Initial plan

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

.github/gallery-agent/agent.go

@@ -11,6 +11,7 @@ import (
 	"slices"
 	"strings"
 
+	"github.com/ghodss/yaml"
 	hfapi "github.com/mudler/LocalAI/pkg/huggingface-api"
 	cogito "github.com/mudler/cogito"
 
@@ -52,6 +53,11 @@ func cleanTextContent(text string) string {
 	return stripThinkingTags(strings.TrimRight(result, "\n"))
 }
 
+type galleryModel struct {
+	Name string   `yaml:"name"`
+	Urls []string `yaml:"urls"`
+}
+
 // isModelExisting checks if a specific model ID exists in the gallery using text search
 func isModelExisting(modelID string) (bool, error) {
 	indexPath := getGalleryIndexPath()
@@ -60,9 +66,20 @@ func isModelExisting(modelID string) (bool, error) {
 		return false, fmt.Errorf("failed to read %s: %w", indexPath, err)
 	}
 
-	contentStr := string(content)
-	// Simple text search - if the model ID appears anywhere in the file, it exists
-	return strings.Contains(contentStr, modelID), nil
+	var galleryModels []galleryModel
+
+	err = yaml.Unmarshal(content, &galleryModels)
+	if err != nil {
+		return false, fmt.Errorf("failed to unmarshal %s: %w", indexPath, err)
+	}
+
+	for _, galleryModel := range galleryModels {
+		if slices.Contains(galleryModel.Urls, modelID) {
+			return true, nil
+		}
+	}
+
+	return false, nil
 }
 
 // filterExistingModels removes models that already exist in the gallery
@@ -134,6 +151,11 @@ func getRealReadme(ctx context.Context, repository string) (string, error) {
 }
 
 func selectMostInterestingModels(ctx context.Context, searchResult *SearchResult) ([]ProcessedModel, error) {
+
+	if len(searchResult.Models) == 1 {
+		return searchResult.Models, nil
+	}
+
 	// Create a conversation fragment
 	fragment := cogito.NewEmptyFragment().
 		AddMessage("user",

.github/gallery-agent/main.go

@@ -119,14 +119,24 @@ func main() {
 	}
 
 	fmt.Println(result.FormattedOutput)
+	var models []ProcessedModel
 
-	// Use AI agent to select the most interesting models
-	fmt.Println("Using AI agent to select the most interesting models...")
-	models, err := selectMostInterestingModels(context.Background(), result)
-	if err != nil {
-		fmt.Fprintf(os.Stderr, "Error in model selection: %v\n", err)
-		// Continue with original result if selection fails
+	if len(result.Models) > 1 {
+		fmt.Println("More than one model found (", len(result.Models), "), using AI agent to select the most interesting models")
+		for _, model := range result.Models {
+			fmt.Println("Model: ", model.ModelID)
+		}
+		// Use AI agent to select the most interesting models
+		fmt.Println("Using AI agent to select the most interesting models...")
+		models, err = selectMostInterestingModels(context.Background(), result)
+		if err != nil {
+			fmt.Fprintf(os.Stderr, "Error in model selection: %v\n", err)
+			// Continue with original result if selection fails
+			models = result.Models
+		}
+	} else if len(result.Models) == 1 {
 		models = result.Models
+		fmt.Println("Only one model found, using it directly")
 	}
 
 	fmt.Print(models)
@@ -315,7 +325,7 @@ func searchAndProcessModels(searchTerm string, limit int, quantization string) (
 				outputBuilder.WriteString(fmt.Sprintf("   README Content Preview: %s\n",
 					processedModel.ReadmeContentPreview))
 			} else {
-				continue
+				fmt.Printf("   Warning: Failed to get real readme: %v\n", err)
 			}
 			fmt.Println("Real readme got", readmeContent)
 

.github/workflows/dependabot_auto.yml

@@ -14,7 +14,7 @@ jobs:
     steps:
       - name: Dependabot metadata
         id: metadata
-        uses: dependabot/fetch-metadata@v2.4.0
+        uses: dependabot/fetch-metadata@v2.5.0
         with:
           github-token: "${{ secrets.GITHUB_TOKEN }}"
           skip-commit-verification: true

.github/workflows/gallery-agent.yaml

@@ -49,12 +49,12 @@ jobs:
           PATH="$PATH:$HOME/go/bin" make protogen-go
       - uses: mudler/localai-github-action@v1.1
         with:
-          model: 'qwen3-4b'
+          model: 'https://huggingface.co/bartowski/Qwen_Qwen3-1.7B-GGUF'
 
       - name: Run gallery agent
         env:
           #OPENAI_MODEL: ${{ secrets.OPENAI_MODEL }}
-          OPENAI_MODE: qwen3-4b
+          OPENAI_MODE: Qwen_Qwen3-1.7B-GGUF
           OPENAI_BASE_URL: "http://localhost:8080"
           OPENAI_KEY: ${{ secrets.OPENAI_KEY }}
           #OPENAI_BASE_URL: ${{ secrets.OPENAI_BASE_URL }}

.github/workflows/generate_grpc_cache.yaml

@@ -16,7 +16,7 @@ jobs:
     strategy:
       matrix:
         include:
-          - grpc-base-image: ubuntu:22.04
+          - grpc-base-image: ubuntu:24.04
             runs-on: 'ubuntu-latest'
             platforms: 'linux/amd64,linux/arm64'
     runs-on: ${{matrix.runs-on}}

.github/workflows/generate_intel_image.yaml

@@ -15,7 +15,7 @@ jobs:
     strategy:
       matrix:
         include:
-          - base-image: intel/oneapi-basekit:2025.2.0-0-devel-ubuntu22.04
+          - base-image: intel/oneapi-basekit:2025.3.0-0-devel-ubuntu24.04
             runs-on: 'arc-runner-set'
             platforms: 'linux/amd64'
     runs-on: ${{matrix.runs-on}}
@@ -53,7 +53,7 @@ jobs:
             BASE_IMAGE=${{ matrix.base-image }}
           context: .
           file: ./Dockerfile
-          tags: quay.io/go-skynet/intel-oneapi-base:latest
+          tags: quay.io/go-skynet/intel-oneapi-base:24.04
           push: true
           target: intel
           platforms: ${{ matrix.platforms }}

.github/workflows/image-pr.yml

@@ -1,94 +1,95 @@
 ---
-name: 'build container images tests'
-
-on:
-  pull_request:
-
-concurrency:
-  group: ci-${{ github.head_ref || github.ref }}-${{ github.repository }}
-  cancel-in-progress: true
-
-jobs:
-  image-build:
-    uses: ./.github/workflows/image_build.yml
-    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 }}
-      grpc-base-image: ${{ matrix.grpc-base-image }}
-      makeflags: ${{ matrix.makeflags }}
-      ubuntu-version: ${{ matrix.ubuntu-version }}
-    secrets:
-      dockerUsername: ${{ secrets.DOCKERHUB_USERNAME }}
-      dockerPassword: ${{ secrets.DOCKERHUB_PASSWORD }}
-      quayUsername: ${{ secrets.LOCALAI_REGISTRY_USERNAME }}
-      quayPassword: ${{ secrets.LOCALAI_REGISTRY_PASSWORD }}
-    strategy:
-      # Pushing with all jobs in parallel
-      # eats the bandwidth of all the nodes
-      max-parallel: ${{ github.event_name != 'pull_request' && 4 || 8 }}
-      fail-fast: false
-      matrix:
-        include:
-          - build-type: 'cublas'
-            cuda-major-version: "12"
-            cuda-minor-version: "0"
-            platforms: 'linux/amd64'
-            tag-latest: 'false'
-            tag-suffix: '-gpu-nvidia-cuda-12'
-            runs-on: 'ubuntu-latest'
-            base-image: "ubuntu:22.04"
-            makeflags: "--jobs=3 --output-sync=target"
-            ubuntu-version: '2204'
-          - build-type: 'cublas'
-            cuda-major-version: "13"
-            cuda-minor-version: "0"
-            platforms: 'linux/amd64'
-            tag-latest: 'false'
-            tag-suffix: '-gpu-nvidia-cuda-13'
-            runs-on: 'ubuntu-latest'
-            base-image: "ubuntu:22.04"
-            makeflags: "--jobs=3 --output-sync=target"
-            ubuntu-version: '2204'
-          - build-type: 'hipblas'
-            platforms: 'linux/amd64'
-            tag-latest: 'false'
-            tag-suffix: '-hipblas'
-            base-image: "rocm/dev-ubuntu-22.04:6.4.3"
-            grpc-base-image: "ubuntu:22.04"
-            runs-on: 'ubuntu-latest'
-            makeflags: "--jobs=3 --output-sync=target"
-            ubuntu-version: '2204'
-          - build-type: 'sycl'
-            platforms: 'linux/amd64'
-            tag-latest: 'false'
-            base-image: "quay.io/go-skynet/intel-oneapi-base:latest"
-            grpc-base-image: "ubuntu:22.04"
-            tag-suffix: 'sycl'
-            runs-on: 'ubuntu-latest'
-            makeflags: "--jobs=3 --output-sync=target"
-            ubuntu-version: '2204'
-          - build-type: 'vulkan'
-            platforms: 'linux/amd64'
-            tag-latest: 'false'
-            tag-suffix: '-vulkan-core'
-            runs-on: 'ubuntu-latest'
-            base-image: "ubuntu:22.04"
-            makeflags: "--jobs=4 --output-sync=target"
-            ubuntu-version: '2204'
-          - build-type: 'cublas'
-            cuda-major-version: "13"
-            cuda-minor-version: "0"
-            platforms: 'linux/arm64'
-            tag-latest: 'false'
-            tag-suffix: '-nvidia-l4t-arm64-cuda-13'
-            base-image: "ubuntu:24.04"
-            runs-on: 'ubuntu-24.04-arm'
-            makeflags: "--jobs=4 --output-sync=target"
-            skip-drivers: 'false'
-            ubuntu-version: '2404'
+  name: 'build container images tests'
+  
+  on:
+    pull_request:
+  
+  concurrency:
+    group: ci-${{ github.head_ref || github.ref }}-${{ github.repository }}
+    cancel-in-progress: true
+  
+  jobs:
+    image-build:
+      uses: ./.github/workflows/image_build.yml
+      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 }}
+        grpc-base-image: ${{ matrix.grpc-base-image }}
+        makeflags: ${{ matrix.makeflags }}
+        ubuntu-version: ${{ matrix.ubuntu-version }}
+      secrets:
+        dockerUsername: ${{ secrets.DOCKERHUB_USERNAME }}
+        dockerPassword: ${{ secrets.DOCKERHUB_PASSWORD }}
+        quayUsername: ${{ secrets.LOCALAI_REGISTRY_USERNAME }}
+        quayPassword: ${{ secrets.LOCALAI_REGISTRY_PASSWORD }}
+      strategy:
+        # Pushing with all jobs in parallel
+        # eats the bandwidth of all the nodes
+        max-parallel: ${{ github.event_name != 'pull_request' && 4 || 8 }}
+        fail-fast: false
+        matrix:
+          include:
+            - build-type: 'cublas'
+              cuda-major-version: "12"
+              cuda-minor-version: "8"
+              platforms: 'linux/amd64'
+              tag-latest: 'false'
+              tag-suffix: '-gpu-nvidia-cuda-12'
+              runs-on: 'ubuntu-latest'
+              base-image: "ubuntu:24.04"
+              makeflags: "--jobs=3 --output-sync=target"
+              ubuntu-version: '2404'
+            - build-type: 'cublas'
+              cuda-major-version: "13"
+              cuda-minor-version: "0"
+              platforms: 'linux/amd64'
+              tag-latest: 'false'
+              tag-suffix: '-gpu-nvidia-cuda-13'
+              runs-on: 'ubuntu-latest'
+              base-image: "ubuntu:22.04"
+              makeflags: "--jobs=3 --output-sync=target"
+              ubuntu-version: '2404'
+            - build-type: 'hipblas'
+              platforms: 'linux/amd64'
+              tag-latest: 'false'
+              tag-suffix: '-hipblas'
+              base-image: "rocm/dev-ubuntu-24.04:6.4.4"
+              grpc-base-image: "ubuntu:24.04"
+              runs-on: 'ubuntu-latest'
+              makeflags: "--jobs=3 --output-sync=target"
+              ubuntu-version: '2404'
+            - build-type: 'sycl'
+              platforms: 'linux/amd64'
+              tag-latest: 'false'
+              base-image: "intel/oneapi-basekit:2025.3.0-0-devel-ubuntu24.04"
+              grpc-base-image: "ubuntu:24.04"
+              tag-suffix: 'sycl'
+              runs-on: 'ubuntu-latest'
+              makeflags: "--jobs=3 --output-sync=target"
+              ubuntu-version: '2404'
+            - build-type: 'vulkan'
+              platforms: 'linux/amd64,linux/arm64'
+              tag-latest: 'false'
+              tag-suffix: '-vulkan-core'
+              runs-on: 'ubuntu-latest'
+              base-image: "ubuntu:24.04"
+              makeflags: "--jobs=4 --output-sync=target"
+              ubuntu-version: '2404'
+            - build-type: 'cublas'
+              cuda-major-version: "13"
+              cuda-minor-version: "0"
+              platforms: 'linux/arm64'
+              tag-latest: 'false'
+              tag-suffix: '-nvidia-l4t-arm64-cuda-13'
+              base-image: "ubuntu:24.04"
+              runs-on: 'ubuntu-24.04-arm'
+              makeflags: "--jobs=4 --output-sync=target"
+              skip-drivers: 'false'
+              ubuntu-version: '2404'
+  

