b1
+  uri: huggingface://openbmb/MiniCPM-V-2_6-gguf/ggml-model-Q4_K_M.gguf
+- filename: minicpm-v-2_6-mmproj-f16.gguf
+  uri: huggingface://openbmb/MiniCPM-V-2_6-gguf/mmproj-model-f16.gguf
+  sha256: 4485f68a0f1aa404c391e788ea88ea653c100d8e98fe572698f701e5809711fd

aio/intel/embeddings.yaml

@@ -1,6 +1,5 @@
 embeddings: true
 name: text-embedding-ada-002
-backend: llama-cpp
 parameters:
   model: huggingface://bartowski/granite-embedding-107m-multilingual-GGUF/granite-embedding-107m-multilingual-f16.gguf
 

aio/intel/rerank.yaml

@@ -1,13 +1,7 @@
 name: jina-reranker-v1-base-en
-reranking: true
-f16: true
+backend: rerankers
 parameters:
-  model: jina-reranker-v1-tiny-en.f16.gguf
-backend: llama-cpp
-download_files:
-  - filename: jina-reranker-v1-tiny-en.f16.gguf
-    sha256: 5f696cf0d0f3d347c4a279eee8270e5918554cdac0ed1f632f2619e4e8341407
-    uri: huggingface://mradermacher/jina-reranker-v1-tiny-en-GGUF/jina-reranker-v1-tiny-en.f16.gguf 
+  model: cross-encoder
 
 usage: |
     You can test this model with curl like this:

aio/intel/text-to-speech.yaml

@@ -2,7 +2,7 @@ name: tts-1
 download_files:
   - filename: voice-en-us-amy-low.tar.gz
     uri: https://github.com/rhasspy/piper/releases/download/v0.0.2/voice-en-us-amy-low.tar.gz
-backend: piper
+
 parameters:
   model: en-us-amy-low.onnx
 

aio/intel/text-to-text.yaml

@@ -1,6 +1,5 @@
 context_size: 4096
 f16: true
-backend: llama-cpp
 function:
   capture_llm_results:
   - (?s)<Thought>(.*?)</Thought>

aio/intel/vision.yaml

@@ -1,11 +1,10 @@
 context_size: 4096
-backend: llama-cpp
 f16: true
 mmap: true
-mmproj: minicpm-v-4_5-mmproj-f16.gguf
+mmproj: minicpm-v-2_6-mmproj-f16.gguf
 name: gpt-4o
 parameters:
-  model: minicpm-v-4_5-Q4_K_M.gguf
+  model: minicpm-v-2_6-Q4_K_M.gguf
 stopwords:
 - <|im_end|>
 - <dummy32000>
@@ -43,9 +42,9 @@ template:
 
 
 download_files:
-- filename: minicpm-v-4_5-Q4_K_M.gguf
-  sha256: c1c3c33100b15b4caf7319acce4e23c0eb0ce1cbd12f70e8d24f05aa67b7512f
-  uri: huggingface://openbmb/MiniCPM-V-4_5-gguf/ggml-model-Q4_K_M.gguf
-- filename: minicpm-v-4_5-mmproj-f16.gguf
-  uri: huggingface://openbmb/MiniCPM-V-4_5-gguf/mmproj-model-f16.gguf
-  sha256: 7a7225a32e8d453aaa3d22d8c579b5bf833c253f784cdb05c99c9a76fd616df8
+- filename: minicpm-v-2_6-Q4_K_M.gguf
+  sha256: 3a4078d53b46f22989adbf998ce5a3fd090b6541f112d7e936eb4204a04100b1
+  uri: huggingface://openbmb/MiniCPM-V-2_6-gguf/ggml-model-Q4_K_M.gguf
+- filename: minicpm-v-2_6-mmproj-f16.gguf
+  uri: huggingface://openbmb/MiniCPM-V-2_6-gguf/mmproj-model-f16.gguf
+  sha256: 4485f68a0f1aa404c391e788ea88ea653c100d8e98fe572698f701e5809711fd

assets.go

@@ -0,0 +1,15 @@
+package main
+
+import (
+	rice "github.com/GeertJohan/go.rice"
+)
+
+var backendAssets *rice.Box
+
+func init() {
+	var err error
+	backendAssets, err = rice.FindBox("backend-assets")
+	if err != nil {
+		panic(err)
+	}
+}

backend/Dockerfile.golang

@@ -1,131 +0,0 @@
-ARG BASE_IMAGE=ubuntu:22.04
-
-FROM ${BASE_IMAGE} AS builder
-ARG BACKEND=rerankers
-ARG BUILD_TYPE
-ENV BUILD_TYPE=${BUILD_TYPE}
-ARG CUDA_MAJOR_VERSION
-ARG CUDA_MINOR_VERSION
-ARG SKIP_DRIVERS=false
-ENV CUDA_MAJOR_VERSION=${CUDA_MAJOR_VERSION}
-ENV CUDA_MINOR_VERSION=${CUDA_MINOR_VERSION}
-ENV DEBIAN_FRONTEND=noninteractive
-ARG TARGETARCH
-ARG TARGETVARIANT
-ARG GO_VERSION=1.22.6
-
-RUN apt-get update && \
-    apt-get install -y --no-install-recommends \
-        build-essential \
-        git ccache \
-        ca-certificates \
-        make cmake \
-        curl unzip \
-        libssl-dev && \
-    apt-get clean && \
-    rm -rf /var/lib/apt/lists/*
-
-
-# Cuda
-ENV PATH=/usr/local/cuda/bin:${PATH}
-
-# HipBLAS requirements
-ENV PATH=/opt/rocm/bin:${PATH}
-
-# Vulkan requirements
-RUN <<EOT bash
-    if [ "${BUILD_TYPE}" = "vulkan" ] && [ "${SKIP_DRIVERS}" = "false" ]; then
-        apt-get update && \
-        apt-get install -y  --no-install-recommends \
-            software-properties-common pciutils wget gpg-agent && \
-        wget -qO - https://packages.lunarg.com/lunarg-signing-key-pub.asc | apt-key add - && \
-        wget -qO /etc/apt/sources.list.d/lunarg-vulkan-jammy.list https://packages.lunarg.com/vulkan/lunarg-vulkan-jammy.list && \
-        apt-get update && \
-        apt-get install -y \
-            vulkan-sdk && \
-        apt-get clean && \
-        rm -rf /var/lib/apt/lists/*
-    fi
-EOT
-
-# CuBLAS requirements
-RUN <<EOT bash
-    if [ "${BUILD_TYPE}" = "cublas" ] && [ "${SKIP_DRIVERS}" = "false" ]; then
-        apt-get update && \
-        apt-get install -y  --no-install-recommends \
-            software-properties-common pciutils
-        if [ "amd64" = "$TARGETARCH" ]; then
-            curl -O https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/cuda-keyring_1.1-1_all.deb
-        fi
-        if [ "arm64" = "$TARGETARCH" ]; then
-            curl -O https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/arm64/cuda-keyring_1.1-1_all.deb
-        fi
-        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 --no-install-recommends \
-            cuda-nvcc-${CUDA_MAJOR_VERSION}-${CUDA_MINOR_VERSION} \
-            libcufft-dev-${CUDA_MAJOR_VERSION}-${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 && \
-        rm -rf /var/lib/apt/lists/*
-    fi
-EOT
-
-# If we are building with clblas support, we need the libraries for the builds
-RUN if [ "${BUILD_TYPE}" = "clblas" ] && [ "${SKIP_DRIVERS}" = "false" ]; then \
-        apt-get update && \
-        apt-get install -y --no-install-recommends \
-            libclblast-dev && \
-        apt-get clean && \
-        rm -rf /var/lib/apt/lists/* \
-    ; fi
-
-RUN if [ "${BUILD_TYPE}" = "hipblas" ] && [ "${SKIP_DRIVERS}" = "false" ]; then \
-        apt-get update && \
-        apt-get install -y --no-install-recommends \
-            hipblas-dev \
-            rocblas-dev && \
-        apt-get clean && \
-        rm -rf /var/lib/apt/lists/* && \
-        # I have no idea why, but the ROCM lib packages don't trigger ldconfig after they install, which results in local-ai and others not being able
-        # to locate the libraries. We run ldconfig ourselves to work around this packaging deficiency
-        ldconfig \
-    ; fi
-
-# 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:/root/go/bin:/usr/local/go/bin:/usr/local/bin
-
-# Install grpc compilers
-RUN go install google.golang.org/protobuf/cmd/protoc-gen-go@v1.34.2 && \
-    go install google.golang.org/grpc/cmd/protoc-gen-go-grpc@1958fcbe2ca8bd93af633f11e97d44e567e945af
-RUN echo "TARGETARCH: $TARGETARCH"
-
-# We need protoc installed, and the version in 22.04 is too old.  We will create one as part installing the GRPC build below
-# but that will also being in a newer version of absl which stablediffusion cannot compile with.  This version of protoc is only
-# here so that we can generate the grpc code for the stablediffusion build
-RUN <<EOT bash
-    if [ "amd64" = "$TARGETARCH" ]; then
-        curl -L -s https://github.com/protocolbuffers/protobuf/releases/download/v27.1/protoc-27.1-linux-x86_64.zip -o protoc.zip && \
-        unzip -j -d /usr/local/bin protoc.zip bin/protoc && \
-        rm protoc.zip
-    fi
-    if [ "arm64" = "$TARGETARCH" ]; then
-        curl -L -s https://github.com/protocolbuffers/protobuf/releases/download/v27.1/protoc-27.1-linux-aarch_64.zip -o protoc.zip && \
-        unzip -j -d /usr/local/bin protoc.zip bin/protoc && \
-        rm protoc.zip
-    fi
-EOT
-
-COPY . /LocalAI
-
-RUN cd /LocalAI && make protogen-go && make -C /LocalAI/backend/go/${BACKEND} build
-
-FROM scratch
-ARG BACKEND=rerankers
-
-COPY --from=builder /LocalAI/backend/go/${BACKEND}/package/. ./

backend/Dockerfile.llama-cpp

@@ -1,207 +0,0 @@
-ARG BASE_IMAGE=ubuntu:22.04
-ARG GRPC_BASE_IMAGE=${BASE_IMAGE}
-
-
-# The grpc target does one thing, it builds and installs GRPC.  This is in it's own layer so that it can be effectively cached by CI.
-# You probably don't need to change anything here, and if you do, make sure that CI is adjusted so that the cache continues to work.
-FROM ${GRPC_BASE_IMAGE} AS grpc
-
-# This is a bit of a hack, but it's required in order to be able to effectively cache this layer in CI
-ARG GRPC_MAKEFLAGS="-j4 -Otarget"
-ARG GRPC_VERSION=v1.65.0
-ARG CMAKE_FROM_SOURCE=false
-ARG CMAKE_VERSION=3.26.4
-
-ENV MAKEFLAGS=${GRPC_MAKEFLAGS}
-
-WORKDIR /build
-
-RUN apt-get update && \
-    apt-get install -y --no-install-recommends \
-        ca-certificates \
-        build-essential curl libssl-dev \
-        git && \
-    apt-get clean && \
-    rm -rf /var/lib/apt/lists/*
-
-# Install CMake (the version in 22.04 is too old)
-RUN <<EOT bash
-    if [ "${CMAKE_FROM_SOURCE}}" = "true" ]; then
-        curl -L -s https://github.com/Kitware/CMake/releases/download/v${CMAKE_VERSION}/cmake-${CMAKE_VERSION}.tar.gz -o cmake.tar.gz && tar xvf cmake.tar.gz && cd cmake-${CMAKE_VERSION} && ./configure && make && make install
-    else
-        apt-get update && \
-        apt-get install -y \
-            cmake && \
-        apt-get clean && \
-        rm -rf /var/lib/apt/lists/*
-    fi
-EOT
-
-# We install GRPC to a different prefix here so that we can copy in only the build artifacts later
-# saves several hundred MB on the final docker image size vs copying in the entire GRPC source tree
-# and running make install in the target container
-RUN git clone --recurse-submodules --jobs 4 -b ${GRPC_VERSION} --depth 1 --shallow-submodules https://github.com/grpc/grpc && \
-    mkdir -p /build/grpc/cmake/build && \
-    cd /build/grpc/cmake/build && \
-    sed -i "216i\  TESTONLY" "../../third_party/abseil-cpp/absl/container/CMakeLists.txt" && \
-    cmake -DgRPC_INSTALL=ON -DgRPC_BUILD_TESTS=OFF -DCMAKE_INSTALL_PREFIX:PATH=/opt/grpc ../.. && \
-    make && \
-    make install && \
-    rm -rf /build
-
-FROM ${BASE_IMAGE} AS builder
-ARG BACKEND=rerankers
-ARG BUILD_TYPE
-ENV BUILD_TYPE=${BUILD_TYPE}
-ARG CUDA_MAJOR_VERSION
-ARG CUDA_MINOR_VERSION
-ARG SKIP_DRIVERS=false
-ENV CUDA_MAJOR_VERSION=${CUDA_MAJOR_VERSION}
-ENV CUDA_MINOR_VERSION=${CUDA_MINOR_VERSION}
-ENV DEBIAN_FRONTEND=noninteractive
-ARG TARGETARCH
-ARG TARGETVARIANT
-ARG GO_VERSION=1.22.6
-
-RUN apt-get update && \
-    apt-get install -y --no-install-recommends \
-        build-essential \
-        ccache git \
-        ca-certificates \
-        make \
-        curl unzip \
-        libssl-dev && \
-    apt-get clean && \
-    rm -rf /var/lib/apt/lists/*
-
-# Cuda
-ENV PATH=/usr/local/cuda/bin:${PATH}
-
-# HipBLAS requirements
-ENV PATH=/opt/rocm/bin:${PATH}
-
-# Vulkan requirements
-RUN <<EOT bash
-    if [ "${BUILD_TYPE}" = "vulkan" ] && [ "${SKIP_DRIVERS}" = "false" ]; then
-        apt-get update && \
-        apt-get install -y  --no-install-recommends \
-            software-properties-common pciutils wget gpg-agent && \
-        wget -qO - https://packages.lunarg.com/lunarg-signing-key-pub.asc | apt-key add - && \
-        wget -qO /etc/apt/sources.list.d/lunarg-vulkan-jammy.list https://packages.lunarg.com/vulkan/lunarg-vulkan-jammy.list && \
-        apt-get update && \
-        apt-get install -y \
-            vulkan-sdk && \
-        apt-get clean && \
-        rm -rf /var/lib/apt/lists/*
-    fi
-EOT
-
-# CuBLAS requirements
-RUN <<EOT bash
-    if [ "${BUILD_TYPE}" = "cublas" ] && [ "${SKIP_DRIVERS}" = "false" ]; then
-        apt-get update && \
-        apt-get install -y  --no-install-recommends \
-            software-properties-common pciutils
-        if [ "amd64" = "$TARGETARCH" ]; then
-            curl -O https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/cuda-keyring_1.1-1_all.deb
-        fi
-        if [ "arm64" = "$TARGETARCH" ]; then
-            curl -O https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/arm64/cuda-keyring_1.1-1_all.deb
-        fi
-        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 --no-install-recommends \
-            cuda-nvcc-${CUDA_MAJOR_VERSION}-${CUDA_MINOR_VERSION} \
-            libcufft-dev-${CUDA_MAJOR_VERSION}-${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 && \
-        rm -rf /var/lib/apt/lists/*
-    fi
-EOT
-
-# If we are building with clblas support, we need the libraries for the builds
-RUN if [ "${BUILD_TYPE}" = "clblas" ] && [ "${SKIP_DRIVERS}" = "false" ]; then \
-        apt-get update && \
-        apt-get install -y --no-install-recommends \
-            libclblast-dev && \
-        apt-get clean && \
-        rm -rf /var/lib/apt/lists/* \
-    ; fi
-
-RUN if [ "${BUILD_TYPE}" = "hipblas" ] && [ "${SKIP_DRIVERS}" = "false" ]; then \
-        apt-get update && \
-        apt-get install -y --no-install-recommends \
-            hipblas-dev \
-            rocblas-dev && \
-        apt-get clean && \
-        rm -rf /var/lib/apt/lists/* && \
-        # I have no idea why, but the ROCM lib packages don't trigger ldconfig after they install, which results in local-ai and others not being able
-        # to locate the libraries. We run ldconfig ourselves to work around this packaging deficiency
-        ldconfig \
-    ; fi
-
-RUN echo "TARGETARCH: $TARGETARCH"
-
-# We need protoc installed, and the version in 22.04 is too old.  We will create one as part installing the GRPC build below
-# but that will also being in a newer version of absl which stablediffusion cannot compile with.  This version of protoc is only
-# here so that we can generate the grpc code for the stablediffusion build
-RUN <<EOT bash
-    if [ "amd64" = "$TARGETARCH" ]; then
-        curl -L -s https://github.com/protocolbuffers/protobuf/releases/download/v27.1/protoc-27.1-linux-x86_64.zip -o protoc.zip && \
-        unzip -j -d /usr/local/bin protoc.zip bin/protoc && \
-        rm protoc.zip
-    fi
-    if [ "arm64" = "$TARGETARCH" ]; then
-        curl -L -s https://github.com/protocolbuffers/protobuf/releases/download/v27.1/protoc-27.1-linux-aarch_64.zip -o protoc.zip && \
-        unzip -j -d /usr/local/bin protoc.zip bin/protoc && \
-        rm protoc.zip
-    fi
-EOT
-
-# Install CMake (the version in 22.04 is too old)
-RUN <<EOT bash
-    if [ "${CMAKE_FROM_SOURCE}}" = "true" ]; then
-        curl -L -s https://github.com/Kitware/CMake/releases/download/v${CMAKE_VERSION}/cmake-${CMAKE_VERSION}.tar.gz -o cmake.tar.gz && tar xvf cmake.tar.gz && cd cmake-${CMAKE_VERSION} && ./configure && make && make install
-    else
-        apt-get update && \
-        apt-get install -y \
-            cmake && \
-        apt-get clean && \
-        rm -rf /var/lib/apt/lists/*
-    fi
-EOT
-
-COPY --from=grpc /opt/grpc /usr/local
-
-
-COPY . /LocalAI
-
-## Otherwise just run the normal build
-RUN <<EOT bash
-if [ "${TARGETARCH}" = "arm64" ] || [ "${BUILD_TYPE}" = "hipblas" ]; then \
-        cd /LocalAI/backend/cpp/llama-cpp && make llama-cpp-fallback && \
-        make llama-cpp-grpc && make llama-cpp-rpc-server; \
-    else \
-        cd /LocalAI/backend/cpp/llama-cpp && make llama-cpp-avx && \
-        make llama-cpp-avx2 && \
-        make llama-cpp-avx512 && \
-        make llama-cpp-fallback && \
-        make llama-cpp-grpc && \
-        make llama-cpp-rpc-server; \
-    fi  
-EOT
-
-
-# Copy libraries using a script to handle architecture differences
-RUN make -C /LocalAI/backend/cpp/llama-cpp package
-
-
-FROM scratch
-
-
-# Copy all available binaries (the build process only creates the appropriate ones for the target architecture)
-COPY --from=builder /LocalAI/backend/cpp/llama-cpp/package/. ./

