feat(diffusers): add builds for nvidia-l4t

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
2026-02-03 03:02:38 -05:00 · 2025-08-08 22:50:40 +02:00
275 changed files with 2712 additions and 13819 deletions
--- a/.dockerignore
+++ b/.dockerignore
@@ -6,10 +6,6 @@ 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*
--- a/.github/workflows/backend.yml
+++ b/.github/workflows/backend.yml
@@ -89,8 +89,8 @@ jobs:
            context: "./backend"
          - build-type: 'l4t'
            cuda-major-version: "12"
-            cuda-minor-version: "8"
-            platforms: 'linux/arm64'
+            cuda-minor-version: "0"
+            platforms: 'linux/amd64'
            tag-latest: 'auto'
            tag-suffix: '-gpu-nvidia-l4t-diffusers'
            runs-on: 'ubuntu-24.04-arm'
@@ -99,30 +99,6 @@ jobs:
            backend: "diffusers"
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
-          - build-type: ''
-            cuda-major-version: ""
-            cuda-minor-version: ""
-            platforms: 'linux/amd64'
-            tag-latest: 'auto'
-            tag-suffix: '-cpu-diffusers'
-            runs-on: 'ubuntu-latest'
-            base-image: "ubuntu:22.04"
-            skip-drivers: 'true'
-            backend: "diffusers"
-            dockerfile: "./backend/Dockerfile.python"
-            context: "./backend"
-          - build-type: ''
-            cuda-major-version: ""
-            cuda-minor-version: ""
-            platforms: 'linux/amd64'
-            tag-latest: 'auto'
-            tag-suffix: '-cpu-chatterbox'
-            runs-on: 'ubuntu-latest'
-            base-image: "ubuntu:22.04"
-            skip-drivers: 'true'
-            backend: "chatterbox"
-            dockerfile: "./backend/Dockerfile.python"
-            context: "./backend"
          # CUDA 11 additional backends
          - build-type: 'cublas'
            cuda-major-version: "11"
@@ -187,7 +163,7 @@ jobs:
          # CUDA 12 builds
          - build-type: 'cublas'
            cuda-major-version: "12"
-            cuda-minor-version: "8"
+            cuda-minor-version: "0"
            platforms: 'linux/amd64'
            tag-latest: 'auto'
            tag-suffix: '-gpu-nvidia-cuda-12-rerankers'
@@ -199,7 +175,7 @@ jobs:
            context: "./backend"
          - build-type: 'cublas'
            cuda-major-version: "12"
-            cuda-minor-version: "8"
+            cuda-minor-version: "0"
            platforms: 'linux/amd64'
            tag-latest: 'auto'
            tag-suffix: '-gpu-nvidia-cuda-12-llama-cpp'
@@ -211,11 +187,11 @@ jobs:
            context: "./"
          - build-type: 'cublas'
            cuda-major-version: "12"
-            cuda-minor-version: "8"
+            cuda-minor-version: "0"
            platforms: 'linux/amd64'
            tag-latest: 'auto'
            tag-suffix: '-gpu-nvidia-cuda-12-vllm'
-            runs-on: 'arc-runner-set'
+            runs-on: 'ubuntu-latest'
            base-image: "ubuntu:22.04"
            skip-drivers: 'false'
            backend: "vllm"
@@ -223,7 +199,7 @@ jobs:
            context: "./backend"
          - build-type: 'cublas'
            cuda-major-version: "12"
-            cuda-minor-version: "8"
+            cuda-minor-version: "0"
            platforms: 'linux/amd64'
            tag-latest: 'auto'
            tag-suffix: '-gpu-nvidia-cuda-12-transformers'
@@ -235,20 +211,20 @@ jobs:
            context: "./backend"
          - build-type: 'cublas'
            cuda-major-version: "12"
-            cuda-minor-version: "8"
+            cuda-minor-version: "0"
            platforms: 'linux/amd64'
            tag-latest: 'auto'
            tag-suffix: '-gpu-nvidia-cuda-12-diffusers'
            runs-on: 'ubuntu-latest'
            base-image: "ubuntu:22.04"
            skip-drivers: 'false'
-            backend: "diffusers"
+            backend: "diffusers"            
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          # CUDA 12 additional backends
          - build-type: 'cublas'
            cuda-major-version: "12"
-            cuda-minor-version: "8"
+            cuda-minor-version: "0"
            platforms: 'linux/amd64'
            tag-latest: 'auto'
            tag-suffix: '-gpu-nvidia-cuda-12-kokoro'
@@ -260,7 +236,7 @@ jobs:
            context: "./backend"
          - build-type: 'cublas'
            cuda-major-version: "12"
-            cuda-minor-version: "8"
+            cuda-minor-version: "0"
            platforms: 'linux/amd64'
            tag-latest: 'auto'
            tag-suffix: '-gpu-nvidia-cuda-12-faster-whisper'
@@ -272,7 +248,7 @@ jobs:
            context: "./backend"
          - build-type: 'cublas'
            cuda-major-version: "12"
-            cuda-minor-version: "8"
+            cuda-minor-version: "0"
            platforms: 'linux/amd64'
            tag-latest: 'auto'
            tag-suffix: '-gpu-nvidia-cuda-12-coqui'
@@ -284,7 +260,7 @@ jobs:
            context: "./backend"
          - build-type: 'cublas'
            cuda-major-version: "12"
-            cuda-minor-version: "8"
+            cuda-minor-version: "0"
            platforms: 'linux/amd64'
            tag-latest: 'auto'
            tag-suffix: '-gpu-nvidia-cuda-12-bark'
@@ -296,7 +272,7 @@ jobs:
            context: "./backend"
          - build-type: 'cublas'
            cuda-major-version: "12"
-            cuda-minor-version: "8"
+            cuda-minor-version: "0"
            platforms: 'linux/amd64'
            tag-latest: 'auto'
            tag-suffix: '-gpu-nvidia-cuda-12-chatterbox'
@@ -314,7 +290,7 @@ jobs:
            tag-latest: 'auto'
            tag-suffix: '-gpu-rocm-hipblas-rerankers'
            runs-on: 'ubuntu-latest'
-            base-image: "rocm/dev-ubuntu-22.04:6.4.3"
+            base-image: "rocm/dev-ubuntu-22.04:6.1"
            skip-drivers: 'false'
            backend: "rerankers"
            dockerfile: "./backend/Dockerfile.python"
@@ -326,7 +302,7 @@ jobs:
            tag-latest: 'auto'
            tag-suffix: '-gpu-rocm-hipblas-llama-cpp'
            runs-on: 'ubuntu-latest'
-            base-image: "rocm/dev-ubuntu-22.04:6.4.3"
+            base-image: "rocm/dev-ubuntu-22.04:6.1"
            skip-drivers: 'false'
            backend: "llama-cpp"
            dockerfile: "./backend/Dockerfile.llama-cpp"
@@ -337,8 +313,8 @@ jobs:
            platforms: 'linux/amd64'
            tag-latest: 'auto'
            tag-suffix: '-gpu-rocm-hipblas-vllm'
-            runs-on: 'arc-runner-set'
-            base-image: "rocm/dev-ubuntu-22.04:6.4.3"
+            runs-on: 'ubuntu-latest'
+            base-image: "rocm/dev-ubuntu-22.04:6.1"
            skip-drivers: 'false'
            backend: "vllm"
            dockerfile: "./backend/Dockerfile.python"
@@ -350,7 +326,7 @@ jobs:
            tag-latest: 'auto'
            tag-suffix: '-gpu-rocm-hipblas-transformers'
            runs-on: 'arc-runner-set'
-            base-image: "rocm/dev-ubuntu-22.04:6.4.3"
+            base-image: "rocm/dev-ubuntu-22.04:6.1"
            skip-drivers: 'false'
            backend: "transformers"
            dockerfile: "./backend/Dockerfile.python"
@@ -362,7 +338,7 @@ jobs:
            tag-latest: 'auto'
            tag-suffix: '-gpu-rocm-hipblas-diffusers'
            runs-on: 'arc-runner-set'
-            base-image: "rocm/dev-ubuntu-22.04:6.4.3"
+            base-image: "rocm/dev-ubuntu-22.04:6.1"
            skip-drivers: 'false'
            backend: "diffusers"
            dockerfile: "./backend/Dockerfile.python"
@@ -375,7 +351,7 @@ jobs:
            tag-latest: 'auto'
            tag-suffix: '-gpu-rocm-hipblas-kokoro'
            runs-on: 'arc-runner-set'
-            base-image: "rocm/dev-ubuntu-22.04:6.4.3"
+            base-image: "rocm/dev-ubuntu-22.04:6.1"
            skip-drivers: 'false'
            backend: "kokoro"
            dockerfile: "./backend/Dockerfile.python"
@@ -387,7 +363,7 @@ jobs:
            tag-latest: 'auto'
            tag-suffix: '-gpu-rocm-hipblas-faster-whisper'
            runs-on: 'ubuntu-latest'
-            base-image: "rocm/dev-ubuntu-22.04:6.4.3"
+            base-image: "rocm/dev-ubuntu-22.04:6.1"
            skip-drivers: 'false'
            backend: "faster-whisper"
            dockerfile: "./backend/Dockerfile.python"
@@ -399,7 +375,7 @@ jobs:
            tag-latest: 'auto'
            tag-suffix: '-gpu-rocm-hipblas-coqui'
            runs-on: 'ubuntu-latest'
-            base-image: "rocm/dev-ubuntu-22.04:6.4.3"
+            base-image: "rocm/dev-ubuntu-22.04:6.1"
            skip-drivers: 'false'
            backend: "coqui"
            dockerfile: "./backend/Dockerfile.python"
@@ -411,7 +387,7 @@ jobs:
            tag-latest: 'auto'
            tag-suffix: '-gpu-rocm-hipblas-bark'
            runs-on: 'arc-runner-set'
-            base-image: "rocm/dev-ubuntu-22.04:6.4.3"
+            base-image: "rocm/dev-ubuntu-22.04:6.1"
            skip-drivers: 'false'
            backend: "bark"
            dockerfile: "./backend/Dockerfile.python"
@@ -459,7 +435,7 @@ jobs:
            platforms: 'linux/amd64'
            tag-latest: 'auto'
            tag-suffix: '-gpu-intel-vllm'
-            runs-on: 'arc-runner-set'
+            runs-on: 'ubuntu-latest'
            base-image: "quay.io/go-skynet/intel-oneapi-base:latest"
            skip-drivers: 'false'
            backend: "vllm"
@@ -578,7 +554,7 @@ jobs:
            context: "./"
          - build-type: 'cublas'
            cuda-major-version: "12"
-            cuda-minor-version: "8"
+            cuda-minor-version: "0"
            platforms: 'linux/arm64'
            skip-drivers: 'true'
            tag-latest: 'auto'
@@ -615,7 +591,7 @@ jobs:
            context: "./"
          - build-type: 'cublas'
            cuda-major-version: "12"
-            cuda-minor-version: "8"
+            cuda-minor-version: "0"
            platforms: 'linux/amd64'
            tag-latest: 'auto'
            tag-suffix: '-gpu-nvidia-cuda-12-stablediffusion-ggml'
@@ -675,7 +651,7 @@ jobs:
            context: "./"
          - build-type: 'cublas'
            cuda-major-version: "12"
-            cuda-minor-version: "8"
+            cuda-minor-version: "0"
            platforms: 'linux/arm64'
            skip-drivers: 'true'
            tag-latest: 'auto'
@@ -700,7 +676,7 @@ jobs:
            context: "./"
          - build-type: 'cublas'
            cuda-major-version: "12"
-            cuda-minor-version: "8"
+            cuda-minor-version: "0"
            platforms: 'linux/amd64'
            tag-latest: 'auto'
            tag-suffix: '-gpu-nvidia-cuda-12-whisper'
@@ -760,7 +736,7 @@ jobs:
            context: "./"
          - build-type: 'cublas'
            cuda-major-version: "12"
-            cuda-minor-version: "8"
+            cuda-minor-version: "0"
            platforms: 'linux/arm64'
            skip-drivers: 'true'
            tag-latest: 'auto'
@@ -776,7 +752,7 @@ jobs:
            platforms: 'linux/amd64'
            tag-latest: 'auto'
            tag-suffix: '-gpu-hipblas-whisper'
-            base-image: "rocm/dev-ubuntu-22.04:6.4.3"
+            base-image: "rocm/dev-ubuntu-22.04:6.1"
            runs-on: 'ubuntu-latest'
            skip-drivers: 'false'
            backend: "whisper"
@@ -836,7 +812,7 @@ jobs:
            context: "./backend"
          - build-type: 'cublas'
            cuda-major-version: "12"
-            cuda-minor-version: "8"
+            cuda-minor-version: "0"
            platforms: 'linux/amd64'
            tag-latest: 'auto'
            tag-suffix: '-gpu-nvidia-cuda-12-rfdetr'
@@ -872,7 +848,7 @@ jobs:
            context: "./backend"
          - build-type: 'cublas'
            cuda-major-version: "12"
-            cuda-minor-version: "8"
+            cuda-minor-version: "0"
            platforms: 'linux/arm64'
            skip-drivers: 'true'
            tag-latest: 'auto'
@@ -897,7 +873,7 @@ jobs:
            context: "./backend"
          - build-type: 'cublas'
            cuda-major-version: "12"
-            cuda-minor-version: "8"
+            cuda-minor-version: "0"
            platforms: 'linux/amd64'
            tag-latest: 'auto'
            tag-suffix: '-gpu-nvidia-cuda-12-exllama2'
@@ -938,7 +914,7 @@ jobs:
            skip-drivers: 'true'
            tag-latest: 'auto'
            tag-suffix: '-gpu-hipblas-exllama2'
-            base-image: "rocm/dev-ubuntu-22.04:6.4.3"
+            base-image: "rocm/dev-ubuntu-22.04:6.1"
            runs-on: 'ubuntu-latest'
            backend: "exllama2"
            dockerfile: "./backend/Dockerfile.python"
@@ -950,7 +926,7 @@ jobs:
          #   platforms: 'linux/amd64'
          #   tag-latest: 'auto'
          #   tag-suffix: '-gpu-hipblas-rfdetr'
-          #   base-image: "rocm/dev-ubuntu-22.04:6.4.3"
+          #   base-image: "rocm/dev-ubuntu-22.04:6.1"
          #   runs-on: 'ubuntu-latest'
          #   skip-drivers: 'false'
          #   backend: "rfdetr"
@@ -969,47 +945,6 @@ jobs:
            backend: "kitten-tts"
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
-  backend-jobs-darwin:
-    uses: ./.github/workflows/backend_build_darwin.yml
-    strategy:
-      matrix:
-        include:
-          - backend: "diffusers"
-            tag-suffix: "-metal-darwin-arm64-diffusers"
-            build-type: "mps"
-          - backend: "mlx"
-            tag-suffix: "-metal-darwin-arm64-mlx"
-            build-type: "mps"
-          - backend: "chatterbox"
-            tag-suffix: "-metal-darwin-arm64-chatterbox"
-            build-type: "mps"
-          - backend: "mlx-vlm"
-            tag-suffix: "-metal-darwin-arm64-mlx-vlm"
-            build-type: "mps"
-          - backend: "mlx-audio"
-            tag-suffix: "-metal-darwin-arm64-mlx-audio"
-            build-type: "mps"
-          - backend: "stablediffusion-ggml"
-            tag-suffix: "-metal-darwin-arm64-stablediffusion-ggml"
-            build-type: "metal"
-            lang: "go"
-          - backend: "whisper"
-            tag-suffix: "-metal-darwin-arm64-whisper"
-            build-type: "metal"
-            lang: "go"
-    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:
-      dockerUsername: ${{ secrets.DOCKERHUB_USERNAME }}
-      dockerPassword: ${{ secrets.DOCKERHUB_PASSWORD }}
-      quayUsername: ${{ secrets.LOCALAI_REGISTRY_USERNAME }}
-      quayPassword: ${{ secrets.LOCALAI_REGISTRY_PASSWORD }}
  llama-cpp-darwin:
    runs-on: macOS-14
    strategy:
@@ -1017,7 +952,7 @@ 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 }}
@@ -1034,19 +969,22 @@ jobs:
      - name: Build llama-cpp-darwin
        run: |
          make protogen-go
-          make backends/llama-cpp-darwin
+          make build
+          bash scripts/build-llama-cpp-darwin.sh
+          ls -la build/darwin.tar
+          mv build/darwin.tar build/llama-cpp.tar
      - name: Upload llama-cpp.tar
        uses: actions/upload-artifact@v4
        with:
          name: llama-cpp-tar
-          path: backend-images/llama-cpp.tar
+          path: build/llama-cpp.tar
  llama-cpp-darwin-publish:
    needs: llama-cpp-darwin
    if: github.event_name != 'pull_request'
    runs-on: ubuntu-latest
    steps:
      - name: Download llama-cpp.tar
-        uses: actions/download-artifact@v5
+        uses: actions/download-artifact@v4
        with:
          name: llama-cpp-tar
          path: .
@@ -1103,7 +1041,7 @@ 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 }}
@@ -1122,19 +1060,21 @@ jobs:
          make protogen-go
          make build
          export PLATFORMARCH=darwin/amd64
-          make backends/llama-cpp-darwin
+          bash scripts/build-llama-cpp-darwin.sh
+          ls -la build/darwin.tar
+          mv build/darwin.tar build/llama-cpp.tar
      - name: Upload llama-cpp.tar
        uses: actions/upload-artifact@v4
        with:
          name: llama-cpp-tar-x86
-          path: backend-images/llama-cpp.tar
+          path: build/llama-cpp.tar
  llama-cpp-darwin-x86-publish:
    if: github.event_name != 'pull_request'
    needs: llama-cpp-darwin-x86
    runs-on: ubuntu-latest
    steps:
      - name: Download llama-cpp.tar
-        uses: actions/download-artifact@v5
+        uses: actions/download-artifact@v4
        with:
          name: llama-cpp-tar-x86
          path: .
--- a/.github/workflows/backend_build.yml
+++ b/.github/workflows/backend_build.yml
@@ -55,9 +55,9 @@ on:
        type: string
    secrets:
      dockerUsername:
-        required: false
+        required: true
      dockerPassword:
-        required: false
+        required: true
      quayUsername:
        required: true
      quayPassword:
@@ -66,8 +66,6 @@ on:
 jobs:
  backend-build:
    runs-on: ${{ inputs.runs-on }}
-    env:
-        quay_username: ${{ secrets.quayUsername }}
    steps:


@@ -97,7 +95,7 @@ jobs:
          && 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'
@@ -189,7 +187,7 @@ jobs:
          password: ${{ secrets.dockerPassword }}

      - name: Login to Quay.io
-        if: ${{ env.quay_username != '' }}
+        # if: github.event_name != 'pull_request'
        uses: docker/login-action@v3
        with:
          registry: quay.io
@@ -232,7 +230,7 @@ jobs:
          file: ${{ inputs.dockerfile }}
          cache-from: type=gha
          platforms: ${{ inputs.platforms }}
-          push: ${{ env.quay_username != '' }}
+          push: true
          tags: ${{ steps.meta_pull_request.outputs.tags }}
          labels: ${{ steps.meta_pull_request.outputs.labels }}

@@ -240,4 +238,4 @@ jobs:

      - name: job summary
        run: |
-          echo "Built image: ${{ steps.meta.outputs.labels }}" >> $GITHUB_STEP_SUMMARY
+          echo "Built image: ${{ steps.meta.outputs.labels }}" >> $GITHUB_STEP_SUMMARY 
--- a/.github/workflows/backend_build_darwin.yml
+++ b/.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 $tag
-          done
--- a/.github/workflows/backend_pr.yml
+++ b/.github/workflows/backend_pr.yml
@@ -1,78 +0,0 @@
-name: 'build backend container images (PR-filtered)'
-
-on:
-  pull_request:
-
-concurrency:
-  group: ci-backends-pr-${{ 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) }}
--- a/.github/workflows/build-test.yaml
+++ b/.github/workflows/build-test.yaml
@@ -11,7 +11,7 @@ jobs:
    runs-on: ubuntu-latest
    steps:
      - name: Checkout
-        uses: actions/checkout@v5
+        uses: actions/checkout@v4
        with:
          fetch-depth: 0
      - name: Set up Go
@@ -21,47 +21,3 @@ jobs:
      - 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
--- a/.github/workflows/bump_deps.yaml
+++ b/.github/workflows/bump_deps.yaml
@@ -31,7 +31,7 @@ jobs:
            file: "backend/go/piper/Makefile"
    runs-on: ubuntu-latest
    steps:
-      - uses: actions/checkout@v5
+      - uses: actions/checkout@v4
      - name: Bump dependencies 🔧
        id: bump
        run: |
--- a/.github/workflows/bump_docs.yaml
+++ b/.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 }}
--- a/.github/workflows/checksum_checker.yaml
+++ b/.github/workflows/checksum_checker.yaml
@@ -15,7 +15,7 @@ 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
--- a/.github/workflows/dependabot_auto.yml
+++ b/.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: |
--- a/.github/workflows/deploy-explorer.yaml
+++ b/.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
--- a/.github/workflows/generate_grpc_cache.yaml
+++ b/.github/workflows/generate_grpc_cache.yaml
@@ -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
--- a/.github/workflows/generate_intel_image.yaml
+++ b/.github/workflows/generate_intel_image.yaml
@@ -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
--- a/.github/workflows/image-pr.yml
+++ b/.github/workflows/image-pr.yml
@@ -36,7 +36,7 @@ jobs:
        include:
          - build-type: 'cublas'
            cuda-major-version: "12"
-            cuda-minor-version: "8"
+            cuda-minor-version: "0"
            platforms: 'linux/amd64'
            tag-latest: 'false'
            tag-suffix: '-gpu-nvidia-cuda-12'
@@ -47,7 +47,7 @@ jobs:
            platforms: 'linux/amd64'
            tag-latest: 'false'
            tag-suffix: '-hipblas'
-            base-image: "rocm/dev-ubuntu-22.04:6.4.3"
+            base-image: "rocm/dev-ubuntu-22.04:6.1"
            grpc-base-image: "ubuntu:22.04"
            runs-on: 'ubuntu-latest'
            makeflags: "--jobs=3 --output-sync=target"
--- a/.github/workflows/image.yml
+++ b/.github/workflows/image.yml
@@ -39,7 +39,7 @@ jobs:
            platforms: 'linux/amd64'
            tag-latest: 'auto'
            tag-suffix: '-gpu-hipblas'
-            base-image: "rocm/dev-ubuntu-22.04:6.4.3"
+            base-image: "rocm/dev-ubuntu-22.04:6.1"
            grpc-base-image: "ubuntu:22.04"
            runs-on: 'ubuntu-latest'
            makeflags: "--jobs=3 --output-sync=target"
@@ -91,7 +91,7 @@ jobs:
            aio: "-aio-gpu-nvidia-cuda-11"
          - build-type: 'cublas'
            cuda-major-version: "12"
-            cuda-minor-version: "8"
+            cuda-minor-version: "0"
            platforms: 'linux/amd64'
            tag-latest: 'auto'
            tag-suffix: '-gpu-nvidia-cuda-12'
@@ -144,7 +144,7 @@ jobs:
        include:
          - build-type: 'cublas'
            cuda-major-version: "12"
-            cuda-minor-version: "8"
+            cuda-minor-version: "0"
            platforms: 'linux/arm64'
            tag-latest: 'auto'
            tag-suffix: '-nvidia-l4t-arm64'
--- a/.github/workflows/image_build.yml
+++ b/.github/workflows/image_build.yml
@@ -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'
--- a/.github/workflows/labeler.yml
+++ b/.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
--- a/.github/workflows/localaibot_automerge.yml
+++ b/.github/workflows/localaibot_automerge.yml
@@ -13,7 +13,7 @@ jobs:
    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: |
--- a/.github/workflows/notify-models.yaml
+++ b/.github/workflows/notify-models.yaml
@@ -11,7 +11,7 @@ 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
@@ -90,7 +90,7 @@ 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
--- a/.github/workflows/release.yaml
+++ b/.github/workflows/release.yaml
@@ -10,7 +10,7 @@ jobs:
    runs-on: ubuntu-latest
    steps:
      - name: Checkout
-        uses: actions/checkout@v5
+        uses: actions/checkout@v4
        with:
          fetch-depth: 0
      - name: Set up Go
@@ -23,42 +23,4 @@ jobs:
          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
-        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
-        with:
-          files: ./local-ai-launcher-linux.tar.xz
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }} 
--- a/.github/workflows/secscan.yaml
+++ b/.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@v2.22.8
+        uses: securego/gosec@v2.22.7
        with:
          # we let the report trigger content trigger a failure using the GitHub Security features.
          args: '-no-fail -fmt sarif -out results.sarif ./...'
--- a/.github/workflows/stalebot.yml
+++ b/.github/workflows/stalebot.yml
@@ -10,7 +10,7 @@ jobs:
  stale:
    runs-on: ubuntu-latest
    steps:
-      - uses: actions/stale@3a9db7e6a41a89f618792c92c0e97cc736e1b13f # v9
+      - uses: actions/stale@5bef64f19d7facfb25b37b414482c7164d639639 # 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.'
--- a/.github/workflows/test-extra.yml
+++ b/.github/workflows/test-extra.yml
@@ -19,7 +19,7 @@ jobs:
  #   runs-on: ubuntu-latest
  #   steps:
  #     - name: Clone
-  #       uses: actions/checkout@v5
+  #       uses: actions/checkout@v4
  #       with:
  #         submodules: true
  #     - name: Dependencies
@@ -40,7 +40,7 @@ jobs:
    runs-on: ubuntu-latest
    steps:
      - name: Clone
-        uses: actions/checkout@v5
+        uses: actions/checkout@v4
        with:
          submodules: true
      - name: Dependencies
@@ -61,7 +61,7 @@ jobs:
    runs-on: ubuntu-latest
    steps:
      - name: Clone
-        uses: actions/checkout@v5
+        uses: actions/checkout@v4
        with:
          submodules: true
      - name: Dependencies
@@ -83,7 +83,7 @@ jobs:
    runs-on: ubuntu-latest
    steps:
      - name: Clone
-        uses: actions/checkout@v5
+        uses: actions/checkout@v4
        with:
          submodules: true
      - name: Dependencies
@@ -104,7 +104,7 @@ jobs:
  #  runs-on: ubuntu-latest
  #  steps:
  #    - name: Clone
-  #      uses: actions/checkout@v5
+  #      uses: actions/checkout@v4
  #      with:
  #        submodules: true
  #    - name: Dependencies
@@ -124,7 +124,7 @@ jobs:
  #   runs-on: ubuntu-latest
  #   steps:
  #     - name: Clone
-  #       uses: actions/checkout@v5
+  #       uses: actions/checkout@v4
  #       with:
  #         submodules: true
  #     - name: Dependencies
@@ -186,7 +186,7 @@ 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 +211,7 @@ jobs:
  #   runs-on: ubuntu-latest
  #   steps:
  #     - name: Clone
-  #       uses: actions/checkout@v5
+  #       uses: actions/checkout@v4
  #       with:
  #         submodules: true
  #     - name: Dependencies
@@ -232,7 +232,7 @@ jobs:
    runs-on: ubuntu-latest
    steps:
      - name: Clone
-        uses: actions/checkout@v5
+        uses: actions/checkout@v4
        with:
          submodules: true
      - name: Dependencies
--- a/.github/workflows/test.yml
+++ b/.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 }}
@@ -166,7 +152,7 @@ jobs:
          sudo rm -rfv build || true
          df -h
      - name: Clone
-        uses: actions/checkout@v5
+        uses: actions/checkout@v4
        with:
          submodules: true
      - name: Dependencies
@@ -196,7 +182,7 @@ jobs:
        go-version: ['1.21.x']
    steps:
      - name: Clone
-        uses: actions/checkout@v5
+        uses: actions/checkout@v4
        with:
          submodules: true
      - name: Setup Go ${{ matrix.go-version }}
@@ -214,7 +200,11 @@ jobs:
      - name: Build llama-cpp-darwin
        run: |
          make protogen-go
-          make backends/llama-cpp-darwin
+          make build
+          bash scripts/build-llama-cpp-darwin.sh
+          ls -la build/darwin.tar
+          mv build/darwin.tar build/llama-cpp.tar
+          ./local-ai backends install "ocifile://$PWD/build/llama-cpp.tar"
      - name: Test
        run: |
          export C_INCLUDE_PATH=/usr/local/include
--- a/.github/workflows/update_swagger.yaml
+++ b/.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'
--- a/.gitignore
+++ b/.gitignore
@@ -24,7 +24,7 @@ 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
--- a/.goreleaser.yaml
+++ b/.goreleaser.yaml
@@ -8,7 +8,7 @@ source:
  enabled: true
  name_template: '{{ .ProjectName }}-{{ .Tag }}-source'
 builds:
-  - main: ./cmd/local-ai
+  -
    env:
      - CGO_ENABLED=0
    ldflags:
--- a/8
+++ b/8
@@ -9,7 +9,7 @@ ENV DEBIAN_FRONTEND=noninteractive
 RUN apt-get update && \
    apt-get install -y --no-install-recommends \
        ca-certificates curl wget espeak-ng libgomp1 \
-        ffmpeg libopenblas-base libopenblas-dev && \
+        python3 python-is-python3 ffmpeg libopenblas-base libopenblas-dev && \
    apt-get clean && \
    rm -rf /var/lib/apt/lists/*