.github/workflows/image.yml

@@ -1,187 +1,187 @@
 ---
-name: 'build container images'
-
-on:
-  push:
-    branches:
-      - master
-    tags:
-      - '*'
-
-concurrency:
-  group: ci-${{ github.head_ref || github.ref }}-${{ github.repository }}
-  cancel-in-progress: true
-
-jobs:
-  hipblas-jobs:
-    uses: ./.github/workflows/image_build.yml
-    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 }}
-      grpc-base-image: ${{ matrix.grpc-base-image }}
-      aio: ${{ matrix.aio }}
-      makeflags: ${{ matrix.makeflags }}
-      ubuntu-version: ${{ matrix.ubuntu-version }}
-    secrets:
-      dockerUsername: ${{ secrets.DOCKERHUB_USERNAME }}
-      dockerPassword: ${{ secrets.DOCKERHUB_PASSWORD }}
-      quayUsername: ${{ secrets.LOCALAI_REGISTRY_USERNAME }}
-      quayPassword: ${{ secrets.LOCALAI_REGISTRY_PASSWORD }}
-    strategy:
-      matrix:
-        include:
-          - build-type: 'hipblas'
-            platforms: 'linux/amd64'
-            tag-latest: 'auto'
-            tag-suffix: '-gpu-hipblas'
-            base-image: "rocm/dev-ubuntu-22.04:6.4.3"
-            grpc-base-image: "ubuntu:22.04"
-            runs-on: 'ubuntu-latest'
-            makeflags: "--jobs=3 --output-sync=target"
-            aio: "-aio-gpu-hipblas"
-            ubuntu-version: '2204'
-
-  core-image-build:
-    uses: ./.github/workflows/image_build.yml
-    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 }}
-      aio: ${{ matrix.aio }}
-      base-image: ${{ matrix.base-image }}
-      grpc-base-image: ${{ matrix.grpc-base-image }}
-      makeflags: ${{ matrix.makeflags }}
-      skip-drivers: ${{ matrix.skip-drivers }}
-      ubuntu-version: ${{ matrix.ubuntu-version }}
-    secrets:
-      dockerUsername: ${{ secrets.DOCKERHUB_USERNAME }}
-      dockerPassword: ${{ secrets.DOCKERHUB_PASSWORD }}
-      quayUsername: ${{ secrets.LOCALAI_REGISTRY_USERNAME }}
-      quayPassword: ${{ secrets.LOCALAI_REGISTRY_PASSWORD }}
-    strategy:
-      #max-parallel: ${{ github.event_name != 'pull_request' && 2 || 4 }}
-      matrix:
-        include:
-          - build-type: ''
-            platforms: 'linux/amd64,linux/arm64'
-            tag-latest: 'auto'
-            tag-suffix: ''
-            base-image: "ubuntu:22.04"
-            runs-on: 'ubuntu-latest'
-            aio: "-aio-cpu"
-            makeflags: "--jobs=4 --output-sync=target"
-            skip-drivers: 'false'
-            ubuntu-version: '2204'
-          - build-type: 'cublas'
-            cuda-major-version: "11"
-            cuda-minor-version: "7"
-            platforms: 'linux/amd64'
-            tag-latest: 'auto'
-            tag-suffix: '-gpu-nvidia-cuda-11'
-            runs-on: 'ubuntu-latest'
-            base-image: "ubuntu:22.04"
-            makeflags: "--jobs=4 --output-sync=target"
-            skip-drivers: 'false'
-            aio: "-aio-gpu-nvidia-cuda-11"
-            ubuntu-version: '2204'
-          - build-type: 'cublas'
-            cuda-major-version: "12"
-            cuda-minor-version: "0"
-            platforms: 'linux/amd64'
-            tag-latest: 'auto'
-            tag-suffix: '-gpu-nvidia-cuda-12'
-            runs-on: 'ubuntu-latest'
-            base-image: "ubuntu:22.04"
-            skip-drivers: 'false'
-            makeflags: "--jobs=4 --output-sync=target"
-            aio: "-aio-gpu-nvidia-cuda-12"
-            ubuntu-version: '2204'
-          - build-type: 'cublas'
-            cuda-major-version: "13"
-            cuda-minor-version: "0"
-            platforms: 'linux/amd64'
-            tag-latest: 'auto'
-            tag-suffix: '-gpu-nvidia-cuda-13'
-            runs-on: 'ubuntu-latest'
-            base-image: "ubuntu:22.04"
-            skip-drivers: 'false'
-            makeflags: "--jobs=4 --output-sync=target"
-            aio: "-aio-gpu-nvidia-cuda-13"
-            ubuntu-version: '2204'
-          - build-type: 'vulkan'
-            platforms: 'linux/amd64'
-            tag-latest: 'auto'
-            tag-suffix: '-gpu-vulkan'
-            runs-on: 'ubuntu-latest'
-            base-image: "ubuntu:22.04"
-            skip-drivers: 'false'
-            makeflags: "--jobs=4 --output-sync=target"
-            aio: "-aio-gpu-vulkan"
-            ubuntu-version: '2204'
-          - build-type: 'intel'
-            platforms: 'linux/amd64'
-            tag-latest: 'auto'
-            base-image: "quay.io/go-skynet/intel-oneapi-base:latest"
-            grpc-base-image: "ubuntu:22.04"
-            tag-suffix: '-gpu-intel'
-            runs-on: 'ubuntu-latest'
-            makeflags: "--jobs=3 --output-sync=target"
-            aio: "-aio-gpu-intel"
-            ubuntu-version: '2204'
-
-  gh-runner:
-    uses: ./.github/workflows/image_build.yml
-    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 }}
-      aio: ${{ matrix.aio }}
-      base-image: ${{ matrix.base-image }}
-      grpc-base-image: ${{ matrix.grpc-base-image }}
-      makeflags: ${{ matrix.makeflags }}
-      skip-drivers: ${{ matrix.skip-drivers }}
-      ubuntu-version: ${{ matrix.ubuntu-version }}
-    secrets:
-      dockerUsername: ${{ secrets.DOCKERHUB_USERNAME }}
-      dockerPassword: ${{ secrets.DOCKERHUB_PASSWORD }}
-      quayUsername: ${{ secrets.LOCALAI_REGISTRY_USERNAME }}
-      quayPassword: ${{ secrets.LOCALAI_REGISTRY_PASSWORD }}
-    strategy:
-      matrix:
-        include:
-          - build-type: 'cublas'
-            cuda-major-version: "12"
-            cuda-minor-version: "0"
-            platforms: 'linux/arm64'
-            tag-latest: 'auto'
-            tag-suffix: '-nvidia-l4t-arm64'
-            base-image: "nvcr.io/nvidia/l4t-jetpack:r36.4.0"
-            runs-on: 'ubuntu-24.04-arm'
-            makeflags: "--jobs=4 --output-sync=target"
-            skip-drivers: 'true'
-            ubuntu-version: "2204"
-          - build-type: 'cublas'
-            cuda-major-version: "13"
-            cuda-minor-version: "0"
-            platforms: 'linux/arm64'
-            tag-latest: 'auto'
-            tag-suffix: '-nvidia-l4t-arm64-cuda-13'
-            base-image: "ubuntu:24.04"
-            runs-on: 'ubuntu-24.04-arm'
-            makeflags: "--jobs=4 --output-sync=target"
-            skip-drivers: 'false'
-            ubuntu-version: '2404'
+  name: 'build container images'
+  
+  on:
+    push:
+      branches:
+        - master
+      tags:
+        - '*'
+  
+  concurrency:
+    group: ci-${{ github.head_ref || github.ref }}-${{ github.repository }}
+    cancel-in-progress: true
+  
+  jobs:
+    hipblas-jobs:
+      uses: ./.github/workflows/image_build.yml
+      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 }}
+        grpc-base-image: ${{ matrix.grpc-base-image }}
+        aio: ${{ matrix.aio }}
+        makeflags: ${{ matrix.makeflags }}
+        ubuntu-version: ${{ matrix.ubuntu-version }}
+        ubuntu-codename: ${{ matrix.ubuntu-codename }}
+      secrets:
+        dockerUsername: ${{ secrets.DOCKERHUB_USERNAME }}
+        dockerPassword: ${{ secrets.DOCKERHUB_PASSWORD }}