backend/Dockerfile.python

@@ -1,123 +0,0 @@
-ARG BASE_IMAGE=ubuntu:22.04
-
-FROM ${BASE_IMAGE} AS builder
-ARG BACKEND=rerankers
-ARG BUILD_TYPE
-ENV BUILD_TYPE=${BUILD_TYPE}
-ARG CUDA_MAJOR_VERSION
-ARG CUDA_MINOR_VERSION
-ARG SKIP_DRIVERS=false
-ENV CUDA_MAJOR_VERSION=${CUDA_MAJOR_VERSION}
-ENV CUDA_MINOR_VERSION=${CUDA_MINOR_VERSION}
-ENV DEBIAN_FRONTEND=noninteractive
-ARG TARGETARCH
-ARG TARGETVARIANT
-
-RUN apt-get update && \
-    apt-get install -y --no-install-recommends \
-        build-essential \
-        ccache \
-        ca-certificates \
-        espeak-ng \
-        curl \
-        libssl-dev \
-        git \
-        git-lfs \
-        unzip clang \
-        upx-ucl \
-        curl python3-pip \
-        python-is-python3 \
-        python3-dev llvm \
-        python3-venv make && \
-    apt-get clean && \
-    rm -rf /var/lib/apt/lists/* && \
-    pip install --upgrade pip
-
-
-# Cuda
-ENV PATH=/usr/local/cuda/bin:${PATH}
-
-# HipBLAS requirements
-ENV PATH=/opt/rocm/bin:${PATH}
-
-# Vulkan requirements
-RUN <<EOT bash
-    if [ "${BUILD_TYPE}" = "vulkan" ] && [ "${SKIP_DRIVERS}" = "false" ]; then
-        apt-get update && \
-        apt-get install -y  --no-install-recommends \
-            software-properties-common pciutils wget gpg-agent && \
-        wget -qO - https://packages.lunarg.com/lunarg-signing-key-pub.asc | apt-key add - && \
-        wget -qO /etc/apt/sources.list.d/lunarg-vulkan-jammy.list https://packages.lunarg.com/vulkan/lunarg-vulkan-jammy.list && \
-        apt-get update && \
-        apt-get install -y \
-            vulkan-sdk && \
-        apt-get clean && \
-        rm -rf /var/lib/apt/lists/*
-    fi
-EOT
-
-# CuBLAS requirements
-RUN <<EOT bash
-    if [ "${BUILD_TYPE}" = "cublas" ] && [ "${SKIP_DRIVERS}" = "false" ]; then
-        apt-get update && \
-        apt-get install -y  --no-install-recommends \
-            software-properties-common pciutils
-        if [ "amd64" = "$TARGETARCH" ]; then
-            curl -O https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/cuda-keyring_1.1-1_all.deb
-        fi
-        if [ "arm64" = "$TARGETARCH" ]; then
-            curl -O https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/arm64/cuda-keyring_1.1-1_all.deb
-        fi
-        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 --no-install-recommends \
-            cuda-nvcc-${CUDA_MAJOR_VERSION}-${CUDA_MINOR_VERSION} \
-            libcufft-dev-${CUDA_MAJOR_VERSION}-${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 && \
-        rm -rf /var/lib/apt/lists/*
-    fi
-EOT
-
-# If we are building with clblas support, we need the libraries for the builds
-RUN if [ "${BUILD_TYPE}" = "clblas" ] && [ "${SKIP_DRIVERS}" = "false" ]; then \
-        apt-get update && \
-        apt-get install -y --no-install-recommends \
-            libclblast-dev && \
-        apt-get clean && \
-        rm -rf /var/lib/apt/lists/* \
-    ; fi
-
-RUN if [ "${BUILD_TYPE}" = "hipblas" ] && [ "${SKIP_DRIVERS}" = "false" ]; then \
-        apt-get update && \
-        apt-get install -y --no-install-recommends \
-            hipblas-dev \
-            rocblas-dev && \
-        apt-get clean && \
-        rm -rf /var/lib/apt/lists/* && \
-        # I have no idea why, but the ROCM lib packages don't trigger ldconfig after they install, which results in local-ai and others not being able
-        # to locate the libraries. We run ldconfig ourselves to work around this packaging deficiency
-        ldconfig \
-    ; fi
-# Install uv as a system package
-RUN curl -LsSf https://astral.sh/uv/install.sh | UV_INSTALL_DIR=/usr/bin sh
-ENV PATH="/root/.cargo/bin:${PATH}"
-
-RUN curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y
-
-# Install grpcio-tools (the version in 22.04 is too old)
-RUN pip install --user grpcio-tools==1.71.0 grpcio==1.71.0
-
-COPY python/${BACKEND} /${BACKEND}
-COPY backend.proto /${BACKEND}/backend.proto
-COPY python/common/ /${BACKEND}/common
-
-RUN cd /${BACKEND} && PORTABLE_PYTHON=true make
-
-FROM scratch
-ARG BACKEND=rerankers
-COPY --from=builder /${BACKEND}/ /

backend/README.md

@@ -1,213 +0,0 @@
-# LocalAI Backend Architecture
-
-This directory contains the core backend infrastructure for LocalAI, including the gRPC protocol definition, multi-language Dockerfiles, and language-specific backend implementations.
-
-## Overview
-
-LocalAI uses a unified gRPC-based architecture that allows different programming languages to implement AI backends while maintaining consistent interfaces and capabilities. The backend system supports multiple hardware acceleration targets and provides a standardized way to integrate various AI models and frameworks.
-
-## Architecture Components
-
-### 1. Protocol Definition (`backend.proto`)
-
-The `backend.proto` file defines the gRPC service interface that all backends must implement. This ensures consistency across different language implementations and provides a contract for communication between LocalAI core and backend services.
-
-#### Core Services
-
-- **Text Generation**: `Predict`, `PredictStream` for LLM inference
-- **Embeddings**: `Embedding` for text vectorization
-- **Image Generation**: `GenerateImage` for stable diffusion and image models
-- **Audio Processing**: `AudioTranscription`, `TTS`, `SoundGeneration`
-- **Video Generation**: `GenerateVideo` for video synthesis
-- **Object Detection**: `Detect` for computer vision tasks
-- **Vector Storage**: `StoresSet`, `StoresGet`, `StoresFind` for RAG operations
-- **Reranking**: `Rerank` for document relevance scoring
-- **Voice Activity Detection**: `VAD` for audio segmentation
-
-#### Key Message Types
-
-- **`PredictOptions`**: Comprehensive configuration for text generation
-- **`ModelOptions`**: Model loading and configuration parameters
-- **`Result`**: Standardized response format
-- **`StatusResponse`**: Backend health and memory usage information
-
-### 2. Multi-Language Dockerfiles
-
-The backend system provides language-specific Dockerfiles that handle the build environment and dependencies for different programming languages:
-
-- `Dockerfile.python`
-- `Dockerfile.golang`
-- `Dockerfile.llama-cpp`
-
-### 3. Language-Specific Implementations
-
-#### Python Backends (`python/`)
-- **transformers**: Hugging Face Transformers framework
-- **vllm**: High-performance LLM inference
-- **mlx**: Apple Silicon optimization
-- **diffusers**: Stable Diffusion models
-- **Audio**: bark, coqui, faster-whisper, kitten-tts
-- **Vision**: mlx-vlm, rfdetr
-- **Specialized**: rerankers, chatterbox, kokoro
-
-#### Go Backends (`go/`)
-- **whisper**: OpenAI Whisper speech recognition in Go with GGML cpp backend (whisper.cpp)
-- **stablediffusion-ggml**: Stable Diffusion in Go with GGML Cpp backend
-- **huggingface**: Hugging Face model integration
-- **piper**: Text-to-speech synthesis Golang with C bindings using rhaspy/piper
-- **bark-cpp**: Bark TTS models Golang with Cpp bindings
-- **local-store**: Vector storage backend
-
-#### C++ Backends (`cpp/`)
-- **llama-cpp**: Llama.cpp integration
-- **grpc**: GRPC utilities and helpers
-
-## Hardware Acceleration Support
-
-### CUDA (NVIDIA)
-- **Versions**: CUDA 11.x, 12.x
-- **Features**: cuBLAS, cuDNN, TensorRT optimization
-- **Targets**: x86_64, ARM64 (Jetson)
-
-### ROCm (AMD)
-- **Features**: HIP, rocBLAS, MIOpen
-- **Targets**: AMD GPUs with ROCm support
-
-### Intel
-- **Features**: oneAPI, Intel Extension for PyTorch
-- **Targets**: Intel GPUs, XPUs, CPUs
-
-### Vulkan
-- **Features**: Cross-platform GPU acceleration
-- **Targets**: Windows, Linux, Android, macOS
-
-### Apple Silicon
-- **Features**: MLX framework, Metal Performance Shaders
-- **Targets**: M1/M2/M3 Macs
-
-## Backend Registry (`index.yaml`)
-
-The `index.yaml` file serves as a central registry for all available backends, providing:
-
-- **Metadata**: Name, description, license, icons
-- **Capabilities**: Hardware targets and optimization profiles
-- **Tags**: Categorization for discovery
-- **URLs**: Source code and documentation links
-
-## Building Backends
-
-### Prerequisites
-- Docker with multi-architecture support
-- Appropriate hardware drivers (CUDA, ROCm, etc.)
-- Build tools (make, cmake, compilers)
-
-### Build Commands
-
-Example of build commands with Docker
-
-```bash
-# Build Python backend
-docker build -f backend/Dockerfile.python \
-  --build-arg BACKEND=transformers \
-  --build-arg BUILD_TYPE=cublas12 \
-  --build-arg CUDA_MAJOR_VERSION=12 \
-  --build-arg CUDA_MINOR_VERSION=0 \
-  -t localai-backend-transformers .
-
-# Build Go backend
-docker build -f backend/Dockerfile.golang \
-  --build-arg BACKEND=whisper \
-  --build-arg BUILD_TYPE=cpu \
-  -t localai-backend-whisper .
-
-# Build C++ backend
-docker build -f backend/Dockerfile.llama-cpp \
-  --build-arg BACKEND=llama-cpp \
-  --build-arg BUILD_TYPE=cublas12 \
-  -t localai-backend-llama-cpp .
-```
-
-For ARM64/Mac builds, docker can't be used, and the makefile in the respective backend has to be used.
-
-### Build Types
-
-- **`cpu`**: CPU-only optimization
-- **`cublas11`**: CUDA 11.x with cuBLAS
-- **`cublas12`**: CUDA 12.x with cuBLAS
-- **`hipblas`**: ROCm with rocBLAS
-- **`intel`**: Intel oneAPI optimization
-- **`vulkan`**: Vulkan-based acceleration
-- **`metal`**: Apple Metal optimization
-
-## Backend Development
-
-### Creating a New Backend
-
-1. **Choose Language**: Select Python, Go, or C++ based on requirements
-2. **Implement Interface**: Implement the gRPC service defined in `backend.proto`
-3. **Add Dependencies**: Create appropriate requirements files
-4. **Configure Build**: Set up Dockerfile and build scripts
-5. **Register Backend**: Add entry to `index.yaml`
-6. **Test Integration**: Verify gRPC communication and functionality
-
-### Backend Structure
-
-```
-backend-name/
-├── backend.py/go/cpp    # Main implementation
-├── requirements.txt      # Dependencies
-├── Dockerfile           # Build configuration
-├── install.sh           # Installation script
-├── run.sh              # Execution script
-├── test.sh             # Test script
-└── README.md           # Backend documentation
-```
-
-### Required gRPC Methods
-
-At minimum, backends must implement:
-- `Health()` - Service health check
-- `LoadModel()` - Model loading and initialization
-- `Predict()` - Main inference endpoint
-- `Status()` - Backend status and metrics
-
-## Integration with LocalAI Core
-
-Backends communicate with LocalAI core through gRPC:
-
-1. **Service Discovery**: Core discovers available backends
-2. **Model Loading**: Core requests model loading via `LoadModel`
-3. **Inference**: Core sends requests via `Predict` or specialized endpoints
-4. **Streaming**: Core handles streaming responses for real-time generation
-5. **Monitoring**: Core tracks backend health and performance
-
-## Performance Optimization
-
-### Memory Management
-- **Model Caching**: Efficient model loading and caching
-- **Batch Processing**: Optimize for multiple concurrent requests
-- **Memory Pinning**: GPU memory optimization for CUDA/ROCm
-
-### Hardware Utilization
-- **Multi-GPU**: Support for tensor parallelism
-- **Mixed Precision**: FP16/BF16 for memory efficiency
-- **Kernel Fusion**: Optimized CUDA/ROCm kernels
-
-## Troubleshooting
-
-### Common Issues
-
-1. **GRPC Connection**: Verify backend service is running and accessible
-2. **Model Loading**: Check model paths and dependencies
-3. **Hardware Detection**: Ensure appropriate drivers and libraries
-4. **Memory Issues**: Monitor GPU memory usage and model sizes
-
-## Contributing
-
-When contributing to the backend system:
-
-1. **Follow Protocol**: Implement the exact gRPC interface
-2. **Add Tests**: Include comprehensive test coverage
-3. **Document**: Provide clear usage examples
-4. **Optimize**: Consider performance and resource usage
-5. **Validate**: Test across different hardware targets

backend/backend.proto

@@ -20,7 +20,6 @@ service Backend {
   rpc SoundGeneration(SoundGenerationRequest) returns (Result) {}
   rpc TokenizeString(PredictOptions) returns (TokenizationResponse) {}
   rpc Status(HealthMessage) returns (StatusResponse) {}
-  rpc Detect(DetectOptions) returns (DetectResponse) {}
 
   rpc StoresSet(StoresSetOptions) returns (Result) {}
   rpc StoresDelete(StoresDeleteOptions) returns (Result) {}
@@ -163,7 +162,6 @@ message Reply {
   int32 prompt_tokens = 3;
   double timing_prompt_processing = 4;
   double timing_token_generation = 5;
-  bytes audio = 6;
 }
 
 message GrammarTrigger {
@@ -186,6 +184,7 @@ message ModelOptions {
   string MainGPU = 13;
   string TensorSplit = 14;
   int32 Threads = 15;
+  string LibrarySearchPath = 16;
   float RopeFreqBase = 17;
   float RopeFreqScale = 18;
   float RMSNormEps = 19;
@@ -242,7 +241,7 @@ message ModelOptions {
 
   string Type = 49;
 
-  string FlashAttention = 56;
+  bool FlashAttention = 56;
   bool NoKVOffload = 57;
 
   string ModelPath = 59;
@@ -256,10 +255,6 @@ message ModelOptions {
   string CacheTypeValue = 64;
 
   repeated GrammarTrigger GrammarTriggers = 65;
-
-  bool Reranking = 71;
-
-  repeated string Overrides = 72;
 }
 
 message Result {
@@ -276,7 +271,6 @@ message TranscriptRequest {
   string language = 3;
   uint32 threads = 4;
   bool translate = 5;
-  bool diarize = 6;
 }
 
 message TranscriptResult {
@@ -306,24 +300,19 @@ message GenerateImageRequest {
   // Diffusers
   string EnableParameters = 10;
   int32 CLIPSkip = 11;
-
-  // Reference images for models that support them (e.g., Flux Kontext)
-  repeated string ref_images = 12;
 }
 
 message GenerateVideoRequest {
   string prompt = 1;
-  string negative_prompt = 2;  // Negative prompt for video generation
-  string start_image = 3;  // Path or base64 encoded image for the start frame
-  string end_image = 4;    // Path or base64 encoded image for the end frame
-  int32 width = 5;
-  int32 height = 6;
-  int32 num_frames = 7;    // Number of frames to generate
-  int32 fps = 8;          // Frames per second
-  int32 seed = 9;
-  float cfg_scale = 10;    // Classifier-free guidance scale
-  int32 step = 11;         // Number of inference steps
-  string dst = 12;        // Output path for the generated video
+  string start_image = 2;  // Path or base64 encoded image for the start frame
+  string end_image = 3;    // Path or base64 encoded image for the end frame
+  int32 width = 4;
+  int32 height = 5;
+  int32 num_frames = 6;    // Number of frames to generate
+  int32 fps = 7;          // Frames per second
+  int32 seed = 8;
+  float cfg_scale = 9;    // Classifier-free guidance scale
+  string dst = 10;        // Output path for the generated video
 }
 
 message TTSRequest {
@@ -383,20 +372,3 @@ message Message {
   string role = 1;
   string content = 2;
 }
-
-message DetectOptions {
-  string src = 1;
-}
-
-message Detection {
-  float x = 1;
-  float y = 2;
-  float width = 3;
-  float height = 4;
-  float confidence = 5;
-  string class_name = 6;
-}
-
-message DetectResponse {
-  repeated Detection Detections = 1;
-}