@@ -18,7 +18,7 @@ FROM requirements AS requirements-drivers

 ARG BUILD_TYPE
 ARG CUDA_MAJOR_VERSION=12
-ARG CUDA_MINOR_VERSION=8
+ARG CUDA_MINOR_VERSION=0
 ARG SKIP_DRIVERS=false
 ARG TARGETARCH
 ARG TARGETVARIANT
@@ -100,10 +100,6 @@ RUN if [ "${BUILD_TYPE}" = "hipblas" ] && [ "${SKIP_DRIVERS}" = "false" ]; then
        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
--- a/133
+++ b/133
@@ -2,7 +2,6 @@ GOCMD=go
 GOTEST=$(GOCMD) test
 GOVET=$(GOCMD) vet
 BINARY_NAME=local-ai
-LAUNCHER_BINARY_NAME=local-ai-launcher

 GORELEASER?=

@@ -91,17 +90,7 @@ build: protogen-go install-go-tools ## Build the project
 	$(info ${GREEN}I LD_FLAGS: ${YELLOW}$(LD_FLAGS)${RESET})
 	$(info ${GREEN}I UPX: ${YELLOW}$(UPX)${RESET})
 	rm -rf $(BINARY_NAME) || true
-	CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o $(BINARY_NAME) ./cmd/local-ai
-
-build-launcher: ## Build the launcher application
-	$(info ${GREEN}I local-ai launcher build info:${RESET})
-	$(info ${GREEN}I BUILD_TYPE: ${YELLOW}$(BUILD_TYPE)${RESET})
-	$(info ${GREEN}I GO_TAGS: ${YELLOW}$(GO_TAGS)${RESET})
-	$(info ${GREEN}I LD_FLAGS: ${YELLOW}$(LD_FLAGS)${RESET})
-	rm -rf $(LAUNCHER_BINARY_NAME) || true
-	CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o $(LAUNCHER_BINARY_NAME) ./cmd/launcher
-
-build-all: build build-launcher ## Build both server and launcher
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o $(BINARY_NAME) ./

 dev-dist:
 	$(GORELEASER) build --snapshot --clean
@@ -143,6 +132,39 @@ test: test-models/testmodel.ggml protogen-go
 	$(MAKE) test-tts
 	$(MAKE) test-stablediffusion

+backends/diffusers: docker-build-diffusers docker-save-diffusers build
+	./local-ai backends install "ocifile://$(abspath ./backend-images/diffusers.tar)"
+
+backends/llama-cpp: docker-build-llama-cpp docker-save-llama-cpp build
+	./local-ai backends install "ocifile://$(abspath ./backend-images/llama-cpp.tar)"
+
+backends/piper: docker-build-piper docker-save-piper build
+	./local-ai backends install "ocifile://$(abspath ./backend-images/piper.tar)"
+
+backends/stablediffusion-ggml: docker-build-stablediffusion-ggml docker-save-stablediffusion-ggml build
+	./local-ai backends install "ocifile://$(abspath ./backend-images/stablediffusion-ggml.tar)"
+
+backends/whisper: docker-build-whisper docker-save-whisper build
+	./local-ai backends install "ocifile://$(abspath ./backend-images/whisper.tar)"
+
+backends/silero-vad: docker-build-silero-vad docker-save-silero-vad build
+	./local-ai backends install "ocifile://$(abspath ./backend-images/silero-vad.tar)"
+
+backends/local-store: docker-build-local-store docker-save-local-store build
+	./local-ai backends install "ocifile://$(abspath ./backend-images/local-store.tar)"
+
+backends/huggingface: docker-build-huggingface docker-save-huggingface build
+	./local-ai backends install "ocifile://$(abspath ./backend-images/huggingface.tar)"
+
+backends/rfdetr: docker-build-rfdetr docker-save-rfdetr build
+	./local-ai backends install "ocifile://$(abspath ./backend-images/rfdetr.tar)"
+
+backends/kitten-tts: docker-build-kitten-tts docker-save-kitten-tts build
+	./local-ai backends install "ocifile://$(abspath ./backend-images/kitten-tts.tar)"
+
+backends/kokoro: docker-build-kokoro docker-save-kokoro build
+	./local-ai backends install "ocifile://$(abspath ./backend-images/kokoro.tar)"
+
 ########################################################
 ## AIO tests
 ########################################################
@@ -170,7 +192,7 @@ prepare-e2e:
 	mkdir -p $(TEST_DIR)
 	cp -rfv $(abspath ./tests/e2e-fixtures)/gpu.yaml $(TEST_DIR)/gpu.yaml
 	test -e $(TEST_DIR)/ggllm-test-model.bin || wget -q https://huggingface.co/TheBloke/CodeLlama-7B-Instruct-GGUF/resolve/main/codellama-7b-instruct.Q2_K.gguf -O $(TEST_DIR)/ggllm-test-model.bin
-	docker build --build-arg IMAGE_TYPE=core --build-arg BUILD_TYPE=$(BUILD_TYPE) --build-arg CUDA_MAJOR_VERSION=12 --build-arg CUDA_MINOR_VERSION=8 -t localai-tests .
+	docker build --build-arg IMAGE_TYPE=core --build-arg BUILD_TYPE=$(BUILD_TYPE) --build-arg CUDA_MAJOR_VERSION=12 --build-arg CUDA_MINOR_VERSION=0 -t localai-tests .

 run-e2e-image:
 	ls -liah $(abspath ./tests/e2e-fixtures)
@@ -335,73 +357,6 @@ docker-image-intel:
 ## Backends
 ########################################################

-
-backends/diffusers: docker-build-diffusers docker-save-diffusers build
-	./local-ai backends install "ocifile://$(abspath ./backend-images/diffusers.tar)"
-
-backends/llama-cpp: docker-build-llama-cpp docker-save-llama-cpp build
-	./local-ai backends install "ocifile://$(abspath ./backend-images/llama-cpp.tar)"
-
-backends/piper: docker-build-piper docker-save-piper build
-	./local-ai backends install "ocifile://$(abspath ./backend-images/piper.tar)"
-
-backends/stablediffusion-ggml: docker-build-stablediffusion-ggml docker-save-stablediffusion-ggml build
-	./local-ai backends install "ocifile://$(abspath ./backend-images/stablediffusion-ggml.tar)"
-
-backends/whisper: docker-build-whisper docker-save-whisper build
-	./local-ai backends install "ocifile://$(abspath ./backend-images/whisper.tar)"
-
-backends/silero-vad: docker-build-silero-vad docker-save-silero-vad build
-	./local-ai backends install "ocifile://$(abspath ./backend-images/silero-vad.tar)"
-
-backends/local-store: docker-build-local-store docker-save-local-store build
-	./local-ai backends install "ocifile://$(abspath ./backend-images/local-store.tar)"
-
-backends/huggingface: docker-build-huggingface docker-save-huggingface build
-	./local-ai backends install "ocifile://$(abspath ./backend-images/huggingface.tar)"
-
-backends/rfdetr: docker-build-rfdetr docker-save-rfdetr build
-	./local-ai backends install "ocifile://$(abspath ./backend-images/rfdetr.tar)"
-
-backends/kitten-tts: docker-build-kitten-tts docker-save-kitten-tts build
-	./local-ai backends install "ocifile://$(abspath ./backend-images/kitten-tts.tar)"
-
-backends/kokoro: docker-build-kokoro docker-save-kokoro build
-	./local-ai backends install "ocifile://$(abspath ./backend-images/kokoro.tar)"
-
-backends/chatterbox: docker-build-chatterbox docker-save-chatterbox build
-	./local-ai backends install "ocifile://$(abspath ./backend-images/chatterbox.tar)"
-
-backends/llama-cpp-darwin: build
-	bash ./scripts/build/llama-cpp-darwin.sh
-	./local-ai backends install "ocifile://$(abspath ./backend-images/llama-cpp.tar)"
-
-build-darwin-python-backend: build
-	bash ./scripts/build/python-darwin.sh
-
-build-darwin-go-backend: build
-	bash ./scripts/build/golang-darwin.sh
-
-backends/mlx:
-	BACKEND=mlx $(MAKE) build-darwin-python-backend
-	./local-ai backends install "ocifile://$(abspath ./backend-images/mlx.tar)"
-
-backends/diffuser-darwin:
-	BACKEND=diffusers $(MAKE) build-darwin-python-backend
-	./local-ai backends install "ocifile://$(abspath ./backend-images/diffusers.tar)"
-
-backends/mlx-vlm:
-	BACKEND=mlx-vlm $(MAKE) build-darwin-python-backend
-	./local-ai backends install "ocifile://$(abspath ./backend-images/mlx-vlm.tar)"
-
-backends/mlx-audio:
-	BACKEND=mlx-audio $(MAKE) build-darwin-python-backend
-	./local-ai backends install "ocifile://$(abspath ./backend-images/mlx-audio.tar)"
-
-backends/stablediffusion-ggml-darwin:
-	BACKEND=stablediffusion-ggml BUILD_TYPE=metal $(MAKE) build-darwin-go-backend
-	./local-ai backends install "ocifile://$(abspath ./backend-images/stablediffusion-ggml.tar)"
-
 backend-images:
 	mkdir -p backend-images

@@ -496,7 +451,7 @@ docker-build-bark:
 	docker build --build-arg BUILD_TYPE=$(BUILD_TYPE) --build-arg BASE_IMAGE=$(BASE_IMAGE) -t local-ai-backend:bark -f backend/Dockerfile.python --build-arg BACKEND=bark .

 docker-build-chatterbox:
-	docker build --build-arg BUILD_TYPE=$(BUILD_TYPE) --build-arg BASE_IMAGE=$(BASE_IMAGE) -t local-ai-backend:chatterbox -f backend/Dockerfile.python --build-arg BACKEND=chatterbox ./backend
+	docker build --build-arg BUILD_TYPE=$(BUILD_TYPE) --build-arg BASE_IMAGE=$(BASE_IMAGE) -t local-ai-backend:chatterbox -f backend/Dockerfile.python --build-arg BACKEND=chatterbox .

 docker-build-exllama2:
 	docker build --build-arg BUILD_TYPE=$(BUILD_TYPE) --build-arg BASE_IMAGE=$(BASE_IMAGE) -t local-ai-backend:exllama2 -f backend/Dockerfile.python --build-arg BACKEND=exllama2 .
@@ -532,19 +487,3 @@ docs-clean:
 .PHONY: docs
 docs: docs/static/gallery.html
 	cd docs && hugo serve
-
-########################################################
-## Platform-specific builds
-########################################################
-
-## fyne cross-platform build
-build-launcher-darwin: build-launcher
-	go run github.com/tiagomelo/macos-dmg-creator/cmd/createdmg@latest \
-	--appName "LocalAI" \
-	--appBinaryPath "$(LAUNCHER_BINARY_NAME)" \
-	--bundleIdentifier "com.localai.launcher" \
-	--iconPath "core/http/static/logo.png" \
-	--outputDir "dist/"
-
-build-launcher-linux:
-	cd cmd/launcher && go run fyne.io/tools/cmd/fyne@latest package -os linux -icon ../../core/http/static/logo.png --executable $(LAUNCHER_BINARY_NAME)-linux && mv launcher.tar.xz ../../$(LAUNCHER_BINARY_NAME)-linux.tar.xz
--- a/README.md
+++ b/README.md
@@ -110,12 +110,6 @@ 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:
@@ -197,7 +191,6 @@ For more information, see [💻 Getting started](https://localai.io/basics/getti

 ## 📰 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).
@@ -239,60 +232,6 @@ Roadmap items: [List of issues](https://github.com/mudler/LocalAI/issues?q=is%3A
 - 🔊 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 +246,6 @@ 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
--- a/backend/Dockerfile.python
+++ b/backend/Dockerfile.python
@@ -23,7 +23,7 @@ RUN apt-get update && \
        libssl-dev \
        git \
        git-lfs \
-        unzip clang \
+        unzip \
        upx-ucl \
        curl python3-pip \
        python-is-python3 \
@@ -116,7 +116,7 @@ COPY python/${BACKEND} /${BACKEND}
 COPY backend.proto /${BACKEND}/backend.proto
 COPY python/common/ /${BACKEND}/common

-RUN cd /${BACKEND} && PORTABLE_PYTHON=true make
+RUN cd /${BACKEND} && make

 FROM scratch
 ARG BACKEND=rerankers
--- a/backend/README.md
+++ b/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=8 \
-  -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
--- a/backend/backend.proto
+++ b/backend/backend.proto
@@ -242,7 +242,7 @@ message ModelOptions {

  string Type = 49;

-  string FlashAttention = 56;
+  bool FlashAttention = 56;
  bool NoKVOffload = 57;

  string ModelPath = 59;
@@ -276,7 +276,6 @@ message TranscriptRequest {
  string language = 3;
  uint32 threads = 4;
  bool translate = 5;
-  bool diarize = 6;
 }

 message TranscriptResult {
@@ -306,24 +305,22 @@ 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 {
--- a/backend/cpp/llama-cpp/Makefile
+++ b/backend/cpp/llama-cpp/Makefile
@@ -1,6 +1,6 @@

-LLAMA_VERSION?=fe4eb4f8ec25a1239b0923f1c7f87adf5730c3e5
-LLAMA_REPO?=https://github.com/JohannesGaessler/llama.cpp
+LLAMA_VERSION?=a0552c8beef74e843bb085c8ef0c63f9ed7a2b27
+LLAMA_REPO?=https://github.com/ggerganov/llama.cpp

 CMAKE_ARGS?=
 BUILD_TYPE?=
@@ -32,8 +32,10 @@ else ifeq ($(BUILD_TYPE),hipblas)
 	ROCM_PATH ?= /opt/rocm
 	export CXX=$(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)
+#	GPU_TARGETS ?= gfx803,gfx900,gfx906,gfx908,gfx90a,gfx942,gfx1010,gfx1030,gfx1032,gfx1100,gfx1101,gfx1102
+#	AMDGPU_TARGETS ?= "$(GPU_TARGETS)"
+	CMAKE_ARGS+=-DGGML_HIP=ON
+#	CMAKE_ARGS+=-DGGML_HIP=ON -DAMDGPU_TARGETS="$(AMDGPU_TARGETS)" -DGPU_TARGETS="$(GPU_TARGETS)"
 else ifeq ($(BUILD_TYPE),vulkan)
 	CMAKE_ARGS+=-DGGML_VULKAN=1
 else ifeq ($(OS),Darwin)
--- a/backend/cpp/llama-cpp/grpc-server.cpp
+++ b/backend/cpp/llama-cpp/grpc-server.cpp
@@ -53,9 +53,9 @@ static void start_llama_server(server_context& ctx_server) {
    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);
+    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());

    // Reset the chat templates
    // TODO: We should make this configurable by respecting the option that is already present in LocalAI for vLLM
@@ -304,15 +304,7 @@ static void params_parse(const backend::ModelOptions* request,
    }
    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.flash_attn = request->flashattention();
    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)

@@ -445,7 +437,24 @@ public:
                }
            }

+            // process files
+            mtmd::bitmaps bitmaps;
            const bool has_mtmd = ctx_server.mctx != nullptr;
+            {
+                if (!has_mtmd && !files.empty()) {
+                    throw std::runtime_error("This server does not support multimodal");
+                }
+                for (auto & file : files) {
+                    mtmd::bitmap bmp(mtmd_helper_bitmap_init_from_buf(ctx_server.mctx, file.data(), file.size()));
+                    if (!bmp.ptr) {
+                        throw std::runtime_error("Failed to load image/audio");
+                    }
+                    // calculate bitmap hash (for KV caching)
+                    std::string hash = fnv_hash(bmp.data(), bmp.n_bytes());
+                    bmp.set_id(hash.c_str());
+                    bitmaps.entries.push_back(std::move(bmp));
+                }
+            }

            // process prompt
            std::vector<server_tokens> inputs;
@@ -455,10 +464,32 @@ public:

            if (has_mtmd) {
                // multimodal
-                inputs.push_back(process_mtmd_prompt(ctx_server.mctx, prompt.get<std::string>(), files));
+                std::string prompt_str = prompt.get<std::string>();
+                mtmd_input_text inp_txt = {
+                    prompt_str.c_str(),
+                    /* add_special */   true,
+                    /* parse_special */ true,
+                };
+                mtmd::input_chunks chunks(mtmd_input_chunks_init());
+                auto bitmaps_c_ptr = bitmaps.c_ptr();
+                int32_t tokenized = mtmd_tokenize(ctx_server.mctx,
+                                                    chunks.ptr.get(),
+                                                    &inp_txt,
+                                                    bitmaps_c_ptr.data(),
+                                                    bitmaps_c_ptr.size());
+                if (tokenized != 0) {
+                    throw std::runtime_error("Failed to tokenize prompt");
+                }
+
+                server_tokens tmp(chunks, true);
+                inputs.push_back(std::move(tmp));
            } else {
-                 // Everything else, including multimodal completions.
-                inputs = tokenize_input_prompts(ctx_server.vocab, ctx_server.mctx, prompt, true, true);
+                // non-multimodal version
+                auto tokenized_prompts = tokenize_input_prompts(ctx_server.vocab, prompt, true, true);
+                for (auto & p : tokenized_prompts) {
+                    auto tmp = server_tokens(p, ctx_server.mctx != nullptr);
+                    inputs.push_back(std::move(tmp));
+                }
            }

            tasks.reserve(inputs.size());
@@ -599,7 +630,23 @@ public:
            }

            // process files
+            mtmd::bitmaps bitmaps;
            const bool has_mtmd = ctx_server.mctx != nullptr;
+            {
+                if (!has_mtmd && !files.empty()) {
+                    throw std::runtime_error("This server does not support multimodal");
+                }
+                for (auto & file : files) {
+                    mtmd::bitmap bmp(mtmd_helper_bitmap_init_from_buf(ctx_server.mctx, file.data(), file.size()));
+                    if (!bmp.ptr) {
+                        throw std::runtime_error("Failed to load image/audio");
+                    }
+                    // calculate bitmap hash (for KV caching)
+                    std::string hash = fnv_hash(bmp.data(), bmp.n_bytes());
+                    bmp.set_id(hash.c_str());
+                    bitmaps.entries.push_back(std::move(bmp));
+                }
+            }

            // process prompt
            std::vector<server_tokens> inputs;
@@ -610,10 +657,33 @@ public:

            if (has_mtmd) {
                // multimodal
-                inputs.push_back(process_mtmd_prompt(ctx_server.mctx, prompt.get<std::string>(), files));
+                std::string prompt_str = prompt.get<std::string>();
+                mtmd_input_text inp_txt = {
+                    prompt_str.c_str(),
+                    /* add_special */   true,
+                    /* parse_special */ true,
+                };
+                mtmd::input_chunks chunks(mtmd_input_chunks_init());
+                auto bitmaps_c_ptr = bitmaps.c_ptr();
+                int32_t tokenized = mtmd_tokenize(ctx_server.mctx,
+                                                    chunks.ptr.get(),
+                                                    &inp_txt,
+                                                    bitmaps_c_ptr.data(),
+                                                    bitmaps_c_ptr.size());
+                if (tokenized != 0) {
+                    std::cout << "[PREDICT] Failed to tokenize prompt" << std::endl;
+                    throw std::runtime_error("Failed to tokenize prompt");
+                }
+
+                server_tokens tmp(chunks, true);
+                inputs.push_back(std::move(tmp));
            } else {
-                 // Everything else, including multimodal completions.
-                inputs = tokenize_input_prompts(ctx_server.vocab, ctx_server.mctx, prompt, true, true);
+                // non-multimodal version
+                auto tokenized_prompts = tokenize_input_prompts(ctx_server.vocab, prompt, true, true);
+                for (auto & p : tokenized_prompts) {
+                    auto tmp = server_tokens(p, ctx_server.mctx != nullptr);
+                    inputs.push_back(std::move(tmp));
+                }
            }

            tasks.reserve(inputs.size());
@@ -704,7 +774,7 @@ public:
        json prompt = body.at("prompt");


-        auto tokenized_prompts = tokenize_input_prompts(ctx_server.vocab, ctx_server.mctx, prompt, true, true);
+        auto tokenized_prompts = tokenize_input_prompts(ctx_server.vocab, 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()) {
@@ -723,7 +793,7 @@ public:

                task.id            = ctx_server.queue_tasks.get_new_id();
                task.index         = i;
-                task.prompt_tokens = std::move(tokenized_prompts[i]);
+                task.prompt_tokens = server_tokens(tokenized_prompts[i], ctx_server.mctx != nullptr);