+        quayUsername: ${{ secrets.LOCALAI_REGISTRY_USERNAME }}
+        quayPassword: ${{ secrets.LOCALAI_REGISTRY_PASSWORD }}
+      strategy:
+        matrix:
+          include:
+            - build-type: 'hipblas'
+              platforms: 'linux/amd64'
+              tag-latest: 'auto'
+              tag-suffix: '-gpu-hipblas'
+              base-image: "rocm/dev-ubuntu-24.04:6.4.4"
+              grpc-base-image: "ubuntu:24.04"
+              runs-on: 'ubuntu-latest'
+              makeflags: "--jobs=3 --output-sync=target"
+              aio: "-aio-gpu-hipblas"
+              ubuntu-version: '2404'
+              ubuntu-codename: 'noble'
+  
+    core-image-build:
+      uses: ./.github/workflows/image_build.yml
+      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 }}
+        aio: ${{ matrix.aio }}
+        base-image: ${{ matrix.base-image }}
+        grpc-base-image: ${{ matrix.grpc-base-image }}
+        makeflags: ${{ matrix.makeflags }}
+        skip-drivers: ${{ matrix.skip-drivers }}
+        ubuntu-version: ${{ matrix.ubuntu-version }}
+        ubuntu-codename: ${{ matrix.ubuntu-codename }}
+      secrets:
+        dockerUsername: ${{ secrets.DOCKERHUB_USERNAME }}
+        dockerPassword: ${{ secrets.DOCKERHUB_PASSWORD }}
+        quayUsername: ${{ secrets.LOCALAI_REGISTRY_USERNAME }}
+        quayPassword: ${{ secrets.LOCALAI_REGISTRY_PASSWORD }}
+      strategy:
+        #max-parallel: ${{ github.event_name != 'pull_request' && 2 || 4 }}
+        matrix:
+          include:
+            - build-type: ''
+              platforms: 'linux/amd64,linux/arm64'
+              tag-latest: 'auto'
+              tag-suffix: ''
+              base-image: "ubuntu:24.04"
+              runs-on: 'ubuntu-latest'
+              aio: "-aio-cpu"
+              makeflags: "--jobs=4 --output-sync=target"
+              skip-drivers: 'false'
+              ubuntu-version: '2404'
+              ubuntu-codename: 'noble'
+            - build-type: 'cublas'
+              cuda-major-version: "12"
+              cuda-minor-version: "8"
+              platforms: 'linux/amd64'
+              tag-latest: 'auto'
+              tag-suffix: '-gpu-nvidia-cuda-12'
+              runs-on: 'ubuntu-latest'
+              base-image: "ubuntu:24.04"
+              skip-drivers: 'false'
+              makeflags: "--jobs=4 --output-sync=target"
+              aio: "-aio-gpu-nvidia-cuda-12"
+              ubuntu-version: '2404'
+              ubuntu-codename: 'noble'
+            - build-type: 'cublas'
+              cuda-major-version: "13"
+              cuda-minor-version: "0"
+              platforms: 'linux/amd64'
+              tag-latest: 'auto'
+              tag-suffix: '-gpu-nvidia-cuda-13'
+              runs-on: 'ubuntu-latest'
+              base-image: "ubuntu:22.04"
+              skip-drivers: 'false'
+              makeflags: "--jobs=4 --output-sync=target"
+              aio: "-aio-gpu-nvidia-cuda-13"
+              ubuntu-version: '2404'
+              ubuntu-codename: 'noble'
+            - build-type: 'vulkan'
+              platforms: 'linux/amd64,linux/arm64'
+              tag-latest: 'auto'
+              tag-suffix: '-gpu-vulkan'
+              runs-on: 'ubuntu-latest'
+              base-image: "ubuntu:24.04"
+              skip-drivers: 'false'
+              makeflags: "--jobs=4 --output-sync=target"
+              aio: "-aio-gpu-vulkan"
+              ubuntu-version: '2404'
+              ubuntu-codename: 'noble'
+            - build-type: 'intel'
+              platforms: 'linux/amd64'
+              tag-latest: 'auto'
+              base-image: "intel/oneapi-basekit:2025.3.0-0-devel-ubuntu24.04"
+              grpc-base-image: "ubuntu:24.04"
+              tag-suffix: '-gpu-intel'
+              runs-on: 'ubuntu-latest'
+              makeflags: "--jobs=3 --output-sync=target"
+              aio: "-aio-gpu-intel"
+              ubuntu-version: '2404'
+              ubuntu-codename: 'noble'
+  
+    gh-runner:
+      uses: ./.github/workflows/image_build.yml
+      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 }}
+        aio: ${{ matrix.aio }}
+        base-image: ${{ matrix.base-image }}
+        grpc-base-image: ${{ matrix.grpc-base-image }}
+        makeflags: ${{ matrix.makeflags }}
+        skip-drivers: ${{ matrix.skip-drivers }}
+        ubuntu-version: ${{ matrix.ubuntu-version }}
+        ubuntu-codename: ${{ matrix.ubuntu-codename }}
+      secrets:
+        dockerUsername: ${{ secrets.DOCKERHUB_USERNAME }}
+        dockerPassword: ${{ secrets.DOCKERHUB_PASSWORD }}
+        quayUsername: ${{ secrets.LOCALAI_REGISTRY_USERNAME }}
+        quayPassword: ${{ secrets.LOCALAI_REGISTRY_PASSWORD }}
+      strategy:
+        matrix:
+          include:
+            - build-type: 'cublas'
+              cuda-major-version: "12"
+              cuda-minor-version: "0"
+              platforms: 'linux/arm64'
+              tag-latest: 'auto'
+              tag-suffix: '-nvidia-l4t-arm64'
+              base-image: "nvcr.io/nvidia/l4t-jetpack:r36.4.0"
+              runs-on: 'ubuntu-24.04-arm'
+              makeflags: "--jobs=4 --output-sync=target"
+              skip-drivers: 'true'
+              ubuntu-version: "2204"
+              ubuntu-codename: 'jammy'
+            - build-type: 'cublas'
+              cuda-major-version: "13"
+              cuda-minor-version: "0"
+              platforms: 'linux/arm64'
+              tag-latest: 'auto'
+              tag-suffix: '-nvidia-l4t-arm64-cuda-13'
+              base-image: "ubuntu:24.04"
+              runs-on: 'ubuntu-24.04-arm'
+              makeflags: "--jobs=4 --output-sync=target"
+              skip-drivers: 'false'
+              ubuntu-version: '2404'
+              ubuntu-codename: 'noble'
+  

.github/workflows/image_build.yml

@@ -23,7 +23,7 @@ on:
         type: string
       cuda-minor-version:
         description: 'CUDA minor version'
-        default: "4"
+        default: "9"
         type: string
       platforms:
         description: 'Platforms'
@@ -61,6 +61,11 @@ on:
         required: false
         default: '2204'
         type: string
+      ubuntu-codename:
+        description: 'Ubuntu codename'
+        required: false
+        default: 'noble'
+        type: string
     secrets:
       dockerUsername:
         required: true
@@ -244,6 +249,7 @@ jobs:
             MAKEFLAGS=${{ inputs.makeflags }}
             SKIP_DRIVERS=${{ inputs.skip-drivers }}
             UBUNTU_VERSION=${{ inputs.ubuntu-version }}
+            UBUNTU_CODENAME=${{ inputs.ubuntu-codename }}
           context: .
           file: ./Dockerfile
           cache-from: type=gha
@@ -272,6 +278,7 @@ jobs:
             MAKEFLAGS=${{ inputs.makeflags }}
             SKIP_DRIVERS=${{ inputs.skip-drivers }}
             UBUNTU_VERSION=${{ inputs.ubuntu-version }}
+            UBUNTU_CODENAME=${{ inputs.ubuntu-codename }}
           context: .
           file: ./Dockerfile
           cache-from: type=gha