backend/cpp/llama-cpp/Makefile

@@ -1,166 +0,0 @@
-
-LLAMA_VERSION?=d64c8104f090b27b1f99e8da5995ffcfa6b726e2
-LLAMA_REPO?=https://github.com/ggerganov/llama.cpp
-
-CMAKE_ARGS?=
-BUILD_TYPE?=
-NATIVE?=false
-ONEAPI_VARS?=/opt/intel/oneapi/setvars.sh
-TARGET?=--target grpc-server
-JOBS?=$(shell nproc)
-
-# Disable Shared libs as we are linking on static gRPC and we can't mix shared and static
-CMAKE_ARGS+=-DBUILD_SHARED_LIBS=OFF -DLLAMA_CURL=OFF
-
-CURRENT_MAKEFILE_DIR := $(dir $(abspath $(lastword $(MAKEFILE_LIST))))
-ifeq ($(NATIVE),false)
-	CMAKE_ARGS+=-DGGML_NATIVE=OFF -DLLAMA_OPENSSL=OFF
-endif
-# If build type is cublas, then we set -DGGML_CUDA=ON to CMAKE_ARGS automatically
-ifeq ($(BUILD_TYPE),cublas)
-	CMAKE_ARGS+=-DGGML_CUDA=ON
-# If build type is openblas then we set -DGGML_BLAS=ON -DGGML_BLAS_VENDOR=OpenBLAS
-# to CMAKE_ARGS automatically
-else ifeq ($(BUILD_TYPE),openblas)
-	CMAKE_ARGS+=-DGGML_BLAS=ON -DGGML_BLAS_VENDOR=OpenBLAS
-# If build type is clblas (openCL) we set -DGGML_CLBLAST=ON -DCLBlast_DIR=/some/path
-else ifeq ($(BUILD_TYPE),clblas)
-	CMAKE_ARGS+=-DGGML_CLBLAST=ON -DCLBlast_DIR=/some/path
-# If it's hipblas we do have also to set CC=/opt/rocm/llvm/bin/clang CXX=/opt/rocm/llvm/bin/clang++
-else ifeq ($(BUILD_TYPE),hipblas)
-	ROCM_HOME ?= /opt/rocm
-	ROCM_PATH ?= /opt/rocm
-	export CXX=$(ROCM_HOME)/llvm/bin/clang++
-	export CC=$(ROCM_HOME)/llvm/bin/clang
-	AMDGPU_TARGETS?=gfx803,gfx900,gfx906,gfx908,gfx90a,gfx942,gfx1010,gfx1030,gfx1032,gfx1100,gfx1101,gfx1102,gfx1200,gfx1201
-	CMAKE_ARGS+=-DGGML_HIP=ON -DAMDGPU_TARGETS=$(AMDGPU_TARGETS)
-else ifeq ($(BUILD_TYPE),vulkan)
-	CMAKE_ARGS+=-DGGML_VULKAN=1
-else ifeq ($(OS),Darwin)
-	ifeq ($(BUILD_TYPE),)
-		BUILD_TYPE=metal
-	endif
-	ifneq ($(BUILD_TYPE),metal)
-		CMAKE_ARGS+=-DGGML_METAL=OFF
-	else
-		CMAKE_ARGS+=-DGGML_METAL=ON
-		CMAKE_ARGS+=-DGGML_METAL_EMBED_LIBRARY=ON
-		CMAKE_ARGS+=-DGGML_METAL_USE_BF16=ON
-		CMAKE_ARGS+=-DGGML_OPENMP=OFF
-	endif
-	TARGET+=--target ggml-metal
-endif
-
-ifeq ($(BUILD_TYPE),sycl_f16)
-	CMAKE_ARGS+=-DGGML_SYCL=ON \
-		-DCMAKE_C_COMPILER=icx \
-		-DCMAKE_CXX_COMPILER=icpx \
-		-DCMAKE_CXX_FLAGS="-fsycl" \
-		-DGGML_SYCL_F16=ON
-endif
-
-ifeq ($(BUILD_TYPE),sycl_f32)
-	CMAKE_ARGS+=-DGGML_SYCL=ON \
-		-DCMAKE_C_COMPILER=icx \
-		-DCMAKE_CXX_COMPILER=icpx \
-		-DCMAKE_CXX_FLAGS="-fsycl"
-endif
-
-INSTALLED_PACKAGES=$(CURDIR)/../grpc/installed_packages
-INSTALLED_LIB_CMAKE=$(INSTALLED_PACKAGES)/lib/cmake
-ADDED_CMAKE_ARGS=-Dabsl_DIR=${INSTALLED_LIB_CMAKE}/absl \
-				 -DProtobuf_DIR=${INSTALLED_LIB_CMAKE}/protobuf \
-				 -Dutf8_range_DIR=${INSTALLED_LIB_CMAKE}/utf8_range \
-				 -DgRPC_DIR=${INSTALLED_LIB_CMAKE}/grpc \
-				 -DCMAKE_CXX_STANDARD_INCLUDE_DIRECTORIES=${INSTALLED_PACKAGES}/include
-build-llama-cpp-grpc-server:
-# Conditionally build grpc for the llama backend to use if needed
-ifdef BUILD_GRPC_FOR_BACKEND_LLAMA
-	$(MAKE) -C ../../grpc build
-	_PROTOBUF_PROTOC=${INSTALLED_PACKAGES}/bin/proto \
-	_GRPC_CPP_PLUGIN_EXECUTABLE=${INSTALLED_PACKAGES}/bin/grpc_cpp_plugin \
-	PATH="${INSTALLED_PACKAGES}/bin:${PATH}" \
-	CMAKE_ARGS="${CMAKE_ARGS} ${ADDED_CMAKE_ARGS}" \
-	LLAMA_VERSION=$(LLAMA_VERSION) \
-	$(MAKE) -C $(CURRENT_MAKEFILE_DIR)/../$(VARIANT) grpc-server
-else
-	echo "BUILD_GRPC_FOR_BACKEND_LLAMA is not defined."
-	LLAMA_VERSION=$(LLAMA_VERSION) $(MAKE) -C $(CURRENT_MAKEFILE_DIR)/../$(VARIANT) grpc-server
-endif
-
-llama-cpp-avx2: llama.cpp
-	cp -rf $(CURRENT_MAKEFILE_DIR)/../llama-cpp $(CURRENT_MAKEFILE_DIR)/../llama-cpp-avx2-build
-	$(MAKE) -C $(CURRENT_MAKEFILE_DIR)/../llama-cpp-avx2-build purge
-	$(info ${GREEN}I llama-cpp build info:avx2${RESET})
-	CMAKE_ARGS="$(CMAKE_ARGS) -DGGML_AVX=on -DGGML_AVX2=on -DGGML_AVX512=off -DGGML_FMA=on -DGGML_F16C=on" $(MAKE) VARIANT="llama-cpp-avx2-build" build-llama-cpp-grpc-server
-	cp -rfv $(CURRENT_MAKEFILE_DIR)/../llama-cpp-avx2-build/grpc-server llama-cpp-avx2
-
-llama-cpp-avx512: llama.cpp
-	cp -rf $(CURRENT_MAKEFILE_DIR)/../llama-cpp $(CURRENT_MAKEFILE_DIR)/../llama-cpp-avx512-build
-	$(MAKE) -C $(CURRENT_MAKEFILE_DIR)/../llama-cpp-avx512-build purge
-	$(info ${GREEN}I llama-cpp build info:avx512${RESET})
-	CMAKE_ARGS="$(CMAKE_ARGS) -DGGML_AVX=on -DGGML_AVX2=off -DGGML_AVX512=on -DGGML_FMA=on -DGGML_F16C=on" $(MAKE) VARIANT="llama-cpp-avx512-build" build-llama-cpp-grpc-server
-	cp -rfv $(CURRENT_MAKEFILE_DIR)/../llama-cpp-avx512-build/grpc-server llama-cpp-avx512
-
-llama-cpp-avx: llama.cpp
-	cp -rf $(CURRENT_MAKEFILE_DIR)/../llama-cpp $(CURRENT_MAKEFILE_DIR)/../llama-cpp-avx-build
-	$(MAKE) -C $(CURRENT_MAKEFILE_DIR)/../llama-cpp-avx-build purge
-	$(info ${GREEN}I llama-cpp build info:avx${RESET})
-	CMAKE_ARGS="$(CMAKE_ARGS) -DGGML_AVX=on -DGGML_AVX2=off -DGGML_AVX512=off -DGGML_FMA=off -DGGML_F16C=off" $(MAKE) VARIANT="llama-cpp-avx-build" build-llama-cpp-grpc-server
-	cp -rfv $(CURRENT_MAKEFILE_DIR)/../llama-cpp-avx-build/grpc-server llama-cpp-avx
-
-llama-cpp-fallback: llama.cpp
-	cp -rf $(CURRENT_MAKEFILE_DIR)/../llama-cpp $(CURRENT_MAKEFILE_DIR)/../llama-cpp-fallback-build
-	$(MAKE) -C $(CURRENT_MAKEFILE_DIR)/../llama-cpp-fallback-build purge
-	$(info ${GREEN}I llama-cpp build info:fallback${RESET})
-	CMAKE_ARGS="$(CMAKE_ARGS) -DGGML_AVX=off -DGGML_AVX2=off -DGGML_AVX512=off -DGGML_FMA=off -DGGML_F16C=off" $(MAKE) VARIANT="llama-cpp-fallback-build" build-llama-cpp-grpc-server
-	cp -rfv $(CURRENT_MAKEFILE_DIR)/../llama-cpp-fallback-build/grpc-server llama-cpp-fallback
-
-llama-cpp-grpc: llama.cpp
-	cp -rf $(CURRENT_MAKEFILE_DIR)/../llama-cpp $(CURRENT_MAKEFILE_DIR)/../llama-cpp-grpc-build
-	$(MAKE) -C $(CURRENT_MAKEFILE_DIR)/../llama-cpp-grpc-build purge
-	$(info ${GREEN}I llama-cpp build info:grpc${RESET})
-	CMAKE_ARGS="$(CMAKE_ARGS) -DGGML_RPC=ON -DGGML_AVX=off -DGGML_AVX2=off -DGGML_AVX512=off -DGGML_FMA=off -DGGML_F16C=off" TARGET="--target grpc-server --target rpc-server" $(MAKE) VARIANT="llama-cpp-grpc-build" build-llama-cpp-grpc-server
-	cp -rfv $(CURRENT_MAKEFILE_DIR)/../llama-cpp-grpc-build/grpc-server llama-cpp-grpc
-
-llama-cpp-rpc-server: llama-cpp-grpc
-	cp -rf $(CURRENT_MAKEFILE_DIR)/../llama-cpp-grpc-build/llama.cpp/build/bin/rpc-server llama-cpp-rpc-server
-
-llama.cpp:
-	mkdir -p llama.cpp
-	cd llama.cpp && \
-	git init && \
-	git remote add origin $(LLAMA_REPO)  && \
-	git fetch origin && \
-	git checkout -b build $(LLAMA_VERSION) && \
-	git submodule update --init --recursive --depth 1 --single-branch
-
-llama.cpp/tools/grpc-server: llama.cpp
-	mkdir -p llama.cpp/tools/grpc-server
-	bash prepare.sh
-
-rebuild:
-	bash prepare.sh
-	rm -rf grpc-server
-	$(MAKE) grpc-server
-
-package:
-	bash package.sh
-
-purge:
-	rm -rf llama.cpp/build
-	rm -rf llama.cpp/tools/grpc-server
-	rm -rf grpc-server
-
-clean: purge
-	rm -rf llama.cpp
-
-grpc-server: llama.cpp llama.cpp/tools/grpc-server
-	@echo "Building grpc-server with $(BUILD_TYPE) build type and $(CMAKE_ARGS)"
-ifneq (,$(findstring sycl,$(BUILD_TYPE)))
-	+bash -c "source $(ONEAPI_VARS); \
-	cd llama.cpp && mkdir -p build && cd build && cmake .. $(CMAKE_ARGS) && cmake --build . --config Release -j $(JOBS) $(TARGET)"
-else
-	+cd llama.cpp && mkdir -p build && cd build && cmake .. $(CMAKE_ARGS) && cmake --build . --config Release -j $(JOBS) $(TARGET)
-endif
-	cp llama.cpp/build/bin/grpc-server .