                // OAI-compat
                task.params.oaicompat = OAICOMPAT_TYPE_EMBEDDING;
@@ -779,10 +849,8 @@ public:
        }

        // Tokenize the query
-        auto tokenized_query = tokenize_input_prompts(ctx_server.vocab, ctx_server.mctx, request->query(), /* add_special */ false, true);
-        if (tokenized_query.size() != 1) {
-            return grpc::Status(grpc::StatusCode::INVALID_ARGUMENT, "\"query\" must contain only a single prompt");
-        }
+        llama_tokens tokenized_query = tokenize_input_prompts(ctx_server.vocab, request->query(), /* add_special */ false, true)[0];
+
        // Create and queue the task
        json responses = json::array();
        bool error = false;
@@ -794,14 +862,14 @@ public:
                documents.push_back(request->documents(i));
            }
            
-            auto tokenized_docs = tokenize_input_prompts(ctx_server.vocab, ctx_server.mctx, documents, /* add_special */ false, true);
+            auto tokenized_docs = tokenize_input_prompts(ctx_server.vocab, documents, /* add_special */ false, true);
            tasks.reserve(tokenized_docs.size());
            for (size_t i = 0; i < tokenized_docs.size(); i++) {
-                auto tmp = format_rerank(ctx_server.vocab, tokenized_query[0], tokenized_docs[i]);
+                auto tmp = format_rerank(ctx_server.vocab, tokenized_query, tokenized_docs[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);
+                task.prompt_tokens = server_tokens(tmp, ctx_server.mctx != nullptr);
                tasks.push_back(std::move(task));
            }

--- a/backend/cpp/llama-cpp/run.sh
+++ b/backend/cpp/llama-cpp/run.sh
@@ -42,8 +42,7 @@ 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:$DYLD_FALLBACK_LIBRARY_PATH
+	DYLD_FALLBACK_LIBRARY_PATH=$CURDIR/lib:$DYLD_FALLBACK_LIBRARY_PATH
 else
 	export LD_LIBRARY_PATH=$CURDIR/lib:$LD_LIBRARY_PATH
 fi
@@ -58,5 +57,5 @@ fi
 echo "Using binary: $BINARY"
 exec $CURDIR/$BINARY "$@"

-# We should never reach this point, however just in case we do, run fallback
+# In case we fail execing, just run fallback
 exec $CURDIR/llama-cpp-fallback "$@"
--- a/backend/go/stablediffusion-ggml/.gitignore
+++ b/backend/go/stablediffusion-ggml/.gitignore
@@ -1,6 +0,0 @@
-package/
-sources/
-.cache/
-build/
-libgosd.so
-stablediffusion-ggml
--- a/backend/go/stablediffusion-ggml/CMakeLists.txt
+++ b/backend/go/stablediffusion-ggml/CMakeLists.txt
@@ -1,20 +0,0 @@
-cmake_minimum_required(VERSION 3.12)
-project(gosd LANGUAGES C CXX)
-set(CMAKE_POSITION_INDEPENDENT_CODE ON)
-
-add_subdirectory(./sources/stablediffusion-ggml.cpp)
-
-add_library(gosd MODULE gosd.cpp)
-target_link_libraries(gosd PRIVATE stable-diffusion ggml)
-
-if(CMAKE_CXX_COMPILER_ID MATCHES "GNU" AND CMAKE_CXX_COMPILER_VERSION VERSION_LESS 9.0)
-    target_link_libraries(gosd PRIVATE stdc++fs)
-endif()
-
-target_include_directories(gosd PUBLIC
-    stable-diffusion.cpp
-    stable-diffusion.cpp/thirdparty
-)
-
-set_property(TARGET gosd PROPERTY CXX_STANDARD 17)
-set_target_properties(gosd PROPERTIES LIBRARY_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR})
--- a/backend/go/stablediffusion-ggml/Makefile
+++ b/backend/go/stablediffusion-ggml/Makefile
@@ -1,16 +1,28 @@
+INCLUDE_PATH := $(abspath ./)
+LIBRARY_PATH := $(abspath ./)
+
+AR?=ar
 CMAKE_ARGS?=
 BUILD_TYPE?=
 NATIVE?=false
+CUDA_LIBPATH?=/usr/local/cuda/lib64/
+ONEAPI_VARS?=/opt/intel/oneapi/setvars.sh
+# keep standard at C11 and C++11
+CXXFLAGS = -I. -I$(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?=
-JOBS?=$(shell nproc --ignore=1)
+LD_FLAGS?=

 # stablediffusion.cpp (ggml)
 STABLEDIFFUSION_GGML_REPO?=https://github.com/leejet/stable-diffusion.cpp
-STABLEDIFFUSION_GGML_VERSION?=b0179181069254389ccad604e44f17a2c25b4094
+STABLEDIFFUSION_GGML_VERSION?=5900ef6605c6fbf7934239f795c13c97bc993853

-CMAKE_ARGS+=-DGGML_MAX_NAME=128
+# Disable Shared libs as we are linking on static gRPC and we can't mix shared and static
+CMAKE_ARGS+=-DBUILD_SHARED_LIBS=OFF -DGGML_MAX_NAME=128 -DSD_USE_SYSTEM_GGML=OFF

 ifeq ($(NATIVE),false)
 	CMAKE_ARGS+=-DGGML_NATIVE=OFF
@@ -19,6 +31,7 @@ endif
 # If build type is cublas, then we set -DGGML_CUDA=ON to CMAKE_ARGS automatically
 ifeq ($(BUILD_TYPE),cublas)
 	CMAKE_ARGS+=-DSD_CUDA=ON -DGGML_CUDA=ON
+	CGO_LDFLAGS+=-lcublas -lcudart -L$(CUDA_LIBPATH) -L$(CUDA_LIBPATH)/stubs/ -lcuda
 # If build type is openblas then we set -DGGML_BLAS=ON -DGGML_BLAS_VENDOR=OpenBLAS
 # to CMAKE_ARGS automatically
 else ifeq ($(BUILD_TYPE),openblas)
@@ -29,14 +42,18 @@ else ifeq ($(BUILD_TYPE),clblas)
 # If it's hipblas we do have also to set CC=/opt/rocm/llvm/bin/clang CXX=/opt/rocm/llvm/bin/clang++
 else ifeq ($(BUILD_TYPE),hipblas)
 	CMAKE_ARGS+=-DSD_HIPBLAS=ON -DGGML_HIPBLAS=ON
+# If it's OSX, DO NOT embed the metal library - -DGGML_METAL_EMBED_LIBRARY=ON requires further investigation
+# But if it's OSX without metal, disable it here
 else ifeq ($(BUILD_TYPE),vulkan)
 	CMAKE_ARGS+=-DSD_VULKAN=ON -DGGML_VULKAN=ON
+	CGO_LDFLAGS+=-lvulkan
 else ifeq ($(OS),Darwin)
 	ifneq ($(BUILD_TYPE),metal)
 		CMAKE_ARGS+=-DSD_METAL=OFF -DGGML_METAL=OFF
 	else
 		CMAKE_ARGS+=-DSD_METAL=ON -DGGML_METAL=ON
 		CMAKE_ARGS+=-DGGML_METAL_EMBED_LIBRARY=ON
+		TARGET+=--target ggml-metal
 	endif
 endif

@@ -46,6 +63,12 @@ ifeq ($(BUILD_TYPE),sycl_f16)
 		-DCMAKE_CXX_COMPILER=icpx \
 		-DSD_SYCL=ON \
 		-DGGML_SYCL_F16=ON
+	export CC=icx
+	export CXX=icpx
+	CGO_LDFLAGS_SYCL += -fsycl -L${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 ($(BUILD_TYPE),sycl_f32)
@@ -53,29 +76,78 @@ ifeq ($(BUILD_TYPE),sycl_f32)
 		-DCMAKE_C_COMPILER=icx \
 		-DCMAKE_CXX_COMPILER=icpx \
 		-DSD_SYCL=ON
+	export CC=icx
+	export 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')
+ALL_OBJS := $(shell find $(GGML_ARCHIVE_DIR) -type f -name '*.o')
+
+# Name of the single merged library
+COMBINED_LIB := libggmlall.a
+
+# Instead of using the archives generated by GGML, use the object files directly to avoid overwriting objects with the same base name
+$(COMBINED_LIB): $(ALL_ARCHIVES)
+	@echo "Merging all .o into $(COMBINED_LIB): $(ALL_OBJS)"
+	rm -f $@
+	ar -qc $@ $(ALL_OBJS)
+	# Ensure we have a proper index
+	ranlib $@
+
+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
+
+## stablediffusion (ggml)
 sources/stablediffusion-ggml.cpp:
 	git clone --recursive $(STABLEDIFFUSION_GGML_REPO) sources/stablediffusion-ggml.cpp && \
 	cd sources/stablediffusion-ggml.cpp && \
 	git checkout $(STABLEDIFFUSION_GGML_VERSION) && \
 	git submodule update --init --recursive --depth 1 --single-branch

-libgosd.so: sources/stablediffusion-ggml.cpp CMakeLists.txt gosd.cpp gosd.h
-	mkdir -p build && \
-	cd build && \
-	cmake .. $(CMAKE_ARGS) && \
-	cmake --build . --config Release -j$(JOBS) && \
-	cd .. && \
-	mv build/libgosd.so ./
+libsd.a: sources/stablediffusion-ggml.cpp build/libstable-diffusion.a gosd.o
+	cp $(INCLUDE_PATH)/build/libstable-diffusion.a ./libsd.a
+	$(AR) rcs libsd.a gosd.o

-stablediffusion-ggml: main.go gosd.go libgosd.so
-	CGO_ENABLED=0 $(GOCMD) build -tags "$(GO_TAGS)" -o stablediffusion-ggml ./
+stablediffusion-ggml: libsd.a
+	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 stablediffusion-ggml ./

-package: stablediffusion-ggml
+package:
 	bash package.sh

-build: package
+build: stablediffusion-ggml package

 clean:
-	rm -rf libgosd.so build stablediffusion-ggml package sources
+	rm -rf gosd.o libsd.a build $(COMBINED_LIB)
--- a/backend/go/stablediffusion-ggml/gosd.cpp
+++ b/backend/go/stablediffusion-ggml/gosd.cpp
@@ -1,4 +1,3 @@
-#include <cstdint>
 #define GGML_MAX_NAME 128

 #include <stdio.h>
@@ -44,7 +43,7 @@ const char* sample_method_str[] = {
 };

 // Names of the sigma schedule overrides, same order as sample_schedule in stable-diffusion.h
-const char* schedulers[] = {
+const char* schedule_str[] = {
    "default",
    "discrete",
    "karras",
@@ -54,13 +53,11 @@ const char* schedulers[] = {
 };

 sd_ctx_t* sd_c;
-// Moved from the context (load time) to generation time params
-scheduler_t scheduler = scheduler_t::DEFAULT;

 sample_method_t sample_method;

 // Copied from the upstream CLI
-static void sd_log_cb(enum sd_log_level_t level, const char* log, void* data) {
+void sd_log_cb(enum sd_log_level_t level, const char* log, void* data) {
    //SDParams* params = (SDParams*)data;
    const char* level_str;

@@ -91,33 +88,33 @@ static void sd_log_cb(enum sd_log_level_t level, const char* log, void* data) {
    fflush(stderr);
 }

-int load_model(const char *model, char *model_path, char* options[], int threads, int diff) {
-    fprintf (stderr, "Loading model: %p=%s\n", model, model);
+int load_model(char *model, char *model_path, char* options[], int threads, int diff) {
+    fprintf (stderr, "Loading model!\n");

    sd_set_log_callback(sd_log_cb, NULL);

-    const char *stableDiffusionModel = "";
+    char *stableDiffusionModel = "";
    if (diff == 1 ) {
        stableDiffusionModel = model;
        model = "";
    }

    // decode options. Options are in form optname:optvale, or if booleans only optname.
-    const char *clip_l_path  = "";
-    const char *clip_g_path  = "";
-    const char *t5xxl_path  = "";
-    const char *vae_path  = "";
-    const char *scheduler_str = "";
-    const char *sampler = "";
+    char *clip_l_path  = "";
+    char *clip_g_path  = "";
+    char *t5xxl_path  = "";
+    char *vae_path  = "";
+    char *scheduler = "";
+    char *sampler = "";
    char *lora_dir = model_path;
    bool lora_dir_allocated = false;

-    fprintf(stderr, "parsing options: %p\n", options);
+    fprintf(stderr, "parsing options\n");

    // If options is not NULL, parse options
    for (int i = 0; options[i] != NULL; i++) {
-        const char *optname = strtok(options[i], ":");
-        const char *optval = strtok(NULL, ":");
+        char *optname = strtok(options[i], ":");
+        char *optval = strtok(NULL, ":");
        if (optval == NULL) {
            optval = "true";
        }
@@ -135,7 +132,7 @@ int load_model(const char *model, char *model_path, char* options[], int threads
            vae_path = optval;
        }
        if (!strcmp(optname, "scheduler")) {
-            scheduler_str = optval;
+            scheduler = optval;
        }
        if (!strcmp(optname, "sampler")) {
            sampler = optval;
@@ -150,8 +147,7 @@ int load_model(const char *model, char *model_path, char* options[], int threads
                lora_dir_allocated = true;
                fprintf(stderr, "Lora dir resolved to: %s\n", lora_dir);
            } else {
-                lora_dir = strdup(optval);
-                lora_dir_allocated = true;
+                lora_dir = optval;
                fprintf(stderr, "No model path provided, using lora dir as-is: %s\n", lora_dir);
            }
        }
@@ -172,13 +168,22 @@ int load_model(const char *model, char *model_path, char* options[], int threads
    }
    sample_method = (sample_method_t)sample_method_found;

+    int schedule_found            = -1;
    for (int d = 0; d < SCHEDULE_COUNT; d++) {
-        if (!strcmp(scheduler_str, schedulers[d])) {
-            scheduler = (scheduler_t)d;
-            fprintf (stderr, "Found scheduler: %s\n", scheduler_str);
+        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_params_t ctx_params;
    sd_ctx_params_init(&ctx_params);
@@ -198,6 +203,7 @@ int load_model(const char *model, char *model_path, char* options[], int threads
    ctx_params.free_params_immediately = false;
    ctx_params.n_threads = threads;
    ctx_params.rng_type = STD_DEFAULT_RNG;
+    ctx_params.schedule = schedule;
    sd_ctx_t* sd_ctx = new_sd_ctx(&ctx_params);

    if (sd_ctx == NULL) {
@@ -220,7 +226,7 @@ int load_model(const char *model, char *model_path, char* options[], int threads
    return 0;
 }

-int gen_image(char *text, char *negativeText, int width, int height, int steps, int64_t seed, char *dst, float cfg_scale, char *src_image, float strength, char *mask_image, char **ref_images, int ref_images_count) {
+int gen_image(char *text, char *negativeText, int width, int height, int steps, int seed , char *dst, float cfg_scale, char *src_image, float strength, char *mask_image, char **ref_images, int ref_images_count) {

    sd_image_t* results;

@@ -233,28 +239,27 @@ int gen_image(char *text, char *negativeText, int width, int height, int steps,

    p.prompt = text;
    p.negative_prompt = negativeText;
-    p.sample_params.guidance.txt_cfg = cfg_scale;
-    p.sample_params.guidance.slg.layers = skip_layers.data();
-    p.sample_params.guidance.slg.layer_count = skip_layers.size();
+    p.guidance.txt_cfg = cfg_scale;
+    p.guidance.slg.layers = skip_layers.data();
+    p.guidance.slg.layer_count = skip_layers.size();
    p.width = width;
    p.height = height;
-    p.sample_params.sample_method = sample_method;
-    p.sample_params.sample_steps = steps;
+    p.sample_method = sample_method;
+    p.sample_steps = steps;
    p.seed = seed;
    p.input_id_images_path = "";
-    p.sample_params.scheduler = scheduler;

    // Handle input image for img2img
    bool has_input_image = (src_image != NULL && strlen(src_image) > 0);
    bool has_mask_image = (mask_image != NULL && strlen(mask_image) > 0);
-
+    
    uint8_t* input_image_buffer = NULL;
    uint8_t* mask_image_buffer = NULL;
    std::vector<uint8_t> default_mask_image_vec;
-
+    
    if (has_input_image) {
        fprintf(stderr, "Loading input image: %s\n", src_image);
-
+        
        int c = 0;
        int img_width = 0;
        int img_height = 0;
@@ -268,29 +273,29 @@ int gen_image(char *text, char *negativeText, int width, int height, int steps,
            free(input_image_buffer);
            return 1;
        }
-
+        
        // Resize input image if dimensions don't match
        if (img_width != width || img_height != height) {
            fprintf(stderr, "Resizing input image from %dx%d to %dx%d\n", img_width, img_height, width, height);
-
+            
            uint8_t* resized_image_buffer = (uint8_t*)malloc(height * width * 3);
            if (resized_image_buffer == NULL) {
                fprintf(stderr, "Failed to allocate memory for resized image\n");
                free(input_image_buffer);
                return 1;
            }
-
+            
            stbir_resize(input_image_buffer, img_width, img_height, 0,
                         resized_image_buffer, width, height, 0, STBIR_TYPE_UINT8,
                         3, STBIR_ALPHA_CHANNEL_NONE, 0,
                         STBIR_EDGE_CLAMP, STBIR_EDGE_CLAMP,
                         STBIR_FILTER_BOX, STBIR_FILTER_BOX,
                         STBIR_COLORSPACE_SRGB, nullptr);
-
+            
            free(input_image_buffer);
            input_image_buffer = resized_image_buffer;
        }
-
+        
        p.init_image = {(uint32_t)width, (uint32_t)height, 3, input_image_buffer};
        p.strength = strength;
        fprintf(stderr, "Using img2img with strength: %.2f\n", strength);
@@ -299,11 +304,11 @@ int gen_image(char *text, char *negativeText, int width, int height, int steps,
        p.init_image = {(uint32_t)width, (uint32_t)height, 3, NULL};
        p.strength = 0.0f;
    }
-
+    
    // Handle mask image for inpainting
    if (has_mask_image) {
        fprintf(stderr, "Loading mask image: %s\n", mask_image);
-
+        
        int c = 0;
        int mask_width = 0;
        int mask_height = 0;
@@ -313,11 +318,11 @@ int gen_image(char *text, char *negativeText, int width, int height, int steps,
            if (input_image_buffer) free(input_image_buffer);
            return 1;
        }
-
+        
        // Resize mask if dimensions don't match
        if (mask_width != width || mask_height != height) {
            fprintf(stderr, "Resizing mask image from %dx%d to %dx%d\n", mask_width, mask_height, width, height);
-
+            
            uint8_t* resized_mask_buffer = (uint8_t*)malloc(height * width);
            if (resized_mask_buffer == NULL) {
                fprintf(stderr, "Failed to allocate memory for resized mask\n");
@@ -325,18 +330,18 @@ int gen_image(char *text, char *negativeText, int width, int height, int steps,
                if (input_image_buffer) free(input_image_buffer);
                return 1;
            }
-
+            
            stbir_resize(mask_image_buffer, mask_width, mask_height, 0,
                         resized_mask_buffer, width, height, 0, STBIR_TYPE_UINT8,
                         1, STBIR_ALPHA_CHANNEL_NONE, 0,
                         STBIR_EDGE_CLAMP, STBIR_EDGE_CLAMP,
                         STBIR_FILTER_BOX, STBIR_FILTER_BOX,
                         STBIR_COLORSPACE_SRGB, nullptr);
-
+            
            free(mask_image_buffer);
            mask_image_buffer = resized_mask_buffer;
        }
-
+        
        p.mask_image = {(uint32_t)width, (uint32_t)height, 1, mask_image_buffer};
        fprintf(stderr, "Using inpainting with mask\n");
    } else {
@@ -348,17 +353,17 @@ int gen_image(char *text, char *negativeText, int width, int height, int steps,
    // Handle reference images
    std::vector<sd_image_t> ref_images_vec;
    std::vector<uint8_t*> ref_image_buffers;
-
+    
    if (ref_images_count > 0 && ref_images != NULL) {
        fprintf(stderr, "Loading %d reference images\n", ref_images_count);
-
+        
        for (int i = 0; i < ref_images_count; i++) {
            if (ref_images[i] == NULL || strlen(ref_images[i]) == 0) {
                continue;
            }
-
+            
            fprintf(stderr, "Loading reference image %d: %s\n", i + 1, ref_images[i]);
-
+            
            int c = 0;
            int ref_width = 0;
            int ref_height = 0;
@@ -372,33 +377,33 @@ int gen_image(char *text, char *negativeText, int width, int height, int steps,
                free(ref_image_buffer);
                continue;
            }
-
+            
            // Resize reference image if dimensions don't match
            if (ref_width != width || ref_height != height) {
                fprintf(stderr, "Resizing reference image from %dx%d to %dx%d\n", ref_width, ref_height, width, height);
-
+                
                uint8_t* resized_ref_buffer = (uint8_t*)malloc(height * width * 3);
                if (resized_ref_buffer == NULL) {
                    fprintf(stderr, "Failed to allocate memory for resized reference image\n");
                    free(ref_image_buffer);
                    continue;
                }
-
+                
                stbir_resize(ref_image_buffer, ref_width, ref_height, 0,
                             resized_ref_buffer, width, height, 0, STBIR_TYPE_UINT8,
                             3, STBIR_ALPHA_CHANNEL_NONE, 0,
                             STBIR_EDGE_CLAMP, STBIR_EDGE_CLAMP,
                             STBIR_FILTER_BOX, STBIR_FILTER_BOX,
                             STBIR_COLORSPACE_SRGB, nullptr);
-
+                
                free(ref_image_buffer);
                ref_image_buffer = resized_ref_buffer;
            }
-
+            
            ref_image_buffers.push_back(ref_image_buffer);
            ref_images_vec.push_back({(uint32_t)width, (uint32_t)height, 3, ref_image_buffer});
        }
-
+        
        if (!ref_images_vec.empty()) {
            p.ref_images = ref_images_vec.data();
            p.ref_images_count = ref_images_vec.size();
@@ -449,12 +454,12 @@ int gen_image(char *text, char *negativeText, int width, int height, int steps,
    for (auto buffer : ref_image_buffers) {
        if (buffer) free(buffer);
    }
-    fprintf (stderr, "gen_image is done: %s", dst);
+    fprintf (stderr, "gen_image is done", dst);

    return 0;
 }

 int unload() {
    free_sd_ctx(sd_c);
-    return 0;
 }
+
--- a/backend/go/stablediffusion-ggml/gosd.go
+++ b/backend/go/stablediffusion-ggml/gosd.go
@@ -1,10 +1,15 @@
 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"
-	"runtime"
 	"strings"
 	"unsafe"

@@ -20,34 +25,25 @@ type SDGGML struct {
 	cfgScale     float32
 }

-var (
-	LoadModel func(model, model_apth string, options []uintptr, threads int32, diff int) int
-	GenImage  func(text, negativeText string, width, height, steps int, seed int64, dst string, cfgScale float32, srcImage string, strength float32, maskImage string, refImages []string, refImagesCount int) int
-)
-
-// Copied from Purego internal/strings
-// TODO: We should upstream sending []string
-func hasSuffix(s, suffix string) bool {
-	return len(s) >= len(suffix) && s[len(s)-len(suffix):] == suffix
-}
-
-func CString(name string) *byte {
-	if hasSuffix(name, "\x00") {
-		return &(*(*[]byte)(unsafe.Pointer(&name)))[0]
-	}
-	b := make([]byte, len(name)+1)
-	copy(b, name)
-	return &b[0]
-}
-
 func (sd *SDGGML) Load(opts *pb.ModelOptions) error {

 	sd.threads = int(opts.Threads)

 	modelPath := opts.ModelPath

-	modelFile := opts.ModelFile
-	modelPathC := modelPath
+	modelFile := C.CString(opts.ModelFile)
+	defer C.free(unsafe.Pointer(modelFile))
+
+	modelPathC := C.CString(modelPath)
+	defer C.free(unsafe.Pointer(modelPathC))
+
+	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 + 1) * size))
+	view := (*[1 << 30]*C.char)(unsafe.Pointer(options))[0 : len(opts.Options)+1 : len(opts.Options)+1]

 	var diffusionModel int

@@ -72,55 +68,81 @@ func (sd *SDGGML) Load(opts *pb.ModelOptions) error {

 	fmt.Fprintf(os.Stderr, "Options: %+v\n", oo)