.github/workflows/test-extra.yml

@@ -238,7 +238,7 @@ jobs:
       - name: Dependencies
         run: |
           sudo apt-get update
-          sudo apt-get install build-essential ffmpeg
+          sudo apt-get install -y build-essential ffmpeg
           sudo apt-get install -y ca-certificates cmake curl patch espeak espeak-ng python3-pip
           # Install UV
           curl -LsSf https://astral.sh/uv/install.sh | sh
@@ -247,3 +247,98 @@ jobs:
         run: |
           make --jobs=5 --output-sync=target -C backend/python/coqui
           make --jobs=5 --output-sync=target -C backend/python/coqui test
+  tests-moonshine:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Clone
+        uses: actions/checkout@v6
+        with:
+          submodules: true
+      - name: Dependencies
+        run: |
+          sudo apt-get update
+          sudo apt-get install -y build-essential ffmpeg
+          sudo apt-get install -y ca-certificates cmake curl patch python3-pip
+          # Install UV
+          curl -LsSf https://astral.sh/uv/install.sh | sh
+          pip install --user --no-cache-dir grpcio-tools==1.64.1
+      - name: Test moonshine
+        run: |
+          make --jobs=5 --output-sync=target -C backend/python/moonshine
+          make --jobs=5 --output-sync=target -C backend/python/moonshine test
+  tests-pocket-tts:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Clone
+        uses: actions/checkout@v6
+        with:
+          submodules: true
+      - name: Dependencies
+        run: |
+          sudo apt-get update
+          sudo apt-get install -y build-essential ffmpeg
+          sudo apt-get install -y ca-certificates cmake curl patch python3-pip
+          # Install UV
+          curl -LsSf https://astral.sh/uv/install.sh | sh
+          pip install --user --no-cache-dir grpcio-tools==1.64.1
+      - name: Test pocket-tts
+        run: |
+          make --jobs=5 --output-sync=target -C backend/python/pocket-tts
+          make --jobs=5 --output-sync=target -C backend/python/pocket-tts test
+  tests-qwen-tts:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Clone
+        uses: actions/checkout@v6
+        with:
+          submodules: true
+      - name: Dependencies
+        run: |
+          sudo apt-get update
+          sudo apt-get install -y build-essential ffmpeg
+          sudo apt-get install -y ca-certificates cmake curl patch python3-pip
+          # Install UV
+          curl -LsSf https://astral.sh/uv/install.sh | sh
+          pip install --user --no-cache-dir grpcio-tools==1.64.1
+      - name: Test qwen-tts
+        run: |
+          make --jobs=5 --output-sync=target -C backend/python/qwen-tts
+          make --jobs=5 --output-sync=target -C backend/python/qwen-tts test
+  tests-qwen-asr:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Clone
+        uses: actions/checkout@v6
+        with:
+          submodules: true
+      - name: Dependencies
+        run: |
+          sudo apt-get update
+          sudo apt-get install -y build-essential ffmpeg sox
+          sudo apt-get install -y ca-certificates cmake curl patch python3-pip
+          # Install UV
+          curl -LsSf https://astral.sh/uv/install.sh | sh
+          pip install --user --no-cache-dir grpcio-tools==1.64.1
+      - name: Test qwen-asr
+        run: |
+          make --jobs=5 --output-sync=target -C backend/python/qwen-asr
+          make --jobs=5 --output-sync=target -C backend/python/qwen-asr test
+  tests-voxcpm:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Clone
+        uses: actions/checkout@v6
+        with:
+          submodules: true
+      - name: Dependencies
+        run: |
+          sudo apt-get update
+          sudo apt-get install build-essential ffmpeg
+          sudo apt-get install -y ca-certificates cmake curl patch python3-pip
+          # Install UV
+          curl -LsSf https://astral.sh/uv/install.sh | sh
+          pip install --user --no-cache-dir grpcio-tools==1.64.1
+      - name: Test voxcpm
+        run: |
+          make --jobs=5 --output-sync=target -C backend/python/voxcpm
+          make --jobs=5 --output-sync=target -C backend/python/voxcpm test

.github/workflows/tests-e2e.yml

@@ -0,0 +1,56 @@
+---
+name: 'E2E Backend Tests'
+
+on:
+  pull_request:
+  push:
+    branches:
+      - master
+    tags:
+      - '*'
+
+concurrency:
+  group: ci-tests-e2e-backend-${{ github.head_ref || github.ref }}-${{ github.repository }}
+  cancel-in-progress: true
+
+jobs:
+  tests-e2e-backend:
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        go-version: ['1.25.x']
+    steps:
+      - name: Clone
+        uses: actions/checkout@v6
+        with:
+          submodules: true
+      - name: Setup Go ${{ matrix.go-version }}
+        uses: actions/setup-go@v5
+        with:
+          go-version: ${{ matrix.go-version }}
+          cache: false
+      - name: Display Go version
+        run: go version
+      - name: Proto Dependencies
+        run: |
+          # Install protoc
+          curl -L -s https://github.com/protocolbuffers/protobuf/releases/download/v26.1/protoc-26.1-linux-x86_64.zip -o protoc.zip && \
+          unzip -j -d /usr/local/bin protoc.zip bin/protoc && \
+          rm protoc.zip
+          go install google.golang.org/protobuf/cmd/protoc-gen-go@v1.34.2
+          go install google.golang.org/grpc/cmd/protoc-gen-go-grpc@1958fcbe2ca8bd93af633f11e97d44e567e945af
+          PATH="$PATH:$HOME/go/bin" make protogen-go
+      - name: Dependencies
+        run: |
+          sudo apt-get update
+          sudo apt-get install -y build-essential
+      - name: Test Backend E2E
+        run: |
+          PATH="$PATH:$HOME/go/bin" make build-mock-backend test-e2e
+      - name: Setup tmate session if tests fail
+        if: ${{ failure() }}
+        uses: mxschmitt/action-tmate@v3.23
+        with:
+          detached: true
+          connect-timeout-seconds: 180
+          limit-access-to-actor: true

.gitignore

@@ -25,6 +25,7 @@ go-bert
 # LocalAI build binary
 LocalAI
 /local-ai
+/local-ai-launcher
 # prevent above rules from omitting the helm chart
 !charts/*
 # prevent above rules from omitting the api/localai folder
@@ -35,6 +36,8 @@ LocalAI
 models/*
 test-models/
 test-dir/
+tests/e2e-aio/backends
+tests/e2e-aio/models
 
 release/
 

AGENTS.md

@@ -2,6 +2,163 @@
 
 Building and testing the project depends on the components involved and the platform where development is taking place. Due to the amount of context required it's usually best not to try building or testing the project unless the user requests it. If you must build the project then inspect the Makefile in the project root and the Makefiles of any backends that are effected by changes you are making. In addition the workflows in .github/workflows can be used as a reference when it is unclear how to build or test a component. The primary Makefile contains targets for building inside or outside Docker, if the user has not previously specified a preference then ask which they would like to use.
 