backend/cpp/llama-cpp/grpc-server.cpp

@@ -1,996 +0,0 @@
-// llama.cpp gRPC C++ backend server
-//
-// Ettore Di Giacinto <mudler@localai.io> and llama.cpp authors
-//
-// This is a gRPC server for llama.cpp compatible with the LocalAI proto
-// Note: this is a re-adaptation of the original llama.cpp example/server.cpp for HTTP (https://github.com/ggerganov/llama.cpp/tree/master/examples/server), 
-// but modified to work with gRPC
-//
-
-#include "server.cpp"
-// LocalAI
-
-#include "backend.pb.h"
-#include "backend.grpc.pb.h"
-#include "common.h"
-#include <getopt.h>
-#include <grpcpp/ext/proto_server_reflection_plugin.h>
-#include <grpcpp/grpcpp.h>
-#include <grpcpp/health_check_service_interface.h>
-#include <regex>
-
-
-using grpc::Server;
-using grpc::ServerBuilder;
-using grpc::ServerContext;
-using grpc::Status;
-// END LocalAI
-
-
-
-
-/////////////////////////////////
-////////////////////////////////
-//////// LOCALAI code starts below here
-/////////////////////////////////
-////////////////////////////////
-
-bool loaded_model; // TODO: add a mutex for this, but happens only once loading the model
-
-static void start_llama_server(server_context& ctx_server) {
-
-    LOG_INF("%s: starting llama server\n", __func__);
-
-    LOG_INF("%s: waiting for model to be loaded\n", __func__);
-    // Wait for model to be loaded first
-    while (!loaded_model) {
-        std::this_thread::sleep_for(std::chrono::milliseconds(100));
-    }
-
-    ctx_server.init();
-    //state.store(SERVER_STATE_READY);
-
-    LOG_INF("%s: model loaded\n", __func__);
-
-    // print sample chat example to make it clear which template is used
-    // LOG_INF("%s: chat template, chat_template: %s, example_format: '%s'\n", __func__,
-    //     common_chat_templates_source(ctx_server.chat_templates.get()),
-    //     common_chat_format_example(ctx_server.chat_templates.get(), ctx_server.params_base.use_jinja).c_str(), ctx_server.params_base.default_template_kwargs);
-
-    // Reset the chat templates
-    // TODO: We should make this configurable by respecting the option that is already present in LocalAI for vLLM
-    ctx_server.chat_templates.reset();
-
-    ctx_server.queue_tasks.on_new_task([&ctx_server](server_task && task) {
-        ctx_server.process_single_task(std::move(task));
-    });
-
-    ctx_server.queue_tasks.on_update_slots([&ctx_server]() {
-        ctx_server.update_slots();
-    });
-
-    shutdown_handler = [&](int) {
-        // this will unblock start_loop()
-        ctx_server.queue_tasks.terminate();
-    };
-
-#if defined (__unix__) || (defined (__APPLE__) && defined (__MACH__))
-    struct sigaction sigint_action;
-    sigint_action.sa_handler = signal_handler;
-    sigemptyset (&sigint_action.sa_mask);
-    sigint_action.sa_flags = 0;
-    sigaction(SIGINT, &sigint_action, NULL);
-    sigaction(SIGTERM, &sigint_action, NULL);
-#elif defined (_WIN32)
-    auto console_ctrl_handler = +[](DWORD ctrl_type) -> BOOL {
-        return (ctrl_type == CTRL_C_EVENT) ? (signal_handler(SIGINT), true) : false;
-    };
-    SetConsoleCtrlHandler(reinterpret_cast<PHANDLER_ROUTINE>(console_ctrl_handler), true);
-#endif
-
-    // this call blocks the main thread until queue_tasks.terminate() is called
-    ctx_server.queue_tasks.start_loop();
-}
-
-json parse_options(bool streaming, const backend::PredictOptions* predict)
-{
-    
-    // Create now a json data from the prediction options instead
-    //
-    json data;
-    data["stream"] = streaming;
-    data["cache_prompt"] = predict->promptcacheall();
-    data["n_predict"] = predict->tokens() == 0 ? -1 : predict->tokens();
-    data["top_k"] = predict->topk();
-    data["top_p"] = predict->topp();
-    data["typical_p"] = predict->typicalp();
-    data["temperature"] = predict->temperature();
-    data["repeat_last_n"] = predict->repeat();
-    data["repeat_penalty"] = predict->penalty();
-    data["frequency_penalty"] = predict->frequencypenalty();
-    data["presence_penalty"] = predict->presencepenalty();
-    data["mirostat"] = predict->mirostat();
-    data["mirostat_tau"] = predict->mirostattau();
-    data["mirostat_eta"] = predict->mirostateta();
-    data["n_keep"] = predict->nkeep();
-    data["seed"] = predict->seed();
-    data["grammar"] = predict->grammar();
-    data["prompt"] = predict->prompt();
-    data["ignore_eos"] = predict->ignoreeos();
-    data["embeddings"] = predict->embeddings();
-    // TODO: add back json_schema and let this be controlled by the user
-    // data["json_schema"] = predict->jsonschema();
-
-    // Add the correlationid to json data
-    data["correlation_id"] = predict->correlationid();
-
-    // for each image in the request, add the image data
-    //
-    for (int i = 0; i < predict->images_size(); i++) {
-        data["image_data"].push_back(json
-            {
-                {"id", i},
-                {"data",    predict->images(i)},
-            });
-    }
-
-    // for each audio in the request, add the audio data
-    for (int i = 0; i < predict->audios_size(); i++) {
-        data["audio_data"].push_back(json
-            {
-                {"id", i},
-                {"data",    predict->audios(i)},
-            });
-    }
-
-    data["stop"] = predict->stopprompts();
-    // data["n_probs"] = predict->nprobs();
-    //TODO: images,
-
-    return data;
-}
-
-
-const std::vector<ggml_type> kv_cache_types = {
-    GGML_TYPE_F32,
-    GGML_TYPE_F16,
-    GGML_TYPE_BF16,
-    GGML_TYPE_Q8_0,
-    GGML_TYPE_Q4_0,
-    GGML_TYPE_Q4_1,
-    GGML_TYPE_IQ4_NL,
-    GGML_TYPE_Q5_0,
-    GGML_TYPE_Q5_1,
-};
-
-static ggml_type kv_cache_type_from_str(const std::string & s) {
-    for (const auto & type : kv_cache_types) {
-        if (ggml_type_name(type) == s) {
-            return type;
-        }
-    }
-    throw std::runtime_error("Unsupported cache type: " + s);
-}
-
-static std::string get_all_kv_cache_types() {
-    std::ostringstream msg;
-    for (const auto & type : kv_cache_types) {
-        msg << ggml_type_name(type) << (&type == &kv_cache_types.back() ? "" : ", ");
-    }
-    return msg.str();
-}
-
-
-// Adds an RPC server
-// https://github.com/ggerganov/llama.cpp/compare/4dbc8b9cb71876e005724f4e8f73a3544646bcf5..3edfa7d3753c29e44b964c0ff424d2ea8d5fdee6
-static void add_rpc_devices(std::string servers) {
-    auto rpc_servers = string_split<std::string>(servers, ',');
-    if (rpc_servers.empty()) {
-        throw std::invalid_argument("no RPC servers specified");
-    }
-    ggml_backend_reg_t rpc_reg = ggml_backend_reg_by_name("RPC");
-    if (!rpc_reg) {
-        throw std::invalid_argument("failed to find RPC backend");
-    }
-    typedef ggml_backend_dev_t (*ggml_backend_rpc_add_device_t)(const char * endpoint);
-    ggml_backend_rpc_add_device_t ggml_backend_rpc_add_device_fn = (ggml_backend_rpc_add_device_t) ggml_backend_reg_get_proc_address(rpc_reg, "ggml_backend_rpc_add_device");
-    if (!ggml_backend_rpc_add_device_fn) {
-        throw std::invalid_argument("failed to find RPC device add function");
-    }
-    for (const auto & server : rpc_servers) {
-        ggml_backend_dev_t dev = ggml_backend_rpc_add_device_fn(server.c_str());
-        if (dev) {
-            ggml_backend_device_register(dev);
-        } else {
-            throw std::invalid_argument("failed to register RPC device");
-        }
-    }
-}
-
-static void params_parse(const backend::ModelOptions* request,
-                                common_params & params) {
-   
-    // this is comparable to: https://github.com/ggerganov/llama.cpp/blob/d9b33fe95bd257b36c84ee5769cc048230067d6f/examples/server/server.cpp#L1809
-
-    params.model.path = request->modelfile();
-    if (!request->mmproj().empty()) {
-    // get the directory of modelfile
-      std::string model_dir = params.model.path.substr(0, params.model.path.find_last_of("/\\"));
-      params.mmproj.path = model_dir + "/"+ request->mmproj();
-    }
-    //  params.model_alias ??
-    params.model_alias =  request->modelfile();
-    if (!request->cachetypekey().empty()) {
-        params.cache_type_k = kv_cache_type_from_str(request->cachetypekey());
-    }
-    if (!request->cachetypevalue().empty()) {
-        params.cache_type_v = kv_cache_type_from_str(request->cachetypevalue());
-    }
-    params.n_ctx = request->contextsize();
-    //params.memory_f16 = request->f16memory();
-    params.cpuparams.n_threads = request->threads();
-    params.n_gpu_layers = request->ngpulayers();
-    params.n_batch = request->nbatch();
-    params.n_ubatch = request->nbatch(); // fixes issue with reranking models being limited to 512 tokens (the default n_ubatch size); allows for setting the maximum input amount of tokens thereby avoiding this error "input is too large to process. increase the physical batch size"
-    // Set params.n_parallel by environment variable (LLAMA_PARALLEL), defaults to 1
-    //params.n_parallel = 1;
-    const char *env_parallel = std::getenv("LLAMACPP_PARALLEL");
-    if (env_parallel != NULL) {
-        params.n_parallel = std::stoi(env_parallel);
-        params.cont_batching = true;
-    } else {
-        params.n_parallel = 1;
-    }
-
-
-    const char *llama_grpc_servers = std::getenv("LLAMACPP_GRPC_SERVERS");
-    if (llama_grpc_servers != NULL) {
-        add_rpc_devices(std::string(llama_grpc_servers));
-    }
-    
-     // decode options. Options are in form optname:optvale, or if booleans only optname.
-    for (int i = 0; i < request->options_size(); i++) {
-        std::string opt = request->options(i);
-        char *optname = strtok(&opt[0], ":");
-        char *optval = strtok(NULL, ":");
-        if (optval == NULL) {
-            optval = "true";
-        }
-
-        if (!strcmp(optname, "gpu")) {
-          //  llama.has_gpu = true;
-        }
-    }
-
-    // Add kv_overrides
-    if (request->overrides_size() > 0) {
-        for (int i = 0; i < request->overrides_size(); i++) {
-            string_parse_kv_override(request->overrides(i).c_str(), params.kv_overrides);
-        }
-    }
-
-    // TODO: Add yarn
-
-    if (!request->tensorsplit().empty()) {
-        std::string arg_next = request->tensorsplit();
-
-        // split string by , and /
-        const std::regex regex{ R"([,/]+)" };
-        std::sregex_token_iterator it{ arg_next.begin(), arg_next.end(), regex, -1 };
-        std::vector<std::string> split_arg{ it, {} };
-
-        GGML_ASSERT(split_arg.size() <= llama_max_devices());
-
-        for (size_t i_device = 0; i_device < llama_max_devices(); ++i_device) {
-            if (i_device < split_arg.size()) {
-                params.tensor_split[i_device] = std::stof(split_arg[i_device]);
-            }
-            else {
-                params.tensor_split[i_device] = 0.0f;
-            }
-        }
-    }
-
-    if (!request->maingpu().empty()) {
-        params.main_gpu = std::stoi(request->maingpu());
-    }
-    if (!request->loraadapter().empty() && !request->lorabase().empty()) {
-     float scale_factor = 1.0f;
-     if (request->lorascale() != 0.0f) {
-        scale_factor = request->lorascale();
-     }
-     // get the directory of modelfile
-     std::string model_dir = params.model.path.substr(0, params.model.path.find_last_of("/\\"));
-     params.lora_adapters.push_back({ model_dir + "/"+request->loraadapter(), scale_factor });
-    }
-    params.use_mlock = request->mlock();
-    params.use_mmap = request->mmap();
-
-    if (request->flashattention() == "on" || request->flashattention() == "enabled") {
-        params.flash_attn_type = LLAMA_FLASH_ATTN_TYPE_ENABLED;
-    } else if (request->flashattention() == "off" || request->flashattention() == "disabled") {
-        params.flash_attn_type = LLAMA_FLASH_ATTN_TYPE_DISABLED;
-    } else if (request->flashattention() == "auto") {
-        params.flash_attn_type = LLAMA_FLASH_ATTN_TYPE_AUTO;
-    }
-
-    params.no_kv_offload = request->nokvoffload();
-    params.ctx_shift = false; // We control context-shifting in any case (and we disable it as it could just lead to infinite loops)
-
-    params.embedding = request->embeddings() || request->reranking();
-    if (request->reranking()) {
-        params.pooling_type = LLAMA_POOLING_TYPE_RANK;
-    }
-
-    
-    if (request->ropescaling() == "none")   { params.rope_scaling_type = LLAMA_ROPE_SCALING_TYPE_NONE; }
-    else if (request->ropescaling() == "yarn")   { params.rope_scaling_type = LLAMA_ROPE_SCALING_TYPE_YARN; }
-    else if (request->ropescaling() == "linear")   {  params.rope_scaling_type = LLAMA_ROPE_SCALING_TYPE_LINEAR; }
-
-    if ( request->yarnextfactor() != 0.0f ) {
-        params.yarn_ext_factor = request->yarnextfactor();
-    }
-    if ( request->yarnattnfactor() != 0.0f ) {
-        params.yarn_attn_factor = request->yarnattnfactor();
-    }
-    if ( request->yarnbetafast() != 0.0f ) {
-        params.yarn_beta_fast = request->yarnbetafast();
-    }
-    if ( request->yarnbetaslow() != 0.0f ) {
-        params.yarn_beta_slow = request->yarnbetaslow();
-    }
-    if ( request->ropefreqbase() != 0.0f ) {
-        params.rope_freq_base = request->ropefreqbase();
-    }
-    if ( request->ropefreqscale() != 0.0f ) {
-        params.rope_freq_scale = request->ropefreqscale();
-    }
-
-    if (request->grammartriggers_size() > 0) {
-        params.sampling.grammar_lazy = true;
-        for (int i = 0; i < request->grammartriggers_size(); i++) {
-            common_grammar_trigger trigger;
-	    trigger.type = COMMON_GRAMMAR_TRIGGER_TYPE_WORD;
-            trigger.value = request->grammartriggers(i).word();
-	    // trigger.at_start = request->grammartriggers(i).at_start();
-            params.sampling.grammar_triggers.push_back(trigger);
-           
-        }
-    }
-}
-
-
-// GRPC Server start
-class BackendServiceImpl final : public backend::Backend::Service {
-private:
-    server_context& ctx_server;
-
-public:
-    BackendServiceImpl(server_context& ctx) : ctx_server(ctx) {}
-
-    grpc::Status Health(ServerContext* context, const backend::HealthMessage* request, backend::Reply* reply) {
-        // Implement Health RPC
-        reply->set_message("OK");
-        return Status::OK;
-    }
-
-    grpc::Status LoadModel(ServerContext* context, const backend::ModelOptions* request, backend::Result* result) {
-        // Implement LoadModel RPC
-        common_params params;
-        params_parse(request, params);
-
-        common_init();
-
-        llama_backend_init();
-        llama_numa_init(params.numa);
-
-
-        LOG_INF("system info: n_threads = %d, n_threads_batch = %d, total_threads = %d\n", params.cpuparams.n_threads, params.cpuparams_batch.n_threads, std::thread::hardware_concurrency());
-        LOG_INF("\n");
-        LOG_INF("%s\n", common_params_get_system_info(params).c_str());
-        LOG_INF("\n");
-        // load the model
-        if (!ctx_server.load_model(params)) {
-            result->set_message("Failed loading model");
-            result->set_success(false);
-            return Status::CANCELLED;
-        }
-
-        //ctx_server.init();
-        result->set_message("Loading succeeded");
-        result->set_success(true);
-        loaded_model = true;
-        ctx_server.slot_prompt_similarity = params.slot_prompt_similarity;
-
-        return Status::OK;
-    }
-
-    grpc::Status PredictStream(grpc::ServerContext* context, const backend::PredictOptions* request, grpc::ServerWriter<backend::Reply>* writer) override {
-        json data = parse_options(true, request);
-
-
-        //Raise error if embeddings is set to true
-        if (ctx_server.params_base.embedding) {
-            return grpc::Status(grpc::StatusCode::INVALID_ARGUMENT, "Embedding is not supported in streaming mode");
-        }
-
-
-        auto completion_id = gen_chatcmplid();
-        std::unordered_set<int> task_ids;
-        try {
-            std::vector<server_task> tasks;
-
-            const auto & prompt = data.at("prompt");
-            const auto type = SERVER_TASK_TYPE_COMPLETION;
-            // TODO: this log can become very long, put it behind a flag or think about a more compact format
-            //SRV_DBG("Prompt: %s\n", prompt.is_string() ? prompt.get<std::string>().c_str() : prompt.dump(2).c_str());
-
-            std::vector<raw_buffer> files;
-            const auto &images_data = data.find("image_data");
-            if (images_data != data.end() && images_data->is_array())
-            {
-                for (const auto &img : *images_data)
-                {
-                    auto decoded_data = base64_decode(img["data"].get<std::string>());
-                    files.push_back(decoded_data);
-                }
-            }
-
-            const auto &audio_data = data.find("audio_data");
-            if (audio_data != data.end() && audio_data->is_array())
-            {
-                for (const auto &audio : *audio_data)
-                {
-                    auto decoded_data = base64_decode(audio["data"].get<std::string>());
-                    files.push_back(decoded_data);
-                }
-            }
-
-            const bool has_mtmd = ctx_server.mctx != nullptr;
-
-            // process prompt
-            std::vector<server_tokens> inputs;
-            if (!prompt.is_string()) {
-                throw std::runtime_error("prompt must be a string");
-            }
-
-            if (has_mtmd) {
-                // multimodal
-                inputs.push_back(process_mtmd_prompt(ctx_server.mctx, prompt.get<std::string>(), files));
-            } else {
-                 // Everything else, including multimodal completions.
-                inputs = tokenize_input_prompts(ctx_server.vocab, ctx_server.mctx, prompt, true, true);
-            }
-
-            tasks.reserve(inputs.size());
-            for (size_t i = 0; i < inputs.size(); i++) {
-                server_task task = server_task(type);
-
-                task.id    = ctx_server.queue_tasks.get_new_id();
-                task.index = i;
-
-                task.prompt_tokens    = std::move(inputs[i]);
-                task.params           = server_task::params_from_json_cmpl(
-                        ctx_server.ctx,
-                        ctx_server.params_base,
-                        data);
-                task.id_selected_slot = json_value(data, "id_slot", -1);
-
-                // OAI-compat
-                task.params.oaicompat                 = OAICOMPAT_TYPE_NONE;
-                task.params.oaicompat_cmpl_id         = completion_id;
-                // oaicompat_model is already populated by params_from_json_cmpl
-
-                tasks.push_back(std::move(task));
-            }
-
-            task_ids = server_task::get_list_id(tasks);
-            ctx_server.queue_results.add_waiting_tasks(tasks);
-            ctx_server.queue_tasks.post(std::move(tasks));
-        } catch (const std::exception & e) {
-            return grpc::Status(grpc::StatusCode::INVALID_ARGUMENT, e.what());
-        }
-
-        ctx_server.receive_cmpl_results_stream(task_ids, [&](server_task_result_ptr & result) -> bool {
-            json res_json = result->to_json();
-            if (res_json.is_array()) {
-                for (const auto & res : res_json) {
-                    std::string completion_text = res.value("content", "");
-
-                    backend::Reply reply;
-                    reply.set_message(completion_text);
-                    int32_t tokens_predicted = res.value("tokens_predicted", 0);
-                    reply.set_tokens(tokens_predicted);
-                    int32_t tokens_evaluated = res.value("tokens_evaluated", 0);
-                    reply.set_prompt_tokens(tokens_evaluated);
-
-                    if (res.contains("timings")) {
-                        double timing_prompt_processing = res.at("timings").value("prompt_ms", 0.0);
-                        reply.set_timing_prompt_processing(timing_prompt_processing);
-                        double timing_token_generation = res.at("timings").value("predicted_ms", 0.0);
-                        reply.set_timing_token_generation(timing_token_generation);
-                    }
-
-                    // Log Request Correlation Id
-                
-                    // Send the reply
-                    writer->Write(reply);
-                }
-            } else {
-                std::string completion_text = res_json.value("content", "");
-
-                backend::Reply reply;
-                reply.set_message(completion_text);
-                int32_t tokens_predicted = res_json.value("tokens_predicted", 0);
-                reply.set_tokens(tokens_predicted);
-                int32_t tokens_evaluated = res_json.value("tokens_evaluated", 0);
-                reply.set_prompt_tokens(tokens_evaluated);
-
-                if (res_json.contains("timings")) {
-                    double timing_prompt_processing = res_json.at("timings").value("prompt_ms", 0.0);
-                    reply.set_timing_prompt_processing(timing_prompt_processing);
-                    double timing_token_generation = res_json.at("timings").value("predicted_ms", 0.0);
-                    reply.set_timing_token_generation(timing_token_generation);
-                }
-
-                
-
-                // Send the reply
-                    writer->Write(reply);
-           
-            }
-                 return true;
-        }, [&](const json & error_data) {
-            backend::Reply reply;
-            reply.set_message(error_data.value("content", ""));
-            writer->Write(reply);
-            return true;
-        }, [&]() {
-            // NOTE: we should try to check when the writer is closed here
-            return false;
-        });
-
-        ctx_server.queue_results.remove_waiting_task_ids(task_ids);
-
-        return grpc::Status::OK;
-    }
-
-    grpc::Status Predict(ServerContext* context, const backend::PredictOptions* request, backend::Reply* reply) {
-         json data = parse_options(true, request);
-
-        data["stream"] = false;
-        //Raise error if embeddings is set to true
-        if (ctx_server.params_base.embedding) {
-            return grpc::Status(grpc::StatusCode::INVALID_ARGUMENT, "Embedding is not supported in Predict mode");
-        }
-        std::cout << "[PREDICT] Received result: " << data.dump(2) << std::endl;
-        auto completion_id = gen_chatcmplid();
-        std::unordered_set<int> task_ids;
-        try {
-            std::vector<server_task> tasks;
-
-            const auto & prompt = data.at("prompt");
-            const auto type = SERVER_TASK_TYPE_COMPLETION;
-            // TODO: this log can become very long, put it behind a flag or think about a more compact format
-            //SRV_DBG("Prompt: %s\n", prompt.is_string() ? prompt.get<std::string>().c_str() : prompt.dump(2).c_str());
-
-            std::vector<raw_buffer> files;
-            const auto &images_data = data.find("image_data");
-           // std::cout << "[PREDICT] Images data: " << images_data->dump(2) << std::endl;
-           
-            if (images_data != data.end() && images_data->is_array())
-            {
-                std::cout << "[PREDICT] Processing " << images_data->size() << " images" << std::endl;
-                for (const auto &img : *images_data)
-                {
-                    std::cout << "[PREDICT] Processing image" << std::endl;
-                    auto decoded_data = base64_decode(img["data"].get<std::string>());
-                    files.push_back(decoded_data);
-                }
-            }
-
-            const auto &audio_data = data.find("audio_data");
-            if (audio_data != data.end() && audio_data->is_array())
-            {
-                for (const auto &audio : *audio_data)
-                {
-                    auto decoded_data = base64_decode(audio["data"].get<std::string>());
-                    files.push_back(decoded_data);
-                }
-            }
-
-            // process files
-            const bool has_mtmd = ctx_server.mctx != nullptr;
-
-            // process prompt
-            std::vector<server_tokens> inputs;
-            if (!prompt.is_string()) {
-                std::cout << "[PREDICT] Prompt must be a string" << std::endl;
-                throw std::runtime_error("prompt must be a string");
-            }
-
-            if (has_mtmd) {
-                // multimodal
-                inputs.push_back(process_mtmd_prompt(ctx_server.mctx, prompt.get<std::string>(), files));
-            } else {
-                 // Everything else, including multimodal completions.
-                inputs = tokenize_input_prompts(ctx_server.vocab, ctx_server.mctx, prompt, true, true);
-            }
-
-            tasks.reserve(inputs.size());
-            for (size_t i = 0; i < inputs.size(); i++) {
-                server_task task = server_task(type);
-
-                task.id    = ctx_server.queue_tasks.get_new_id();
-                task.index = i;
-
-                task.prompt_tokens    = std::move(inputs[i]);
-                task.params           = server_task::params_from_json_cmpl(
-                        ctx_server.ctx,
-                        ctx_server.params_base,
-                        data);
-                task.id_selected_slot = json_value(data, "id_slot", -1);
-
-                // OAI-compat
-                task.params.oaicompat                 = OAICOMPAT_TYPE_NONE;
-                task.params.oaicompat_cmpl_id         = completion_id;
-                // oaicompat_model is already populated by params_from_json_cmpl
-
-                tasks.push_back(std::move(task));
-            }
-
-            task_ids = server_task::get_list_id(tasks);
-            ctx_server.queue_results.add_waiting_tasks(tasks);
-            ctx_server.queue_tasks.post(std::move(tasks));
-        } catch (const std::exception & e) {
-            return grpc::Status(grpc::StatusCode::INVALID_ARGUMENT, e.what());
-        }
-
-
-        std::cout << "[DEBUG] Waiting for results..." << std::endl;
-        ctx_server.receive_multi_results(task_ids, [&](std::vector<server_task_result_ptr> & results) {
-            std::cout << "[DEBUG] Received " << results.size() << " results" << std::endl;
-            if (results.size() == 1) {
-                // single result
-                reply->set_message(results[0]->to_json().value("content", ""));
-
-                int32_t tokens_predicted = results[0]->to_json().value("tokens_predicted", 0);
-                reply->set_tokens(tokens_predicted);
-                int32_t tokens_evaluated = results[0]->to_json().value("tokens_evaluated", 0);
-                reply->set_prompt_tokens(tokens_evaluated);
-
-                if (results[0]->to_json().contains("timings")) {
-                    double timing_prompt_processing = results[0]->to_json().at("timings").value("prompt_ms", 0.0);
-                    reply->set_timing_prompt_processing(timing_prompt_processing);
-                    double timing_token_generation = results[0]->to_json().at("timings").value("predicted_ms", 0.0);
-                    reply->set_timing_token_generation(timing_token_generation);
-                }
-
-            } else {
-                // multiple results (multitask)
-                json arr = json::array();
-                for (auto & res : results) {
-                    arr.push_back(res->to_json().value("content", ""));
-                }
-                reply->set_message(arr);
-            }
-
-            
-        }, [&](const json & error_data) {
-            std::cout << "[DEBUG] Error in results: " << error_data.value("content", "") << std::endl;
-            reply->set_message(error_data.value("content", ""));
-        }, [&]() {
-            return false;
-        });
-
-        ctx_server.queue_results.remove_waiting_task_ids(task_ids);
-        std::cout << "[DEBUG] Predict request completed successfully" << std::endl;
-
-        return grpc::Status::OK;
-    }
-
-    grpc::Status Embedding(ServerContext* context, const backend::PredictOptions* request, backend::EmbeddingResult* embeddingResult) {
-
-        json body = parse_options(false, request);
-
-        body["stream"] = false;
-
-        /*
-        if (llama_pooling_type(ctx_server.ctx) == LLAMA_POOLING_TYPE_NONE) {
-            return grpc::Status(grpc::StatusCode::INVALID_ARGUMENT, "Pooling type 'none' is not OAI compatible. Please use a different pooling type");
-        }
-        */
-
-        // for the shape of input/content, see tokenize_input_prompts()
-        json prompt = body.at("embeddings");
-
-
-        auto tokenized_prompts = tokenize_input_prompts(ctx_server.vocab, ctx_server.mctx, prompt, true, true);
-        for (const auto & tokens : tokenized_prompts) {
-            // this check is necessary for models that do not add BOS token to the input
-            if (tokens.empty()) {
-                return grpc::Status(grpc::StatusCode::INVALID_ARGUMENT, "Input content cannot be empty");
-            }
-        }
-
-        int embd_normalize = 2; // default to Euclidean/L2 norm
-        // create and queue the task
-        json responses = json::array();
-        bool error = false;
-        std::unordered_set<int> task_ids;
-        {
-            std::vector<server_task> tasks;
-            for (size_t i = 0; i < tokenized_prompts.size(); i++) {
-                server_task task = server_task(SERVER_TASK_TYPE_EMBEDDING);
-
-                task.id            = ctx_server.queue_tasks.get_new_id();
-                task.index         = i;
-                task.prompt_tokens = std::move(tokenized_prompts[i]);
-
-                task.params.oaicompat = OAICOMPAT_TYPE_NONE;
-                task.params.embd_normalize = embd_normalize;
-                tasks.push_back(std::move(task));
-            }
-
-            task_ids = server_task::get_list_id(tasks);
-            ctx_server.queue_results.add_waiting_tasks(tasks);
-            ctx_server.queue_tasks.post(std::move(tasks));
-        }
-
-        // get the result
-        ctx_server.receive_multi_results(task_ids, [&](std::vector<server_task_result_ptr> & results) {
-            for (auto & res : results) {
-                GGML_ASSERT(dynamic_cast<server_task_result_embd*>(res.get()) != nullptr);
-                responses.push_back(res->to_json());
-            }
-        }, [&](const json & error_data) {
-            error = true;
-        }, [&]() {
-            return false;
-        });
-
-        ctx_server.queue_results.remove_waiting_task_ids(task_ids);
-
-        if (error) {
-            return grpc::Status(grpc::StatusCode::INTERNAL, "Error in receiving results");
-        }
-
-        std::cout << "[DEBUG] Responses size: " << responses.size() << std::endl;
-        
-        // Process the responses and extract embeddings
-        for (const auto & response_elem : responses) {
-            // Check if the response has an "embedding" field
-            if (response_elem.contains("embedding")) {
-                json embedding_data = json_value(response_elem, "embedding", json::array());
-                
-                if (embedding_data.is_array() && !embedding_data.empty()) {
-                    for (const auto & embedding_vector : embedding_data) {
-                        if (embedding_vector.is_array()) {
-                            for (const auto & embedding_value : embedding_vector) {
-                                embeddingResult->add_embeddings(embedding_value.get<float>());
-                            }
-                        }
-                    }
-                }
-            } else {
-                // Check if the response itself contains the embedding data directly
-                if (response_elem.is_array()) {
-                    for (const auto & embedding_value : response_elem) {
-                        embeddingResult->add_embeddings(embedding_value.get<float>());
-                    }
-                }
-            }
-        }
-
-
-    
-
-        return grpc::Status::OK;
-    }
-
-    grpc::Status Rerank(ServerContext* context, const backend::RerankRequest* request, backend::RerankResult* rerankResult) {
-        if (!ctx_server.params_base.embedding || ctx_server.params_base.pooling_type != LLAMA_POOLING_TYPE_RANK) {
-            return grpc::Status(grpc::StatusCode::UNIMPLEMENTED, "This server does not support reranking. Start it with `--reranking` and without `--embedding`");
-        }
-
-        // Validate request
-        if (request->query().empty()) {
-            return grpc::Status(grpc::StatusCode::INVALID_ARGUMENT, "\"query\" must be provided");
-        }
-
-        if (request->documents_size() == 0) {
-            return grpc::Status(grpc::StatusCode::INVALID_ARGUMENT, "\"documents\" must be a non-empty string array");
-        }
-
-        // Create and queue the task
-        json responses = json::array();
-        bool error = false;
-        std::unordered_set<int> task_ids;
-        {
-            std::vector<server_task> tasks;
-            std::vector<std::string> documents;
-            for (int i = 0; i < request->documents_size(); i++) {
-                documents.push_back(request->documents(i));
-            }
-            
-            tasks.reserve(documents.size());
-            for (size_t i = 0; i < documents.size(); i++) {
-                auto tmp = format_rerank(ctx_server.model, ctx_server.vocab, ctx_server.mctx, request->query(), documents[i]);
-                server_task task = server_task(SERVER_TASK_TYPE_RERANK);
-                task.id = ctx_server.queue_tasks.get_new_id();
-                task.index = i;
-                task.prompt_tokens = std::move(tmp);
-                tasks.push_back(std::move(task));
-            }
-
-            task_ids = server_task::get_list_id(tasks);
-            ctx_server.queue_results.add_waiting_tasks(tasks);
-            ctx_server.queue_tasks.post(std::move(tasks));
-        }
-
-        // Get the results
-        ctx_server.receive_multi_results(task_ids, [&](std::vector<server_task_result_ptr> & results) {
-            for (auto & res : results) {
-                GGML_ASSERT(dynamic_cast<server_task_result_rerank*>(res.get()) != nullptr);
-                responses.push_back(res->to_json());
-            }
-        }, [&](const json & error_data) {
-            error = true;
-        }, [&]() {
-            return false;
-        });
-
-        ctx_server.queue_results.remove_waiting_task_ids(task_ids);
-
-        if (error) {
-            return grpc::Status(grpc::StatusCode::INTERNAL, "Error in receiving results");
-        }
-
-        // Set usage information
-        backend::Usage* usage = rerankResult->mutable_usage();
-        int total_tokens = 0;
-        int prompt_tokens = 0;
-
-        // Create document results
-        for (const auto& response : responses) {
-            backend::DocumentResult* doc_result = rerankResult->add_results();
-            doc_result->set_index(response.value("index", 0));
-            doc_result->set_text(request->documents(response.value("index", 0)));
-            doc_result->set_relevance_score(response.value("score", 0.0f));
-            
-            // Add tokens evaluated for this document
-            int tokens_evaluated = response.value("tokens_evaluated", 0);
-            total_tokens += tokens_evaluated;
-            prompt_tokens += tokens_evaluated;
-        }
-
-        // Set the total tokens in usage
-        usage->set_total_tokens(total_tokens);
-        usage->set_prompt_tokens(prompt_tokens);
-
-        return grpc::Status::OK;
-    }
-
-    grpc::Status TokenizeString(ServerContext* context, const backend::PredictOptions* request, backend::TokenizationResponse* response) {
-        json body = parse_options(false, request);
-        body["stream"] = false;
-        
-        json tokens_response = json::array();
-        if (body.count("prompt") != 0) {
-            const bool add_special = json_value(body, "add_special", false);
-            const bool with_pieces = json_value(body, "with_pieces", false);
-
-            llama_tokens tokens = tokenize_mixed(ctx_server.vocab, body.at("content"), add_special, true);
-
-
-            for (const auto& token : tokens) {
-                std::string piece = common_token_to_piece(ctx_server.ctx, token);
-                response->add_tokens(token);
-            }
-        }
-
-        return grpc::Status::OK;
-    }
-
-    grpc::Status GetMetrics(ServerContext* context, const backend::MetricsRequest* request, backend::MetricsResponse* response) {
-
-// request slots data using task queue
-        int task_id = ctx_server.queue_tasks.get_new_id();
-        {
-            server_task task(SERVER_TASK_TYPE_METRICS);
-            task.id = task_id;
-            ctx_server.queue_results.add_waiting_task_id(task_id);
-            ctx_server.queue_tasks.post(std::move(task), true); // high-priority task
-        }
-
-        // get the result
-        server_task_result_ptr result = ctx_server.queue_results.recv(task_id);
-        ctx_server.queue_results.remove_waiting_task_id(task_id);
-
-        if (result->is_error()) {
-            // Handle case when no active slot exists
-            response->set_slot_id(0);
-            response->set_prompt_json_for_slot("");
-            response->set_tokens_per_second(0);
-            response->set_tokens_generated(0);
-            response->set_prompt_tokens_processed(0);
-            return grpc::Status(grpc::StatusCode::INTERNAL, "Error in receiving results");
-        }
-
-        // TODO: get rid of this dynamic_cast
-        auto res_metrics = dynamic_cast<server_task_result_metrics*>(result.get());
-        GGML_ASSERT(res_metrics != nullptr);    
-
-        // Populate the response with metrics
-        response->set_slot_id(0);
-        response->set_prompt_json_for_slot("");
-        response->set_tokens_per_second(res_metrics->n_prompt_tokens_processed ? 1.e3 / res_metrics->t_prompt_processing * res_metrics->n_prompt_tokens_processed : 0.);
-        response->set_tokens_generated(res_metrics->n_tokens_predicted_total);
-        response->set_prompt_tokens_processed(res_metrics->n_prompt_tokens_processed_total);
-        
-
-        return grpc::Status::OK;
-    }
-};
-
-
-int main(int argc, char** argv) {
-  std::string server_address("localhost:50051");
-
-  // Define long and short options
-  struct option long_options[] = {
-      {"addr", required_argument, nullptr, 'a'},
-      {nullptr, 0, nullptr, 0}
-  };
-
-  // Parse command-line arguments
-  int option;
-  int option_index = 0;
-  while ((option = getopt_long(argc, argv, "a:", long_options, &option_index)) != -1) {
-    switch (option) {
-      case 'a':
-        server_address = optarg;
-        break;
-      default:
-        std::cerr << "Usage: " << argv[0] << " [--addr=<address>] or [-a <address>]" << std::endl;
-        return 1;
-    }
-  }
-  
-    server_context ctx_server;
-    BackendServiceImpl service(ctx_server);
-
-    ServerBuilder builder;
-    builder.AddListeningPort(server_address, grpc::InsecureServerCredentials());
-    builder.RegisterService(&service);
-    builder.SetMaxMessageSize(50 * 1024 * 1024); // 50MB
-    builder.SetMaxSendMessageSize(50 * 1024 * 1024); // 50MB
-    builder.SetMaxReceiveMessageSize(50 * 1024 * 1024); // 50MB
-    std::unique_ptr<Server> server(builder.BuildAndStart());
-   // run the HTTP server in a thread - see comment below
-    std::thread t([&]()
-    {
-        std::cout << "Server listening on " << server_address << std::endl;
-        server->Wait();
-        return 0;
-    });
-
-    // clean up function, to be called before exit
-    auto clean_up = [&server, &ctx_server]() {
-        SRV_INF("%s: cleaning up before exit...\n", __func__);
-        server->Shutdown();
-        ctx_server.queue_results.terminate();
-        llama_backend_free();
-    };
-
-
-    //);
-    start_llama_server(ctx_server);
-    std::cout << "stopping" << std::endl;
-
-
-    clean_up();
-    t.join();
-
-    return 0;
-}