-	// At the time of writing Purego doesn't recurse into slices and convert Go strings to pointers so we need to do that
-	var keepAlive []any
-	options := make([]uintptr, len(oo), len(oo)+1)
-	for i, op := range oo {
-		bytep := CString(op)
-		options[i] = uintptr(unsafe.Pointer(bytep))
-		keepAlive = append(keepAlive, bytep)
+	for i, x := range oo {
+		view[i] = C.CString(x)
 	}
+	view[len(oo)] = nil

 	sd.cfgScale = opts.CFGScale

-	ret := LoadModel(modelFile, modelPathC, options, opts.Threads, diffusionModel)
+	ret := C.load_model(modelFile, modelPathC, options, C.int(opts.Threads), C.int(diffusionModel))
 	if ret != 0 {
 		return fmt.Errorf("could not load model")
 	}

-	runtime.KeepAlive(keepAlive)
-
 	return nil
 }

 func (sd *SDGGML) GenerateImage(opts *pb.GenerateImageRequest) error {
-	t := opts.PositivePrompt
-	dst := opts.Dst
-	negative := opts.NegativePrompt
-	srcImage := opts.Src
+	t := C.CString(opts.PositivePrompt)
+	defer C.free(unsafe.Pointer(t))

-	var maskImage string
+	dst := C.CString(opts.Dst)
+	defer C.free(unsafe.Pointer(dst))
+
+	negative := C.CString(opts.NegativePrompt)
+	defer C.free(unsafe.Pointer(negative))
+
+	// Handle source image path
+	var srcImage *C.char
+	if opts.Src != "" {
+		srcImage = C.CString(opts.Src)
+		defer C.free(unsafe.Pointer(srcImage))
+	}
+
+	// Handle mask image path
+	var maskImage *C.char
 	if opts.EnableParameters != "" {
+		// Parse EnableParameters for mask path if provided
+		// This is a simple approach - in a real implementation you might want to parse JSON
 		if strings.Contains(opts.EnableParameters, "mask:") {
 			parts := strings.Split(opts.EnableParameters, "mask:")
 			if len(parts) > 1 {
 				maskPath := strings.TrimSpace(parts[1])
 				if maskPath != "" {
-					maskImage = maskPath
+					maskImage = C.CString(maskPath)
+					defer C.free(unsafe.Pointer(maskImage))
 				}
 			}
 		}
 	}

-	refImagesCount := len(opts.RefImages)
-	refImages := make([]string, refImagesCount, refImagesCount+1)
-	copy(refImages, opts.RefImages)
-	*(*uintptr)(unsafe.Add(unsafe.Pointer(&refImages), refImagesCount)) = 0
+	// Handle reference images
+	var refImages **C.char
+	var refImagesCount C.int
+	if len(opts.RefImages) > 0 {
+		refImagesCount = C.int(len(opts.RefImages))
+		// Allocate array of C strings
+		size := C.size_t(unsafe.Sizeof((*C.char)(nil)))
+		refImages = (**C.char)(C.malloc((C.size_t(len(opts.RefImages)) + 1) * size))
+		view := (*[1 << 30]*C.char)(unsafe.Pointer(refImages))[0 : len(opts.RefImages)+1 : len(opts.RefImages)+1]
+
+		for i, refImagePath := range opts.RefImages {
+			view[i] = C.CString(refImagePath)
+			defer C.free(unsafe.Pointer(view[i]))
+		}
+		view[len(opts.RefImages)] = nil
+	}

 	// Default strength for img2img (0.75 is a good default)
-	strength := float32(0.75)
+	strength := C.float(0.75)
+	if opts.Src != "" {
+		// If we have a source image, use img2img mode
+		// You could also parse strength from EnableParameters if needed
+		strength = C.float(0.75)
+	}

-	ret := GenImage(t, negative, int(opts.Width), int(opts.Height), int(opts.Step), int64(opts.Seed), dst, sd.cfgScale, srcImage, strength, maskImage, refImages, refImagesCount)
+	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), srcImage, strength, maskImage, refImages, refImagesCount)
 	if ret != 0 {
 		return fmt.Errorf("inference failed")
 	}
--- a/backend/go/stablediffusion-ggml/gosd.h
+++ b/backend/go/stablediffusion-ggml/gosd.h
@@ -1,8 +1,8 @@
 #ifdef __cplusplus
 extern "C" {
 #endif
-int load_model(const char *model, char *model_path, char* options[], int threads, int diffusionModel);
-int gen_image(char *text, char *negativeText, int width, int height, int steps, int64_t seed, char *dst, float cfg_scale, char *src_image, float strength, char *mask_image, char **ref_images, int ref_images_count);
+int load_model(char *model, char *model_path, 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, char *src_image, float strength, char *mask_image, char **ref_images, int ref_images_count);
 #ifdef __cplusplus
 }
-#endif
+#endif
--- a/backend/go/stablediffusion-ggml/main.go
+++ b/backend/go/stablediffusion-ggml/main.go
@@ -1,9 +1,9 @@
 package main

+// Note: this is started internally by LocalAI and a server is allocated for each model
 import (
 	"flag"

-	"github.com/ebitengine/purego"
 	grpc "github.com/mudler/LocalAI/pkg/grpc"
 )

@@ -12,14 +12,6 @@ var (
 )

 func main() {
-	gosd, err := purego.Dlopen("./libgosd.so", purego.RTLD_NOW|purego.RTLD_GLOBAL)
-	if err != nil {
-		panic(err)
-	}
-
-	purego.RegisterLibFunc(&LoadModel, gosd, "load_model")
-	purego.RegisterLibFunc(&GenImage, gosd, "gen_image")
-
 	flag.Parse()

 	if err := grpc.StartServer(*addr, &SDGGML{}); err != nil {
--- a/backend/go/stablediffusion-ggml/package.sh
+++ b/backend/go/stablediffusion-ggml/package.sh
@@ -10,9 +10,8 @@ CURDIR=$(dirname "$(realpath $0)")
 # Create lib directory
 mkdir -p $CURDIR/package/lib

-cp -avf $CURDIR/libgosd.so $CURDIR/package/
-cp -avf $CURDIR/stablediffusion-ggml $CURDIR/package/
-cp -fv $CURDIR/run.sh $CURDIR/package/
+cp -avrf $CURDIR/stablediffusion-ggml $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
@@ -43,13 +42,11 @@ elif [ -f "/lib/ld-linux-aarch64.so.1" ]; then
    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
-elif [ $(uname -s) = "Darwin" ]; then
-    echo "Detected Darwin"
 else
    echo "Error: Could not detect architecture"
    exit 1
 fi

-echo "Packaging completed successfully"
+echo "Packaging completed successfully" 
 ls -liah $CURDIR/package/
-ls -liah $CURDIR/package/lib/
+ls -liah $CURDIR/package/lib/
--- a/backend/go/whisper/.gitignore
+++ b/backend/go/whisper/.gitignore
@@ -1,7 +0,0 @@
-.cache/
-sources/
-build/
-package/
-whisper
-libgowhisper.so
-
--- a/backend/go/whisper/CMakeLists.txt
+++ b/backend/go/whisper/CMakeLists.txt
@@ -1,16 +0,0 @@
-cmake_minimum_required(VERSION 3.12)
-project(gowhisper LANGUAGES C CXX)
-set(CMAKE_POSITION_INDEPENDENT_CODE ON)
-set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
-
-add_subdirectory(./sources/whisper.cpp)
-
-add_library(gowhisper MODULE gowhisper.cpp)
-target_link_libraries(gowhisper PRIVATE whisper ggml)
-
-if(CMAKE_CXX_COMPILER_ID MATCHES "GNU" AND CMAKE_CXX_COMPILER_VERSION VERSION_LESS 9.0)
-    target_link_libraries(gosd PRIVATE stdc++fs)
-endif()
-
-set_property(TARGET gowhisper PROPERTY CXX_STANDARD 17)
-set_target_properties(gowhisper PROPERTIES LIBRARY_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR})
--- a/backend/go/whisper/Makefile
+++ b/backend/go/whisper/Makefile
@@ -1,53 +1,110 @@
-CMAKE_ARGS?=
-BUILD_TYPE?=
+GOCMD=go
 NATIVE?=false

-GOCMD?=go
-GO_TAGS?=
-JOBS?=$(shell nproc --ignore=1)
+BUILD_TYPE?=
+CMAKE_ARGS?=

 # whisper.cpp version
 WHISPER_REPO?=https://github.com/ggml-org/whisper.cpp
-WHISPER_CPP_VERSION?=edea8a9c3cf0eb7676dcdb604991eb2f95c3d984
+WHISPER_CPP_VERSION?=4245c77b654cd384ad9f53a4a302be716b3e5861

-CMAKE_ARGS+=-DBUILD_SHARED_LIBS=OFF
+export WHISPER_CMAKE_ARGS?=-DBUILD_SHARED_LIBS=OFF
+export WHISPER_DIR=$(abspath ./sources/whisper.cpp)
+export WHISPER_INCLUDE_PATH=$(WHISPER_DIR)/include:$(WHISPER_DIR)/ggml/include
+export WHISPER_LIBRARY_PATH=$(WHISPER_DIR)/build/src/:$(WHISPER_DIR)/build/ggml/src

+CGO_LDFLAGS_WHISPER?=
+CGO_LDFLAGS_WHISPER+=-lggml
+CMAKE_ARGS+=-DBUILD_SHARED_LIBS=OFF -DLLAMA_CURL=OFF
+CUDA_LIBPATH?=/usr/local/cuda/lib64/
+
+ONEAPI_VERSION?=2025.2
+
+# IF native is false, we add -DGGML_NATIVE=OFF to CMAKE_ARGS
+ifeq ($(NATIVE),false)
+	CMAKE_ARGS+=-DGGML_NATIVE=OFF
+	WHISPER_CMAKE_ARGS+=-DGGML_NATIVE=OFF
+endif
+CURRENT_MAKEFILE_DIR := $(dir $(abspath $(lastword $(MAKEFILE_LIST))))
 ifeq ($(NATIVE),false)
 	CMAKE_ARGS+=-DGGML_NATIVE=OFF
 endif
-
+# If build type is cublas, then we set -DGGML_CUDA=ON to CMAKE_ARGS automatically
 ifeq ($(BUILD_TYPE),cublas)
+	CGO_LDFLAGS+=-lcublas -lcudart -L$(CUDA_LIBPATH) -L$(CUDA_LIBPATH)/stubs/ -lcuda
 	CMAKE_ARGS+=-DGGML_CUDA=ON
+	CGO_LDFLAGS_WHISPER+=-lcufft -lggml-cuda
+	export WHISPER_LIBRARY_PATH:=$(WHISPER_LIBRARY_PATH):$(WHISPER_DIR)/build/ggml/src/ggml-cuda/
+# If build type is openblas then we set -DGGML_BLAS=ON -DGGML_BLAS_VENDOR=OpenBLAS
+# to CMAKE_ARGS automatically
 else ifeq ($(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_HIPBLAS=ON
+	ROCM_HOME ?= /opt/rocm
+	ROCM_PATH ?= /opt/rocm
+	LD_LIBRARY_PATH ?= /opt/rocm/lib:/opt/rocm/llvm/lib
+	export STABLE_BUILD_TYPE=
+	export CXX=$(ROCM_HOME)/llvm/bin/clang++
+	export CC=$(ROCM_HOME)/llvm/bin/clang
+#	GPU_TARGETS ?= gfx803,gfx900,gfx906,gfx908,gfx90a,gfx942,gfx1010,gfx1030,gfx1032,gfx1100,gfx1101,gfx1102
+#	AMDGPU_TARGETS ?= "$(GPU_TARGETS)"
+	CMAKE_ARGS+=-DGGML_HIP=ON
+	CGO_LDFLAGS += -O3 --rtlib=compiler-rt -unwindlib=libgcc -lhipblas -lrocblas --hip-link -L${ROCM_HOME}/lib/llvm/lib -L$(CURRENT_MAKEFILE_DIR)/sources/whisper.cpp/build/ggml/src/ggml-hip/ -lggml-hip
+#	CMAKE_ARGS+=-DGGML_HIP=ON -DAMDGPU_TARGETS="$(AMDGPU_TARGETS)" -DGPU_TARGETS="$(GPU_TARGETS)"
 else ifeq ($(BUILD_TYPE),vulkan)
-	CMAKE_ARGS+=-DGGML_VULKAN=ON
+	CMAKE_ARGS+=-DGGML_VULKAN=1
+	CGO_LDFLAGS_WHISPER+=-lggml-vulkan -lvulkan
+	export WHISPER_LIBRARY_PATH:=$(WHISPER_LIBRARY_PATH):$(WHISPER_DIR)/build/ggml/src/ggml-vulkan/
 else ifeq ($(OS),Darwin)
+	ifeq ($(BUILD_TYPE),)
+		BUILD_TYPE=metal
+	endif
 	ifneq ($(BUILD_TYPE),metal)
 		CMAKE_ARGS+=-DGGML_METAL=OFF
+		CGO_LDFLAGS_WHISPER+=-lggml-blas
+		export WHISPER_LIBRARY_PATH:=$(WHISPER_LIBRARY_PATH):$(WHISPER_DIR)/build/ggml/src/ggml-blas
 	else
 		CMAKE_ARGS+=-DGGML_METAL=ON
 		CMAKE_ARGS+=-DGGML_METAL_EMBED_LIBRARY=ON
+		CMAKE_ARGS+=-DGGML_METAL_USE_BF16=ON
+		CMAKE_ARGS+=-DGGML_OPENMP=OFF
+		CMAKE_ARGS+=-DWHISPER_BUILD_EXAMPLES=OFF
+		CMAKE_ARGS+=-DWHISPER_BUILD_TESTS=OFF
+		CMAKE_ARGS+=-DWHISPER_BUILD_SERVER=OFF
+		CGO_LDFLAGS += -framework Accelerate
+		CGO_LDFLAGS_WHISPER+=-lggml-metal -lggml-blas
+		export WHISPER_LIBRARY_PATH:=$(WHISPER_LIBRARY_PATH):$(WHISPER_DIR)/build/ggml/src/ggml-metal/:$(WHISPER_DIR)/build/ggml/src/ggml-blas
 	endif
+	TARGET+=--target ggml-metal
 endif

-ifeq ($(BUILD_TYPE),sycl_f16)
+ifneq (,$(findstring sycl,$(BUILD_TYPE)))
+	export CC=icx
+	export CXX=icpx
+	CGO_LDFLAGS_WHISPER += -fsycl -L${DNNLROOT}/lib -rpath ${ONEAPI_ROOT}/${ONEAPI_VERSION}/lib -ldnnl ${MKLROOT}/lib/intel64/libmkl_sycl.a -fiopenmp -fopenmp-targets=spir64 -lOpenCL -lggml-sycl
+	CGO_LDFLAGS_WHISPER += $(shell pkg-config --libs mkl-static-lp64-gomp)
+	CGO_CXXFLAGS_WHISPER += -fiopenmp -fopenmp-targets=spir64
+	CGO_CXXFLAGS_WHISPER += $(shell pkg-config --cflags mkl-static-lp64-gomp )
+	export WHISPER_LIBRARY_PATH:=$(WHISPER_LIBRARY_PATH):$(WHISPER_DIR)/build/ggml/src/ggml-sycl/
 	CMAKE_ARGS+=-DGGML_SYCL=ON \
 		-DCMAKE_C_COMPILER=icx \
 		-DCMAKE_CXX_COMPILER=icpx \
-		-DGGML_SYCL_F16=ON
+		-DCMAKE_CXX_FLAGS="-fsycl"
 endif

-ifeq ($(BUILD_TYPE),sycl_f32)
-	CMAKE_ARGS+=-DGGML_SYCL=ON \
-		-DCMAKE_C_COMPILER=icx \
-		-DCMAKE_CXX_COMPILER=icpx
+ifeq ($(BUILD_TYPE),sycl_f16)
+	CMAKE_ARGS+=-DGGML_SYCL_F16=ON
 endif

+ifneq ($(OS),Darwin)
+	CGO_LDFLAGS_WHISPER+=-lgomp
+endif
+
+## whisper
 sources/whisper.cpp:
 	mkdir -p sources/whisper.cpp
 	cd sources/whisper.cpp && \
@@ -57,21 +114,18 @@ sources/whisper.cpp:
 	git checkout $(WHISPER_CPP_VERSION) && \
 	git submodule update --init --recursive --depth 1 --single-branch

-libgowhisper.so: sources/whisper.cpp CMakeLists.txt gowhisper.cpp gowhisper.h
-	mkdir -p build && \
-	cd build && \
-	cmake .. $(CMAKE_ARGS) && \
-	cmake --build . --config Release -j$(JOBS) && \
-	cd .. && \
-	mv build/libgowhisper.so ./
+sources/whisper.cpp/build/src/libwhisper.a: sources/whisper.cpp
+	cd sources/whisper.cpp && cmake $(CMAKE_ARGS) $(WHISPER_CMAKE_ARGS) . -B ./build
+	cd sources/whisper.cpp/build && cmake --build . --config Release

-whisper: main.go gowhisper.go libgowhisper.so
-	CGO_ENABLED=0 $(GOCMD) build -tags "$(GO_TAGS)" -o whisper ./
+whisper: sources/whisper.cpp sources/whisper.cpp/build/src/libwhisper.a
+	$(GOCMD) mod edit -replace github.com/ggerganov/whisper.cpp=$(CURDIR)/sources/whisper.cpp
+	$(GOCMD) mod edit -replace github.com/ggerganov/whisper.cpp/bindings/go=$(CURDIR)/sources/whisper.cpp/bindings/go
+	CGO_LDFLAGS="$(CGO_LDFLAGS) $(CGO_LDFLAGS_WHISPER)" C_INCLUDE_PATH="${WHISPER_INCLUDE_PATH}" LIBRARY_PATH="${WHISPER_LIBRARY_PATH}" LD_LIBRARY_PATH="${WHISPER_LIBRARY_PATH}" \
+	CGO_CXXFLAGS="$(CGO_CXXFLAGS_WHISPER)" \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o whisper ./

-package: whisper
+package:
 	bash package.sh

-build: package
-
-clean:
-	rm -rf libgowhisper.o build whisper
+build: whisper package
--- a/backend/go/whisper/gowhisper.cpp
+++ b/backend/go/whisper/gowhisper.cpp
@@ -1,154 +0,0 @@
-#include "gowhisper.h"
-#include "ggml-backend.h"
-#include "whisper.h"
-#include <vector>
-
-static struct whisper_vad_context *vctx;
-static struct whisper_context *ctx;
-static std::vector<float> flat_segs;
-
-static void ggml_log_cb(enum ggml_log_level level, const char *log,
-                        void *data) {
-  const char *level_str;
-
-  if (!log) {
-    return;
-  }
-
-  switch (level) {
-  case GGML_LOG_LEVEL_DEBUG:
-    level_str = "DEBUG";
-    break;
-  case GGML_LOG_LEVEL_INFO:
-    level_str = "INFO";
-    break;
-  case GGML_LOG_LEVEL_WARN:
-    level_str = "WARN";
-    break;
-  case GGML_LOG_LEVEL_ERROR:
-    level_str = "ERROR";
-    break;
-  default: /* Potential future-proofing */
-    level_str = "?????";
-    break;
-  }
-
-  fprintf(stderr, "[%-5s] ", level_str);
-  fputs(log, stderr);
-  fflush(stderr);
-}
-
-int load_model(const char *const model_path) {
-  whisper_log_set(ggml_log_cb, nullptr);
-  ggml_backend_load_all();
-
-  struct whisper_context_params cparams = whisper_context_default_params();
-
-  ctx = whisper_init_from_file_with_params(model_path, cparams);
-  if (ctx == nullptr) {
-    fprintf(stderr, "error: Also failed to init model as transcriber\n");
-    return 1;
-  }
-
-  return 0;
-}
-
-int load_model_vad(const char *const model_path) {
-  whisper_log_set(ggml_log_cb, nullptr);
-  ggml_backend_load_all();
-
-  struct whisper_vad_context_params vcparams =
-      whisper_vad_default_context_params();
-
-  // XXX: Overridden to false in upstream due to performance?
-  // vcparams.use_gpu = true;
-
-  vctx = whisper_vad_init_from_file_with_params(model_path, vcparams);
-  if (vctx == nullptr) {
-    fprintf(stderr, "error: Failed to init model as VAD\n");
-    return 1;
-  }
-
-  return 0;
-}
-
-int vad(float pcmf32[], size_t pcmf32_len, float **segs_out,
-        size_t *segs_out_len) {
-  if (!whisper_vad_detect_speech(vctx, pcmf32, pcmf32_len)) {
-    fprintf(stderr, "error: failed to detect speech\n");
-    return 1;
-  }
-
-  struct whisper_vad_params params = whisper_vad_default_params();
-  struct whisper_vad_segments *segs =
-      whisper_vad_segments_from_probs(vctx, params);
-  size_t segn = whisper_vad_segments_n_segments(segs);
-
-  // fprintf(stderr, "Got segments %zd\n", segn);
-
-  flat_segs.clear();
-
-  for (int i = 0; i < segn; i++) {
-    flat_segs.push_back(whisper_vad_segments_get_segment_t0(segs, i));
-    flat_segs.push_back(whisper_vad_segments_get_segment_t1(segs, i));
-  }
-
-  // fprintf(stderr, "setting out variables: %p=%p -> %p, %p=%zx -> %zx\n",
-  //         segs_out, *segs_out, flat_segs.data(), segs_out_len, *segs_out_len,
-  //         flat_segs.size());
-  *segs_out = flat_segs.data();
-  *segs_out_len = flat_segs.size();
-
-  // fprintf(stderr, "freeing segs\n");
-  whisper_vad_free_segments(segs);
-
-  // fprintf(stderr, "returning\n");
-  return 0;
-}
-
-int transcribe(uint32_t threads, char *lang, bool translate, bool tdrz,
-               float pcmf32[], size_t pcmf32_len, size_t *segs_out_len) {
-  whisper_full_params wparams =
-      whisper_full_default_params(WHISPER_SAMPLING_GREEDY);
-
-  wparams.n_threads = threads;
-  if (*lang != '\0')
-    wparams.language = lang;
-  else {
-    wparams.language = nullptr;
-  }
-
-  wparams.translate = translate;
-  wparams.debug_mode = true;
-  wparams.print_progress = true;
-  wparams.tdrz_enable = tdrz;
-
-  fprintf(stderr, "info: Enable tdrz: %d\n", tdrz);
-
-  if (whisper_full(ctx, wparams, pcmf32, pcmf32_len)) {
-    fprintf(stderr, "error: transcription failed\n");
-    return 1;
-  }
-
-  *segs_out_len = whisper_full_n_segments(ctx);
-
-  return 0;
-}
-
-const char *get_segment_text(int i) {
-  return whisper_full_get_segment_text(ctx, i);
-}
-
-int64_t get_segment_t0(int i) { return whisper_full_get_segment_t0(ctx, i); }
-
-int64_t get_segment_t1(int i) { return whisper_full_get_segment_t1(ctx, i); }
-
-int n_tokens(int i) { return whisper_full_n_tokens(ctx, i); }
-
-int32_t get_token_id(int i, int j) {
-  return whisper_full_get_token_id(ctx, i, j);
-}
-
-bool get_segment_speaker_turn_next(int i) {
-  return whisper_full_get_segment_speaker_turn_next(ctx, i);
-}
--- a/backend/go/whisper/gowhisper.go
+++ b/backend/go/whisper/gowhisper.go
@@ -1,161 +0,0 @@
-package main
-
-import (
-	"fmt"
-	"os"
-	"path/filepath"
-	"strings"
-	"unsafe"
-
-	"github.com/go-audio/wav"
-	"github.com/mudler/LocalAI/pkg/grpc/base"
-	pb "github.com/mudler/LocalAI/pkg/grpc/proto"
-	"github.com/mudler/LocalAI/pkg/utils"
-)
-
-var (
-	CppLoadModel                 func(modelPath string) int
-	CppLoadModelVAD              func(modelPath string) int
-	CppVAD                       func(pcmf32 []float32, pcmf32Size uintptr, segsOut unsafe.Pointer, segsOutLen unsafe.Pointer) int
-	CppTranscribe                func(threads uint32, lang string, translate bool, diarize bool, pcmf32 []float32, pcmf32Len uintptr, segsOutLen unsafe.Pointer) int
-	CppGetSegmentText            func(i int) string
-	CppGetSegmentStart           func(i int) int64
-	CppGetSegmentEnd             func(i int) int64
-	CppNTokens                   func(i int) int
-	CppGetTokenID                func(i int, j int) int
-	CppGetSegmentSpeakerTurnNext func(i int) bool
-)
-
-type Whisper struct {
-	base.SingleThread
-}
-
-func (w *Whisper) Load(opts *pb.ModelOptions) error {
-	vadOnly := false
-
-	for _, oo := range opts.Options {
-		if oo == "vad_only" {
-			vadOnly = true
-		} else {
-			fmt.Fprintf(os.Stderr, "Unrecognized option: %v\n", oo)
-		}
-	}
-
-	if vadOnly {
-		if ret := CppLoadModelVAD(opts.ModelFile); ret != 0 {
-			return fmt.Errorf("Failed to load Whisper VAD model")
-		}
-
-		return nil
-	}
-
-	if ret := CppLoadModel(opts.ModelFile); ret != 0 {
-		return fmt.Errorf("Failed to load Whisper transcription model")
-	}
-
-	return nil
-}
-
-func (w *Whisper) VAD(req *pb.VADRequest) (pb.VADResponse, error) {
-	audio := req.Audio
-	// We expect 0xdeadbeef to be overwritten and if we see it in a stack trace we know it wasn't
-	segsPtr, segsLen := uintptr(0xdeadbeef), uintptr(0xdeadbeef)
-	segsPtrPtr, segsLenPtr := unsafe.Pointer(&segsPtr), unsafe.Pointer(&segsLen)
-
-	if ret := CppVAD(audio, uintptr(len(audio)), segsPtrPtr, segsLenPtr); ret != 0 {
-		return pb.VADResponse{}, fmt.Errorf("Failed VAD")
-	}
-
-	// Happens when CPP vector has not had any elements pushed to it
-	if segsPtr == 0 {
-		return pb.VADResponse{
-			Segments: []*pb.VADSegment{},
-		}, nil
-	}
-
-	// unsafeptr warning is caused by segsPtr being on the stack and therefor being subject to stack copying AFAICT
-	// however the stack shouldn't have grown between setting segsPtr and now, also the memory pointed to is allocated by C++
-	segs := unsafe.Slice((*float32)(unsafe.Pointer(segsPtr)), segsLen)
-
-	vadSegments := []*pb.VADSegment{}
-	for i := range len(segs) >> 1 {
-		s := segs[2*i] / 100
-		t := segs[2*i+1] / 100
-		vadSegments = append(vadSegments, &pb.VADSegment{
-			Start: s,
-			End:   t,
-		})
-	}
-
-	return pb.VADResponse{
-		Segments: vadSegments,
-	}, nil
-}
-
-func (w *Whisper) AudioTranscription(opts *pb.TranscriptRequest) (pb.TranscriptResult, error) {
-	dir, err := os.MkdirTemp("", "whisper")
-	if err != nil {
-		return pb.TranscriptResult{}, err
-	}
-	defer os.RemoveAll(dir)
-
-	convertedPath := filepath.Join(dir, "converted.wav")
-
-	if err := utils.AudioToWav(opts.Dst, convertedPath); err != nil {
-		return pb.TranscriptResult{}, err
-	}
-
-	// Open samples
-	fh, err := os.Open(convertedPath)
-	if err != nil {
-		return pb.TranscriptResult{}, err
-	}
-	defer fh.Close()
-
-	// Read samples
-	d := wav.NewDecoder(fh)
-	buf, err := d.FullPCMBuffer()
-	if err != nil {
-		return pb.TranscriptResult{}, err
-	}
-
-	data := buf.AsFloat32Buffer().Data
-	segsLen := uintptr(0xdeadbeef)
-	segsLenPtr := unsafe.Pointer(&segsLen)
-
-	if ret := CppTranscribe(opts.Threads, opts.Language, opts.Translate, opts.Diarize, data, uintptr(len(data)), segsLenPtr); ret != 0 {
-		return pb.TranscriptResult{}, fmt.Errorf("Failed Transcribe")
-	}
-
-	segments := []*pb.TranscriptSegment{}
-	text := ""
-	for i := range int(segsLen) {
-		s := CppGetSegmentStart(i)
-		t := CppGetSegmentEnd(i)
-		txt := strings.Clone(CppGetSegmentText(i))
-		tokens := make([]int32, CppNTokens(i))
-
-		if opts.Diarize && CppGetSegmentSpeakerTurnNext(i) {
-			txt += " [SPEAKER_TURN]"
-		}
-
-		for j := range tokens {
-			tokens[j] = int32(CppGetTokenID(i, j))
-		}
-		segment := &pb.TranscriptSegment{
-			Id:    int32(i),
-			Text:  txt,
-			Start: s, End: t,
-			Tokens: tokens,
-		}
-
-		segments = append(segments, segment)
-
-		text += " " + strings.TrimSpace(txt)
-	}
-
-	return pb.TranscriptResult{
-		Segments: segments,
-		Text:     strings.TrimSpace(text),
-	}, nil
-}
--- a/backend/go/whisper/gowhisper.h
+++ b/backend/go/whisper/gowhisper.h
@@ -1,17 +0,0 @@
-#include <cstddef>
-#include <cstdint>
-
-extern "C" {
-int load_model(const char *const model_path);
-int load_model_vad(const char *const model_path);
-int vad(float pcmf32[], size_t pcmf32_size, float **segs_out,
-        size_t *segs_out_len);
-int transcribe(uint32_t threads, char *lang, bool translate, bool tdrz,
-               float pcmf32[], size_t pcmf32_len, size_t *segs_out_len);
-const char *get_segment_text(int i);
-int64_t get_segment_t0(int i);
-int64_t get_segment_t1(int i);
-int n_tokens(int i);
-int32_t get_token_id(int i, int j);
-bool get_segment_speaker_turn_next(int i);
-}
--- a/backend/go/whisper/main.go
+++ b/backend/go/whisper/main.go
@@ -1,10 +1,10 @@
 package main