+## Building a specified backend
+
+Let's say the user wants to build a particular backend for a given platform. For example let's say they want to build coqui for ROCM/hipblas
+
+- The Makefile has targets like `docker-build-coqui` created with `generate-docker-build-target` at the time of writing. Recently added backends may require a new target.
+- At a minimum we need to set the BUILD_TYPE, BASE_IMAGE build-args
+  - Use .github/workflows/backend.yml as a reference it lists the needed args in the `include` job strategy matrix
+  - l4t and cublas also requires the CUDA major and minor version
+- You can pretty print a command like `DOCKER_MAKEFLAGS=-j$(nproc --ignore=1) BUILD_TYPE=hipblas BASE_IMAGE=rocm/dev-ubuntu-24.04:6.4.4 make docker-build-coqui`
+- Unless the user specifies that they want you to run the command, then just print it because not all agent frontends handle long running jobs well and the output may overflow your context
+- The user may say they want to build AMD or ROCM instead of hipblas, or Intel instead of SYCL or NVIDIA insted of l4t or cublas. Ask for confirmation if there is ambiguity.
+- Sometimes the user may need extra parameters to be added to `docker build` (e.g. `--platform` for cross-platform builds or `--progress` to view the full logs), in which case you can generate the `docker build` command directly.
+
+## Adding a New Backend
+
+When adding a new backend to LocalAI, you need to update several files to ensure the backend is properly built, tested, and registered. Here's a step-by-step guide based on the pattern used for adding backends like `moonshine`:
+
+### 1. Create Backend Directory Structure
+
+Create the backend directory under the appropriate location:
+- **Python backends**: `backend/python/<backend-name>/`
+- **Go backends**: `backend/go/<backend-name>/`
+- **C++ backends**: `backend/cpp/<backend-name>/`
+
+For Python backends, you'll typically need:
+- `backend.py` - Main gRPC server implementation
+- `Makefile` - Build configuration
+- `install.sh` - Installation script for dependencies
+- `protogen.sh` - Protocol buffer generation script
+- `requirements.txt` - Python dependencies
+- `run.sh` - Runtime script
+- `test.py` / `test.sh` - Test files
+
+### 2. Add Build Configurations to `.github/workflows/backend.yml`
+
+Add build matrix entries for each platform/GPU type you want to support. Look at similar backends (e.g., `chatterbox`, `faster-whisper`) for reference.
+
+**Placement in file:**
+- CPU builds: Add after other CPU builds (e.g., after `cpu-chatterbox`)
+- CUDA 12 builds: Add after other CUDA 12 builds (e.g., after `gpu-nvidia-cuda-12-chatterbox`)
+- CUDA 13 builds: Add after other CUDA 13 builds (e.g., after `gpu-nvidia-cuda-13-chatterbox`)
+
+**Additional build types you may need:**
+- ROCm/HIP: Use `build-type: 'hipblas'` with `base-image: "rocm/dev-ubuntu-24.04:6.4.4"`
+- Intel/SYCL: Use `build-type: 'intel'` or `build-type: 'sycl_f16'`/`sycl_f32` with `base-image: "intel/oneapi-basekit:2025.3.0-0-devel-ubuntu24.04"`
+- L4T (ARM): Use `build-type: 'l4t'` with `platforms: 'linux/arm64'` and `runs-on: 'ubuntu-24.04-arm'`
+
+### 3. Add Backend Metadata to `backend/index.yaml`
+
+**Step 3a: Add Meta Definition**
+
+Add a YAML anchor definition in the `## metas` section (around line 2-300). Look for similar backends to use as a template such as `diffusers` or `chatterbox`
+
+**Step 3b: Add Image Entries**
+
+Add image entries at the end of the file, following the pattern of similar backends such as `diffusers` or `chatterbox`. Include both `latest` (production) and `master` (development) tags.
+
+### 4. Update the Makefile
+
+The Makefile needs to be updated in several places to support building and testing the new backend:
+
+**Step 4a: Add to `.NOTPARALLEL`**
+
+Add `backends/<backend-name>` to the `.NOTPARALLEL` line (around line 2) to prevent parallel execution conflicts:
+
+```makefile
+.NOTPARALLEL: ... backends/<backend-name>
+```
+
+**Step 4b: Add to `prepare-test-extra`**
+
+Add the backend to the `prepare-test-extra` target (around line 312) to prepare it for testing:
+
+```makefile
+prepare-test-extra: protogen-python
+	...
+	$(MAKE) -C backend/python/<backend-name>
+```
+
+**Step 4c: Add to `test-extra`**
+
+Add the backend to the `test-extra` target (around line 319) to run its tests:
+
+```makefile
+test-extra: prepare-test-extra
+	...
+	$(MAKE) -C backend/python/<backend-name> test
+```
+
+**Step 4d: Add Backend Definition**
+
+Add a backend definition variable in the backend definitions section (around line 428-457). The format depends on the backend type:
+
+**For Python backends with root context** (like `faster-whisper`, `coqui`):
+```makefile
+BACKEND_<BACKEND_NAME> = <backend-name>|python|.|false|true
+```
+
+**For Python backends with `./backend` context** (like `chatterbox`, `moonshine`):
+```makefile
+BACKEND_<BACKEND_NAME> = <backend-name>|python|./backend|false|true
+```
+
+**For Go backends**:
+```makefile
+BACKEND_<BACKEND_NAME> = <backend-name>|golang|.|false|true
+```
+
+**Step 4e: Generate Docker Build Target**
+
+Add an eval call to generate the docker-build target (around line 480-501):
+
+```makefile
+$(eval $(call generate-docker-build-target,$(BACKEND_<BACKEND_NAME>)))
+```
+
+**Step 4f: Add to `docker-build-backends`**
+
+Add `docker-build-<backend-name>` to the `docker-build-backends` target (around line 507):
+
+```makefile
+docker-build-backends: ... docker-build-<backend-name>
+```
+
+**Determining the Context:**
+
+- If the backend is in `backend/python/<backend-name>/` and uses `./backend` as context in the workflow file, use `./backend` context
+- If the backend is in `backend/python/<backend-name>/` but uses `.` as context in the workflow file, use `.` context
+- Check similar backends to determine the correct context
+
+### 5. Verification Checklist
+
+After adding a new backend, verify:
+
+- [ ] Backend directory structure is complete with all necessary files
+- [ ] Build configurations added to `.github/workflows/backend.yml` for all desired platforms
+- [ ] Meta definition added to `backend/index.yaml` in the `## metas` section
+- [ ] Image entries added to `backend/index.yaml` for all build variants (latest + development)
+- [ ] Tag suffixes match between workflow file and index.yaml
+- [ ] Makefile updated with all 6 required changes (`.NOTPARALLEL`, `prepare-test-extra`, `test-extra`, backend definition, docker-build target eval, `docker-build-backends`)
+- [ ] No YAML syntax errors (check with linter)
+- [ ] No Makefile syntax errors (check with linter)
+- [ ] Follows the same pattern as similar backends (e.g., if it's a transcription backend, follow `faster-whisper` pattern)
+
+### 6. Example: Adding a Python Backend
+
+For reference, when `moonshine` was added:
+- **Files created**: `backend/python/moonshine/{backend.py, Makefile, install.sh, protogen.sh, requirements.txt, run.sh, test.py, test.sh}`
+- **Workflow entries**: 3 build configurations (CPU, CUDA 12, CUDA 13)
+- **Index entries**: 1 meta definition + 6 image entries (cpu, cuda12, cuda13 × latest/development)
+- **Makefile updates**: 
+  - Added to `.NOTPARALLEL` line
+  - Added to `prepare-test-extra` and `test-extra` targets
+  - Added `BACKEND_MOONSHINE = moonshine|python|./backend|false|true`
+  - Added eval for docker-build target generation
+  - Added `docker-build-moonshine` to `docker-build-backends`
+
 # Coding style
 
 - The project has the following .editorconfig
@@ -77,3 +234,57 @@ When fixing compilation errors after upstream changes:
 - HTTP server uses `server_routes` with HTTP handlers
 - Both use the same `server_context` and task queue infrastructure
 - gRPC methods: `LoadModel`, `Predict`, `PredictStream`, `Embedding`, `Rerank`, `TokenizeString`, `GetMetrics`, `Health`
+
+## Tool Call Parsing Maintenance
+
+When working on JSON/XML tool call parsing functionality, always check llama.cpp for reference implementation and updates:
+
+### Checking for XML Parsing Changes
+
+1. **Review XML Format Definitions**: Check `llama.cpp/common/chat-parser-xml-toolcall.h` for `xml_tool_call_format` struct changes
+2. **Review Parsing Logic**: Check `llama.cpp/common/chat-parser-xml-toolcall.cpp` for parsing algorithm updates
+3. **Review Format Presets**: Check `llama.cpp/common/chat-parser.cpp` for new XML format presets (search for `xml_tool_call_format form`)
+4. **Review Model Lists**: Check `llama.cpp/common/chat.h` for `COMMON_CHAT_FORMAT_*` enum values that use XML parsing:
+   - `COMMON_CHAT_FORMAT_GLM_4_5`
+   - `COMMON_CHAT_FORMAT_MINIMAX_M2`
+   - `COMMON_CHAT_FORMAT_KIMI_K2`
+   - `COMMON_CHAT_FORMAT_QWEN3_CODER_XML`
+   - `COMMON_CHAT_FORMAT_APRIEL_1_5`
+   - `COMMON_CHAT_FORMAT_XIAOMI_MIMO`
+   - Any new formats added
+
+### Model Configuration Options
+
+Always check `llama.cpp` for new model configuration options that should be supported in LocalAI:
+
+1. **Check Server Context**: Review `llama.cpp/tools/server/server-context.cpp` for new parameters
+2. **Check Chat Params**: Review `llama.cpp/common/chat.h` for `common_chat_params` struct changes
+3. **Check Server Options**: Review `llama.cpp/tools/server/server.cpp` for command-line argument changes
+4. **Examples of options to check**:
+   - `ctx_shift` - Context shifting support
+   - `parallel_tool_calls` - Parallel tool calling
+   - `reasoning_format` - Reasoning format options
+   - Any new flags or parameters
+
+### Implementation Guidelines
+
+1. **Feature Parity**: Always aim for feature parity with llama.cpp's implementation
+2. **Test Coverage**: Add tests for new features matching llama.cpp's behavior
+3. **Documentation**: Update relevant documentation when adding new formats or options
+4. **Backward Compatibility**: Ensure changes don't break existing functionality
+
+### Files to Monitor
+
+- `llama.cpp/common/chat-parser-xml-toolcall.h` - Format definitions
+- `llama.cpp/common/chat-parser-xml-toolcall.cpp` - Parsing logic
+- `llama.cpp/common/chat-parser.cpp` - Format presets and model-specific handlers
+- `llama.cpp/common/chat.h` - Format enums and parameter structures
+- `llama.cpp/tools/server/server-context.cpp` - Server configuration options
+
+# Documentation
+
+The project documentation is located in `docs/content`. When adding new features or changing existing functionality, it is crucial to update the documentation to reflect these changes. This helps users understand how to use the new capabilities and ensures the documentation stays relevant.
+
+- **Feature Documentation**: If you add a new feature (like a new backend or API endpoint), create a new markdown file in `docs/content/features/` explaining what it is, how to configure it, and how to use it.
+- **Configuration**: If you modify configuration options, update the relevant sections in `docs/content/`.
+- **Examples**: providing concrete examples (like YAML configuration blocks) is highly encouraged to help users get started quickly.

CONTRIBUTING.md

@@ -78,6 +78,20 @@ LOCALAI_IMAGE_TAG=test LOCALAI_IMAGE=local-ai-aio make run-e2e-aio
 
 We are welcome the contribution of the documents, please open new PR or create a new issue. The documentation is available under `docs/` https://github.com/mudler/LocalAI/tree/master/docs
 
+### Gallery YAML Schema
+
+LocalAI provides a JSON Schema for gallery model YAML files at:
+
+`core/schema/gallery-model.schema.json`
+
+This schema mirrors the internal gallery model configuration and can be used by editors (such as VS Code) to enable autocomplete, validation, and inline documentation when creating or modifying gallery files.
+
+To use it with the YAML language server, add the following comment at the top of a gallery YAML file:
+
+```yaml
+# yaml-language-server: $schema=../core/schema/gallery-model.schema.json
+```
+
 ## Community and Communication
 
 - You can reach out via the Github issue tracker.