backend/cpp/llama-cpp/package.sh

@@ -1,42 +0,0 @@
-#!/bin/bash
-
-# Script to copy the appropriate libraries based on architecture
-# This script is used in the final stage of the Dockerfile
-
-set -e
-
-CURDIR=$(dirname "$(realpath $0)")
-
-# Create lib directory
-mkdir -p $CURDIR/package/lib
-
-cp -avrf $CURDIR/llama-cpp-* $CURDIR/package/
-cp -rfv $CURDIR/run.sh $CURDIR/package/
-
-# Detect architecture and copy appropriate libraries
-if [ -f "/lib64/ld-linux-x86-64.so.2" ]; then
-    # x86_64 architecture
-    echo "Detected x86_64 architecture, copying x86_64 libraries..."
-    cp -arfLv /lib64/ld-linux-x86-64.so.2 $CURDIR/package/lib/ld.so
-    cp -arfLv /lib/x86_64-linux-gnu/libc.so.6 $CURDIR/package/lib/libc.so.6
-    cp -arfLv /lib/x86_64-linux-gnu/libgcc_s.so.1 $CURDIR/package/lib/libgcc_s.so.1
-    cp -arfLv /lib/x86_64-linux-gnu/libstdc++.so.6 $CURDIR/package/lib/libstdc++.so.6
-    cp -arfLv /lib/x86_64-linux-gnu/libm.so.6 $CURDIR/package/lib/libm.so.6
-    cp -arfLv /lib/x86_64-linux-gnu/libgomp.so.1 $CURDIR/package/lib/libgomp.so.1
-elif [ -f "/lib/ld-linux-aarch64.so.1" ]; then
-    # ARM64 architecture
-    echo "Detected ARM64 architecture, copying ARM64 libraries..."
-    cp -arfLv /lib/ld-linux-aarch64.so.1 $CURDIR/package/lib/ld.so
-    cp -arfLv /lib/aarch64-linux-gnu/libc.so.6 $CURDIR/package/lib/libc.so.6
-    cp -arfLv /lib/aarch64-linux-gnu/libgcc_s.so.1 $CURDIR/package/lib/libgcc_s.so.1
-    cp -arfLv /lib/aarch64-linux-gnu/libstdc++.so.6 $CURDIR/package/lib/libstdc++.so.6
-    cp -arfLv /lib/aarch64-linux-gnu/libm.so.6 $CURDIR/package/lib/libm.so.6
-    cp -arfLv /lib/aarch64-linux-gnu/libgomp.so.1 $CURDIR/package/lib/libgomp.so.1
-else
-    echo "Error: Could not detect architecture"
-    exit 1
-fi
-
-echo "Packaging completed successfully" 
-ls -liah $CURDIR/package/
-ls -liah $CURDIR/package/lib/

backend/cpp/llama-cpp/prepare.sh

@@ -1,52 +0,0 @@
-#!/bin/bash
-
-## Patches
-## Apply patches from the `patches` directory
-for patch in $(ls patches); do
-    echo "Applying patch $patch"
-    patch -d llama.cpp/ -p1 < patches/$patch
-done 
-
-set -e
-
-cp -r CMakeLists.txt llama.cpp/tools/grpc-server/
-cp -r grpc-server.cpp llama.cpp/tools/grpc-server/
-cp -rfv llama.cpp/vendor/nlohmann/json.hpp llama.cpp/tools/grpc-server/
-cp -rfv llama.cpp/tools/server/utils.hpp llama.cpp/tools/grpc-server/
-cp -rfv llama.cpp/vendor/cpp-httplib/httplib.h llama.cpp/tools/grpc-server/
-
-set +e
-if grep -q "grpc-server" llama.cpp/tools/CMakeLists.txt; then
-    echo "grpc-server already added"
-else
-    echo "add_subdirectory(grpc-server)" >> llama.cpp/tools/CMakeLists.txt
-fi
-set -e
-
-# Now to keep maximum compatibility with the original server.cpp, we need to remove the index.html.gz.hpp and loading.html.hpp includes
-# and remove the main function
-# TODO: upstream this to the original server.cpp by extracting the upstream main function to a separate file
-awk '
-/int[ \t]+main[ \t]*\(/ {          # If the line starts the main function
-    in_main=1;                     # Set a flag
-    open_braces=0;                 # Track number of open braces
-}
-in_main {
-    open_braces += gsub(/\{/, "{"); # Count opening braces
-    open_braces -= gsub(/\}/, "}"); # Count closing braces
-    if (open_braces == 0) {         # If all braces are closed
-        in_main=0;                  # End skipping
-    }
-    next;                           # Skip lines inside main
-}
-!in_main                           # Print lines not inside main
-' "llama.cpp/tools/server/server.cpp" > llama.cpp/tools/grpc-server/server.cpp
-
-# remove index.html.gz.hpp and loading.html.hpp includes
-if [[ "$OSTYPE" == "darwin"* ]]; then
-    # macOS
-    sed -i '' '/#include "index\.html\.gz\.hpp"/d; /#include "loading\.html\.hpp"/d' llama.cpp/tools/grpc-server/server.cpp
-else
-    # Linux and others
-    sed -i '/#include "index\.html\.gz\.hpp"/d; /#include "loading\.html\.hpp"/d' llama.cpp/tools/grpc-server/server.cpp
-fi

backend/cpp/llama-cpp/run.sh

@@ -1,62 +0,0 @@
-#!/bin/bash
-set -ex
-
-# Get the absolute current dir where the script is located
-CURDIR=$(dirname "$(realpath $0)")
-
-cd /
-
-echo "CPU info:"
-grep -e "model\sname" /proc/cpuinfo | head -1
-grep -e "flags" /proc/cpuinfo | head -1
-
-BINARY=llama-cpp-fallback
-
-if grep -q -e "\savx\s" /proc/cpuinfo ; then
-	echo "CPU:    AVX    found OK"
-	if [ -e $CURDIR/llama-cpp-avx ]; then
-		BINARY=llama-cpp-avx
-	fi
-fi
-
-if grep -q -e "\savx2\s" /proc/cpuinfo ; then
-	echo "CPU:    AVX2   found OK"
-	if [ -e $CURDIR/llama-cpp-avx2 ]; then
-		BINARY=llama-cpp-avx2
-	fi
-fi
-
-# Check avx 512
-if grep -q -e "\savx512f\s" /proc/cpuinfo ; then
-	echo "CPU:    AVX512F found OK"
-	if [ -e $CURDIR/llama-cpp-avx512 ]; then
-		BINARY=llama-cpp-avx512
-	fi
-fi
-
-if [ -n "$LLAMACPP_GRPC_SERVERS" ]; then
-	if [ -e $CURDIR/llama-cpp-grpc ]; then
-		BINARY=llama-cpp-grpc
-	fi
-fi
- 
-# Extend ld library path with the dir where this script is located/lib
-if [ "$(uname)" == "Darwin" ]; then
-	export DYLD_LIBRARY_PATH=$CURDIR/lib:$DYLD_LIBRARY_PATH
-	#export DYLD_FALLBACK_LIBRARY_PATH=$CURDIR/lib:$DYLD_FALLBACK_LIBRARY_PATH
-else
-	export LD_LIBRARY_PATH=$CURDIR/lib:$LD_LIBRARY_PATH
-fi
-
-# If there is a lib/ld.so, use it
-if [ -f $CURDIR/lib/ld.so ]; then
-	echo "Using lib/ld.so"
-	echo "Using binary: $BINARY"
-	exec $CURDIR/lib/ld.so $CURDIR/$BINARY "$@"
-fi
-
-echo "Using binary: $BINARY"
-exec $CURDIR/$BINARY "$@"
-
-# We should never reach this point, however just in case we do, run fallback
-exec $CURDIR/llama-cpp-fallback "$@"

backend/cpp/llama/CMakeLists.txt

@@ -1,3 +1,20 @@
+
+## 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 clip-impl.h llava.cpp llava.h)
+# install(TARGETS ${TARGET} LIBRARY)
+# target_include_directories(myclip PUBLIC .)
+# target_include_directories(myclip PUBLIC ../..)
+# target_include_directories(myclip PUBLIC ../../common)
+# target_link_libraries(${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)
 set(CMAKE_CXX_STANDARD 17)
 cmake_minimum_required(VERSION 3.15)
@@ -57,7 +74,7 @@ add_library(hw_grpc_proto
   ${hw_proto_srcs}
   ${hw_proto_hdrs} )
 
-add_executable(${TARGET} grpc-server.cpp utils.hpp json.hpp httplib.h)
+add_executable(${TARGET} grpc-server.cpp utils.hpp json.hpp)
 
 target_include_directories(${TARGET} PRIVATE ../llava)
 target_include_directories(${TARGET} PRIVATE ${CMAKE_SOURCE_DIR})

backend/cpp/llama/Makefile

@@ -0,0 +1,87 @@
+
+LLAMA_VERSION?=
+LLAMA_REPO?=https://github.com/ggerganov/llama.cpp
+
+CMAKE_ARGS?=
+BUILD_TYPE?=
+ONEAPI_VARS?=/opt/intel/oneapi/setvars.sh
+TARGET?=--target grpc-server
+
+# Disable Shared libs as we are linking on static gRPC and we can't mix shared and static
+CMAKE_ARGS+=-DBUILD_SHARED_LIBS=OFF -DLLAMA_CURL=OFF
+
+# If build type is cublas, then we set -DGGML_CUDA=ON to CMAKE_ARGS automatically
+ifeq ($(BUILD_TYPE),cublas)
+	CMAKE_ARGS+=-DGGML_CUDA=ON
+# If build type is openblas then we set -DGGML_BLAS=ON -DGGML_BLAS_VENDOR=OpenBLAS
+# to CMAKE_ARGS automatically
+else ifeq ($(BUILD_TYPE),openblas)
+	CMAKE_ARGS+=-DGGML_BLAS=ON -DGGML_BLAS_VENDOR=OpenBLAS
+# If build type is clblas (openCL) we set -DGGML_CLBLAST=ON -DCLBlast_DIR=/some/path
+else ifeq ($(BUILD_TYPE),clblas)
+	CMAKE_ARGS+=-DGGML_CLBLAST=ON -DCLBlast_DIR=/some/path
+# If it's hipblas we do have also to set CC=/opt/rocm/llvm/bin/clang CXX=/opt/rocm/llvm/bin/clang++ 
+else ifeq ($(BUILD_TYPE),hipblas)
+	CMAKE_ARGS+=-DGGML_HIP=ON
+# If it's OSX, DO NOT embed the metal library - -DGGML_METAL_EMBED_LIBRARY=ON requires further investigation
+# But if it's OSX without metal, disable it here
+else ifeq ($(OS),Darwin)
+	ifneq ($(BUILD_TYPE),metal)
+		CMAKE_ARGS+=-DGGML_METAL=OFF
+	else
+		CMAKE_ARGS+=-DGGML_METAL=ON
+		CMAKE_ARGS+=-DGGML_METAL_EMBED_LIBRARY=ON
+		TARGET+=--target ggml-metal
+	endif
+endif
+
+ifeq ($(BUILD_TYPE),sycl_f16)
+	CMAKE_ARGS+=-DGGML_SYCL=ON \
+		-DCMAKE_C_COMPILER=icx \
+		-DCMAKE_CXX_COMPILER=icpx \
+		-DCMAKE_CXX_FLAGS="-fsycl" \
+		-DGGML_SYCL_F16=ON
+endif
+
+ifeq ($(BUILD_TYPE),sycl_f32)
+	CMAKE_ARGS+=-DGGML_SYCL=ON \
+		-DCMAKE_C_COMPILER=icx \
+		-DCMAKE_CXX_COMPILER=icpx \
+		-DCMAKE_CXX_FLAGS="-fsycl"
+endif
+
+llama.cpp:
+	mkdir -p llama.cpp
+	cd llama.cpp && \
+	git init && \
+	git remote add origin $(LLAMA_REPO)  && \
+	git fetch origin && \
+	git checkout -b build $(LLAMA_VERSION) && \
+	git submodule update --init --recursive --depth 1 --single-branch
+
+llama.cpp/tools/grpc-server: llama.cpp
+	mkdir -p llama.cpp/tools/grpc-server
+	bash prepare.sh
+
+rebuild:
+	bash prepare.sh
+	rm -rf grpc-server
+	$(MAKE) grpc-server
+
+purge:
+	rm -rf llama.cpp/build
+	rm -rf llama.cpp/tools/grpc-server
+	rm -rf grpc-server
+
+clean: purge
+	rm -rf llama.cpp
+
+grpc-server: llama.cpp llama.cpp/tools/grpc-server
+	@echo "Building grpc-server with $(BUILD_TYPE) build type and $(CMAKE_ARGS)"
+ifneq (,$(findstring sycl,$(BUILD_TYPE)))
+	+bash -c "source $(ONEAPI_VARS); \
+	cd llama.cpp && mkdir -p build && cd build && cmake .. $(CMAKE_ARGS) && cmake --build . --config Release $(TARGET)"
+else
+	+cd llama.cpp && mkdir -p build && cd build && cmake .. $(CMAKE_ARGS) && cmake --build . --config Release $(TARGET)
+endif
+	cp llama.cpp/build/bin/grpc-server .