 // Note: this is started internally by LocalAI and a server is allocated for each model
+
 import (
 	"flag"

-	"github.com/ebitengine/purego"
 	grpc "github.com/mudler/LocalAI/pkg/grpc"
 )

@@ -12,34 +12,7 @@ var (
 	addr = flag.String("addr", "localhost:50051", "the address to connect to")
 )

-type LibFuncs struct {
-	FuncPtr any
-	Name    string
-}
-
 func main() {
-	gosd, err := purego.Dlopen("./libgowhisper.so", purego.RTLD_NOW|purego.RTLD_GLOBAL)
-	if err != nil {
-		panic(err)
-	}
-
-	libFuncs := []LibFuncs{
-		{&CppLoadModel, "load_model"},
-		{&CppLoadModelVAD, "load_model_vad"},
-		{&CppVAD, "vad"},
-		{&CppTranscribe, "transcribe"},
-		{&CppGetSegmentText, "get_segment_text"},
-		{&CppGetSegmentStart, "get_segment_t0"},
-		{&CppGetSegmentEnd, "get_segment_t1"},
-		{&CppNTokens, "n_tokens"},
-		{&CppGetTokenID, "get_token_id"},
-		{&CppGetSegmentSpeakerTurnNext, "get_segment_speaker_turn_next"},
-	}
-
-	for _, lf := range libFuncs {
-		purego.RegisterLibFunc(lf.FuncPtr, gosd, lf.Name)
-	}
-
 	flag.Parse()

 	if err := grpc.StartServer(*addr, &Whisper{}); err != nil {
--- a/backend/go/whisper/package.sh
+++ b/backend/go/whisper/package.sh
@@ -10,8 +10,8 @@ CURDIR=$(dirname "$(realpath $0)")
 # Create lib directory
 mkdir -p $CURDIR/package/lib

-cp -avf $CURDIR/whisper $CURDIR/libgowhisper.so $CURDIR/package/
-cp -fv $CURDIR/run.sh $CURDIR/package/
+cp -avrf $CURDIR/whisper $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
@@ -42,13 +42,11 @@ elif [ -f "/lib/ld-linux-aarch64.so.1" ]; then
    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
-elif [ $(uname -s) = "Darwin" ]; then
-    echo "Detected Darwin"
 else
    echo "Error: Could not detect architecture"
    exit 1
 fi

-echo "Packaging completed successfully"
+echo "Packaging completed successfully" 
 ls -liah $CURDIR/package/
-ls -liah $CURDIR/package/lib/
+ls -liah $CURDIR/package/lib/
--- a/backend/go/whisper/whisper.go
+++ b/backend/go/whisper/whisper.go
@@ -0,0 +1,105 @@
+package main
+
+// This is a wrapper to statisfy the GRPC service interface
+// It is meant to be used by the main executable that is the server for the specific backend type (falcon, gpt3, etc)
+import (
+	"os"
+	"path/filepath"
+
+	"github.com/ggerganov/whisper.cpp/bindings/go/pkg/whisper"
+	"github.com/go-audio/wav"
+	"github.com/mudler/LocalAI/pkg/grpc/base"
+	pb "github.com/mudler/LocalAI/pkg/grpc/proto"
+	"github.com/mudler/LocalAI/pkg/utils"
+)
+
+type Whisper struct {
+	base.SingleThread
+	whisper whisper.Model
+}
+
+func (sd *Whisper) Load(opts *pb.ModelOptions) error {
+	// Note: the Model here is a path to a directory containing the model files
+	w, err := whisper.New(opts.ModelFile)
+	sd.whisper = w
+	return err
+}
+
+func (sd *Whisper) AudioTranscription(opts *pb.TranscriptRequest) (pb.TranscriptResult, error) {
+
+	dir, err := os.MkdirTemp("", "whisper")
+	if err != nil {
+		return pb.TranscriptResult{}, err
+	}
+	defer os.RemoveAll(dir)
+
+	convertedPath := filepath.Join(dir, "converted.wav")
+
+	if err := utils.AudioToWav(opts.Dst, convertedPath); err != nil {
+		return pb.TranscriptResult{}, err
+	}
+
+	// Open samples
+	fh, err := os.Open(convertedPath)
+	if err != nil {
+		return pb.TranscriptResult{}, err
+	}
+	defer fh.Close()
+
+	// Read samples
+	d := wav.NewDecoder(fh)
+	buf, err := d.FullPCMBuffer()
+	if err != nil {
+		return pb.TranscriptResult{}, err
+	}
+
+	data := buf.AsFloat32Buffer().Data
+
+	// Process samples
+	context, err := sd.whisper.NewContext()
+	if err != nil {
+		return pb.TranscriptResult{}, err
+
+	}
+
+	context.SetThreads(uint(opts.Threads))
+
+	if opts.Language != "" {
+		context.SetLanguage(opts.Language)
+	} else {
+		context.SetLanguage("auto")
+	}
+
+	if opts.Translate {
+		context.SetTranslate(true)
+	}
+
+	if err := context.Process(data, nil, nil, nil); err != nil {
+		return pb.TranscriptResult{}, err
+	}
+
+	segments := []*pb.TranscriptSegment{}
+	text := ""
+	for {
+		s, err := context.NextSegment()
+		if err != nil {
+			break
+		}
+
+		var tokens []int32
+		for _, t := range s.Tokens {
+			tokens = append(tokens, int32(t.Id))
+		}
+
+		segment := &pb.TranscriptSegment{Id: int32(s.Num), Text: s.Text, Start: int64(s.Start), End: int64(s.End), Tokens: tokens}
+		segments = append(segments, segment)
+
+		text += s.Text
+	}
+
+	return pb.TranscriptResult{
+		Segments: segments,
+		Text:     text,
+	}, nil
+
+}
--- a/backend/index.yaml
+++ b/backend/index.yaml
@@ -45,7 +45,6 @@
    default: "cpu-whisper"
    nvidia: "cuda12-whisper"
    intel: "intel-sycl-f16-whisper"
-    metal: "metal-whisper"
    amd: "rocm-whisper"
    vulkan: "vulkan-whisper"
    nvidia-l4t: "nvidia-l4t-arm64-whisper"
@@ -72,7 +71,7 @@
    # amd: "rocm-stablediffusion-ggml"
    vulkan: "vulkan-stablediffusion-ggml"
    nvidia-l4t: "nvidia-l4t-arm64-stablediffusion-ggml"
-    metal: "metal-stablediffusion-ggml"
+    # metal: "metal-stablediffusion-ggml"
    # darwin-x86: "darwin-x86-stablediffusion-ggml"
 - &rfdetr
  name: "rfdetr"
@@ -128,55 +127,6 @@
    nvidia: "cuda12-vllm"
    amd: "rocm-vllm"
    intel: "intel-vllm"
- &mlx
-  name: "mlx"
-  uri: "quay.io/go-skynet/local-ai-backends:latest-metal-darwin-arm64-mlx"
-  icon: https://avatars.githubusercontent.com/u/102832242?s=200&v=4
-  urls:
-    - https://github.com/ml-explore/mlx-lm
-  mirrors:
-    - localai/localai-backends:latest-metal-darwin-arm64-mlx
-  license: MIT
-  description: |
-      Run LLMs with MLX
-  tags:
-    - text-to-text
-    - LLM
-    - MLX
- &mlx-vlm
-  name: "mlx-vlm"
-  uri: "quay.io/go-skynet/local-ai-backends:latest-metal-darwin-arm64-mlx-vlm"
-  icon: https://avatars.githubusercontent.com/u/102832242?s=200&v=4
-  urls:
-    - https://github.com/Blaizzy/mlx-vlm
-  mirrors:
-    - localai/localai-backends:latest-metal-darwin-arm64-mlx-vlm
-  license: MIT
-  description: |
-      Run Vision-Language Models with MLX
-  tags:
-    - text-to-text
-    - multimodal
-    - vision-language
-    - LLM
-    - MLX
- &mlx-audio
-  name: "mlx-audio"
-  uri: "quay.io/go-skynet/local-ai-backends:latest-metal-darwin-arm64-mlx-audio"
-  icon: https://avatars.githubusercontent.com/u/102832242?s=200&v=4
-  urls:
-    - https://github.com/Blaizzy/mlx-audio
-  mirrors:
-    - localai/localai-backends:latest-metal-darwin-arm64-mlx-audio
-  license: MIT
-  description: |
-      Run Audio Models with MLX
-  tags:
-    - audio-to-text
-    - audio-generation
-    - text-to-audio
-    - LLM
-    - MLX
 - &rerankers
  name: "rerankers"
  alias: "rerankers"
@@ -219,8 +169,6 @@
    intel: "intel-diffusers"
    amd: "rocm-diffusers"
    nvidia-l4t: "nvidia-l4t-diffusers"
-    metal: "metal-diffusers"
-    default: "cpu-diffusers"
 - &exllama2
  name: "exllama2"
  urls:
@@ -350,8 +298,6 @@
  alias: "chatterbox"
  capabilities:
    nvidia: "cuda12-chatterbox"
-    metal: "metal-chatterbox"
-    default: "cpu-chatterbox"
 - &piper
  name: "piper"
  uri: "quay.io/go-skynet/local-ai-backends:latest-piper"
@@ -425,21 +371,6 @@
    - text-to-speech
    - TTS
  license: apache-2.0
- !!merge <<: *mlx
-  name: "mlx-development"
-  uri: "quay.io/go-skynet/local-ai-backends:master-metal-darwin-arm64-mlx"
-  mirrors:
-    - localai/localai-backends:master-metal-darwin-arm64-mlx
- !!merge <<: *mlx-vlm
-  name: "mlx-vlm-development"
-  uri: "quay.io/go-skynet/local-ai-backends:master-metal-darwin-arm64-mlx-vlm"
-  mirrors:
-    - localai/localai-backends:master-metal-darwin-arm64-mlx-vlm
- !!merge <<: *mlx-audio
-  name: "mlx-audio-development"
-  uri: "quay.io/go-skynet/local-ai-backends:master-metal-darwin-arm64-mlx-audio"
-  mirrors:
-    - localai/localai-backends:master-metal-darwin-arm64-mlx-audio
 - !!merge <<: *kitten-tts
  name: "kitten-tts-development"
  uri: "quay.io/go-skynet/local-ai-backends:master-kitten-tts"
@@ -582,16 +513,6 @@
  uri: "quay.io/go-skynet/local-ai-backends:latest-cpu-whisper"
  mirrors:
    - localai/localai-backends:latest-cpu-whisper
- !!merge <<: *whispercpp
-  name: "metal-whisper"
-  uri: "quay.io/go-skynet/local-ai-backends:latest-metal-darwin-arm64-whisper"
-  mirrors:
-    - localai/localai-backends:latest-metal-darwin-arm64-whisper
- !!merge <<: *whispercpp
-  name: "metal-whisper-development"
-  uri: "quay.io/go-skynet/local-ai-backends:master-metal-darwin-arm64-whisper"
-  mirrors:
-    - localai/localai-backends:master-metal-darwin-arm64-whisper
 - !!merge <<: *whispercpp
  name: "cpu-whisper-development"
  uri: "quay.io/go-skynet/local-ai-backends:master-cpu-whisper"
@@ -678,16 +599,6 @@
  uri: "quay.io/go-skynet/local-ai-backends:master-cpu-stablediffusion-ggml"
  mirrors:
    - localai/localai-backends:master-cpu-stablediffusion-ggml
- !!merge <<: *stablediffusionggml
-  name: "metal-stablediffusion-ggml"
-  uri: "quay.io/go-skynet/local-ai-backends:latest-metal-darwin-arm64-stablediffusion-ggml"
-  mirrors:
-    - localai/localai-backends:latest-metal-darwin-arm64-stablediffusion-ggml
- !!merge <<: *stablediffusionggml
-  name: "metal-stablediffusion-ggml-development"
-  uri: "quay.io/go-skynet/local-ai-backends:master-metal-darwin-arm64-stablediffusion-ggml"
-  mirrors:
-    - localai/localai-backends:master-metal-darwin-arm64-stablediffusion-ggml
 - !!merge <<: *stablediffusionggml
  name: "vulkan-stablediffusion-ggml"
  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-vulkan-stablediffusion-ggml"
@@ -944,18 +855,6 @@
    intel: "intel-diffusers-development"
    amd: "rocm-diffusers-development"
    nvidia-l4t: "nvidia-l4t-diffusers-development"
-    metal: "metal-diffusers-development"
-    default: "cpu-diffusers-development"
- !!merge <<: *diffusers
-  name: "cpu-diffusers"
-  uri: "quay.io/go-skynet/local-ai-backends:latest-cpu-diffusers"
-  mirrors:
-    - localai/localai-backends:latest-cpu-diffusers
- !!merge <<: *diffusers
-  name: "cpu-diffusers-development"
-  uri: "quay.io/go-skynet/local-ai-backends:master-cpu-diffusers"
-  mirrors:
-    - localai/localai-backends:master-cpu-diffusers
 - !!merge <<: *diffusers
  name: "nvidia-l4t-diffusers"
  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-l4t-diffusers"
@@ -1006,16 +905,6 @@
  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-intel-diffusers"
  mirrors:
    - localai/localai-backends:master-gpu-intel-diffusers
- !!merge <<: *diffusers
-  name: "metal-diffusers"
-  uri: "quay.io/go-skynet/local-ai-backends:latest-metal-darwin-arm64-diffusers"
-  mirrors:
-    - localai/localai-backends:latest-metal-darwin-arm64-diffusers
- !!merge <<: *diffusers
-  name: "metal-diffusers-development"
-  uri: "quay.io/go-skynet/local-ai-backends:master-metal-darwin-arm64-diffusers"
-  mirrors:
-    - localai/localai-backends:master-metal-darwin-arm64-diffusers
  ## exllama2
 - !!merge <<: *exllama2
  name: "exllama2-development"
@@ -1225,28 +1114,6 @@
  name: "chatterbox-development"
  capabilities:
    nvidia: "cuda12-chatterbox-development"
-    metal: "metal-chatterbox-development"
-    default: "cpu-chatterbox-development"
- !!merge <<: *chatterbox
-  name: "cpu-chatterbox"
-  uri: "quay.io/go-skynet/local-ai-backends:latest-cpu-chatterbox"
-  mirrors:
-    - localai/localai-backends:latest-cpu-chatterbox
- !!merge <<: *chatterbox
-  name: "cpu-chatterbox-development"
-  uri: "quay.io/go-skynet/local-ai-backends:master-cpu-chatterbox"
-  mirrors:
-    - localai/localai-backends:master-cpu-chatterbox
- !!merge <<: *chatterbox
-  name: "metal-chatterbox"
-  uri: "quay.io/go-skynet/local-ai-backends:latest-metal-darwin-arm64-chatterbox"
-  mirrors:
-    - localai/localai-backends:latest-metal-darwin-arm64-chatterbox
- !!merge <<: *chatterbox
-  name: "metal-chatterbox-development"
-  uri: "quay.io/go-skynet/local-ai-backends:master-metal-darwin-arm64-chatterbox"
-  mirrors:
-    - localai/localai-backends:master-metal-darwin-arm64-chatterbox
 - !!merge <<: *chatterbox
  name: "cuda12-chatterbox-development"
  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-12-chatterbox"
--- a/backend/python/README.md
+++ b/backend/python/README.md
@@ -1,190 +1,38 @@
-# Python Backends for LocalAI
+# Common commands about conda environment

-This directory contains Python-based AI backends for LocalAI, providing support for various AI models and hardware acceleration targets.
+## Create a new empty conda environment

-## Overview
+```
+conda create --name <env-name> python=<your version> -y

-The Python backends use a unified build system based on `libbackend.sh` that provides:
- **Automatic virtual environment management** with support for both `uv` and `pip`
- **Hardware-specific dependency installation** (CPU, CUDA, Intel, MLX, etc.)
- **Portable Python support** for standalone deployments
- **Consistent backend execution** across different environments
-
-## Available Backends
-
-### Core AI Models
- **transformers** - Hugging Face Transformers framework (PyTorch-based)
- **vllm** - High-performance LLM inference engine
- **mlx** - Apple Silicon optimized ML framework
- **exllama2** - ExLlama2 quantized models
-
-### Audio & Speech
- **bark** - Text-to-speech synthesis
- **coqui** - Coqui TTS models
- **faster-whisper** - Fast Whisper speech recognition
- **kitten-tts** - Lightweight TTS
- **mlx-audio** - Apple Silicon audio processing
- **chatterbox** - TTS model
- **kokoro** - TTS models
-
-### Computer Vision
- **diffusers** - Stable Diffusion and image generation
- **mlx-vlm** - Vision-language models for Apple Silicon
- **rfdetr** - Object detection models
-
-### Specialized
-
- **rerankers** - Text reranking models
-
-## Quick Start
-
-### Prerequisites
- Python 3.10+ (default: 3.10.18)
- `uv` package manager (recommended) or `pip`
- Appropriate hardware drivers for your target (CUDA, Intel, etc.)
-
-### Installation
-
-Each backend can be installed individually:
-
-```bash
-# Navigate to a specific backend
-cd backend/python/transformers
-
-# Install dependencies
-make transformers
-# or
-bash install.sh
-
-# Run the backend
-make run
-# or
-bash run.sh
+conda create --name autogptq python=3.11 -y
 ```

-### Using the Unified Build System
+## To activate the environment

-The `libbackend.sh` script provides consistent commands across all backends:
-
-```bash
-# Source the library in your backend script
-source $(dirname $0)/../common/libbackend.sh
-
-# Install requirements (automatically handles hardware detection)
-installRequirements
-
-# Start the backend server
-startBackend $@
-
-# Run tests
-runUnittests
+As of conda 4.4
+```
+conda activate autogptq
 ```

-## Hardware Targets
+The conda version older than 4.4

-The build system automatically detects and configures for different hardware:
-
- **CPU** - Standard CPU-only builds
- **CUDA** - NVIDIA GPU acceleration (supports CUDA 11/12)
- **Intel** - Intel XPU/GPU optimization
- **MLX** - Apple Silicon (M1/M2/M3) optimization
- **HIP** - AMD GPU acceleration
-
-### Target-Specific Requirements
-
-Backends can specify hardware-specific dependencies:
- `requirements.txt` - Base requirements
- `requirements-cpu.txt` - CPU-specific packages
- `requirements-cublas11.txt` - CUDA 11 packages
- `requirements-cublas12.txt` - CUDA 12 packages
- `requirements-intel.txt` - Intel-optimized packages
- `requirements-mps.txt` - Apple Silicon packages
-
-## Configuration Options
-
-### Environment Variables
-
- `PYTHON_VERSION` - Python version (default: 3.10)
- `PYTHON_PATCH` - Python patch version (default: 18)
- `BUILD_TYPE` - Force specific build target
- `USE_PIP` - Use pip instead of uv (default: false)
- `PORTABLE_PYTHON` - Enable portable Python builds
- `LIMIT_TARGETS` - Restrict backend to specific targets
-
-### Example: CUDA 12 Only Backend
-
-```bash
-# In your backend script
-LIMIT_TARGETS="cublas12"
-source $(dirname $0)/../common/libbackend.sh
+```
+source activate autogptq
 ```

-### Example: Intel-Optimized Backend
+## Install the packages to your environment

-```bash
-# In your backend script
-LIMIT_TARGETS="intel"
-source $(dirname $0)/../common/libbackend.sh
+Sometimes you need to install the packages from the conda-forge channel
+
+By using `conda`
+```
+conda install <your-package-name>
+
+conda install -c conda-forge <your package-name>
 ```

-## Development
-
-### Adding a New Backend
-
-1. Create a new directory in `backend/python/`
-2. Copy the template structure from `common/template/`
-3. Implement your `backend.py` with the required gRPC interface
-4. Add appropriate requirements files for your target hardware
-5. Use `libbackend.sh` for consistent build and execution
-
-### Testing
-
-```bash
-# Run backend tests
-make test
-# or
-bash test.sh
+Or by using `pip`
 ```
-
-### Building
-
-```bash
-# Install dependencies
-make <backend-name>
-
-# Clean build artifacts
-make clean
+pip install <your-package-name>
 ```
-
-## Architecture
-
-Each backend follows a consistent structure:
-```
-backend-name/
-├── backend.py          # Main backend implementation
-├── requirements.txt    # Base dependencies
-├── requirements-*.txt  # Hardware-specific dependencies
-├── install.sh         # Installation script
-├── run.sh            # Execution script
-├── test.sh           # Test script
-├── Makefile          # Build targets
-└── test.py           # Unit tests
-```
-
-## Troubleshooting
-
-### Common Issues
-
-1. **Missing dependencies**: Ensure all requirements files are properly configured
-2. **Hardware detection**: Check that `BUILD_TYPE` matches your system
-3. **Python version**: Verify Python 3.10+ is available
-4. **Virtual environment**: Use `ensureVenv` to create/activate environments
-
-## Contributing
-
-When adding new backends or modifying existing ones:
-1. Follow the established directory structure
-2. Use `libbackend.sh` for consistent behavior
-3. Include appropriate requirements files for all target hardware
-4. Add comprehensive tests
-5. Update this README if adding new backend types
--- a/backend/python/bark/Makefile
+++ b/backend/python/bark/Makefile
@@ -1,23 +1,29 @@
 .PHONY: ttsbark
-ttsbark:
+ttsbark: protogen
 	bash install.sh

 .PHONY: run
-run: ttsbark
+run: protogen
 	@echo "Running bark..."
 	bash run.sh
 	@echo "bark run."