Dockerfile

@@ -1,6 +1,7 @@
-ARG BASE_IMAGE=ubuntu:22.04
+ARG BASE_IMAGE=ubuntu:24.04
 ARG GRPC_BASE_IMAGE=${BASE_IMAGE}
 ARG INTEL_BASE_IMAGE=${BASE_IMAGE}
+ARG UBUNTU_CODENAME=noble
 
 FROM ${BASE_IMAGE} AS requirements
 
@@ -9,7 +10,7 @@ ENV DEBIAN_FRONTEND=noninteractive
 RUN apt-get update && \
     apt-get install -y --no-install-recommends \
         ca-certificates curl wget espeak-ng libgomp1 \
-        ffmpeg && \
+        ffmpeg libopenblas0 libopenblas-dev sox && \
     apt-get clean && \
     rm -rf /var/lib/apt/lists/*
 
@@ -23,7 +24,7 @@ ARG SKIP_DRIVERS=false
 ARG TARGETARCH
 ARG TARGETVARIANT
 ENV BUILD_TYPE=${BUILD_TYPE}
-ARG UBUNTU_VERSION=2204
+ARG UBUNTU_VERSION=2404
 
 RUN mkdir -p /run/localai
 RUN echo "default" > /run/localai/capability
@@ -34,11 +35,45 @@ RUN <<EOT bash
         apt-get update && \
         apt-get install -y  --no-install-recommends \
             software-properties-common pciutils wget gpg-agent && \
-        wget -qO - https://packages.lunarg.com/lunarg-signing-key-pub.asc | apt-key add - && \
-        wget -qO /etc/apt/sources.list.d/lunarg-vulkan-jammy.list https://packages.lunarg.com/vulkan/lunarg-vulkan-jammy.list && \
-        apt-get update && \
-        apt-get install -y \
-            vulkan-sdk && \
+        apt-get install -y libglm-dev cmake libxcb-dri3-0 libxcb-present0 libpciaccess0 \
+            libpng-dev libxcb-keysyms1-dev libxcb-dri3-dev libx11-dev g++ gcc \
+            libwayland-dev libxrandr-dev libxcb-randr0-dev libxcb-ewmh-dev \
+            git python-is-python3 bison libx11-xcb-dev liblz4-dev libzstd-dev \
+            ocaml-core ninja-build pkg-config libxml2-dev wayland-protocols python3-jsonschema \
+            clang-format qtbase5-dev qt6-base-dev libxcb-glx0-dev sudo xz-utils mesa-vulkan-drivers
+        if [ "amd64" = "$TARGETARCH" ]; then
+            wget "https://sdk.lunarg.com/sdk/download/1.4.335.0/linux/vulkansdk-linux-x86_64-1.4.335.0.tar.xz" && \
+            tar -xf vulkansdk-linux-x86_64-1.4.335.0.tar.xz && \
+            rm vulkansdk-linux-x86_64-1.4.335.0.tar.xz && \
+            mkdir -p /opt/vulkan-sdk && \
+            mv 1.4.335.0 /opt/vulkan-sdk/ && \
+            cd /opt/vulkan-sdk/1.4.335.0 && \
+            ./vulkansdk --no-deps --maxjobs \
+                vulkan-loader \
+                vulkan-validationlayers \
+                vulkan-extensionlayer \
+                vulkan-tools \
+                shaderc && \
+            cp -rfv /opt/vulkan-sdk/1.4.335.0/x86_64/bin/* /usr/bin/ && \
+            cp -rfv /opt/vulkan-sdk/1.4.335.0/x86_64/lib/* /usr/lib/x86_64-linux-gnu/ && \
+            cp -rfv /opt/vulkan-sdk/1.4.335.0/x86_64/include/* /usr/include/ && \
+            cp -rfv /opt/vulkan-sdk/1.4.335.0/x86_64/share/* /usr/share/ && \
+            rm -rf /opt/vulkan-sdk
+        fi
+        if [ "arm64" = "$TARGETARCH" ]; then
+            mkdir vulkan && cd vulkan && \
+            curl -L -o vulkan-sdk.tar.xz https://github.com/mudler/vulkan-sdk-arm/releases/download/1.4.335.0/vulkansdk-ubuntu-24.04-arm-1.4.335.0.tar.xz && \
+            tar -xvf vulkan-sdk.tar.xz && \
+            rm vulkan-sdk.tar.xz && \
+            cd 1.4.335.0 && \
+            cp -rfv aarch64/bin/* /usr/bin/ && \
+            cp -rfv aarch64/lib/* /usr/lib/aarch64-linux-gnu/ && \
+            cp -rfv aarch64/include/* /usr/include/ && \
+            cp -rfv aarch64/share/* /usr/share/ && \
+            cd ../.. && \
+            rm -rf vulkan
+        fi
+        ldconfig && \
         apt-get clean && \
         rm -rf /var/lib/apt/lists/* && \
         echo "vulkan" > /run/localai/capability
@@ -141,13 +176,12 @@ ENV PATH=/opt/rocm/bin:${PATH}
 # The requirements-core target is common to all images.  It should not be placed in requirements-core unless every single build will use it.
 FROM requirements-drivers AS build-requirements
 
-ARG GO_VERSION=1.22.6
-ARG CMAKE_VERSION=3.26.4
+ARG GO_VERSION=1.25.4
+ARG CMAKE_VERSION=3.31.10
 ARG CMAKE_FROM_SOURCE=false
 ARG TARGETARCH
 ARG TARGETVARIANT
 
-
 RUN apt-get update && \
     apt-get install -y --no-install-recommends \
         build-essential \
@@ -204,9 +238,10 @@ WORKDIR /build
 # https://community.intel.com/t5/Intel-oneAPI-Math-Kernel-Library/APT-Repository-not-working-signatures-invalid/m-p/1599436/highlight/true#M36143
 # This is a temporary workaround until Intel fixes their repository
 FROM ${INTEL_BASE_IMAGE} AS intel
+ARG UBUNTU_CODENAME=noble
 RUN wget -qO - https://repositories.intel.com/gpu/intel-graphics.key | \
 gpg --yes --dearmor --output /usr/share/keyrings/intel-graphics.gpg
-RUN echo "deb [arch=amd64 signed-by=/usr/share/keyrings/intel-graphics.gpg] https://repositories.intel.com/gpu/ubuntu jammy/lts/2350 unified" > /etc/apt/sources.list.d/intel-graphics.list
+RUN echo "deb [arch=amd64 signed-by=/usr/share/keyrings/intel-graphics.gpg] https://repositories.intel.com/gpu/ubuntu ${UBUNTU_CODENAME}/lts/2350 unified" > /etc/apt/sources.list.d/intel-graphics.list
 RUN apt-get update && \
     apt-get install -y --no-install-recommends \
         intel-oneapi-runtime-libs && \

Dockerfile.aio

@@ -1,4 +1,4 @@
-ARG BASE_IMAGE=ubuntu:22.04
+ARG BASE_IMAGE=ubuntu:24.04
 
 FROM ${BASE_IMAGE} 
 

Makefile

@@ -1,15 +1,20 @@
+# Disable parallel execution for backend builds
+.NOTPARALLEL: backends/diffusers backends/llama-cpp backends/piper backends/stablediffusion-ggml backends/whisper backends/faster-whisper backends/silero-vad backends/local-store backends/huggingface backends/rfdetr backends/kitten-tts backends/kokoro backends/chatterbox backends/llama-cpp-darwin backends/neutts build-darwin-python-backend build-darwin-go-backend backends/mlx backends/diffuser-darwin backends/mlx-vlm backends/mlx-audio backends/stablediffusion-ggml-darwin backends/vllm backends/vllm-omni backends/moonshine backends/pocket-tts backends/qwen-tts backends/qwen-asr backends/voxcpm backends/whisperx
+
 GOCMD=go
 GOTEST=$(GOCMD) test
 GOVET=$(GOCMD) vet
 BINARY_NAME=local-ai
 LAUNCHER_BINARY_NAME=local-ai-launcher
 
-CUDA_MAJOR_VERSION?=13
-CUDA_MINOR_VERSION?=0
+UBUNTU_VERSION?=2404
+UBUNTU_CODENAME?=noble
 
 GORELEASER?=
 
 export BUILD_TYPE?=
+export CUDA_MAJOR_VERSION?=13
+export CUDA_MINOR_VERSION?=0
 
 GO_TAGS?=
 BUILD_ID?=
@@ -155,7 +160,17 @@ test: test-models/testmodel.ggml protogen-go
 ########################################################
 
 docker-build-aio:
-	docker build --build-arg MAKEFLAGS="--jobs=5 --output-sync=target" -t local-ai:tests -f Dockerfile .
+	docker build \
+		--build-arg MAKEFLAGS="--jobs=5 --output-sync=target" \
+		--build-arg BASE_IMAGE=$(BASE_IMAGE) \
+		--build-arg IMAGE_TYPE=$(IMAGE_TYPE) \
+		--build-arg BUILD_TYPE=$(BUILD_TYPE) \
+		--build-arg CUDA_MAJOR_VERSION=$(CUDA_MAJOR_VERSION) \
+		--build-arg CUDA_MINOR_VERSION=$(CUDA_MINOR_VERSION) \
+		--build-arg UBUNTU_VERSION=$(UBUNTU_VERSION) \
+		--build-arg UBUNTU_CODENAME=$(UBUNTU_CODENAME) \
+		--build-arg GO_TAGS="$(GO_TAGS)" \
+		-t local-ai:tests -f Dockerfile .
 	BASE_IMAGE=local-ai:tests DOCKER_AIO_IMAGE=local-ai-aio:test $(MAKE) docker-aio
 
 e2e-aio:
@@ -174,20 +189,29 @@ run-e2e-aio: protogen-go
 ########################################################
 
 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=0 -t localai-tests .
+	docker build \
+		--build-arg IMAGE_TYPE=core \
+		--build-arg BUILD_TYPE=$(BUILD_TYPE) \
+		--build-arg BASE_IMAGE=$(BASE_IMAGE) \
+		--build-arg CUDA_MAJOR_VERSION=$(CUDA_MAJOR_VERSION) \
+		--build-arg CUDA_MINOR_VERSION=$(CUDA_MINOR_VERSION) \
+		--build-arg UBUNTU_VERSION=$(UBUNTU_VERSION) \
+		--build-arg UBUNTU_CODENAME=$(UBUNTU_CODENAME) \
+		--build-arg GO_TAGS="$(GO_TAGS)" \
+		--build-arg MAKEFLAGS="$(DOCKER_MAKEFLAGS)" \
+		-t localai-tests .
 
 run-e2e-image:
-	ls -liah $(abspath ./tests/e2e-fixtures)
-	docker run -p 5390:8080 -e MODELS_PATH=/models -e THREADS=1 -e DEBUG=true -d --rm -v $(TEST_DIR):/models --gpus all --name e2e-tests-$(RANDOM) localai-tests
+	docker run -p 5390:8080 -e MODELS_PATH=/models -e THREADS=1 -e DEBUG=true -d --rm -v $(TEST_DIR):/models --name e2e-tests-$(RANDOM) localai-tests
 