backend/cpp/llama/prepare.sh

@@ -0,0 +1,28 @@
+#!/bin/bash
+
+## Patches
+## Apply patches from the `patches` directory
+for patch in $(ls patches); do
+    echo "Applying patch $patch"
+    patch -d llama.cpp/ -p1 < patches/$patch
+done 
+
+cp -r CMakeLists.txt llama.cpp/tools/grpc-server/
+cp -r grpc-server.cpp llama.cpp/tools/grpc-server/
+cp -rfv json.hpp llama.cpp/tools/grpc-server/
+cp -rfv utils.hpp llama.cpp/tools/grpc-server/
+    
+if grep -q "grpc-server" llama.cpp/tools/CMakeLists.txt; then
+    echo "grpc-server already added"
+else
+    echo "add_subdirectory(grpc-server)" >> llama.cpp/tools/CMakeLists.txt
+fi
+
+## 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/tools/mtmd/clip.h llama.cpp/tools/grpc-server/clip.h
+# cp -rfv llama.cpp/tools/mtmd/clip-impl.h llama.cpp/tools/grpc-server/clip-impl.h
+# cp -rfv llama.cpp/tools/mtmd/llava.cpp llama.cpp/tools/grpc-server/llava.cpp
+# echo '#include "llama.h"' > llama.cpp/tools/grpc-server/llava.h
+# cat llama.cpp/tools/mtmd/llava.h >> llama.cpp/tools/grpc-server/llava.h
+# cp -rfv llama.cpp/tools/mtmd/clip.cpp llama.cpp/tools/grpc-server/clip.cpp

backend/cpp/llama/utils.hpp

@@ -0,0 +1,910 @@
+// https://github.com/ggerganov/llama.cpp/blob/master/tools/server/utils.hpp
+
+#pragma once
+
+#include <string>
+#include <vector>
+#include <set>
+#include <mutex>
+#include <condition_variable>
+#include <unordered_map>
+
+#include "json.hpp"
+
+#include "../mtmd/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;
+
+}
+
+
+
+//
+// tokenizer and input processing utils
+//
+
+static bool json_is_array_of_numbers(const json & data) {
+    if (data.is_array()) {
+        for (const auto & e : data) {
+            if (!e.is_number_integer()) {
+                return false;
+            }
+        }
+        return true;
+    }
+    return false;
+}
+
+// is array having BOTH numbers & strings?
+static bool json_is_array_of_mixed_numbers_strings(const json & data) {
+    bool seen_string = false;
+    bool seen_number = false;
+    if (data.is_array()) {
+        for (const auto & e : data) {
+            seen_string |= e.is_string();
+            seen_number |= e.is_number_integer();
+            if (seen_number && seen_string) {
+                return true;
+            }
+        }
+    }
+    return false;
+}
+
+// get value by path(key1 / key2)
+static json json_get_nested_values(const std::vector<std::string> & paths, const json & js) {
+    json result = json::object();
+
+    for (const std::string & path : paths) {
+        json current = js;
+        const auto keys = string_split<std::string>(path, /*separator*/ '/');
+        bool valid_path = true;
+        for (const std::string & k : keys) {
+            if (valid_path && current.is_object() && current.contains(k)) {
+                current = current[k];
+            } else {
+                valid_path = false;
+            }
+        }
+        if (valid_path) {
+            result[path] = current;
+        }
+    }
+    return result;
+}
+
+
+/**
+ * this handles 2 cases:
+ * - only string, example: "string"
+ * - mixed string and tokens, example: [12, 34, "string", 56, 78]
+ */
+static llama_tokens tokenize_mixed(const llama_vocab * vocab, const json & json_prompt, bool add_special, bool parse_special) {
+    // If `add_bos` is true, we only add BOS, when json_prompt is a string,
+    // or the first element of the json_prompt array is a string.
+    llama_tokens prompt_tokens;
+
+    if (json_prompt.is_array()) {
+        bool first = true;
+        for (const auto & p : json_prompt) {
+            if (p.is_string()) {
+                auto s = p.template get<std::string>();
+
+                llama_tokens p;
+                if (first) {
+                    p = common_tokenize(vocab, s, add_special, parse_special);
+                    first = false;
+                } else {
+                    p = common_tokenize(vocab, s, false, parse_special);
+                }
+
+                prompt_tokens.insert(prompt_tokens.end(), p.begin(), p.end());
+            } else {
+                if (first) {
+                    first = false;
+                }
+
+                prompt_tokens.push_back(p.template get<llama_token>());
+            }
+        }
+    } else {
+        auto s = json_prompt.template get<std::string>();
+        prompt_tokens = common_tokenize(vocab, s, add_special, parse_special);
+    }
+
+    return prompt_tokens;
+}
+
+/**
+ * break the input "prompt" object into multiple prompt if needed, then tokenize them
+ * this supports these cases:
+ * - "prompt": "string"
+ * - "prompt": [12, 34, 56]
+ * - "prompt": [12, 34, "string", 56, 78]
+ * and multiple prompts (multi-tasks):
+ * - "prompt": ["string1", "string2"]
+ * - "prompt": ["string1", [12, 34, 56]]
+ * - "prompt": [[12, 34, 56], [78, 90, 12]]
+ * - "prompt": [[12, 34, "string", 56, 78], [12, 34, 56]]
+ */
+static std::vector<llama_tokens> tokenize_input_prompts(const llama_vocab * vocab, const json & json_prompt, bool add_special, bool parse_special) {
+    std::vector<llama_tokens> result;
+    if (json_prompt.is_string() || json_is_array_of_mixed_numbers_strings(json_prompt)) {
+        // string or mixed
+        result.push_back(tokenize_mixed(vocab, json_prompt, add_special, parse_special));
+    } else if (json_is_array_of_numbers(json_prompt)) {
+        // array of tokens
+        result.push_back(json_prompt.get<llama_tokens>());
+    } else if (json_prompt.is_array()) {
+        // array of prompts
+        result.reserve(json_prompt.size());
+        for (const auto & p : json_prompt) {
+            if (p.is_string() || json_is_array_of_mixed_numbers_strings(p)) {
+                result.push_back(tokenize_mixed(vocab, p, add_special, parse_special));
+            } else if (json_is_array_of_numbers(p)) {
+                // array of tokens
+                result.push_back(p.get<llama_tokens>());
+            } else {
+                throw std::runtime_error("element of \"prompt\" must be a string, an list of tokens, or a list of mixed strings & tokens");
+            }
+        }
+    } else {
+        throw std::runtime_error("\"prompt\" must be a string, an list of tokens, a list of mixed strings & tokens, or a list of prompts");
+    }
+    if (result.empty()) {
+        throw std::runtime_error("\"prompt\" must not be empty");
+    }
+    return result;
+}
+
+
+
+
+//
+// utils for interacting with libmtmd
+// (may need to refactor in near future)
+//
+
+/**
+ * server_tokens is a helper to manage the input tokens and image for the server.
+ * it is made this way to simplify the logic of KV cache management.
+ */
+struct server_tokens {
+    bool has_mtmd = false;
+
+private: // disallow accessing these members directly, risking out-of-sync
+
+    // map a **start** position in tokens to the image chunk
+    std::unordered_map<llama_pos, mtmd::input_chunk_ptr> map_pos_to_image;
+
+    // list of tokens
+    // it can include LLAMA_TOKEN_NULL, which is used to indicate a token that is not a text token
+    // a mtmd_input_chunk can occupy multiple tokens, one llama_token per **position**
+    // important: for models using mrope, an image can contain multiple tokens but will use only one **position**
+    llama_tokens tokens;
+
+    // for ex. with input of 5 text tokens and 2 images:
+    //      [0] [1] [2] [3] [4] [img0] [img0] [img0] [img1] [img1]
+    // pos  0   1   2   3   4   5      6      7      8      9
+    // map_pos_to_image will contain: {5, img0}, {8, img1}
+
+public:
+    server_tokens() = default;
+    ~server_tokens() = default;
+
+    // Prevent copying
+    server_tokens(const server_tokens&) = delete;
+    server_tokens& operator=(const server_tokens&) = delete;
+
+    // Allow moving (usually implicitly generated if members are movable)
+    server_tokens(server_tokens&&) = default;
+    server_tokens& operator=(server_tokens&&) = default;
+
+    // Allow accessing elements using [] operator
+    llama_token operator[](size_t index) { return tokens[index]; }
+    const llama_token& operator[](size_t index) const { return tokens[index]; }
+
+    server_tokens(mtmd::input_chunks & mtmd_chunks, bool has_mtmd) : has_mtmd(has_mtmd) {
+        for (size_t i = 0; i < mtmd_chunks.size(); ++i) {
+            push_back(mtmd_chunks[i]);
+        }
+    }
+
+    server_tokens(llama_tokens & tokens, bool has_mtmd) : has_mtmd(has_mtmd), tokens(tokens) {}
+
+    // for debugging
+    std::string str() const {
+        std::ostringstream oss;
+        oss << "tokens: ";
+        for (const auto & t : tokens) {
+            if (t == LLAMA_TOKEN_NULL) {
+                oss << "<embd> ";
+            } else {
+                oss << t << " ";
+            }
+        }
+        oss << "\n";
+        oss << "image pos: ";
+        for (const auto & it : map_pos_to_image) {
+            oss << it.first << ", ";
+        }
+        return oss.str();
+    }
+
+    const mtmd::input_chunk_ptr & find_chunk(llama_pos pos) const {
+        auto it = map_pos_to_image.find(pos);
+        if (it != map_pos_to_image.end()) {
+            return it->second;
+        } else {
+            throw std::runtime_error("Chunk not found");
+        }
+    }
+
+    void push_back(llama_token tok) {
+        if (tok == LLAMA_TOKEN_NULL) {
+            throw std::runtime_error("Invalid token");
+        }
+        tokens.emplace_back(tok);
+    }
+
+    // will create a copy of the chunk if it contains non-text data
+    void push_back(const mtmd_input_chunk * chunk) {
+        auto type = mtmd_input_chunk_get_type(chunk);
+        if (type == MTMD_INPUT_CHUNK_TYPE_IMAGE) {
+            GGML_ASSERT(has_mtmd);
+            auto img_tokens = mtmd_input_chunk_get_tokens_image(chunk);
+            const int n_pos = mtmd_image_tokens_get_n_pos(img_tokens);
+            llama_pos start_pos = tokens.size();
+            for (int i = 0; i < n_pos; ++i) {
+                tokens.emplace_back(LLAMA_TOKEN_NULL);
+            }
+            mtmd::input_chunk_ptr new_chunk(mtmd_input_chunk_copy(chunk));
+            map_pos_to_image[start_pos] = std::move(new_chunk);
+        } else if (type == MTMD_INPUT_CHUNK_TYPE_TEXT) {
+            size_t n_tokens;
+            auto text_tokens = mtmd_input_chunk_get_tokens_text(chunk, &n_tokens);
+            for (size_t i = 0; i < n_tokens; ++i) {
+                push_back(text_tokens[i]);
+            }
+        } else {
+            GGML_ABORT("Invalid chunk type");
+        }
+    }
+
+    // for compatibility with context shift and prompt truncation
+    void insert(const llama_tokens & inp_tokens) {
+        GGML_ASSERT(!has_mtmd); // only allow this if mtmd is disabled
+        tokens.insert(tokens.end(), inp_tokens.begin(), inp_tokens.end());
+    }
+
+    // for compatibility with speculative decoding, ctx shift, slot save/load
+    const llama_tokens & get_text_tokens() const {
+        GGML_ASSERT(!has_mtmd); // only allow this if mtmd is disabled
+        return tokens;
+    }
+
+    // for compatibility with speculative decoding
+    void set_token(llama_pos pos, llama_token id) {
+        GGML_ASSERT(!has_mtmd); // only allow this if mtmd is disabled
+        tokens[pos] = id;
+    }
+
+    size_t size() const {
+        return tokens.size();
+    }
+
+    bool empty() const {
+        return tokens.empty();
+    }
+
+    void clear() {
+        tokens.clear();
+    }
+
+    void resize(size_t n) {
+        GGML_ASSERT(n <= tokens.size());
+        if (has_mtmd) {
+            // we throw an error if we try to remove a token in the middle of an image
+            // for ex. with input of 5 text tokens and 2 images:
+            //    [0] [1] [2] [3] [4] [img0] [img0] [img0] [img1] [img1]
+            // n  1   2   3   4   5   6      7      8      9      10
+            // allowed to resize      ^                    ^
+            // disallowed to resize          ^      ^             ^
+            if (n > 0) {
+                llama_token last_token = tokens[n - 1];
+                // make sure we never remove tokens in the middle of an image
+                if (last_token == LLAMA_TOKEN_NULL) {
+                    find_chunk(n - 1); // will throw an error if the token is not begin-of-chunk
+                }
+            }
+            // remove all image chunks that are not used anymore
+            for (auto it = map_pos_to_image.begin(); it != map_pos_to_image.end(); ) {
+                llama_pos pos = it->first;
+                if (pos >= (llama_pos)n) {
+                    it = map_pos_to_image.erase(it);
+                } else {
+                    ++it;
+                }
+            }
+        }
+        tokens.resize(n);
+    }
+
+    std::string detokenize(const llama_context * ctx, bool special) const {
+        llama_tokens text_tokens;
+        text_tokens.reserve(tokens.size());
+        for (const auto & t : tokens) {
+            if (t != LLAMA_TOKEN_NULL) {
+                text_tokens.push_back(t);
+            }
+        }
+        return common_detokenize(ctx, text_tokens, special);
+    }
+
+    size_t get_common_prefix(const server_tokens & b) const {
+        size_t max_idx = std::min(tokens.size(), b.tokens.size());
+        for (size_t i = 0; i < max_idx; ++i) {
+            auto & ai =   tokens[i];
+            auto & bi = b.tokens[i];
+
+            if (ai == LLAMA_TOKEN_NULL && bi == LLAMA_TOKEN_NULL) {
+                GGML_ASSERT(has_mtmd);
+                const auto & a_chunk =   find_chunk(i);
+                const auto & b_chunk = b.find_chunk(i);
+                GGML_ASSERT(a_chunk && b_chunk);
+                const auto * a_img = mtmd_input_chunk_get_tokens_image(a_chunk.get());
+                const auto * b_img = mtmd_input_chunk_get_tokens_image(b_chunk.get());
+                std::string ai_id  = mtmd_image_tokens_get_id(a_img);
+                std::string bi_id  = mtmd_image_tokens_get_id(b_img);
+                size_t a_pos       = mtmd_image_tokens_get_n_pos(a_img);
+                size_t b_pos       = mtmd_image_tokens_get_n_pos(b_img);
+                if (ai_id == bi_id && a_pos == b_pos) {
+                    GGML_ASSERT(a_pos > 0 && "Invalid image token"); // should never happen
+                    i += a_pos - 1; // will be +1 by the for loop
+                    continue;
+                } else {
+                    return i;
+                }
+            } else if (ai == bi) {
+                continue;
+            } else {
+                return i;
+            }
+        }
+        return max_idx; // all tokens are equal
+    }
+
+    // make sure all text tokens are within the vocab range
+    bool validate(const struct llama_context * ctx) const {
+        const llama_model * model = llama_get_model(ctx);
+        const llama_vocab * vocab = llama_model_get_vocab(model);
+        const int32_t n_vocab = llama_vocab_n_tokens(vocab);
+
+        for (size_t i = 0; i < tokens.size(); ++i) {
+            auto & t = tokens[i];
+            if (t == LLAMA_TOKEN_NULL) {
+                try {
+                    const auto & chunk = find_chunk(i);
+                    const auto * img_tokens = mtmd_input_chunk_get_tokens_image(chunk.get());
+                    size_t n_pos = mtmd_image_tokens_get_n_pos(img_tokens);
+                    i += n_pos - 1; // will be +1 by the for loop
+                } catch (const std::exception & e) {
+                    return false;
+                }
+            } else if (t < 0 || t >= n_vocab) {
+                return false;
+            }
+        }
+        return true;
+    }
+
+    // encode and decode the image chunk
+    int32_t process_chunk(
+                llama_context * ctx,
+                mtmd_context * mctx,
+                llama_pos n_past,
+                int32_t seq_id,
+                llama_pos & n_pos_out) {
+        auto it = map_pos_to_image.find(n_past);
+        if (it == map_pos_to_image.end()) {
+            throw std::runtime_error("Chunk not found");
+        }
+     //   SRV_INF("%s\n", "processing image...");
+        int32_t n_batch = llama_n_batch(ctx);
+        int64_t t0 = ggml_time_ms();
+        llama_pos new_n_past = n_past;
+        int32_t result = mtmd_helper_eval_chunk_single(mctx, ctx,
+            it->second.get(), // chunk
+            n_past,
+            seq_id,
+            n_batch,
+            true, // logits last
+            &new_n_past);
+        //SRV_INF("image processed in %" PRId64 " ms\n", ggml_time_ms() - t0);
+        if (result != 0) {
+            LOG_ERR("mtmd_helper_eval failed with status %d", result);
+            n_pos_out = n_past;
+            return result;
+        }
+        n_pos_out = new_n_past;
+        return 0;
+    }
+};
+
+// Computes FNV-1a hash of the data
+static std::string fnv_hash(const uint8_t * data, size_t len) {
+    const uint64_t fnv_prime = 0x100000001b3ULL;
+    uint64_t hash = 0xcbf29ce484222325ULL;
+
+    for (size_t i = 0; i < len; ++i) {
+        hash ^= data[i];
+        hash *= fnv_prime;
+    }
+    return std::to_string(hash);
+}

backend/go/bark-cpp/Makefile

@@ -1,51 +0,0 @@
-INCLUDE_PATH := $(abspath ./)
-LIBRARY_PATH := $(abspath ./)
-
-AR?=ar
-
-CMAKE_ARGS?=-DGGML_NATIVE=OFF
-BUILD_TYPE?=
-GOCMD=go
-# keep standard at C11 and C++11
-CXXFLAGS = -I. -I$(INCLUDE_PATH)/sources/bark.cpp/examples -I$(INCLUDE_PATH)/sources/bark.cpp/encodec.cpp/ggml/include -I$(INCLUDE_PATH)/sources/bark.cpp/spm-headers -I$(INCLUDE_PATH)/sources/bark.cpp -O3 -DNDEBUG -std=c++17 -fPIC
-LDFLAGS  = -L$(LIBRARY_PATH) -L$(LIBRARY_PATH)/sources/bark.cpp/build/examples -lbark -lstdc++ -lm
-
-# bark.cpp
-BARKCPP_REPO?=https://github.com/PABannier/bark.cpp.git
-BARKCPP_VERSION?=5d5be84f089ab9ea53b7a793f088d3fbf7247495
-
-# warnings
-CXXFLAGS += -Wall -Wextra -Wpedantic -Wcast-qual -Wno-unused-function
-
-## bark.cpp
-sources/bark.cpp:
-	git clone --recursive $(BARKCPP_REPO) sources/bark.cpp && \
-	cd sources/bark.cpp && \
-	git checkout $(BARKCPP_VERSION) && \
-	git submodule update --init --recursive --depth 1 --single-branch
-
-sources/bark.cpp/build/libbark.a: sources/bark.cpp
-	cd sources/bark.cpp && \
-	mkdir -p build && \
-	cd build && \
-	cmake $(CMAKE_ARGS) .. && \
-	cmake --build . --config Release
-
-gobark.o:
-	$(CXX) $(CXXFLAGS) gobark.cpp -o gobark.o -c $(LDFLAGS)
-
-libbark.a: sources/bark.cpp/build/libbark.a gobark.o
-	cp $(INCLUDE_PATH)/sources/bark.cpp/build/libbark.a ./
-	$(AR) rcs libbark.a gobark.o
-
-bark-cpp: libbark.a
-	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH="$(CURDIR)" LIBRARY_PATH=$(CURDIR) \
-	$(GOCMD) build -v -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o bark-cpp ./
-
-package:
-	bash package.sh
-
-build: bark-cpp package
-
-clean:
-	rm -f gobark.o libbark.a

backend/go/bark-cpp/package.sh

@@ -1,41 +0,0 @@
-#!/bin/bash
-
-# Script to copy the appropriate libraries based on architecture
-# This script is used in the final stage of the Dockerfile
-
-set -e
-
-CURDIR=$(dirname "$(realpath $0)")
-
-# Create lib directory
-mkdir -p $CURDIR/package/lib
-cp -avrf $CURDIR/bark-cpp $CURDIR/package/
-cp -rfv $CURDIR/run.sh $CURDIR/package/
-
-# Detect architecture and copy appropriate libraries
-if [ -f "/lib64/ld-linux-x86-64.so.2" ]; then
-    # x86_64 architecture
-    echo "Detected x86_64 architecture, copying x86_64 libraries..."
-    cp -arfLv /lib64/ld-linux-x86-64.so.2 $CURDIR/package/lib/ld.so
-    cp -arfLv /lib/x86_64-linux-gnu/libc.so.6 $CURDIR/package/lib/libc.so.6
-    cp -arfLv /lib/x86_64-linux-gnu/libgcc_s.so.1 $CURDIR/package/lib/libgcc_s.so.1
-    cp -arfLv /lib/x86_64-linux-gnu/libstdc++.so.6 $CURDIR/package/lib/libstdc++.so.6
-    cp -arfLv /lib/x86_64-linux-gnu/libm.so.6 $CURDIR/package/lib/libm.so.6
-    cp -arfLv /lib/x86_64-linux-gnu/libgomp.so.1 $CURDIR/package/lib/libgomp.so.1
-elif [ -f "/lib/ld-linux-aarch64.so.1" ]; then
-    # ARM64 architecture
-    echo "Detected ARM64 architecture, copying ARM64 libraries..."
-    cp -arfLv /lib/ld-linux-aarch64.so.1 $CURDIR/package/lib/ld.so
-    cp -arfLv /lib/aarch64-linux-gnu/libc.so.6 $CURDIR/package/lib/libc.so.6
-    cp -arfLv /lib/aarch64-linux-gnu/libgcc_s.so.1 $CURDIR/package/lib/libgcc_s.so.1
-    cp -arfLv /lib/aarch64-linux-gnu/libstdc++.so.6 $CURDIR/package/lib/libstdc++.so.6
-    cp -arfLv /lib/aarch64-linux-gnu/libm.so.6 $CURDIR/package/lib/libm.so.6
-    cp -arfLv /lib/aarch64-linux-gnu/libgomp.so.1 $CURDIR/package/lib/libgomp.so.1
-else
-    echo "Error: Could not detect architecture"
-    exit 1
-fi
-
-echo "Packaging completed successfully" 
-ls -liah $CURDIR/package/
-ls -liah $CURDIR/package/lib/