 .PHONY: test
-test: ttsbark
+test: protogen
 	@echo "Testing bark..."
 	bash test.sh
 	@echo "bark tested."

+.PHONY: protogen
+protogen: backend_pb2_grpc.py backend_pb2.py
+
 .PHONY: protogen-clean
 protogen-clean:
 	$(RM) backend_pb2_grpc.py backend_pb2.py

+backend_pb2_grpc.py backend_pb2.py:
+	python3 -m grpc_tools.protoc -I../.. -I./ --python_out=. --grpc_python_out=. backend.proto
+
 .PHONY: clean
 clean: protogen-clean
 	rm -rf venv __pycache__
--- a/backend/python/bark/requirements-intel.txt
+++ b/backend/python/bark/requirements-intel.txt
@@ -1,5 +1,5 @@
 --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
-intel-extension-for-pytorch==2.8.10+xpu
+intel-extension-for-pytorch==2.3.110+xpu
 torch==2.3.1+cxx11.abi
 torchaudio==2.3.1+cxx11.abi
 oneccl_bind_pt==2.3.100+xpu
--- a/backend/python/bark/requirements.txt
+++ b/backend/python/bark/requirements.txt
@@ -1,4 +1,4 @@
 bark==0.1.5
-grpcio==1.74.0
+grpcio==1.71.0
 protobuf
 certifi
--- a/backend/python/chatterbox/Makefile
+++ b/backend/python/chatterbox/Makefile
@@ -1,23 +1,29 @@
-.PHONY: chatterbox
-chatterbox:
+.PHONY: coqui
+coqui: protogen
 	bash install.sh

 .PHONY: run
-run: chatterbox
+run: protogen
 	@echo "Running coqui..."
 	bash run.sh
 	@echo "coqui run."

 .PHONY: test
-test: chatterbox
+test: protogen
 	@echo "Testing coqui..."
 	bash test.sh
 	@echo "coqui tested."

+.PHONY: protogen
+protogen: backend_pb2_grpc.py backend_pb2.py
+
 .PHONY: protogen-clean
 protogen-clean:
 	$(RM) backend_pb2_grpc.py backend_pb2.py

+backend_pb2_grpc.py backend_pb2.py:
+	python3 -m grpc_tools.protoc -I../.. -I./ --python_out=. --grpc_python_out=. backend.proto
+
 .PHONY: clean
 clean: protogen-clean
 	rm -rf venv __pycache__
--- a/backend/python/chatterbox/backend.py
+++ b/backend/python/chatterbox/backend.py
@@ -41,9 +41,7 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
        else:
            print("CUDA is not available", file=sys.stderr)
            device = "cpu"
-        mps_available = hasattr(torch.backends, "mps") and torch.backends.mps.is_available()
-        if mps_available:
-            device = "mps"
+
        if not torch.cuda.is_available() and request.CUDA:
            return backend_pb2.Result(success=False, message="CUDA is not available")

--- a/backend/python/chatterbox/requirements-cpu.txt
+++ b/backend/python/chatterbox/requirements-cpu.txt
@@ -1,6 +1,5 @@
--extra-index-url https://download.pytorch.org/whl/cpu
 accelerate
 torch==2.6.0
 torchaudio==2.6.0
 transformers==4.46.3
-chatterbox-tts==0.1.2
+chatterbox-tts
--- a/backend/python/chatterbox/requirements-cublas11.txt
+++ b/backend/python/chatterbox/requirements-cublas11.txt
@@ -2,5 +2,5 @@
 torch==2.6.0+cu118
 torchaudio==2.6.0+cu118
 transformers==4.46.3
-chatterbox-tts==0.1.2
+chatterbox-tts
 accelerate
--- a/backend/python/chatterbox/requirements-cublas12.txt
+++ b/backend/python/chatterbox/requirements-cublas12.txt
@@ -1,5 +1,5 @@
 torch==2.6.0
 torchaudio==2.6.0
 transformers==4.46.3
-chatterbox-tts==0.1.2
+chatterbox-tts
 accelerate
--- a/backend/python/chatterbox/requirements-hipblas.txt
+++ b/backend/python/chatterbox/requirements-hipblas.txt
@@ -2,5 +2,5 @@
 torch==2.6.0+rocm6.1
 torchaudio==2.6.0+rocm6.1
 transformers==4.46.3
-chatterbox-tts==0.1.2
+chatterbox-tts
 accelerate
--- a/backend/python/chatterbox/requirements-intel.txt
+++ b/backend/python/chatterbox/requirements-intel.txt
@@ -3,8 +3,9 @@ intel-extension-for-pytorch==2.3.110+xpu
 torch==2.3.1+cxx11.abi
 torchaudio==2.3.1+cxx11.abi
 transformers==4.46.3
-chatterbox-tts==0.1.2
+chatterbox-tts
 accelerate
 oneccl_bind_pt==2.3.100+xpu
 optimum[openvino]
-setuptools
+setuptools
+accelerate
--- a/backend/python/common/libbackend.sh
+++ b/backend/python/common/libbackend.sh
@@ -1,6 +1,6 @@
-#!/usr/bin/env bash
-set -euo pipefail

+
+# init handles the setup of the library
 # 
 # use the library by adding the following line to a script:
 # source $(dirname $0)/../common/libbackend.sh
@@ -17,182 +17,29 @@ set -euo pipefail
 # LIMIT_TARGETS="cublas12"
 # source $(dirname $0)/../common/libbackend.sh
 #
-# You can switch between uv (conda-like) and pip installation methods by setting USE_PIP:
-# USE_PIP=true source $(dirname $0)/../common/libbackend.sh
-#
-# ===================== user-configurable defaults =====================
-PYTHON_VERSION="${PYTHON_VERSION:-3.10}"      # e.g. 3.10 / 3.11 / 3.12 / 3.13
-PYTHON_PATCH="${PYTHON_PATCH:-18}"            # e.g. 18 -> 3.10.18 ; 13 -> 3.11.13
-PY_STANDALONE_TAG="${PY_STANDALONE_TAG:-20250818}"  # release tag date
-# Enable/disable bundling of a portable Python build
-PORTABLE_PYTHON="${PORTABLE_PYTHON:-false}"

-# If you want to fully pin the filename (including tuned CPU targets), set:
-# PORTABLE_PY_FILENAME="cpython-3.10.18+20250818-x86_64_v3-unknown-linux-gnu-install_only.tar.gz"
-: "${PORTABLE_PY_FILENAME:=}"
-: "${PORTABLE_PY_SHA256:=}"  # optional; if set we verify the download
-# =====================================================================
+PYTHON_VERSION="3.10"

-# Default to uv if USE_PIP is not set
-if [ "x${USE_PIP:-}" == "x" ]; then
-    USE_PIP=false
-fi
-
-# ----------------------- helpers -----------------------
-function _is_musl() {
-    # detect musl (Alpine, etc)
-    if command -v ldd >/dev/null 2>&1; then
-        ldd --version 2>&1 | grep -qi musl && return 0
-    fi
-    # busybox-ish fallback
-    if command -v getconf >/dev/null 2>&1; then
-        getconf GNU_LIBC_VERSION >/dev/null 2>&1 || return 0
-    fi
-    return 1
-}
-
-function _triple() {
-    local os="" arch="" libc="gnu"
-    case "$(uname -s)" in
-        Linux*)  os="unknown-linux" ;;
-        Darwin*) os="apple-darwin" ;;
-        MINGW*|MSYS*|CYGWIN*) os="pc-windows-msvc" ;;  # best-effort for Git Bash
-        *) echo "Unsupported OS $(uname -s)"; exit 1;;
-    esac
-
-    case "$(uname -m)" in
-        x86_64) arch="x86_64" ;;
-        aarch64|arm64) arch="aarch64" ;;
-        armv7l) arch="armv7" ;;
-        i686|i386) arch="i686" ;;
-        ppc64le) arch="ppc64le" ;;
-        s390x) arch="s390x" ;;
-        riscv64) arch="riscv64" ;;
-        *) echo "Unsupported arch $(uname -m)"; exit 1;;
-    esac
-
-    if [[ "$os" == "unknown-linux" ]]; then
-        if _is_musl; then
-            libc="musl"
-        else
-            libc="gnu"
-        fi
-        echo "${arch}-${os}-${libc}"
-    else
-        echo "${arch}-${os}"
-    fi
-}
-
-function _portable_dir() {
-    echo "${EDIR}/python"
-}
-
-function _portable_bin() {
-    # python-build-standalone puts python in ./bin
-    echo "$(_portable_dir)/bin"
-}
-
-function _portable_python() {
-    if [ -x "$(_portable_bin)/python3" ]; then
-        echo "$(_portable_bin)/python3"
-    else
-        echo "$(_portable_bin)/python"
-    fi
-}
-
-
-# macOS loader env for the portable CPython
-_macosPortableEnv() {
-  if [ "$(uname -s)" = "Darwin" ]; then
-    export DYLD_LIBRARY_PATH="$(_portable_dir)/lib${DYLD_LIBRARY_PATH:+:${DYLD_LIBRARY_PATH}}"
-    export DYLD_FALLBACK_LIBRARY_PATH="$(_portable_dir)/lib${DYLD_FALLBACK_LIBRARY_PATH:+:${DYLD_FALLBACK_LIBRARY_PATH}}"
-  fi
-}
-
-# Good hygiene on macOS for downloaded/extracted trees
-_unquarantinePortablePython() {
-  if [ "$(uname -s)" = "Darwin" ]; then
-    command -v xattr >/dev/null 2>&1 && xattr -dr com.apple.quarantine "$(_portable_dir)" || true
-  fi
-}
-
-# ------------------ ### PORTABLE PYTHON ------------------
-function ensurePortablePython() {
-    local pdir="$(_portable_dir)"
-    local pbin="$(_portable_bin)"
-    local pyexe
-
-    if [ -x "${pbin}/python3" ] || [ -x "${pbin}/python" ]; then
-        _macosPortableEnv
-        return 0
-    fi
-
-    mkdir -p "${pdir}"
-    local triple="$(_triple)"
-
-    local full_ver="${PYTHON_VERSION}.${PYTHON_PATCH}"
-    local fn=""
-    if [ -n "${PORTABLE_PY_FILENAME}" ]; then
-        fn="${PORTABLE_PY_FILENAME}"
-    else
-        # generic asset name: cpython-<full_ver>+<tag>-<triple>-install_only.tar.gz
-        fn="cpython-${full_ver}+${PY_STANDALONE_TAG}-${triple}-install_only.tar.gz"
-    fi
-
-    local url="https://github.com/astral-sh/python-build-standalone/releases/download/${PY_STANDALONE_TAG}/${fn}"
-    local tmp="${pdir}/${fn}"
-    echo "Downloading portable Python: ${fn}"
-    # curl with retries; fall back to wget if needed
-    if command -v curl >/dev/null 2>&1; then
-        curl -L --fail --retry 3 --retry-delay 1 -o "${tmp}" "${url}"
-    else
-        wget -O "${tmp}" "${url}"
-    fi
-
-    if [ -n "${PORTABLE_PY_SHA256}" ]; then
-        echo "${PORTABLE_PY_SHA256}  ${tmp}" | sha256sum -c -
-    fi
-
-    echo "Extracting ${fn} -> ${pdir}"
-    # always a .tar.gz (we purposely choose install_only)
-    tar -xzf "${tmp}" -C "${pdir}"
-    rm -f "${tmp}"
-
-    # Some archives nest a directory; if so, flatten to ${pdir}
-    # Find the first dir with a 'bin/python*'
-    local inner
-    inner="$(find "${pdir}" -type f -path "*/bin/python*" -maxdepth 3 2>/dev/null | head -n1 || true)"
-    if [ -n "${inner}" ]; then
-        local inner_root
-        inner_root="$(dirname "$(dirname "${inner}")")" # .../bin -> root
-        if [ "${inner_root}" != "${pdir}" ]; then
-            # move contents up one level
-            shopt -s dotglob
-            mv "${inner_root}/"* "${pdir}/"
-            rm -rf "${inner_root}"
-            shopt -u dotglob
-        fi
-    fi
-
-    _unquarantinePortablePython
-    _macosPortableEnv
-    # Make sure it's runnable
-    pyexe="$(_portable_python)"
-    "${pyexe}" -V
-}
-
-# init handles the setup of the library
 function init() {
+    # Name of the backend (directory name)
    BACKEND_NAME=${PWD##*/}
-    MY_DIR=$(realpath "$(dirname "$0")")
+
+    # Path where all backends files are
+    MY_DIR=$(realpath `dirname $0`)
+
+    # Build type
    BUILD_PROFILE=$(getBuildProfile)

+    # Environment directory
    EDIR=${MY_DIR}
-    if [ "x${ENV_DIR:-}" != "x" ]; then
+
+    # Allow to specify a custom env dir for shared environments
+    if [ "x${ENV_DIR}" != "x" ]; then
        EDIR=${ENV_DIR}
    fi

-    if [ ! -z "${LIMIT_TARGETS:-}" ]; then
+    # If a backend has defined a list of valid build profiles...
+    if [ ! -z "${LIMIT_TARGETS}" ]; then
        isValidTarget=$(checkTargets ${LIMIT_TARGETS})
        if [ ${isValidTarget} != true ]; then
            echo "${BACKEND_NAME} can only be used on the following targets: ${LIMIT_TARGETS}"
@@ -203,7 +50,6 @@ function init() {
    echo "Initializing libbackend for ${BACKEND_NAME}"
 }

-
 # getBuildProfile will inspect the system to determine which build profile is appropriate:
 # returns one of the following:
 # - cublas11
@@ -211,140 +57,58 @@ function init() {
 # - hipblas
 # - intel
 function getBuildProfile() {
-    if [ x"${BUILD_TYPE:-}" == "xcublas" ]; then
-        if [ ! -z "${CUDA_MAJOR_VERSION:-}" ]; then
+    if [ "x${BUILD_TYPE}" == "xl4t" ]; then
+        echo "l4t"
+        return 0
+    fi
+
+    # First check if we are a cublas build, and if so report the correct build profile
+    if [ x"${BUILD_TYPE}" == "xcublas" ]; then
+        if [ ! -z ${CUDA_MAJOR_VERSION} ]; then
+            # If we have been given a CUDA version, we trust it
            echo ${BUILD_TYPE}${CUDA_MAJOR_VERSION}
        else
+            # We don't know what version of cuda we are, so we report ourselves as a generic cublas
            echo ${BUILD_TYPE}
        fi
        return 0
    fi

+    # If /opt/intel exists, then we are doing an intel/ARC build
    if [ -d "/opt/intel" ]; then
        echo "intel"
        return 0
    fi

-    if [ -n "${BUILD_TYPE:-}" ]; then
+    # If for any other values of BUILD_TYPE, we don't need any special handling/discovery
+    if [ ! -z ${BUILD_TYPE} ]; then
        echo ${BUILD_TYPE}
        return 0
    fi

+    # If there is no BUILD_TYPE set at all, set a build-profile value of CPU, we aren't building for any GPU targets
    echo "cpu"
 }

-
-# Make the venv relocatable:
-# - rewrite venv/bin/python{,3} to relative symlinks into $(_portable_dir)
-# - normalize entrypoint shebangs to /usr/bin/env python3
-_makeVenvPortable() {
-    local venv_dir="${EDIR}/venv"
-    local vbin="${venv_dir}/bin"
-
-    [ -d "${vbin}" ] || return 0
-
-    # 1) Replace python symlinks with relative ones to ../../python/bin/python3
-    #    (venv/bin -> venv -> EDIR -> python/bin)
-    local rel_py='../../python/bin/python3'
-
-    for name in python3 python; do
-        if [ -e "${vbin}/${name}" ] || [ -L "${vbin}/${name}" ]; then
-            rm -f "${vbin}/${name}"
-        fi
-    done
-    ln -s "${rel_py}" "${vbin}/python3"
-    ln -s "python3" "${vbin}/python"
-
-    # 2) Rewrite shebangs of entry points to use env, so the venv is relocatable
-    #    Only touch text files that start with #! and reference the current venv.
-    local ve_abs="${vbin}/python"
-    local sed_i=(sed -i)
-    # macOS/BSD sed needs a backup suffix; GNU sed doesn't. Make it portable:
-    if sed --version >/dev/null 2>&1; then
-        sed_i=(sed -i)
-    else
-        sed_i=(sed -i '')
-    fi
-
-    for f in "${vbin}"/*; do
-        [ -f "$f" ] || continue
-        # Fast path: check first two bytes (#!)
-        head -c2 "$f" 2>/dev/null | grep -q '^#!' || continue
-        # Only rewrite if the shebang mentions the (absolute) venv python
-        if head -n1 "$f" | grep -Fq "${ve_abs}"; then
-            "${sed_i[@]}" '1s|^#!.*$|#!/usr/bin/env python3|' "$f"
-            chmod +x "$f" 2>/dev/null || true
-        fi
-    done
-}
-
-
 # ensureVenv makes sure that the venv for the backend both exists, and is activated.
 #
 # This function is idempotent, so you can call it as many times as you want and it will
 # always result in an activated virtual environment
 function ensureVenv() {
-    local interpreter=""
-
-    if [ "x${PORTABLE_PYTHON}" == "xtrue" ] || [ -e "$(_portable_python)" ]; then
-        echo "Using portable Python"
-        ensurePortablePython
-        interpreter="$(_portable_python)"
-    else
-        # Prefer system python${PYTHON_VERSION}, else python3, else fall back to bundled
-        if command -v python${PYTHON_VERSION} >/dev/null 2>&1; then
-            interpreter="python${PYTHON_VERSION}"
-        elif command -v python3 >/dev/null 2>&1; then
-            interpreter="python3"
-        else
-            echo "No suitable system Python found, bootstrapping portable build..."
-            ensurePortablePython
-            interpreter="$(_portable_python)"
-        fi
-    fi
-
    if [ ! -d "${EDIR}/venv" ]; then
-        if [ "x${USE_PIP}" == "xtrue" ]; then
-            "${interpreter}" -m venv --copies "${EDIR}/venv"
-            source "${EDIR}/venv/bin/activate"
-            "${interpreter}" -m pip install --upgrade pip
-        else
-            if [ "x${PORTABLE_PYTHON}" == "xtrue" ]; then
-                uv venv --python "${interpreter}" "${EDIR}/venv"
-            else
-                uv venv --python "${PYTHON_VERSION}" "${EDIR}/venv"
-            fi
-        fi
-        if [ "x${PORTABLE_PYTHON}" == "xtrue" ]; then
-            _makeVenvPortable
-        fi
+        uv venv --python ${PYTHON_VERSION} ${EDIR}/venv
+        echo "virtualenv created"
    fi

-    # We call it here to make sure that when we source a venv we can still use python as expected
-    if [ -x "$(_portable_python)" ]; then
-        _macosPortableEnv
+    # Source if we are not already in a Virtual env
+    if [ "x${VIRTUAL_ENV}" != "x${EDIR}/venv" ]; then
+        source ${EDIR}/venv/bin/activate
+        echo "virtualenv activated"
    fi

-    if [ "x${VIRTUAL_ENV:-}" != "x${EDIR}/venv" ]; then
-        source "${EDIR}/venv/bin/activate"
-    fi
+    echo "activated virtualenv has been ensured"
 }

-
-function runProtogen() {
-    ensureVenv
-    if [ "x${USE_PIP}" == "xtrue" ]; then
-        pip install grpcio-tools
-    else
-        uv pip install grpcio-tools
-    fi
-    pushd "${EDIR}" >/dev/null
-        # use the venv python (ensures correct interpreter & sys.path)
-        python -m grpc_tools.protoc -I../../ -I./ --python_out=. --grpc_python_out=. backend.proto
-    popd >/dev/null
-}
-
-
 # installRequirements looks for several requirements files and if they exist runs the install for them in order
 #
 #  - requirements-install.txt
@@ -352,7 +116,7 @@ function runProtogen() {
 #  - requirements-${BUILD_TYPE}.txt
 #  - requirements-${BUILD_PROFILE}.txt
 #
-# BUILD_PROFILE is a more specific version of BUILD_TYPE, ex: cuda-11 or cuda-12
+# BUILD_PROFILE is a pore specific version of BUILD_TYPE, ex: cuda-11 or cuda-12
 # it can also include some options that we do not have BUILD_TYPES for, ex: intel
 #
 # NOTE: for BUILD_PROFILE==intel, this function does NOT automatically use the Intel python package index.
@@ -368,41 +132,36 @@ function runProtogen() {
 # installRequirements
 function installRequirements() {
    ensureVenv
+
+    # These are the requirements files we will attempt to install, in order
    declare -a requirementFiles=(
        "${EDIR}/requirements-install.txt"
        "${EDIR}/requirements.txt"
-        "${EDIR}/requirements-${BUILD_TYPE:-}.txt"
+        "${EDIR}/requirements-${BUILD_TYPE}.txt"
    )

-    if [ "x${BUILD_TYPE:-}" != "x${BUILD_PROFILE}" ]; then
+    if [ "x${BUILD_TYPE}" != "x${BUILD_PROFILE}" ]; then
        requirementFiles+=("${EDIR}/requirements-${BUILD_PROFILE}.txt")
    fi
-    if [ "x${BUILD_TYPE:-}" == "x" ]; then
+
+    # if BUILD_TYPE is empty, we are a CPU build, so we should try to install the CPU requirements
+    if [ "x${BUILD_TYPE}" == "x" ]; then
        requirementFiles+=("${EDIR}/requirements-cpu.txt")
    fi
+
    requirementFiles+=("${EDIR}/requirements-after.txt")
-    if [ "x${BUILD_TYPE:-}" != "x${BUILD_PROFILE}" ]; then
+
+    if [ "x${BUILD_TYPE}" != "x${BUILD_PROFILE}" ]; then
        requirementFiles+=("${EDIR}/requirements-${BUILD_PROFILE}-after.txt")
    fi

-    # This is needed to build wheels that e.g. depends on Python.h
-    if [ "x${PORTABLE_PYTHON}" == "xtrue" ]; then
-        export C_INCLUDE_PATH="${C_INCLUDE_PATH:-}:$(_portable_dir)/include/python${PYTHON_VERSION}"
-    fi
-
    for reqFile in ${requirementFiles[@]}; do
-        if [ -f "${reqFile}" ]; then
+        if [ -f ${reqFile} ]; then
            echo "starting requirements install for ${reqFile}"
-            if [ "x${USE_PIP}" == "xtrue" ]; then
-                pip install ${EXTRA_PIP_INSTALL_FLAGS:-} --requirement "${reqFile}"
-            else
-                uv pip install ${EXTRA_PIP_INSTALL_FLAGS:-} --requirement "${reqFile}"
-            fi
+            uv pip install ${EXTRA_PIP_INSTALL_FLAGS} --requirement ${reqFile}
            echo "finished requirements install for ${reqFile}"
        fi
    done
-
-    runProtogen
 }