-test-e2e:
+test-e2e: build-mock-backend prepare-e2e run-e2e-image
 	@echo 'Running e2e tests'
 	BUILD_TYPE=$(BUILD_TYPE) \
-	LOCALAI_API=http://$(E2E_BRIDGE_IP):5390/v1 \
+	LOCALAI_API=http://$(E2E_BRIDGE_IP):5390 \
 	$(GOCMD) run github.com/onsi/ginkgo/v2/ginkgo --flake-attempts $(TEST_FLAKES) -v -r ./tests/e2e
+	$(MAKE) clean-mock-backend
+	$(MAKE) teardown-e2e
+	docker rmi localai-tests
 
 teardown-e2e:
 	rm -rf $(TEST_DIR) || true
@@ -287,19 +311,33 @@ prepare-test-extra: protogen-python
 	$(MAKE) -C backend/python/diffusers
 	$(MAKE) -C backend/python/chatterbox
 	$(MAKE) -C backend/python/vllm
+	$(MAKE) -C backend/python/vllm-omni
 	$(MAKE) -C backend/python/vibevoice
+	$(MAKE) -C backend/python/moonshine
+	$(MAKE) -C backend/python/pocket-tts
+	$(MAKE) -C backend/python/qwen-tts
+	$(MAKE) -C backend/python/qwen-asr
+	$(MAKE) -C backend/python/voxcpm
+	$(MAKE) -C backend/python/whisperx
 
 test-extra: prepare-test-extra
 	$(MAKE) -C backend/python/transformers test
 	$(MAKE) -C backend/python/diffusers test
 	$(MAKE) -C backend/python/chatterbox test
 	$(MAKE) -C backend/python/vllm test
+	$(MAKE) -C backend/python/vllm-omni test
 	$(MAKE) -C backend/python/vibevoice test
+	$(MAKE) -C backend/python/moonshine test
+	$(MAKE) -C backend/python/pocket-tts test
+	$(MAKE) -C backend/python/qwen-tts test
+	$(MAKE) -C backend/python/qwen-asr test
+	$(MAKE) -C backend/python/voxcpm test
+	$(MAKE) -C backend/python/whisperx test
 
 DOCKER_IMAGE?=local-ai
 DOCKER_AIO_IMAGE?=local-ai-aio
 IMAGE_TYPE?=core
-BASE_IMAGE?=ubuntu:22.04
+BASE_IMAGE?=ubuntu:24.04
 
 docker:
 	docker build \
@@ -308,24 +346,34 @@ docker:
 		--build-arg GO_TAGS="$(GO_TAGS)" \
 		--build-arg MAKEFLAGS="$(DOCKER_MAKEFLAGS)" \
 		--build-arg BUILD_TYPE=$(BUILD_TYPE) \
+		--build-arg CUDA_MAJOR_VERSION=$(CUDA_MAJOR_VERSION) \
+		--build-arg CUDA_MINOR_VERSION=$(CUDA_MINOR_VERSION) \
+		--build-arg UBUNTU_VERSION=$(UBUNTU_VERSION) \
+		--build-arg UBUNTU_CODENAME=$(UBUNTU_CODENAME) \
 		-t $(DOCKER_IMAGE) .
 
-docker-cuda11:
+docker-cuda12:
 	docker build \
-		--build-arg CUDA_MAJOR_VERSION=11 \
-		--build-arg CUDA_MINOR_VERSION=8 \
+		--build-arg CUDA_MAJOR_VERSION=${CUDA_MAJOR_VERSION} \
+		--build-arg CUDA_MINOR_VERSION=${CUDA_MINOR_VERSION} \
 		--build-arg BASE_IMAGE=$(BASE_IMAGE) \
 		--build-arg IMAGE_TYPE=$(IMAGE_TYPE) \
 		--build-arg GO_TAGS="$(GO_TAGS)" \
 		--build-arg MAKEFLAGS="$(DOCKER_MAKEFLAGS)" \
 		--build-arg BUILD_TYPE=$(BUILD_TYPE) \
-		-t $(DOCKER_IMAGE)-cuda-11 .
+		--build-arg UBUNTU_VERSION=$(UBUNTU_VERSION) \
+		--build-arg UBUNTU_CODENAME=$(UBUNTU_CODENAME) \
+		-t $(DOCKER_IMAGE)-cuda-12 .
 
 docker-aio:
 	@echo "Building AIO image with base $(BASE_IMAGE) as $(DOCKER_AIO_IMAGE)"
 	docker build \
 		--build-arg BASE_IMAGE=$(BASE_IMAGE) \
 		--build-arg MAKEFLAGS="$(DOCKER_MAKEFLAGS)" \
+		--build-arg CUDA_MAJOR_VERSION=$(CUDA_MAJOR_VERSION) \
+		--build-arg CUDA_MINOR_VERSION=$(CUDA_MINOR_VERSION) \
+		--build-arg UBUNTU_VERSION=$(UBUNTU_VERSION) \
+		--build-arg UBUNTU_CODENAME=$(UBUNTU_CODENAME) \
 		-t $(DOCKER_AIO_IMAGE) -f Dockerfile.aio .
 
 docker-aio-all:
@@ -334,66 +382,31 @@ docker-aio-all:
 
 docker-image-intel:
 	docker build \
-		--build-arg BASE_IMAGE=quay.io/go-skynet/intel-oneapi-base:latest \
+		--build-arg BASE_IMAGE=intel/oneapi-basekit:2025.3.0-0-devel-ubuntu24.04 \
 		--build-arg IMAGE_TYPE=$(IMAGE_TYPE) \
 		--build-arg GO_TAGS="$(GO_TAGS)" \
 		--build-arg MAKEFLAGS="$(DOCKER_MAKEFLAGS)" \
-		--build-arg BUILD_TYPE=intel -t $(DOCKER_IMAGE) .
+		--build-arg BUILD_TYPE=intel \
+		--build-arg CUDA_MAJOR_VERSION=$(CUDA_MAJOR_VERSION) \
+		--build-arg CUDA_MINOR_VERSION=$(CUDA_MINOR_VERSION) \
+		--build-arg UBUNTU_VERSION=$(UBUNTU_VERSION) \
+		--build-arg UBUNTU_CODENAME=$(UBUNTU_CODENAME) \
+		-t $(DOCKER_IMAGE) .
 
 ########################################################
 ## Backends
 ########################################################
 
+# Pattern rule for standard backends (docker-based)
+# This matches all backends that use docker-build-* and docker-save-*
+backends/%: docker-build-% docker-save-% build
+	./local-ai backends install "ocifile://$(abspath ./backend-images/$*.tar)"
 
-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)"
-
+# Darwin-specific backends (keep as explicit targets since they have special build logic)
 backends/llama-cpp-darwin: build
 	bash ./scripts/build/llama-cpp-darwin.sh
 	./local-ai backends install "ocifile://$(abspath ./backend-images/llama-cpp.tar)"
 
-backends/neutts: docker-build-neutts docker-save-neutts build
-	./local-ai backends install "ocifile://$(abspath ./backend-images/neutts.tar)"
-
-backends/vllm: docker-build-vllm docker-save-vllm build
-	./local-ai backends install "ocifile://$(abspath ./backend-images/vllm.tar)"
-
-backends/vibevoice: docker-build-vibevoice docker-save-vibevoice build
-	./local-ai backends install "ocifile://$(abspath ./backend-images/vibevoice.tar)"
-
 build-darwin-python-backend: build
 	bash ./scripts/build/python-darwin.sh
 
@@ -423,121 +436,102 @@ backends/stablediffusion-ggml-darwin:
 backend-images:
 	mkdir -p backend-images
 
-docker-build-llama-cpp:
-	docker build --build-arg BUILD_TYPE=$(BUILD_TYPE) --build-arg BASE_IMAGE=$(BASE_IMAGE) -t local-ai-backend:llama-cpp -f backend/Dockerfile.llama-cpp .
+# Backend metadata: BACKEND_NAME | DOCKERFILE_TYPE | BUILD_CONTEXT | PROGRESS_FLAG | NEEDS_BACKEND_ARG
+# llama-cpp is special - uses llama-cpp Dockerfile and doesn't need BACKEND arg
+BACKEND_LLAMA_CPP = llama-cpp|llama-cpp|.|false|false
 
-docker-build-bark-cpp:
-	docker build --build-arg BUILD_TYPE=$(BUILD_TYPE) --build-arg BASE_IMAGE=$(BASE_IMAGE) -t local-ai-backend:bark-cpp -f backend/Dockerfile.golang --build-arg BACKEND=bark-cpp .
+# Golang backends
+BACKEND_PIPER = piper|golang|.|false|true
+BACKEND_LOCAL_STORE = local-store|golang|.|false|true
+BACKEND_HUGGINGFACE = huggingface|golang|.|false|true
+BACKEND_SILERO_VAD = silero-vad|golang|.|false|true
+BACKEND_STABLEDIFFUSION_GGML = stablediffusion-ggml|golang|.|--progress=plain|true
+BACKEND_WHISPER = whisper|golang|.|false|true
 
-docker-build-piper:
-	docker build --build-arg BUILD_TYPE=$(BUILD_TYPE) --build-arg BASE_IMAGE=$(BASE_IMAGE) -t local-ai-backend:piper -f backend/Dockerfile.golang --build-arg BACKEND=piper .
+# Python backends with root context
+BACKEND_RERANKERS = rerankers|python|.|false|true
+BACKEND_TRANSFORMERS = transformers|python|.|false|true
+BACKEND_FASTER_WHISPER = faster-whisper|python|.|false|true
+BACKEND_COQUI = coqui|python|.|false|true
+BACKEND_RFDETR = rfdetr|python|.|false|true
+BACKEND_KITTEN_TTS = kitten-tts|python|.|false|true
+BACKEND_NEUTTS = neutts|python|.|false|true
+BACKEND_KOKORO = kokoro|python|.|false|true
+BACKEND_VLLM = vllm|python|.|false|true
+BACKEND_VLLM_OMNI = vllm-omni|python|.|false|true
+BACKEND_DIFFUSERS = diffusers|python|.|--progress=plain|true
+BACKEND_CHATTERBOX = chatterbox|python|.|false|true
+BACKEND_VIBEVOICE = vibevoice|python|.|--progress=plain|true
+BACKEND_MOONSHINE = moonshine|python|.|false|true
+BACKEND_POCKET_TTS = pocket-tts|python|.|false|true
+BACKEND_QWEN_TTS = qwen-tts|python|.|false|true
+BACKEND_QWEN_ASR = qwen-asr|python|.|false|true
+BACKEND_VOXCPM = voxcpm|python|.|false|true
+BACKEND_WHISPERX = whisperx|python|.|false|true
 