backend/go/bark-cpp/run.sh

@@ -1,13 +0,0 @@
-#!/bin/bash
-set -ex
-
-CURDIR=$(dirname "$(realpath $0)")
-export LD_LIBRARY_PATH=$CURDIR/lib:$LD_LIBRARY_PATH
-
-# If there is a lib/ld.so, use it
-if [ -f $CURDIR/lib/ld.so ]; then
-	echo "Using lib/ld.so"
-	exec $CURDIR/lib/ld.so $CURDIR/bark-cpp "$@"
-fi
-
-exec $CURDIR/bark-cpp "$@"

backend/go/bark/Makefile

@@ -0,0 +1,25 @@
+INCLUDE_PATH := $(abspath ./)
+LIBRARY_PATH := $(abspath ./)
+
+AR?=ar
+
+BUILD_TYPE?=
+# keep standard at C11 and C++11
+CXXFLAGS = -I. -I$(INCLUDE_PATH)/../../../sources/bark.cpp/examples -I$(INCLUDE_PATH)/../../../sources/bark.cpp/spm-headers -I$(INCLUDE_PATH)/../../../sources/bark.cpp -O3 -DNDEBUG -std=c++17 -fPIC
+LDFLAGS  = -L$(LIBRARY_PATH) -L$(LIBRARY_PATH)/../../../sources/bark.cpp/build/examples -lbark -lstdc++ -lm
+
+# warnings
+CXXFLAGS += -Wall -Wextra -Wpedantic -Wcast-qual -Wno-unused-function
+
+gobark.o:
+	$(CXX) $(CXXFLAGS) gobark.cpp -o gobark.o -c $(LDFLAGS)
+
+libbark.a: gobark.o
+	cp $(INCLUDE_PATH)/../../../sources/bark.cpp/build/libbark.a ./
+	$(AR) rcs libbark.a gobark.o
+	$(AR) rcs libbark.a $(LIBRARY_PATH)/../../../sources/bark.cpp/build/encodec.cpp/ggml/src/CMakeFiles/ggml.dir/ggml.c.o
+	$(AR) rcs libbark.a $(LIBRARY_PATH)/../../../sources/bark.cpp/build/encodec.cpp/ggml/src/CMakeFiles/ggml.dir/ggml-alloc.c.o
+	$(AR) rcs libbark.a $(LIBRARY_PATH)/../../../sources/bark.cpp/build/encodec.cpp/ggml/src/CMakeFiles/ggml.dir/ggml-backend.c.o
+
+clean:
+	rm -f gobark.o libbark.a

backend/go/bark/gobark.cpp

@@ -48,7 +48,7 @@ int tts(char *text,int  threads, char *dst ) {
 
     // generate audio
     if (!bark_generate_audio(c, text, threads)) {
-        fprintf(stderr, "%s: An error occurred. If the problem persists, feel free to open an issue to report it.\n", __func__);
+        fprintf(stderr, "%s: An error occured. If the problem persists, feel free to open an issue to report it.\n", __func__);
         return 1;
     }
 

backend/go/bark/gobark.go

@@ -1,7 +1,7 @@
 package main
 
-// #cgo CXXFLAGS: -I${SRCDIR}/sources/bark.cpp/ -I${SRCDIR}/sources/bark.cpp/encodec.cpp -I${SRCDIR}/sources/bark.cpp/encodec.cpp/ggml/include -I${SRCDIR}/sources/bark.cpp/examples -I${SRCDIR}/sources/bark.cpp/spm-headers
-// #cgo LDFLAGS: -L${SRCDIR}/ -L${SRCDIR}/sources/bark.cpp/build/examples -L${SRCDIR}/sources/bark.cpp/build/encodec.cpp/ggml/src/ -L${SRCDIR}/sources/bark.cpp/build/encodec.cpp/ -lbark -lencodec -lcommon -lggml -lgomp
+// #cgo CXXFLAGS: -I${SRCDIR}/../../../sources/bark.cpp/ -I${SRCDIR}/../../../sources/bark.cpp/encodec.cpp -I${SRCDIR}/../../../sources/bark.cpp/examples -I${SRCDIR}/../../../sources/bark.cpp/spm-headers
+// #cgo LDFLAGS: -L${SRCDIR}/ -L${SRCDIR}/../../../sources/bark.cpp/build/examples -L${SRCDIR}/../../../sources/bark.cpp/build/encodec.cpp/ -lbark -lencodec -lcommon
 // #include <gobark.h>
 // #include <stdlib.h>
 import "C"

backend/go/huggingface/Makefile

@@ -1,9 +0,0 @@
-GOCMD=go
-
-huggingface:
-	CGO_ENABLED=0 $(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o huggingface ./
-
-package:
-	bash package.sh
-
-build: huggingface package

backend/go/huggingface/package.sh

@@ -1,12 +0,0 @@
-#!/bin/bash
-
-# Script to copy the appropriate libraries based on architecture
-# This script is used in the final stage of the Dockerfile
-
-set -e
-
-CURDIR=$(dirname "$(realpath $0)")
-
-mkdir -p $CURDIR/package
-cp -avrf $CURDIR/huggingface $CURDIR/package/
-cp -rfv $CURDIR/run.sh $CURDIR/package/

backend/go/huggingface/run.sh

@@ -1,6 +0,0 @@
-#!/bin/bash
-set -ex
-
-CURDIR=$(dirname "$(realpath $0)")
-
-exec $CURDIR/huggingface "$@"

backend/go/image/stablediffusion-ggml/Makefile

@@ -0,0 +1,135 @@
+INCLUDE_PATH := $(abspath ./)
+LIBRARY_PATH := $(abspath ./)
+
+AR?=ar
+CMAKE_ARGS?=
+BUILD_TYPE?=
+ONEAPI_VARS?=/opt/intel/oneapi/setvars.sh
+# keep standard at C11 and C++11
+CXXFLAGS = -I. -I$(INCLUDE_PATH)/../../../../sources/stablediffusion-ggml.cpp/thirdparty -I$(INCLUDE_PATH)/../../../../sources/stablediffusion-ggml.cpp/ggml/include -I$(INCLUDE_PATH)/../../../../sources/stablediffusion-ggml.cpp -O3 -DNDEBUG -std=c++17 -fPIC
+
+GOCMD?=go
+CGO_LDFLAGS?=
+# Avoid parent make file overwriting CGO_LDFLAGS which is needed for hipblas
+CGO_LDFLAGS_SYCL=
+GO_TAGS?=
+LD_FLAGS?=
+
+# Disable Shared libs as we are linking on static gRPC and we can't mix shared and static
+CMAKE_ARGS+=-DBUILD_SHARED_LIBS=OFF
+
+# If build type is cublas, then we set -DGGML_CUDA=ON to CMAKE_ARGS automatically
+ifeq ($(BUILD_TYPE),cublas)
+	CMAKE_ARGS+=-DSD_CUDA=ON
+# If build type is openblas then we set -DGGML_BLAS=ON -DGGML_BLAS_VENDOR=OpenBLAS
+# to CMAKE_ARGS automatically
+else ifeq ($(BUILD_TYPE),openblas)
+	CMAKE_ARGS+=-DGGML_BLAS=ON -DGGML_BLAS_VENDOR=OpenBLAS
+# If build type is clblas (openCL) we set -DGGML_CLBLAST=ON -DCLBlast_DIR=/some/path
+else ifeq ($(BUILD_TYPE),clblas)
+	CMAKE_ARGS+=-DGGML_CLBLAST=ON -DCLBlast_DIR=/some/path
+# If it's hipblas we do have also to set CC=/opt/rocm/llvm/bin/clang CXX=/opt/rocm/llvm/bin/clang++
+else ifeq ($(BUILD_TYPE),hipblas)
+	CMAKE_ARGS+=-DSD_HIPBLAS=ON
+# If it's OSX, DO NOT embed the metal library - -DGGML_METAL_EMBED_LIBRARY=ON requires further investigation
+# But if it's OSX without metal, disable it here
+else ifeq ($(OS),Darwin)
+	ifneq ($(BUILD_TYPE),metal)
+		CMAKE_ARGS+=-DSD_METAL=OFF
+	else
+		CMAKE_ARGS+=-DSD_METAL=ON
+		CMAKE_ARGS+=-DGGML_METAL_EMBED_LIBRARY=ON
+		TARGET+=--target ggml-metal
+	endif
+endif
+
+ifeq ($(BUILD_TYPE),sycl_f16)
+	CMAKE_ARGS+=-DGGML_SYCL=ON \
+		-DCMAKE_C_COMPILER=icx \
+		-DCMAKE_CXX_COMPILER=icpx \
+		-DSD_SYCL=ON \
+		-DGGML_SYCL_F16=ON
+	CC=icx
+	CXX=icpx
+	CGO_LDFLAGS_SYCL += -fsycl -L${DNNLROOT}/lib -ldnnl ${MKLROOT}/lib/intel64/libmkl_sycl.a -fiopenmp -fopenmp-targets=spir64 -lOpenCL
+	CGO_LDFLAGS_SYCL += $(shell pkg-config --libs mkl-static-lp64-gomp)
+	CGO_CXXFLAGS += -fiopenmp -fopenmp-targets=spir64
+	CGO_CXXFLAGS += $(shell pkg-config --cflags mkl-static-lp64-gomp )
+endif
+
+ifeq ($(BUILD_TYPE),sycl_f32)
+	CMAKE_ARGS+=-DGGML_SYCL=ON \
+		-DCMAKE_C_COMPILER=icx \
+		-DCMAKE_CXX_COMPILER=icpx \
+		-DSD_SYCL=ON
+	CC=icx
+	CXX=icpx
+	CGO_LDFLAGS_SYCL += -fsycl -L${DNNLROOT}/lib -ldnnl ${MKLROOT}/lib/intel64/libmkl_sycl.a -fiopenmp -fopenmp-targets=spir64 -lOpenCL
+	CGO_LDFLAGS_SYCL += $(shell pkg-config --libs mkl-static-lp64-gomp)
+	CGO_CXXFLAGS += -fiopenmp -fopenmp-targets=spir64
+	CGO_CXXFLAGS += $(shell pkg-config --cflags mkl-static-lp64-gomp )
+endif
+
+# warnings
+# CXXFLAGS += -Wall -Wextra -Wpedantic -Wcast-qual -Wno-unused-function
+
+# Find all .a archives in ARCHIVE_DIR
+# (ggml can have different backends cpu, cuda, etc., each backend generates a .a archive)
+GGML_ARCHIVE_DIR := build/ggml/src/
+ALL_ARCHIVES := $(shell find $(GGML_ARCHIVE_DIR) -type f -name '*.a')
+
+# Name of the single merged library
+COMBINED_LIB := libggmlall.a
+
+# Rule to merge all the .a files into one
+$(COMBINED_LIB): $(ALL_ARCHIVES)
+	@echo "Merging all .a into $(COMBINED_LIB)"
+	rm -f $@
+	mkdir -p merge-tmp
+	for a in $(ALL_ARCHIVES); do \
+		( cd merge-tmp && ar x ../$$a ); \
+	done
+	( cd merge-tmp && ar rcs ../$@ *.o )
+	# Ensure we have a proper index
+	ranlib $@
+	# Clean up
+	rm -rf merge-tmp
+
+build/libstable-diffusion.a:
+	@echo "Building SD with $(BUILD_TYPE) build type and $(CMAKE_ARGS)"
+ifneq (,$(findstring sycl,$(BUILD_TYPE)))
+	+bash -c "source $(ONEAPI_VARS); \
+	mkdir -p build && \
+	cd build && \
+	cmake $(CMAKE_ARGS) ../../../../../sources/stablediffusion-ggml.cpp && \
+	cmake --build . --config Release"
+else
+	mkdir -p build && \
+	cd build && \
+	cmake $(CMAKE_ARGS) ../../../../../sources/stablediffusion-ggml.cpp && \
+	cmake --build . --config Release
+endif
+	$(MAKE) $(COMBINED_LIB)
+
+gosd.o:
+ifneq (,$(findstring sycl,$(BUILD_TYPE)))
+	+bash -c "source $(ONEAPI_VARS); \
+	$(CXX) $(CXXFLAGS) gosd.cpp -o gosd.o -c"
+else
+	$(CXX) $(CXXFLAGS) gosd.cpp -o gosd.o -c
+endif
+
+libsd.a: gosd.o
+	cp $(INCLUDE_PATH)/build/libstable-diffusion.a ./libsd.a
+	$(AR) rcs libsd.a gosd.o
+
+stablediffusion-ggml:
+	CGO_LDFLAGS="$(CGO_LDFLAGS) $(CGO_LDFLAGS_SYCL)" C_INCLUDE_PATH="$(INCLUDE_PATH)" LIBRARY_PATH="$(LIBRARY_PATH)" \
+	CC="$(CC)" CXX="$(CXX)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o ../../../../backend-assets/grpc/stablediffusion-ggml ./
+ifneq ($(UPX),)
+	$(UPX) ../../../../backend-assets/grpc/stablediffusion-ggml
+endif
+
+clean:
+	rm -rf gosd.o libsd.a build $(COMBINED_LIB)

backend/go/image/stablediffusion-ggml/gosd.cpp

@@ -0,0 +1,231 @@
+#include <stdio.h>
+#include <string.h>
+#include <time.h>
+#include <iostream>
+#include <random>
+#include <string>
+#include <vector>
+#include "gosd.h"
+
+// #include "preprocessing.hpp"
+#include "flux.hpp"
+#include "stable-diffusion.h"
+
+#define STB_IMAGE_IMPLEMENTATION
+#define STB_IMAGE_STATIC
+#include "stb_image.h"
+
+#define STB_IMAGE_WRITE_IMPLEMENTATION
+#define STB_IMAGE_WRITE_STATIC
+#include "stb_image_write.h"
+
+#define STB_IMAGE_RESIZE_IMPLEMENTATION
+#define STB_IMAGE_RESIZE_STATIC
+#include "stb_image_resize.h"
+
+// Names of the sampler method, same order as enum sample_method in stable-diffusion.h
+const char* sample_method_str[] = {
+    "euler_a",
+    "euler",
+    "heun",
+    "dpm2",
+    "dpm++2s_a",
+    "dpm++2m",
+    "dpm++2mv2",
+    "ipndm",
+    "ipndm_v",
+    "lcm",
+    "ddim_trailing",
+    "tcd",
+};
+
+// Names of the sigma schedule overrides, same order as sample_schedule in stable-diffusion.h
+const char* schedule_str[] = {
+    "default",
+    "discrete",
+    "karras",
+    "exponential",
+    "ays",
+    "gits",
+};
+
+sd_ctx_t* sd_c;
+
+sample_method_t sample_method;
+
+int load_model(char *model, char* options[], int threads, int diff) {
+    fprintf (stderr, "Loading model!\n");
+
+    char *stableDiffusionModel = "";
+    if (diff == 1 ) {
+        stableDiffusionModel = model;
+        model = "";
+    }
+
+    // decode options. Options are in form optname:optvale, or if booleans only optname.
+    char *clip_l_path  = "";
+    char *clip_g_path  = "";
+    char *t5xxl_path  = "";
+    char *vae_path  = "";
+    char *scheduler = "";
+    char *sampler = "";
+
+    // If options is not NULL, parse options
+    for (int i = 0; options[i] != NULL; i++) {
+        char *optname = strtok(options[i], ":");
+        char *optval = strtok(NULL, ":");
+        if (optval == NULL) {
+            optval = "true";
+        }
+
+        if (!strcmp(optname, "clip_l_path")) {
+            clip_l_path = optval;
+        }
+        if (!strcmp(optname, "clip_g_path")) {
+            clip_g_path = optval;
+        }
+        if (!strcmp(optname, "t5xxl_path")) {
+            t5xxl_path = optval;
+        }
+        if (!strcmp(optname, "vae_path")) {
+            vae_path = optval;
+        }
+        if (!strcmp(optname, "scheduler")) {
+            scheduler = optval;
+        }
+        if (!strcmp(optname, "sampler")) {
+            sampler = optval;
+        }
+    }
+
+    int sample_method_found = -1;
+    for (int m = 0; m < N_SAMPLE_METHODS; m++) {
+        if (!strcmp(sampler, sample_method_str[m])) {
+            sample_method_found = m;
+        }
+    }
+    if (sample_method_found == -1) {
+        fprintf(stderr, "Invalid sample method, default to EULER_A!\n");
+        sample_method_found = EULER_A;
+    }
+    sample_method = (sample_method_t)sample_method_found;
+
+    int schedule_found            = -1;
+    for (int d = 0; d < N_SCHEDULES; d++) {
+        if (!strcmp(scheduler, schedule_str[d])) {
+            schedule_found = d;
+                fprintf (stderr, "Found scheduler: %s\n", scheduler);
+
+        }
+    }
+
+    if (schedule_found == -1) {
+        fprintf (stderr, "Invalid scheduler! using DEFAULT\n");
+        schedule_found = DEFAULT;
+    }
+
+    schedule_t schedule = (schedule_t)schedule_found;
+    
+    fprintf (stderr, "Creating context\n");
+    sd_ctx_t* sd_ctx = new_sd_ctx(model,
+                                  clip_l_path,
+                                  clip_g_path,
+                                  t5xxl_path,
+                                  stableDiffusionModel,
+                                  vae_path,
+                                  "",
+                                  "",
+                                  "",
+                                  "",
+                                  "",
+                                  false,
+                                  false,
+                                  false,
+                                  threads,
+                                  SD_TYPE_COUNT,
+                                  STD_DEFAULT_RNG,
+                                  schedule,
+                                  false,
+                                  false,
+                                  false,
+                                  false);
+
+    if (sd_ctx == NULL) {
+        fprintf (stderr, "failed loading model (generic error)\n");
+        return 1;
+    }
+    fprintf (stderr, "Created context: OK\n");
+
+    sd_c = sd_ctx;
+
+    return 0;
+}
+
+int gen_image(char *text, char *negativeText, int width, int height, int steps, int seed , char *dst, float cfg_scale) {
+
+    sd_image_t* results;
+
+    std::vector<int> skip_layers = {7, 8, 9};
+
+    fprintf (stderr, "Generating image\n");
+
+    results = txt2img(sd_c,
+                            text,
+                            negativeText,
+                            -1, //clip_skip
+                            cfg_scale, // sfg_scale
+                            3.5f,
+			    0, // eta
+                            width,
+                            height,
+                            sample_method, 
+                            steps,
+                            seed,
+                            1,
+                            NULL,
+                            0.9f,
+                            20.f,
+                            false,
+                            "",
+                            skip_layers.data(),
+                            skip_layers.size(),
+                            0,
+                            0.01,
+                            0.2);
+
+    if (results == NULL) {
+        fprintf (stderr, "NO results\n");
+        return 1;
+    }
+
+    if (results[0].data == NULL) {
+        fprintf (stderr, "Results with no data\n");
+        return 1;
+    }
+
+    fprintf (stderr, "Writing PNG\n");
+
+    fprintf (stderr, "DST: %s\n", dst);
+    fprintf (stderr, "Width: %d\n", results[0].width);
+    fprintf (stderr, "Height: %d\n", results[0].height);
+    fprintf (stderr, "Channel: %d\n", results[0].channel);
+    fprintf (stderr, "Data: %p\n", results[0].data);
+
+    stbi_write_png(dst, results[0].width, results[0].height, results[0].channel,
+                       results[0].data, 0, NULL);
+    fprintf (stderr, "Saved resulting image to '%s'\n", dst);
+
+    // TODO: free results. Why does it crash?
+
+    free(results[0].data);
+    results[0].data = NULL;
+    free(results);
+    fprintf (stderr, "gen_image is done", dst);
+
+    return 0;
+}
+
+int unload() {
+    free_sd_ctx(sd_c);
+}
+