 # startBackend discovers and runs the backend GRPC server
@@ -420,18 +179,18 @@ function installRequirements() {
 #  - ${BACKEND_NAME}.py
 function startBackend() {
    ensureVenv
-    if [ ! -z "${BACKEND_FILE:-}" ]; then
-        exec "${EDIR}/venv/bin/python" "${BACKEND_FILE}" "$@"
+
+    if [ ! -z ${BACKEND_FILE} ]; then
+        exec ${EDIR}/venv/bin/python ${BACKEND_FILE} $@
    elif [ -e "${MY_DIR}/server.py" ]; then
-        exec "${EDIR}/venv/bin/python" "${MY_DIR}/server.py" "$@"
+        exec ${EDIR}/venv/bin/python ${MY_DIR}/server.py $@
    elif [ -e "${MY_DIR}/backend.py" ]; then
-        exec "${EDIR}/venv/bin/python" "${MY_DIR}/backend.py" "$@"
+        exec ${EDIR}/venv/bin/python ${MY_DIR}/backend.py $@
    elif [ -e "${MY_DIR}/${BACKEND_NAME}.py" ]; then
-        exec "${EDIR}/venv/bin/python" "${MY_DIR}/${BACKEND_NAME}.py" "$@"
+        exec ${EDIR}/venv/bin/python ${MY_DIR}/${BACKEND_NAME}.py $@
    fi
 }

-
 # runUnittests discovers and runs python unittests
 #
 # You can specify a specific test file to use by setting TEST_FILE before calling runUnittests.
@@ -444,36 +203,41 @@ function startBackend() {
 # be default a file named test.py in the backends directory will be used
 function runUnittests() {
    ensureVenv
-    if [ ! -z "${TEST_FILE:-}" ]; then
-        testDir=$(dirname "$(realpath "${TEST_FILE}")")
-        testFile=$(basename "${TEST_FILE}")
-        pushd "${testDir}" >/dev/null
-        python -m unittest "${testFile}"
-        popd >/dev/null
+
+    if [ ! -z ${TEST_FILE} ]; then
+        testDir=$(dirname `realpath ${TEST_FILE}`)
+        testFile=$(basename ${TEST_FILE})
+        pushd ${testDir}
+        python -m unittest ${testFile}
+        popd
    elif [ -f "${MY_DIR}/test.py" ]; then
-        pushd "${MY_DIR}" >/dev/null
+        pushd ${MY_DIR}
        python -m unittest test.py
-        popd >/dev/null
+        popd
    else
        echo "no tests defined for ${BACKEND_NAME}"
    fi
 }

-
 ##################################################################################
 # Below here are helper functions not intended to be used outside of the library #
 ##################################################################################

 # checkTargets determines if the current BUILD_TYPE or BUILD_PROFILE is in a list of valid targets
 function checkTargets() {
+    # Collect all provided targets into a variable and...
    targets=$@
+    # ...convert it into an array
    declare -a targets=($targets)
+
    for target in ${targets[@]}; do
-        if [ "x${BUILD_TYPE:-}" == "x${target}" ]; then
-            echo true; return 0
+        if [ "x${BUILD_TYPE}" == "x${target}" ]; then
+            echo true
+            return 0
        fi
        if [ "x${BUILD_PROFILE}" == "x${target}" ]; then
-            echo true; return 0
+            echo true
+            return 0
        fi
    done
    echo false
--- a/backend/python/common/template/Makefile
+++ b/backend/python/common/template/Makefile
@@ -3,11 +3,18 @@
 .PHONY: install
 install:
 	bash install.sh
+	$(MAKE) protogen
+
+.PHONY: protogen
+protogen: backend_pb2_grpc.py backend_pb2.py

 .PHONY: protogen-clean
 protogen-clean:
 	$(RM) backend_pb2_grpc.py backend_pb2.py

+backend_pb2_grpc.py backend_pb2.py:
+	bash protogen.sh
+
 .PHONY: clean
 clean: protogen-clean
 	rm -rf venv __pycache__
--- a/backend/python/common/template/protogen.sh
+++ b/backend/python/common/template/protogen.sh
@@ -8,4 +8,6 @@ else
    source $backend_dir/../common/libbackend.sh
 fi

-runProtogen
+ensureVenv
+
+python3 -m grpc_tools.protoc -I../.. -I./ --python_out=. --grpc_python_out=. backend.proto
--- a/backend/python/common/template/requirements-intel.txt
+++ b/backend/python/common/template/requirements-intel.txt
@@ -1,5 +1,5 @@
 --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
-intel-extension-for-pytorch==2.8.10+xpu
-torch==2.8.0
-oneccl_bind_pt==2.8.0+xpu
+intel-extension-for-pytorch==2.3.110+xpu
+torch==2.3.1+cxx11.abi
+oneccl_bind_pt==2.3.100+xpu
 optimum[openvino]
--- a/backend/python/common/template/requirements.txt
+++ b/backend/python/common/template/requirements.txt
@@ -1,3 +1,3 @@
-grpcio==1.74.0
+grpcio==1.71.0
 protobuf
 grpcio-tools
--- a/backend/python/coqui/Makefile
+++ b/backend/python/coqui/Makefile
@@ -1,23 +1,29 @@
 .PHONY: coqui
-coqui:
+coqui: protogen
 	bash install.sh

 .PHONY: run
-run: coqui
+run: protogen
 	@echo "Running coqui..."
 	bash run.sh
 	@echo "coqui run."

 .PHONY: test
-test: coqui
+test: protogen
 	@echo "Testing coqui..."
 	bash test.sh
 	@echo "coqui tested."

+.PHONY: protogen
+protogen: backend_pb2_grpc.py backend_pb2.py
+
 .PHONY: protogen-clean
 protogen-clean:
 	$(RM) backend_pb2_grpc.py backend_pb2.py

+backend_pb2_grpc.py backend_pb2.py:
+	python3 -m grpc_tools.protoc -I../.. -I./ --python_out=. --grpc_python_out=. backend.proto
+
 .PHONY: clean
 clean: protogen-clean
 	rm -rf venv __pycache__
--- a/backend/python/coqui/backend.py
+++ b/backend/python/coqui/backend.py
@@ -40,9 +40,7 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
        else:
            print("CUDA is not available", file=sys.stderr)
            device = "cpu"
-        mps_available = hasattr(torch.backends, "mps") and torch.backends.mps.is_available()
-        if mps_available:
-            device = "mps"
+
        if not torch.cuda.is_available() and request.CUDA:
            return backend_pb2.Result(success=False, message="CUDA is not available")

--- a/backend/python/coqui/requirements.txt
+++ b/backend/python/coqui/requirements.txt
@@ -1,4 +1,4 @@
-grpcio==1.74.0
+grpcio==1.71.0
 protobuf
 certifi
 packaging==24.1
--- a/backend/python/diffusers/Makefile
+++ b/backend/python/diffusers/Makefile
@@ -12,22 +12,28 @@ export SKIP_CONDA=1
 endif

 .PHONY: diffusers
-diffusers:
+diffusers: protogen
 	bash install.sh

 .PHONY: run
-run: diffusers
+run: protogen
 	@echo "Running diffusers..."
 	bash run.sh
 	@echo "Diffusers run."

-test: diffusers
+test: protogen
 	bash test.sh

+.PHONY: protogen
+protogen: backend_pb2_grpc.py backend_pb2.py
+
 .PHONY: protogen-clean
 protogen-clean:
 	$(RM) backend_pb2_grpc.py backend_pb2.py

+backend_pb2_grpc.py backend_pb2.py:
+	python3 -m grpc_tools.protoc -I../.. -I./ --python_out=. --grpc_python_out=. backend.proto
+
 .PHONY: clean
 clean: protogen-clean
 	rm -rf venv __pycache__
--- a/backend/python/diffusers/backend.py
+++ b/backend/python/diffusers/backend.py
@@ -18,7 +18,7 @@ import backend_pb2_grpc
 import grpc

 from diffusers import SanaPipeline, StableDiffusion3Pipeline, StableDiffusionXLPipeline, StableDiffusionDepth2ImgPipeline, DPMSolverMultistepScheduler, StableDiffusionPipeline, DiffusionPipeline, \
-    EulerAncestralDiscreteScheduler, FluxPipeline, FluxTransformer2DModel, QwenImageEditPipeline, AutoencoderKLWan, WanPipeline, WanImageToVideoPipeline
+    EulerAncestralDiscreteScheduler, FluxPipeline, FluxTransformer2DModel
 from diffusers import StableDiffusionImg2ImgPipeline, AutoPipelineForText2Image, ControlNetModel, StableVideoDiffusionPipeline, Lumina2Text2ImgPipeline
 from diffusers.pipelines.stable_diffusion import safety_checker
 from diffusers.utils import load_image, export_to_video
@@ -72,6 +72,13 @@ def is_float(s):
    except ValueError:
        return False

+def is_int(s):
+    try:
+        int(s)
+        return True
+    except ValueError:
+        return False
+
 # The scheduler list mapping was taken from here: https://github.com/neggles/animatediff-cli/blob/6f336f5f4b5e38e85d7f06f1744ef42d0a45f2a7/src/animatediff/schedulers.py#L39
 # Credits to https://github.com/neggles
 # See https://github.com/huggingface/diffusers/issues/4167 for more details on sched mapping from A1111
@@ -177,10 +184,9 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
                key, value = opt.split(":")
                # if value is a number, convert it to the appropriate type
                if is_float(value):
-                    if value.is_integer():
-                        value = int(value)
-                    else:
-                        value = float(value)
+                    value = float(value)
+                elif is_int(value):
+                    value = int(value)
                self.options[key] = value

            # From options, extract if present "torch_dtype" and set it to the appropriate type
@@ -257,9 +263,6 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
            elif request.PipelineType == "DiffusionPipeline":
                self.pipe = DiffusionPipeline.from_pretrained(request.Model,
                                                              torch_dtype=torchType)
-            elif request.PipelineType == "QwenImageEditPipeline":
-                self.pipe = QwenImageEditPipeline.from_pretrained(request.Model,
-                                                                 torch_dtype=torchType)
            elif request.PipelineType == "VideoDiffusionPipeline":
                self.txt2vid = True
                self.pipe = DiffusionPipeline.from_pretrained(request.Model,
@@ -328,32 +331,6 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
                    torch_dtype=torch.bfloat16)
                self.pipe.vae.to(torch.bfloat16)
                self.pipe.text_encoder.to(torch.bfloat16)
-            elif request.PipelineType == "WanPipeline":
-                # WAN2.2 pipeline requires special VAE handling
-                vae = AutoencoderKLWan.from_pretrained(
-                    request.Model, 
-                    subfolder="vae", 
-                    torch_dtype=torch.float32
-                )
-                self.pipe = WanPipeline.from_pretrained(
-                    request.Model,
-                    vae=vae,
-                    torch_dtype=torchType
-                )
-                self.txt2vid = True  # WAN2.2 is a text-to-video pipeline
-            elif request.PipelineType == "WanImageToVideoPipeline":
-                # WAN2.2 image-to-video pipeline
-                vae = AutoencoderKLWan.from_pretrained(
-                    request.Model, 
-                    subfolder="vae", 
-                    torch_dtype=torch.float32
-                )
-                self.pipe = WanImageToVideoPipeline.from_pretrained(
-                    request.Model,
-                    vae=vae,
-                    torch_dtype=torchType
-                )
-                self.img2vid = True  # WAN2.2 image-to-video pipeline

            if CLIPSKIP and request.CLIPSkip != 0:
                self.clip_skip = request.CLIPSkip
@@ -388,9 +365,6 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
            device = "cpu" if not request.CUDA else "cuda"
            if XPU:
                device = "xpu"
-            mps_available = hasattr(torch.backends, "mps") and torch.backends.mps.is_available()
-            if mps_available:
-                device = "mps"
            self.device = device
            if request.LoraAdapter:
                # Check if its a local file and not a directory ( we load lora differently for a safetensor file )
@@ -495,24 +469,11 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
            "num_inference_steps": steps,
        }

-        # Handle image source: prioritize RefImages over request.src
-        image_src = None
-        if hasattr(request, 'ref_images') and request.ref_images and len(request.ref_images) > 0:
-            # Use the first reference image if available
-            image_src = request.ref_images[0]
-            print(f"Using reference image: {image_src}", file=sys.stderr)
-        elif request.src != "":
-            # Fall back to request.src if no ref_images
-            image_src = request.src
-            print(f"Using source image: {image_src}", file=sys.stderr)
-        else:
-            print("No image source provided", file=sys.stderr)
-        
-        if image_src and not self.controlnet and not self.img2vid:
-            image = Image.open(image_src)
+        if request.src != "" and not self.controlnet and not self.img2vid:
+            image = Image.open(request.src)
            options["image"] = image
-        elif self.controlnet and image_src:
-            pose_image = load_image(image_src)
+        elif self.controlnet and request.src:
+            pose_image = load_image(request.src)
            options["image"] = pose_image

        if CLIPSKIP and self.clip_skip != 0:
@@ -554,11 +515,7 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):

        if self.img2vid:
            # Load the conditioning image
-            if image_src:
-                image = load_image(image_src)
-            else:
-                # Fallback to request.src for img2vid if no ref_images
-                image = load_image(request.src)
+            image = load_image(request.src)
            image = image.resize((1024, 576))

            generator = torch.manual_seed(request.seed)
@@ -595,96 +552,6 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):

        return backend_pb2.Result(message="Media generated", success=True)

-    def GenerateVideo(self, request, context):
-        try:
-            prompt = request.prompt
-            if not prompt:
-                return backend_pb2.Result(success=False, message="No prompt provided for video generation")
-
-            # Set default values from request or use defaults
-            num_frames = request.num_frames if request.num_frames > 0 else 81
-            fps = request.fps if request.fps > 0 else 16
-            cfg_scale = request.cfg_scale if request.cfg_scale > 0 else 4.0
-            num_inference_steps = request.step if request.step > 0 else 40
-            
-            # Prepare generation parameters
-            kwargs = {
-                "prompt": prompt,
-                "negative_prompt": request.negative_prompt if request.negative_prompt else "",
-                "height": request.height if request.height > 0 else 720,
-                "width": request.width if request.width > 0 else 1280,
-                "num_frames": num_frames,
-                "guidance_scale": cfg_scale,
-                "num_inference_steps": num_inference_steps,
-            }
-
-            # Add custom options from self.options (including guidance_scale_2 if specified)
-            kwargs.update(self.options)
-
-            # Set seed if provided
-            if request.seed > 0:
-                kwargs["generator"] = torch.Generator(device=self.device).manual_seed(request.seed)
-
-            # Handle start and end images for video generation
-            if request.start_image:
-                kwargs["start_image"] = load_image(request.start_image)
-            if request.end_image:
-                kwargs["end_image"] = load_image(request.end_image)
-
-            print(f"Generating video with {kwargs=}", file=sys.stderr)
-
-            # Generate video frames based on pipeline type
-            if self.PipelineType == "WanPipeline":
-                # WAN2.2 text-to-video generation
-                output = self.pipe(**kwargs)
-                frames = output.frames[0]  # WAN2.2 returns frames in this format
-            elif self.PipelineType == "WanImageToVideoPipeline":
-                # WAN2.2 image-to-video generation
-                if request.start_image:
-                    # Load and resize the input image according to WAN2.2 requirements
-                    image = load_image(request.start_image)
-                    # Use request dimensions or defaults, but respect WAN2.2 constraints
-                    request_height = request.height if request.height > 0 else 480
-                    request_width = request.width if request.width > 0 else 832
-                    max_area = request_height * request_width
-                    aspect_ratio = image.height / image.width
-                    mod_value = self.pipe.vae_scale_factor_spatial * self.pipe.transformer.config.patch_size[1]
-                    height = round((max_area * aspect_ratio) ** 0.5 / mod_value) * mod_value
-                    width = round((max_area / aspect_ratio) ** 0.5 / mod_value) * mod_value
-                    image = image.resize((width, height))
-                    kwargs["image"] = image
-                    kwargs["height"] = height
-                    kwargs["width"] = width
-                
-                output = self.pipe(**kwargs)
-                frames = output.frames[0]
-            elif self.img2vid:
-                # Generic image-to-video generation
-                if request.start_image:
-                    image = load_image(request.start_image)
-                    image = image.resize((request.width if request.width > 0 else 1024, 
-                                       request.height if request.height > 0 else 576))
-                    kwargs["image"] = image
-                
-                output = self.pipe(**kwargs)
-                frames = output.frames[0]
-            elif self.txt2vid:
-                # Generic text-to-video generation
-                output = self.pipe(**kwargs)
-                frames = output.frames[0]
-            else:
-                return backend_pb2.Result(success=False, message=f"Pipeline {self.PipelineType} does not support video generation")
-
-            # Export video
-            export_to_video(frames, request.dst, fps=fps)
-            
-            return backend_pb2.Result(message="Video generated successfully", success=True)
-
-        except Exception as err:
-            print(f"Error generating video: {err}", file=sys.stderr)
-            traceback.print_exc()
-            return backend_pb2.Result(success=False, message=f"Error generating video: {err}")
-

 def serve(address):
    server = grpc.server(futures.ThreadPoolExecutor(max_workers=MAX_WORKERS),
--- a/backend/python/diffusers/requirements-cpu.txt
+++ b/backend/python/diffusers/requirements-cpu.txt
@@ -1,12 +1,9 @@
--extra-index-url https://download.pytorch.org/whl/cpu
 git+https://github.com/huggingface/diffusers
 opencv-python
 transformers
-torchvision==0.22.1
 accelerate
 compel
 peft
 sentencepiece
 torch==2.7.1
-optimum-quanto
-ftfy
+optimum-quanto
--- a/backend/python/diffusers/requirements-cublas11.txt
+++ b/backend/python/diffusers/requirements-cublas11.txt
@@ -1,12 +1,10 @@
 --extra-index-url https://download.pytorch.org/whl/cu118
+torch==2.7.1+cu118
 git+https://github.com/huggingface/diffusers
 opencv-python
 transformers
-torchvision==0.22.1
 accelerate
 compel
 peft
 sentencepiece
-torch==2.7.1
-optimum-quanto
-ftfy
+optimum-quanto
--- a/backend/python/diffusers/requirements-cublas12.txt
+++ b/backend/python/diffusers/requirements-cublas12.txt
@@ -1,12 +1,9 @@
--extra-index-url https://download.pytorch.org/whl/cu121
+torch==2.7.1
 git+https://github.com/huggingface/diffusers
 opencv-python
 transformers
-torchvision
 accelerate
 compel
 peft
 sentencepiece
-torch
-ftfy
-optimum-quanto
+optimum-quanto
--- a/backend/python/diffusers/requirements-hipblas.txt
+++ b/backend/python/diffusers/requirements-hipblas.txt
@@ -8,5 +8,4 @@ accelerate
 compel
 peft
 sentencepiece
-optimum-quanto
-ftfy
+optimum-quanto
--- a/backend/python/diffusers/requirements-intel.txt
+++ b/backend/python/diffusers/requirements-intel.txt
@@ -12,5 +12,4 @@ accelerate
 compel
 peft
 sentencepiece
-optimum-quanto
-ftfy
+optimum-quanto
--- a/backend/python/diffusers/requirements-l4t.txt
+++ b/backend/python/diffusers/requirements-l4t.txt
@@ -6,7 +6,4 @@ accelerate
 compel
 peft
 optimum-quanto
-numpy<2
-sentencepiece
-torchvision
-ftfy
+numpy<2
--- a/backend/python/diffusers/requirements-mps.txt
+++ b/backend/python/diffusers/requirements-mps.txt
@@ -1,11 +0,0 @@
-torch==2.7.1
-torchvision==0.22.1
-git+https://github.com/huggingface/diffusers
-opencv-python
-transformers
-accelerate
-compel
-peft
-sentencepiece
-optimum-quanto
-ftfy
--- a/backend/python/diffusers/requirements.txt
+++ b/backend/python/diffusers/requirements.txt
@@ -1,5 +1,5 @@
 setuptools
-grpcio==1.74.0
+grpcio==1.71.0
 pillow
 protobuf
 certifi
--- a/backend/python/diffusers/run.sh
+++ b/backend/python/diffusers/run.sh
@@ -12,6 +12,4 @@ if [ -d "/opt/intel" ]; then
    export XPU=1
 fi

-export PYTORCH_ENABLE_MPS_FALLBACK=1
-
 startBackend $@
--- a/backend/python/exllama2/Makefile
+++ b/backend/python/exllama2/Makefile
@@ -1,17 +1,23 @@
 .PHONY: exllama2
-exllama2:
+exllama2: protogen
 	bash install.sh

 .PHONY: run
-run: exllama2
+run: protogen
 	@echo "Running exllama2..."
 	bash run.sh
 	@echo "exllama2 run."

+.PHONY: protogen
+protogen: backend_pb2_grpc.py backend_pb2.py
+
 .PHONY: protogen-clean
 protogen-clean:
 	$(RM) backend_pb2_grpc.py backend_pb2.py

+backend_pb2_grpc.py backend_pb2.py:
+	python3 -m grpc_tools.protoc -I../.. -I./ --python_out=. --grpc_python_out=. backend.proto
+
 .PHONY: clean
 clean: protogen-clean
 	$(RM) -r venv source __pycache__
--- a/backend/python/exllama2/requirements.txt
+++ b/backend/python/exllama2/requirements.txt
@@ -1,4 +1,4 @@
-grpcio==1.74.0
+grpcio==1.71.0
 protobuf
 certifi
 wheel
--- a/backend/python/faster-whisper/Makefile
+++ b/backend/python/faster-whisper/Makefile
@@ -3,11 +3,18 @@
 .PHONY: install
 install:
 	bash install.sh
+	$(MAKE) protogen
+
+.PHONY: protogen
+protogen: backend_pb2_grpc.py backend_pb2.py

 .PHONY: protogen-clean
 protogen-clean:
 	$(RM) backend_pb2_grpc.py backend_pb2.py

+backend_pb2_grpc.py backend_pb2.py:
+	bash protogen.sh
+
 .PHONY: clean
 clean: protogen-clean
 	rm -rf venv __pycache__
--- a/backend/python/faster-whisper/backend.py
+++ b/backend/python/faster-whisper/backend.py
@@ -10,7 +10,7 @@ import sys
 import os
 import backend_pb2
 import backend_pb2_grpc
-import torch
+
 from faster_whisper import WhisperModel

 import grpc
@@ -35,9 +35,7 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
        # device = "cuda" if request.CUDA else "cpu"
        if request.CUDA:
            device = "cuda"
-        mps_available = hasattr(torch.backends, "mps") and torch.backends.mps.is_available()
-        if mps_available:
-            device = "mps"
+
        try:
            print("Preparing models, please wait", file=sys.stderr)
            self.model = WhisperModel(request.Model, device=device, compute_type="float16")
--- a/backend/python/kitten-tts/Makefile
+++ b/backend/python/kitten-tts/Makefile
@@ -1,23 +1,29 @@
 .PHONY: kitten-tts
-kitten-tts:
+kitten-tts: protogen
 	bash install.sh

 .PHONY: run
-run: kitten-tts
+run: protogen
 	@echo "Running kitten-tts..."
 	bash run.sh
 	@echo "kitten-tts run."

 .PHONY: test
-test: kitten-tts
+test: protogen
 	@echo "Testing kitten-tts..."
 	bash test.sh
 	@echo "kitten-tts tested."

+.PHONY: protogen
+protogen: backend_pb2_grpc.py backend_pb2.py
+
 .PHONY: protogen-clean
 protogen-clean:
 	$(RM) backend_pb2_grpc.py backend_pb2.py

+backend_pb2_grpc.py backend_pb2.py:
+	python3 -m grpc_tools.protoc -I../.. -I./ --python_out=. --grpc_python_out=. backend.proto
+
 .PHONY: clean
 clean: protogen-clean
 	rm -rf venv __pycache__
--- a/backend/python/kitten-tts/backend.py
+++ b/backend/python/kitten-tts/backend.py
@@ -33,6 +33,18 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
        return backend_pb2.Reply(message=bytes("OK", 'utf-8'))
    def LoadModel(self, request, context):

+        # Get device
+        # device = "cuda" if request.CUDA else "cpu"
+        if torch.cuda.is_available():
+            print("CUDA is available", file=sys.stderr)
+            device = "cuda"
+        else:
+            print("CUDA is not available", file=sys.stderr)
+            device = "cpu"
+
+        if not torch.cuda.is_available() and request.CUDA:
+            return backend_pb2.Result(success=False, message="CUDA is not available")
+
        self.AudioPath = None
        # List available KittenTTS models
        print("Available KittenTTS voices: expr-voice-2-m, expr-voice-2-f, expr-voice-3-m, expr-voice-3-f, expr-voice-4-m, expr-voice-4-f, expr-voice-5-m, expr-voice-5-f")
--- a/backend/python/kokoro/Makefile
+++ b/backend/python/kokoro/Makefile
@@ -1,23 +1,29 @@
 .PHONY: kokoro
-kokoro:
+kokoro: protogen
 	bash install.sh

 .PHONY: run
-run: kokoro
+run: protogen
 	@echo "Running kokoro..."
 	bash run.sh
 	@echo "kokoro run."

 .PHONY: test
-test: kokoro
+test: protogen
 	@echo "Testing kokoro..."
 	bash test.sh
 	@echo "kokoro tested."