backend/go/image/stablediffusion-ggml/gosd.go

@@ -0,0 +1,96 @@
+package main
+
+// #cgo CXXFLAGS: -I${SRCDIR}/../../../../sources/stablediffusion-ggml.cpp/thirdparty -I${SRCDIR}/../../../../sources/stablediffusion-ggml.cpp -I${SRCDIR}/../../../../sources/stablediffusion-ggml.cpp/ggml/include
+// #cgo LDFLAGS: -L${SRCDIR}/ -lsd -lstdc++ -lm -lggmlall -lgomp
+// #include <gosd.h>
+// #include <stdlib.h>
+import "C"
+
+import (
+	"fmt"
+	"os"
+	"path/filepath"
+	"strings"
+	"unsafe"
+
+	"github.com/mudler/LocalAI/pkg/grpc/base"
+	pb "github.com/mudler/LocalAI/pkg/grpc/proto"
+	"github.com/mudler/LocalAI/pkg/utils"
+)
+
+type SDGGML struct {
+	base.SingleThread
+	threads      int
+	sampleMethod string
+	cfgScale     float32
+}
+
+func (sd *SDGGML) Load(opts *pb.ModelOptions) error {
+
+	sd.threads = int(opts.Threads)
+
+	modelFile := C.CString(opts.ModelFile)
+	defer C.free(unsafe.Pointer(modelFile))
+
+	var options **C.char
+	// prepare the options array to pass to C
+
+	size := C.size_t(unsafe.Sizeof((*C.char)(nil)))
+	length := C.size_t(len(opts.Options))
+	options = (**C.char)(C.malloc(length * size))
+	view := (*[1 << 30]*C.char)(unsafe.Pointer(options))[0:len(opts.Options):len(opts.Options)]
+
+	var diffusionModel int
+
+	var oo []string
+	for _, op := range opts.Options {
+		if op == "diffusion_model" {
+			diffusionModel = 1
+			continue
+		}
+
+		// If it's an option path, we resolve absolute path from the model path
+		if strings.Contains(op, ":") && strings.Contains(op, "path") {
+			data := strings.Split(op, ":")
+			data[1] = filepath.Join(opts.ModelPath, data[1])
+			if err := utils.VerifyPath(data[1], opts.ModelPath); err == nil {
+				oo = append(oo, strings.Join(data, ":"))
+			}
+		} else {
+			oo = append(oo, op)
+		}
+	}
+
+	fmt.Fprintf(os.Stderr, "Options: %+v\n", oo)
+
+	for i, x := range oo {
+		view[i] = C.CString(x)
+	}
+
+	sd.cfgScale = opts.CFGScale
+
+	ret := C.load_model(modelFile, options, C.int(opts.Threads), C.int(diffusionModel))
+	if ret != 0 {
+		return fmt.Errorf("could not load model")
+	}
+
+	return nil
+}
+
+func (sd *SDGGML) GenerateImage(opts *pb.GenerateImageRequest) error {
+	t := C.CString(opts.PositivePrompt)
+	defer C.free(unsafe.Pointer(t))
+
+	dst := C.CString(opts.Dst)
+	defer C.free(unsafe.Pointer(dst))
+
+	negative := C.CString(opts.NegativePrompt)
+	defer C.free(unsafe.Pointer(negative))
+
+	ret := C.gen_image(t, negative, C.int(opts.Width), C.int(opts.Height), C.int(opts.Step), C.int(opts.Seed), dst, C.float(sd.cfgScale))
+	if ret != 0 {
+		return fmt.Errorf("inference failed")
+	}
+
+	return nil
+}

backend/go/image/stablediffusion-ggml/gosd.h

@@ -0,0 +1,8 @@
+#ifdef __cplusplus
+extern "C" {
+#endif
+int load_model(char *model, char* options[], int threads, int diffusionModel);
+int gen_image(char *text, char *negativeText, int width, int height, int steps, int seed, char *dst, float cfg_scale);
+#ifdef __cplusplus
+}
+#endif

backend/go/image/stablediffusion-ggml/main.go

@@ -0,0 +1,20 @@
+package main
+
+// Note: this is started internally by LocalAI and a server is allocated for each model
+import (
+	"flag"
+
+	grpc "github.com/mudler/LocalAI/pkg/grpc"
+)
+
+var (
+	addr = flag.String("addr", "localhost:50051", "the address to connect to")
+)
+
+func main() {
+	flag.Parse()
+
+	if err := grpc.StartServer(*addr, &SDGGML{}); err != nil {
+		panic(err)
+	}
+}

backend/go/llm/llama/llama.go

@@ -58,9 +58,6 @@ func (llm *LLM) Load(opts *pb.ModelOptions) error {
 	if opts.Embeddings {
 		llamaOpts = append(llamaOpts, llama.EnableEmbeddings)
 	}
-	if opts.Reranking {
-		llamaOpts = append(llamaOpts, llama.EnableReranking)
-	}
 	if opts.NGPULayers != 0 {
 		llamaOpts = append(llamaOpts, llama.SetGPULayers(int(opts.NGPULayers)))
 	}

backend/go/local-store/Makefile

@@ -1,9 +0,0 @@
-GOCMD=go
-
-local-store:
-	CGO_ENABLED=0 $(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o local-store ./
-
-package:
-	bash package.sh
-
-build: local-store package

backend/go/local-store/package.sh

@@ -1,12 +0,0 @@
-#!/bin/bash
-
-# Script to copy the appropriate libraries based on architecture
-# This script is used in the final stage of the Dockerfile
-
-set -e
-
-CURDIR=$(dirname "$(realpath $0)")
-
-mkdir -p $CURDIR/package
-cp -avrf $CURDIR/local-store $CURDIR/package/
-cp -rfv $CURDIR/run.sh $CURDIR/package/

backend/go/local-store/run.sh

@@ -1,6 +0,0 @@
-#!/bin/bash
-set -ex
-
-CURDIR=$(dirname "$(realpath $0)")
-
-exec $CURDIR/local-store "$@"

backend/go/piper/Makefile

@@ -1,37 +0,0 @@
-
-# go-piper version
-PIPER_REPO?=https://github.com/mudler/go-piper
-PIPER_VERSION?=e10ca041a885d4a8f3871d52924b47792d5e5aa0
-
-CURRENT_DIR=$(abspath ./)
-GOCMD=go
-
-PIPER_CGO_CXXFLAGS+=-I$(CURRENT_DIR)/sources/go-piper/piper/src/cpp -I$(CURRENT_DIR)/sources/go-piper/piper/build/fi/include -I$(CURRENT_DIR)/sources/go-piper/piper/build/pi/include -I$(CURRENT_DIR)/sources/go-piper/piper/build/si/include
-PIPER_CGO_LDFLAGS+=-L$(CURRENT_DIR)/sources/go-piper/piper/build/fi/lib -L$(CURRENT_DIR)/sources/go-piper/piper/build/pi/lib -L$(CURRENT_DIR)/sources/go-piper/piper/build/si/lib -lfmt -lspdlog -lucd
-
-## go-piper
-sources/go-piper:
-	mkdir -p sources/go-piper
-	cd sources/go-piper && \
-	git init && \
-	git remote add origin $(PIPER_REPO) && \
-	git fetch origin && \
-	git checkout $(PIPER_VERSION) && \
-	git submodule update --init --recursive --depth 1 --single-branch
-
-sources/go-piper/libpiper_binding.a: sources/go-piper
-	$(MAKE) -C sources/go-piper libpiper_binding.a example/main piper.o
-
-espeak-ng-data: sources/go-piper sources/go-piper/libpiper_binding.a
-	mkdir -p espeak-ng-data
-	@cp -rf sources/go-piper/piper-phonemize/pi/share/espeak-ng-data/. espeak-ng-data
-
-piper: sources/go-piper sources/go-piper/libpiper_binding.a espeak-ng-data
-	$(GOCMD) mod edit -replace github.com/mudler/go-piper=$(CURRENT_DIR)/sources/go-piper
-	CGO_CXXFLAGS="$(PIPER_CGO_CXXFLAGS)" CGO_LDFLAGS="$(PIPER_CGO_LDFLAGS)" LIBRARY_PATH=$(CURRENT_DIR)/sources/go-piper \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o piper ./
-
-package:
-	bash package.sh
-
-build: piper package

backend/go/piper/package.sh

@@ -1,54 +0,0 @@
-#!/bin/bash
-
-# Script to copy the appropriate libraries based on architecture
-# This script is used in the final stage of the Dockerfile
-
-set -e
-
-CURDIR=$(dirname "$(realpath $0)")
-
-# Create lib directory
-mkdir -p $CURDIR/package/lib
-
-cp -avrf $CURDIR/piper $CURDIR/package/
-cp -avrf $CURDIR/espeak-ng-data $CURDIR/package/
-cp -rfv $CURDIR/run.sh $CURDIR/package/
-cp -rfLv $CURDIR/sources/go-piper/piper-phonemize/pi/lib/* $CURDIR/package/lib/
-
-# Detect architecture and copy appropriate libraries
-if [ -f "/lib64/ld-linux-x86-64.so.2" ]; then
-    # x86_64 architecture
-    echo "Detected x86_64 architecture, copying x86_64 libraries..."
-    cp -arfLv /lib64/ld-linux-x86-64.so.2 $CURDIR/package/lib/ld.so
-    cp -arfLv /lib/x86_64-linux-gnu/libc.so.6 $CURDIR/package/lib/libc.so.6
-    cp -arfLv /lib/x86_64-linux-gnu/libgcc_s.so.1 $CURDIR/package/lib/libgcc_s.so.1
-    cp -arfLv /lib/x86_64-linux-gnu/libstdc++.so.6 $CURDIR/package/lib/libstdc++.so.6
-    cp -arfLv /lib/x86_64-linux-gnu/libm.so.6 $CURDIR/package/lib/libm.so.6
-    cp -arfLv /lib/x86_64-linux-gnu/libgomp.so.1 $CURDIR/package/lib/libgomp.so.1
-    cp -arfLv /lib/x86_64-linux-gnu/libgcc_s.so.1 $CURDIR/package/lib/libgcc_s.so.1
-    cp -arfLv /lib/x86_64-linux-gnu/libstdc++.so.6 $CURDIR/package/lib/libstdc++.so.6
-    cp -arfLv /lib/x86_64-linux-gnu/libdl.so.2 $CURDIR/package/lib/libdl.so.2
-    cp -arfLv /lib/x86_64-linux-gnu/librt.so.1 $CURDIR/package/lib/librt.so.1
-    cp -arfLv /lib/x86_64-linux-gnu/libpthread.so.0 $CURDIR/package/lib/libpthread.so.0
-elif [ -f "/lib/ld-linux-aarch64.so.1" ]; then
-    # ARM64 architecture
-    echo "Detected ARM64 architecture, copying ARM64 libraries..."
-    cp -arfLv /lib/ld-linux-aarch64.so.1 $CURDIR/package/lib/ld.so
-    cp -arfLv /lib/aarch64-linux-gnu/libc.so.6 $CURDIR/package/lib/libc.so.6
-    cp -arfLv /lib/aarch64-linux-gnu/libgcc_s.so.1 $CURDIR/package/lib/libgcc_s.so.1
-    cp -arfLv /lib/aarch64-linux-gnu/libstdc++.so.6 $CURDIR/package/lib/libstdc++.so.6
-    cp -arfLv /lib/aarch64-linux-gnu/libm.so.6 $CURDIR/package/lib/libm.so.6
-    cp -arfLv /lib/aarch64-linux-gnu/libgomp.so.1 $CURDIR/package/lib/libgomp.so.1
-    cp -arfLv /lib/aarch64-linux-gnu/libgcc_s.so.1 $CURDIR/package/lib/libgcc_s.so.1
-    cp -arfLv /lib/aarch64-linux-gnu/libstdc++.so.6 $CURDIR/package/lib/libstdc++.so.6
-    cp -arfLv /lib/aarch64-linux-gnu/libdl.so.2 $CURDIR/package/lib/libdl.so.2
-    cp -arfLv /lib/aarch64-linux-gnu/librt.so.1 $CURDIR/package/lib/librt.so.1
-    cp -arfLv /lib/aarch64-linux-gnu/libpthread.so.0 $CURDIR/package/lib/libpthread.so.0
-else
-    echo "Error: Could not detect architecture"
-    exit 1
-fi
-
-echo "Packaging completed successfully" 
-ls -liah $CURDIR/package/
-ls -liah $CURDIR/package/lib/

backend/go/piper/run.sh

@@ -1,15 +0,0 @@
-#!/bin/bash
-set -ex
-
-CURDIR=$(dirname "$(realpath $0)")
-
-export ESPEAK_NG_DATA=$CURDIR/espeak-ng-data
-export LD_LIBRARY_PATH=$CURDIR/lib:$LD_LIBRARY_PATH
-
-# If there is a lib/ld.so, use it
-if [ -f $CURDIR/lib/ld.so ]; then
-	echo "Using lib/ld.so"
-	exec $CURDIR/lib/ld.so $CURDIR/piper "$@"
-fi
-
-exec $CURDIR/piper "$@"

backend/go/silero-vad/Makefile

@@ -1,47 +0,0 @@
-
-CURRENT_DIR=$(abspath ./)
-GOCMD=go
-
-ONNX_VERSION?=1.20.0
-ONNX_ARCH?=x64
-ONNX_OS?=linux
-
-# Detect if we are running on arm64
-ifneq (,$(findstring aarch64,$(shell uname -m)))
-	ONNX_ARCH=aarch64
-endif
-
-ifeq ($(OS),Darwin)
-	ONNX_OS=osx
-	ifneq (,$(findstring aarch64,$(shell uname -m)))
-		ONNX_ARCH=arm64
-	else ifneq (,$(findstring arm64,$(shell uname -m)))
-		ONNX_ARCH=arm64
-	else
-		ONNX_ARCH=x86_64
-	endif
-endif
-
-sources/onnxruntime:
-	mkdir -p sources/onnxruntime
-	curl -L https://github.com/microsoft/onnxruntime/releases/download/v$(ONNX_VERSION)/onnxruntime-$(ONNX_OS)-$(ONNX_ARCH)-$(ONNX_VERSION).tgz -o sources/onnxruntime/onnxruntime-$(ONNX_OS)-$(ONNX_ARCH)-$(ONNX_VERSION).tgz
-	cd sources/onnxruntime && tar -xvf onnxruntime-$(ONNX_OS)-$(ONNX_ARCH)-$(ONNX_VERSION).tgz && rm onnxruntime-$(ONNX_OS)-$(ONNX_ARCH)-$(ONNX_VERSION).tgz
-	cd sources/onnxruntime && mv onnxruntime-$(ONNX_OS)-$(ONNX_ARCH)-$(ONNX_VERSION)/* ./
-
-backend-assets/lib/libonnxruntime.so.1: sources/onnxruntime
-	mkdir -p backend-assets/lib
-	cp -rfLv sources/onnxruntime/lib/* backend-assets/lib/
-ifeq ($(OS),Darwin)
-	mv backend-assets/lib/libonnxruntime.$(ONNX_VERSION).dylib backend-assets/lib/libonnxruntime.dylib
-else
-	mv backend-assets/lib/libonnxruntime.so.$(ONNX_VERSION) backend-assets/lib/libonnxruntime.so.1
-endif
-
-silero-vad: backend-assets/lib/libonnxruntime.so.1
-	CGO_LDFLAGS="$(CGO_LDFLAGS)" CPATH="$(CPATH):$(CURRENT_DIR)/sources/onnxruntime/include/" LIBRARY_PATH=$(CURRENT_DIR)/backend-assets/lib \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o silero-vad ./
-
-package:
-	bash package.sh
-
-build: silero-vad package

backend/go/silero-vad/package.sh

@@ -1,53 +0,0 @@
-#!/bin/bash
-
-# Script to copy the appropriate libraries based on architecture
-# This script is used in the final stage of the Dockerfile
-
-set -e
-
-CURDIR=$(dirname "$(realpath $0)")
-
-# Create lib directory
-mkdir -p $CURDIR/package/lib
-
-cp -avrf $CURDIR/silero-vad $CURDIR/package/
-cp -avrf $CURDIR/run.sh $CURDIR/package/
-cp -rfLv $CURDIR/backend-assets/lib/* $CURDIR/package/lib/
-
-# Detect architecture and copy appropriate libraries
-if [ -f "/lib64/ld-linux-x86-64.so.2" ]; then
-    # x86_64 architecture
-    echo "Detected x86_64 architecture, copying x86_64 libraries..."
-    cp -arfLv /lib64/ld-linux-x86-64.so.2 $CURDIR/package/lib/ld.so
-    cp -arfLv /lib/x86_64-linux-gnu/libc.so.6 $CURDIR/package/lib/libc.so.6
-    cp -arfLv /lib/x86_64-linux-gnu/libgcc_s.so.1 $CURDIR/package/lib/libgcc_s.so.1
-    cp -arfLv /lib/x86_64-linux-gnu/libstdc++.so.6 $CURDIR/package/lib/libstdc++.so.6
-    cp -arfLv /lib/x86_64-linux-gnu/libm.so.6 $CURDIR/package/lib/libm.so.6
-    cp -arfLv /lib/x86_64-linux-gnu/libgomp.so.1 $CURDIR/package/lib/libgomp.so.1
-    cp -arfLv /lib/x86_64-linux-gnu/libgcc_s.so.1 $CURDIR/package/lib/libgcc_s.so.1
-    cp -arfLv /lib/x86_64-linux-gnu/libstdc++.so.6 $CURDIR/package/lib/libstdc++.so.6
-    cp -arfLv /lib/x86_64-linux-gnu/libdl.so.2 $CURDIR/package/lib/libdl.so.2
-    cp -arfLv /lib/x86_64-linux-gnu/librt.so.1 $CURDIR/package/lib/librt.so.1
-    cp -arfLv /lib/x86_64-linux-gnu/libpthread.so.0 $CURDIR/package/lib/libpthread.so.0
-elif [ -f "/lib/ld-linux-aarch64.so.1" ]; then
-    # ARM64 architecture
-    echo "Detected ARM64 architecture, copying ARM64 libraries..."
-    cp -arfLv /lib/ld-linux-aarch64.so.1 $CURDIR/package/lib/ld.so
-    cp -arfLv /lib/aarch64-linux-gnu/libc.so.6 $CURDIR/package/lib/libc.so.6
-    cp -arfLv /lib/aarch64-linux-gnu/libgcc_s.so.1 $CURDIR/package/lib/libgcc_s.so.1
-    cp -arfLv /lib/aarch64-linux-gnu/libstdc++.so.6 $CURDIR/package/lib/libstdc++.so.6
-    cp -arfLv /lib/aarch64-linux-gnu/libm.so.6 $CURDIR/package/lib/libm.so.6
-    cp -arfLv /lib/aarch64-linux-gnu/libgomp.so.1 $CURDIR/package/lib/libgomp.so.1
-    cp -arfLv /lib/aarch64-linux-gnu/libgcc_s.so.1 $CURDIR/package/lib/libgcc_s.so.1
-    cp -arfLv /lib/aarch64-linux-gnu/libstdc++.so.6 $CURDIR/package/lib/libstdc++.so.6
-    cp -arfLv /lib/aarch64-linux-gnu/libdl.so.2 $CURDIR/package/lib/libdl.so.2
-    cp -arfLv /lib/aarch64-linux-gnu/librt.so.1 $CURDIR/package/lib/librt.so.1
-    cp -arfLv /lib/aarch64-linux-gnu/libpthread.so.0 $CURDIR/package/lib/libpthread.so.0
-else
-    echo "Error: Could not detect architecture"
-    exit 1
-fi
-
-echo "Packaging completed successfully" 
-ls -liah $CURDIR/package/
-ls -liah $CURDIR/package/lib/

backend/go/silero-vad/run.sh

@@ -1,14 +0,0 @@
-#!/bin/bash
-set -ex
-
-CURDIR=$(dirname "$(realpath $0)")
-
-export LD_LIBRARY_PATH=$CURDIR/lib:$LD_LIBRARY_PATH
-
-# If there is a lib/ld.so, use it
-if [ -f $CURDIR/lib/ld.so ]; then
-	echo "Using lib/ld.so"
-	exec $CURDIR/lib/ld.so $CURDIR/silero-vad "$@"
-fi
-
-exec $CURDIR/silero-vad "$@"