+.PHONY: protogen
+protogen: backend_pb2_grpc.py backend_pb2.py
+
 .PHONY: protogen-clean
 protogen-clean:
 	$(RM) backend_pb2_grpc.py backend_pb2.py

+backend_pb2_grpc.py backend_pb2.py:
+	python3 -m grpc_tools.protoc -I../.. -I./ --python_out=. --grpc_python_out=. backend.proto
+
 .PHONY: clean
 clean: protogen-clean
 	rm -rf venv __pycache__
--- a/backend/python/kokoro/backend.py
+++ b/backend/python/kokoro/backend.py
@@ -33,6 +33,17 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
        return backend_pb2.Reply(message=bytes("OK", 'utf-8'))
    
    def LoadModel(self, request, context):
+        # Get device
+        if torch.cuda.is_available():
+            print("CUDA is available", file=sys.stderr)
+            device = "cuda"
+        else:
+            print("CUDA is not available", file=sys.stderr)
+            device = "cpu"
+
+        if not torch.cuda.is_available() and request.CUDA:
+            return backend_pb2.Result(success=False, message="CUDA is not available")
+
        try:
            print("Preparing Kokoro TTS pipeline, please wait", file=sys.stderr)
            # empty dict
--- a/backend/python/mlx-audio/Makefile
+++ b/backend/python/mlx-audio/Makefile
@@ -1,23 +0,0 @@
-.PHONY: mlx-audio
-mlx-audio:
-	bash install.sh
-
-.PHONY: run
-run: mlx-audio
-	@echo "Running mlx-audio..."
-	bash run.sh
-	@echo "mlx run."
-
-.PHONY: test
-test: mlx-audio
-	@echo "Testing mlx-audio..."
-	bash test.sh
-	@echo "mlx tested."
-
-.PHONY: protogen-clean
-protogen-clean:
-	$(RM) backend_pb2_grpc.py backend_pb2.py
-
-.PHONY: clean
-clean: protogen-clean
-	rm -rf venv __pycache__
--- a/backend/python/mlx-audio/backend.py
+++ b/backend/python/mlx-audio/backend.py
@@ -1,459 +0,0 @@
-#!/usr/bin/env python3
-import asyncio
-from concurrent import futures
-import argparse
-import signal
-import sys
-import os
-import shutil
-import glob
-from typing import List
-import time
-import tempfile
-
-import backend_pb2
-import backend_pb2_grpc
-
-import grpc
-from mlx_audio.tts.utils import load_model
-import soundfile as sf
-import numpy as np
-import uuid
-
-_ONE_DAY_IN_SECONDS = 60 * 60 * 24
-
-# If MAX_WORKERS are specified in the environment use it, otherwise default to 1
-MAX_WORKERS = int(os.environ.get('PYTHON_GRPC_MAX_WORKERS', '1'))
-
-# Implement the BackendServicer class with the service methods
-class BackendServicer(backend_pb2_grpc.BackendServicer):
-    """
-    A gRPC servicer that implements the Backend service defined in backend.proto.
-    This backend provides TTS (Text-to-Speech) functionality using MLX-Audio.
-    """
-
-    def _is_float(self, s):
-        """Check if a string can be converted to float."""
-        try:
-            float(s)
-            return True
-        except ValueError:
-            return False
-
-    def Health(self, request, context):
-        """
-        Returns a health check message.
-
-        Args:
-            request: The health check request.
-            context: The gRPC context.
-
-        Returns:
-            backend_pb2.Reply: The health check reply.
-        """
-        return backend_pb2.Reply(message=bytes("OK", 'utf-8'))
-
-    async def LoadModel(self, request, context):
-        """
-        Loads a TTS model using MLX-Audio.
-
-        Args:
-            request: The load model request.
-            context: The gRPC context.
-
-        Returns:
-            backend_pb2.Result: The load model result.
-        """
-        try:
-            print(f"Loading MLX-Audio TTS model: {request.Model}", file=sys.stderr)
-            print(f"Request: {request}", file=sys.stderr)
-            
-            # Parse options like in the kokoro backend
-            options = request.Options
-            self.options = {}
-            
-            # The options are a list of strings in this form optname:optvalue
-            # We store all the options in a dict for later use
-            for opt in options:
-                if ":" not in opt:
-                    continue
-                key, value = opt.split(":", 1)  # Split only on first colon to handle values with colons
-                
-                # Convert numeric values to appropriate types
-                if self._is_float(value):
-                    if float(value).is_integer():
-                        value = int(value)
-                    else:
-                        value = float(value)
-                elif value.lower() in ["true", "false"]:
-                    value = value.lower() == "true"
-                    
-                self.options[key] = value
-            
-            print(f"Options: {self.options}", file=sys.stderr)
-            
-            # Load the model using MLX-Audio's load_model function
-            try:
-                self.tts_model = load_model(request.Model)
-                self.model_path = request.Model
-                print(f"TTS model loaded successfully from {request.Model}", file=sys.stderr)
-            except Exception as model_err:
-                print(f"Error loading TTS model: {model_err}", file=sys.stderr)
-                return backend_pb2.Result(success=False, message=f"Failed to load model: {model_err}")
-                
-        except Exception as err:
-            print(f"Error loading MLX-Audio TTS model {err=}, {type(err)=}", file=sys.stderr)
-            return backend_pb2.Result(success=False, message=f"Error loading MLX-Audio TTS model: {err}")
-
-        print("MLX-Audio TTS model loaded successfully", file=sys.stderr)
-        return backend_pb2.Result(message="MLX-Audio TTS model loaded successfully", success=True)
-
-    def TTS(self, request, context):
-        """
-        Generates TTS audio from text using MLX-Audio.
-
-        Args:
-            request: A TTSRequest object containing text, model, destination, voice, and language.
-            context: A grpc.ServicerContext object that provides information about the RPC.
-
-        Returns:
-            A Result object indicating success or failure.
-        """
-        try:
-            # Check if model is loaded
-            if not hasattr(self, 'tts_model') or self.tts_model is None:
-                return backend_pb2.Result(success=False, message="TTS model not loaded. Please call LoadModel first.")
-            
-            print(f"Generating TTS with MLX-Audio - text: {request.text[:50]}..., voice: {request.voice}, language: {request.language}", file=sys.stderr)
-            
-            # Handle speed parameter based on model type
-            speed_value = self._handle_speed_parameter(request, self.model_path)
-            
-            # Map language names to codes if needed
-            lang_code = self._map_language_code(request.language, request.voice)
-            
-            # Prepare generation parameters
-            gen_params = {
-                "text": request.text,
-                "speed": speed_value,
-                "verbose": False,
-            }
-            
-            # Add model-specific parameters
-            if request.voice and request.voice.strip():
-                gen_params["voice"] = request.voice
-            
-            # Check if model supports language codes (primarily Kokoro)
-            if "kokoro" in self.model_path.lower():
-                gen_params["lang_code"] = lang_code
-            
-            # Add pitch and gender for Spark models
-            if "spark" in self.model_path.lower():
-                gen_params["pitch"] = 1.0  # Default to moderate
-                gen_params["gender"] = "female"  # Default to female
-            
-            print(f"Generation parameters: {gen_params}", file=sys.stderr)
-            
-            # Generate audio using the loaded model
-            try:
-                results = self.tts_model.generate(**gen_params)
-            except Exception as gen_err:
-                print(f"Error during TTS generation: {gen_err}", file=sys.stderr)
-                return backend_pb2.Result(success=False, message=f"TTS generation failed: {gen_err}")
-            
-            # Process the generated audio segments
-            audio_arrays = []
-            for segment in results:
-                audio_arrays.append(segment.audio)
-            
-            # If no segments, return error
-            if not audio_arrays:
-                print("No audio segments generated", file=sys.stderr)
-                return backend_pb2.Result(success=False, message="No audio generated")
-            
-            # Concatenate all segments
-            cat_audio = np.concatenate(audio_arrays, axis=0)
-            
-            # Generate output filename and path
-            if request.dst:
-                output_path = request.dst
-            else:
-                unique_id = str(uuid.uuid4())
-                filename = f"tts_{unique_id}.wav"
-                output_path = filename
-            
-            # Write the audio as a WAV
-            try:
-                sf.write(output_path, cat_audio, 24000)
-                print(f"Successfully wrote audio file to {output_path}", file=sys.stderr)
-                
-                # Verify the file exists and has content
-                if not os.path.exists(output_path):
-                    print(f"File was not created at {output_path}", file=sys.stderr)
-                    return backend_pb2.Result(success=False, message="Failed to create audio file")
-                
-                file_size = os.path.getsize(output_path)
-                if file_size == 0:
-                    print("File was created but is empty", file=sys.stderr)
-                    return backend_pb2.Result(success=False, message="Generated audio file is empty")
-                
-                print(f"Audio file size: {file_size} bytes", file=sys.stderr)
-                
-            except Exception as write_err:
-                print(f"Error writing audio file: {write_err}", file=sys.stderr)
-                return backend_pb2.Result(success=False, message=f"Failed to save audio: {write_err}")
-            
-            return backend_pb2.Result(success=True, message=f"TTS audio generated successfully: {output_path}")
-                
-        except Exception as e:
-            print(f"Error in MLX-Audio TTS: {e}", file=sys.stderr)
-            return backend_pb2.Result(success=False, message=f"TTS generation failed: {str(e)}")
-
-    async def Predict(self, request, context):
-        """
-        Generates TTS audio based on the given prompt using MLX-Audio TTS.
-        This is a fallback method for compatibility with the Predict endpoint.
-
-        Args:
-            request: The predict request.
-            context: The gRPC context.
-
-        Returns:
-            backend_pb2.Reply: The predict result.
-        """
-        try:
-            # Check if model is loaded
-            if not hasattr(self, 'tts_model') or self.tts_model is None:
-                context.set_code(grpc.StatusCode.FAILED_PRECONDITION)
-                context.set_details("TTS model not loaded. Please call LoadModel first.")
-                return backend_pb2.Reply(message=bytes("", encoding='utf-8'))
-            
-            # For TTS, we expect the prompt to contain the text to synthesize
-            if not request.Prompt:
-                context.set_code(grpc.StatusCode.INVALID_ARGUMENT)
-                context.set_details("Prompt is required for TTS generation")
-                return backend_pb2.Reply(message=bytes("", encoding='utf-8'))
-            
-            # Handle speed parameter based on model type
-            speed_value = self._handle_speed_parameter(request, self.model_path)
-            
-            # Map language names to codes if needed
-            lang_code = self._map_language_code(None, None)  # Use defaults for Predict
-            
-            # Prepare generation parameters
-            gen_params = {
-                "text": request.Prompt,
-                "speed": speed_value,
-                "verbose": False,
-            }
-            
-            # Add model-specific parameters
-            if hasattr(self, 'options') and 'voice' in self.options:
-                gen_params["voice"] = self.options['voice']
-            
-            # Check if model supports language codes (primarily Kokoro)
-            if "kokoro" in self.model_path.lower():
-                gen_params["lang_code"] = lang_code
-            
-            print(f"Generating TTS with MLX-Audio - text: {request.Prompt[:50]}..., params: {gen_params}", file=sys.stderr)
-            
-            # Generate audio using the loaded model
-            try:
-                results = self.tts_model.generate(**gen_params)
-            except Exception as gen_err:
-                print(f"Error during TTS generation: {gen_err}", file=sys.stderr)
-                context.set_code(grpc.StatusCode.INTERNAL)
-                context.set_details(f"TTS generation failed: {gen_err}")
-                return backend_pb2.Reply(message=bytes("", encoding='utf-8'))
-            
-            # Process the generated audio segments
-            audio_arrays = []
-            for segment in results:
-                audio_arrays.append(segment.audio)
-            
-            # If no segments, return error
-            if not audio_arrays:
-                print("No audio segments generated", file=sys.stderr)
-                return backend_pb2.Reply(message=bytes("No audio generated", encoding='utf-8'))
-            
-            # Concatenate all segments
-            cat_audio = np.concatenate(audio_arrays, axis=0)
-            duration = len(cat_audio) / 24000  # Assuming 24kHz sample rate
-            
-            # Return success message with audio information
-            response = f"TTS audio generated successfully. Duration: {duration:.2f}s, Sample rate: 24000Hz"
-            return backend_pb2.Reply(message=bytes(response, encoding='utf-8'))
-            
-        except Exception as e:
-            print(f"Error in MLX-Audio TTS Predict: {e}", file=sys.stderr)
-            context.set_code(grpc.StatusCode.INTERNAL)
-            context.set_details(f"TTS generation failed: {str(e)}")
-            return backend_pb2.Reply(message=bytes("", encoding='utf-8'))
-
-    def _handle_speed_parameter(self, request, model_path):
-        """
-        Handle speed parameter based on model type.
-        
-        Args:
-            request: The TTSRequest object.
-            model_path: The model path to determine model type.
-            
-        Returns:
-            float: The processed speed value.
-        """
-        # Get speed from options if available
-        speed = 1.0
-        if hasattr(self, 'options') and 'speed' in self.options:
-            speed = self.options['speed']
-        
-        # Handle speed parameter based on model type
-        if "spark" in model_path.lower():
-            # Spark actually expects float values that map to speed descriptions
-            speed_map = {
-                "very_low": 0.0,
-                "low": 0.5,
-                "moderate": 1.0,
-                "high": 1.5,
-                "very_high": 2.0,
-            }
-            if isinstance(speed, str) and speed in speed_map:
-                speed_value = speed_map[speed]
-            else:
-                # Try to use as float, default to 1.0 (moderate) if invalid
-                try:
-                    speed_value = float(speed)
-                    if speed_value not in [0.0, 0.5, 1.0, 1.5, 2.0]:
-                        speed_value = 1.0  # Default to moderate
-                except:
-                    speed_value = 1.0  # Default to moderate
-        else:
-            # Other models use float speed values
-            try:
-                speed_value = float(speed)
-                if speed_value < 0.5 or speed_value > 2.0:
-                    speed_value = 1.0  # Default to 1.0 if out of range
-            except ValueError:
-                speed_value = 1.0  # Default to 1.0 if invalid
-        
-        return speed_value
-
-    def _map_language_code(self, language, voice):
-        """
-        Map language names to codes if needed.
-        
-        Args:
-            language: The language parameter from the request.
-            voice: The voice parameter from the request.
-            
-        Returns:
-            str: The language code.
-        """
-        if not language:
-            # Default to voice[0] if not found
-            return voice[0] if voice else "a"
-        
-        # Map language names to codes if needed
-        language_map = {
-            "american_english": "a",
-            "british_english": "b",
-            "spanish": "e",
-            "french": "f",
-            "hindi": "h",
-            "italian": "i",
-            "portuguese": "p",
-            "japanese": "j",
-            "mandarin_chinese": "z",
-            # Also accept direct language codes
-            "a": "a", "b": "b", "e": "e", "f": "f", "h": "h", "i": "i", "p": "p", "j": "j", "z": "z",
-        }
-        
-        return language_map.get(language.lower(), language)
-
-    def _build_generation_params(self, request, default_speed=1.0):
-        """
-        Build generation parameters from request attributes and options for MLX-Audio TTS.
-
-        Args:
-            request: The gRPC request.
-            default_speed: Default speed if not specified.
-
-        Returns:
-            dict: Generation parameters for MLX-Audio
-        """
-        # Initialize generation parameters for MLX-Audio TTS
-        generation_params = {
-            'speed': default_speed,
-            'voice': 'af_heart',  # Default voice
-            'lang_code': 'a',     # Default language code
-        }
-        
-        # Extract parameters from request attributes
-        if hasattr(request, 'Temperature') and request.Temperature > 0:
-            # Temperature could be mapped to speed variation
-            generation_params['speed'] = 1.0 + (request.Temperature - 0.5) * 0.5
-        
-        # Override with options if available
-        if hasattr(self, 'options'):
-            # Speed from options
-            if 'speed' in self.options:
-                generation_params['speed'] = self.options['speed']
-            
-            # Voice from options
-            if 'voice' in self.options:
-                generation_params['voice'] = self.options['voice']
-            
-            # Language code from options
-            if 'lang_code' in self.options:
-                generation_params['lang_code'] = self.options['lang_code']
-            
-            # Model-specific parameters
-            param_option_mapping = {
-                'temp': 'speed',
-                'temperature': 'speed',
-                'top_p': 'speed',  # Map top_p to speed variation
-            }
-            
-            for option_key, param_key in param_option_mapping.items():
-                if option_key in self.options:
-                    if param_key == 'speed':
-                        # Ensure speed is within reasonable bounds
-                        speed_val = float(self.options[option_key])
-                        if 0.5 <= speed_val <= 2.0:
-                            generation_params[param_key] = speed_val
-        
-        return generation_params
-
-async def serve(address):
-    # Start asyncio gRPC server
-    server = grpc.aio.server(migration_thread_pool=futures.ThreadPoolExecutor(max_workers=MAX_WORKERS),
-        options=[
-            ('grpc.max_message_length', 50 * 1024 * 1024),  # 50MB
-            ('grpc.max_send_message_length', 50 * 1024 * 1024),  # 50MB
-            ('grpc.max_receive_message_length', 50 * 1024 * 1024),  # 50MB
-        ])
-    # Add the servicer to the server
-    backend_pb2_grpc.add_BackendServicer_to_server(BackendServicer(), server)
-    # Bind the server to the address
-    server.add_insecure_port(address)
-
-    # Gracefully shutdown the server on SIGTERM or SIGINT
-    loop = asyncio.get_event_loop()
-    for sig in (signal.SIGINT, signal.SIGTERM):
-        loop.add_signal_handler(
-            sig, lambda: asyncio.ensure_future(server.stop(5))
-        )
-
-    # Start the server
-    await server.start()
-    print("MLX-Audio TTS Server started. Listening on: " + address, file=sys.stderr)
-    # Wait for the server to be terminated
-    await server.wait_for_termination()
-
-if __name__ == "__main__":
-    parser = argparse.ArgumentParser(description="Run the MLX-Audio TTS gRPC server.")
-    parser.add_argument(
-        "--addr", default="localhost:50051", help="The address to bind the server to."
-    )
-    args = parser.parse_args()
-
-    asyncio.run(serve(args.addr))
--- a/backend/python/mlx-audio/install.sh
+++ b/backend/python/mlx-audio/install.sh
@@ -1,14 +0,0 @@
-#!/bin/bash
-set -e
-
-USE_PIP=true
-
-backend_dir=$(dirname $0)
-
-if [ -d $backend_dir/common ]; then
-    source $backend_dir/common/libbackend.sh
-else
-    source $backend_dir/../common/libbackend.sh
-fi
-
-installRequirements
--- a/backend/python/mlx-audio/requirements-mps.txt
+++ b/backend/python/mlx-audio/requirements-mps.txt
@@ -1 +0,0 @@
-git+https://github.com/Blaizzy/mlx-audio
--- a/backend/python/mlx-audio/requirements.txt
+++ b/backend/python/mlx-audio/requirements.txt
@@ -1,7 +0,0 @@
-grpcio==1.71.0
-protobuf
-certifi
-setuptools
-mlx-audio
-soundfile
-numpy
--- a/backend/python/mlx-audio/run.sh
+++ b/backend/python/mlx-audio/run.sh
@@ -1,11 +0,0 @@
-#!/bin/bash
-
-backend_dir=$(dirname $0)
-
-if [ -d $backend_dir/common ]; then
-    source $backend_dir/common/libbackend.sh
-else
-    source $backend_dir/../common/libbackend.sh
-fi
-
-startBackend $@
--- a/backend/python/mlx-audio/test.py
+++ b/backend/python/mlx-audio/test.py
@@ -1,142 +0,0 @@
-import unittest
-import subprocess
-import time
-import backend_pb2
-import backend_pb2_grpc
-
-import grpc
-
-import unittest
-import subprocess
-import time
-import grpc
-import backend_pb2_grpc
-import backend_pb2
-
-class TestBackendServicer(unittest.TestCase):
-    """
-    TestBackendServicer is the class that tests the gRPC service.
-
-    This class contains methods to test the startup and shutdown of the gRPC service.
-    """
-    def setUp(self):
-        self.service = subprocess.Popen(["python", "backend.py", "--addr", "localhost:50051"])
-        time.sleep(10)
-
-    def tearDown(self) -> None:
-        self.service.terminate()
-        self.service.wait()
-
-    def test_server_startup(self):
-        try:
-            self.setUp()
-            with grpc.insecure_channel("localhost:50051") as channel:
-                stub = backend_pb2_grpc.BackendStub(channel)
-                response = stub.Health(backend_pb2.HealthMessage())
-                self.assertEqual(response.message, b'OK')
-        except Exception as err:
-            print(err)
-            self.fail("Server failed to start")
-        finally:
-            self.tearDown()
-    def test_load_model(self):
-        """
-        This method tests if the TTS model is loaded successfully
-        """
-        try:
-            self.setUp()
-            with grpc.insecure_channel("localhost:50051") as channel:
-                stub = backend_pb2_grpc.BackendStub(channel)
-                response = stub.LoadModel(backend_pb2.ModelOptions(Model="mlx-community/Kokoro-82M-4bit"))
-                self.assertTrue(response.success)
-                self.assertEqual(response.message, "MLX-Audio TTS model loaded successfully")
-        except Exception as err:
-            print(err)
-            self.fail("LoadModel service failed")
-        finally:
-            self.tearDown()
-
-    def test_tts_generation(self):
-        """
-        This method tests if TTS audio is generated successfully
-        """
-        try:
-            self.setUp()
-            with grpc.insecure_channel("localhost:50051") as channel:
-                stub = backend_pb2_grpc.BackendStub(channel)
-                response = stub.LoadModel(backend_pb2.ModelOptions(Model="mlx-community/Kokoro-82M-4bit"))
-                self.assertTrue(response.success)
-                
-                # Test TTS generation
-                tts_req = backend_pb2.TTSRequest(
-                    text="Hello, this is a test of the MLX-Audio TTS system.",
-                    model="mlx-community/Kokoro-82M-4bit",
-                    voice="af_heart",
-                    language="a"
-                )
-                tts_resp = stub.TTS(tts_req)
-                self.assertTrue(tts_resp.success)
-                self.assertIn("TTS audio generated successfully", tts_resp.message)
-        except Exception as err:
-            print(err)
-            self.fail("TTS service failed")
-        finally:
-            self.tearDown()
-
-    def test_tts_with_options(self):
-        """
-        This method tests if TTS works with various options and parameters
-        """
-        try:
-            self.setUp()
-            with grpc.insecure_channel("localhost:50051") as channel:
-                stub = backend_pb2_grpc.BackendStub(channel)
-                response = stub.LoadModel(backend_pb2.ModelOptions(
-                    Model="mlx-community/Kokoro-82M-4bit",
-                    Options=["voice:af_soft", "speed:1.2", "lang_code:b"]
-                ))
-                self.assertTrue(response.success)
-                
-                # Test TTS generation with different voice and language
-                tts_req = backend_pb2.TTSRequest(
-                    text="Hello, this is a test with British English accent.",
-                    model="mlx-community/Kokoro-82M-4bit",
-                    voice="af_soft",
-                    language="b"
-                )
-                tts_resp = stub.TTS(tts_req)
-                self.assertTrue(tts_resp.success)
-                self.assertIn("TTS audio generated successfully", tts_resp.message)
-        except Exception as err:
-            print(err)
-            self.fail("TTS with options service failed")
-        finally:
-            self.tearDown()
-
-
-    def test_tts_multilingual(self):
-        """
-        This method tests if TTS works with different languages
-        """
-        try:
-            self.setUp()
-            with grpc.insecure_channel("localhost:50051") as channel:
-                stub = backend_pb2_grpc.BackendStub(channel)
-                response = stub.LoadModel(backend_pb2.ModelOptions(Model="mlx-community/Kokoro-82M-4bit"))
-                self.assertTrue(response.success)
-                
-                # Test Spanish TTS
-                tts_req = backend_pb2.TTSRequest(
-                    text="Hola, esto es una prueba del sistema TTS MLX-Audio.",
-                    model="mlx-community/Kokoro-82M-4bit",
-                    voice="af_heart",
-                    language="e"
-                )
-                tts_resp = stub.TTS(tts_req)
-                self.assertTrue(tts_resp.success)
-                self.assertIn("TTS audio generated successfully", tts_resp.message)
-        except Exception as err:
-            print(err)
-            self.fail("Multilingual TTS service failed")
-        finally:
-            self.tearDown()
--- a/backend/python/mlx-audio/test.sh
+++ b/backend/python/mlx-audio/test.sh
@@ -1,12 +0,0 @@
-#!/bin/bash
-set -e
-
-backend_dir=$(dirname $0)
-
-if [ -d $backend_dir/common ]; then
-    source $backend_dir/common/libbackend.sh
-else
-    source $backend_dir/../common/libbackend.sh
-fi
-
-runUnittests
--- a/Show More
+++ b/Show More