-docker-build-local-store:
-	docker build --build-arg BUILD_TYPE=$(BUILD_TYPE) --build-arg BASE_IMAGE=$(BASE_IMAGE) -t local-ai-backend:local-store -f backend/Dockerfile.golang --build-arg BACKEND=local-store .
+# Helper function to build docker image for a backend
+# Usage: $(call docker-build-backend,BACKEND_NAME,DOCKERFILE_TYPE,BUILD_CONTEXT,PROGRESS_FLAG,NEEDS_BACKEND_ARG)
+define docker-build-backend
+	docker build $(if $(filter-out false,$(4)),$(4)) \
+		--build-arg BUILD_TYPE=$(BUILD_TYPE) \
+		--build-arg BASE_IMAGE=$(BASE_IMAGE) \
+		--build-arg CUDA_MAJOR_VERSION=$(CUDA_MAJOR_VERSION) \
+		--build-arg CUDA_MINOR_VERSION=$(CUDA_MINOR_VERSION) \
+		--build-arg UBUNTU_VERSION=$(UBUNTU_VERSION) \
+		--build-arg UBUNTU_CODENAME=$(UBUNTU_CODENAME) \
+		$(if $(filter true,$(5)),--build-arg BACKEND=$(1)) \
+		-t local-ai-backend:$(1) -f backend/Dockerfile.$(2) $(3)
+endef
 
-docker-build-huggingface:
-	docker build --build-arg BUILD_TYPE=$(BUILD_TYPE) --build-arg BASE_IMAGE=$(BASE_IMAGE) -t local-ai-backend:huggingface -f backend/Dockerfile.golang --build-arg BACKEND=huggingface .
+# Generate docker-build targets from backend definitions
+define generate-docker-build-target
+docker-build-$(word 1,$(subst |, ,$(1))):
+	$$(call docker-build-backend,$(word 1,$(subst |, ,$(1))),$(word 2,$(subst |, ,$(1))),$(word 3,$(subst |, ,$(1))),$(word 4,$(subst |, ,$(1))),$(word 5,$(subst |, ,$(1))))
+endef
 
-docker-build-rfdetr:
-	docker build --build-arg BUILD_TYPE=$(BUILD_TYPE) --build-arg BASE_IMAGE=$(BASE_IMAGE) -t local-ai-backend:rfdetr -f backend/Dockerfile.python --build-arg BACKEND=rfdetr ./backend
+# Generate all docker-build targets
+$(eval $(call generate-docker-build-target,$(BACKEND_LLAMA_CPP)))
+$(eval $(call generate-docker-build-target,$(BACKEND_PIPER)))
+$(eval $(call generate-docker-build-target,$(BACKEND_LOCAL_STORE)))
+$(eval $(call generate-docker-build-target,$(BACKEND_HUGGINGFACE)))
+$(eval $(call generate-docker-build-target,$(BACKEND_SILERO_VAD)))
+$(eval $(call generate-docker-build-target,$(BACKEND_STABLEDIFFUSION_GGML)))
+$(eval $(call generate-docker-build-target,$(BACKEND_WHISPER)))
+$(eval $(call generate-docker-build-target,$(BACKEND_RERANKERS)))
+$(eval $(call generate-docker-build-target,$(BACKEND_TRANSFORMERS)))
+$(eval $(call generate-docker-build-target,$(BACKEND_FASTER_WHISPER)))
+$(eval $(call generate-docker-build-target,$(BACKEND_COQUI)))
+$(eval $(call generate-docker-build-target,$(BACKEND_RFDETR)))
+$(eval $(call generate-docker-build-target,$(BACKEND_KITTEN_TTS)))
+$(eval $(call generate-docker-build-target,$(BACKEND_NEUTTS)))
+$(eval $(call generate-docker-build-target,$(BACKEND_KOKORO)))
+$(eval $(call generate-docker-build-target,$(BACKEND_VLLM)))
+$(eval $(call generate-docker-build-target,$(BACKEND_VLLM_OMNI)))
+$(eval $(call generate-docker-build-target,$(BACKEND_DIFFUSERS)))
+$(eval $(call generate-docker-build-target,$(BACKEND_CHATTERBOX)))
+$(eval $(call generate-docker-build-target,$(BACKEND_VIBEVOICE)))
+$(eval $(call generate-docker-build-target,$(BACKEND_MOONSHINE)))
+$(eval $(call generate-docker-build-target,$(BACKEND_POCKET_TTS)))
+$(eval $(call generate-docker-build-target,$(BACKEND_QWEN_TTS)))
+$(eval $(call generate-docker-build-target,$(BACKEND_QWEN_ASR)))
+$(eval $(call generate-docker-build-target,$(BACKEND_VOXCPM)))
+$(eval $(call generate-docker-build-target,$(BACKEND_WHISPERX)))
 
-docker-build-kitten-tts:
-	docker build --build-arg BUILD_TYPE=$(BUILD_TYPE) --build-arg BASE_IMAGE=$(BASE_IMAGE) -t local-ai-backend:kitten-tts -f backend/Dockerfile.python --build-arg BACKEND=kitten-tts ./backend
+# Pattern rule for docker-save targets
+docker-save-%: backend-images
+	docker save local-ai-backend:$* -o backend-images/$*.tar
 
-docker-save-kitten-tts: backend-images
-	docker save local-ai-backend:kitten-tts -o backend-images/kitten-tts.tar
+docker-build-backends: docker-build-llama-cpp docker-build-rerankers docker-build-vllm docker-build-vllm-omni docker-build-transformers docker-build-diffusers docker-build-kokoro docker-build-faster-whisper docker-build-coqui docker-build-chatterbox docker-build-vibevoice docker-build-moonshine docker-build-pocket-tts docker-build-qwen-tts docker-build-qwen-asr docker-build-voxcpm docker-build-whisperx
 
-docker-save-chatterbox: backend-images
-	docker save local-ai-backend:chatterbox -o backend-images/chatterbox.tar
+########################################################
+### Mock Backend for E2E Tests
+########################################################
 
-docker-save-vibevoice: backend-images
-	docker save local-ai-backend:vibevoice -o backend-images/vibevoice.tar
+build-mock-backend: protogen-go
+	$(GOCMD) build -o tests/e2e/mock-backend/mock-backend ./tests/e2e/mock-backend
 
-docker-build-neutts:
-	docker build --build-arg BUILD_TYPE=$(BUILD_TYPE) --build-arg BASE_IMAGE=$(BASE_IMAGE) -t local-ai-backend:neutts -f backend/Dockerfile.python --build-arg BACKEND=neutts ./backend
-
-docker-save-neutts: backend-images
-	docker save local-ai-backend:neutts -o backend-images/neutts.tar
-
-docker-build-kokoro:
-	docker build --build-arg BUILD_TYPE=$(BUILD_TYPE) --build-arg BASE_IMAGE=$(BASE_IMAGE) -t local-ai-backend:kokoro -f backend/Dockerfile.python --build-arg BACKEND=kokoro ./backend
-
-docker-build-vllm:
-	docker build --build-arg CUDA_MAJOR_VERSION=$(CUDA_MAJOR_VERSION) --build-arg CUDA_MINOR_VERSION=$(CUDA_MINOR_VERSION) --build-arg BUILD_TYPE=$(BUILD_TYPE) --build-arg BASE_IMAGE=$(BASE_IMAGE) -t local-ai-backend:vllm -f backend/Dockerfile.python --build-arg BACKEND=vllm ./backend
-
-docker-save-vllm: backend-images
-	docker save local-ai-backend:vllm -o backend-images/vllm.tar
-
-docker-save-kokoro: backend-images
-	docker save local-ai-backend:kokoro -o backend-images/kokoro.tar
-
-docker-save-rfdetr: backend-images
-	docker save local-ai-backend:rfdetr -o backend-images/rfdetr.tar
-
-docker-save-huggingface: backend-images
-	docker save local-ai-backend:huggingface -o backend-images/huggingface.tar
-
-docker-save-local-store: backend-images
-	docker save local-ai-backend:local-store -o backend-images/local-store.tar
-
-docker-build-silero-vad:
-	docker build --build-arg BUILD_TYPE=$(BUILD_TYPE) --build-arg BASE_IMAGE=$(BASE_IMAGE) -t local-ai-backend:silero-vad -f backend/Dockerfile.golang --build-arg BACKEND=silero-vad .
-
-docker-save-silero-vad: backend-images
-	docker save local-ai-backend:silero-vad -o backend-images/silero-vad.tar
-
-docker-save-piper: backend-images
-	docker save local-ai-backend:piper -o backend-images/piper.tar
-
-docker-save-llama-cpp: backend-images
-	docker save local-ai-backend:llama-cpp -o backend-images/llama-cpp.tar
-
-docker-save-bark-cpp: backend-images
-	docker save local-ai-backend:bark-cpp -o backend-images/bark-cpp.tar
-
-docker-build-stablediffusion-ggml:
-	docker build --progress=plain --build-arg BUILD_TYPE=$(BUILD_TYPE) --build-arg BASE_IMAGE=$(BASE_IMAGE) --build-arg CUDA_MAJOR_VERSION=$(CUDA_MAJOR_VERSION) --build-arg CUDA_MINOR_VERSION=$(CUDA_MINOR_VERSION) -t local-ai-backend:stablediffusion-ggml -f backend/Dockerfile.golang --build-arg BACKEND=stablediffusion-ggml .
-
-docker-save-stablediffusion-ggml: backend-images
-	docker save local-ai-backend:stablediffusion-ggml -o backend-images/stablediffusion-ggml.tar
-
-docker-build-rerankers:
-	docker build --build-arg BUILD_TYPE=$(BUILD_TYPE) --build-arg BASE_IMAGE=$(BASE_IMAGE) -t local-ai-backend:rerankers -f backend/Dockerfile.python --build-arg BACKEND=rerankers .
-
-docker-build-transformers:
-	docker build --build-arg BUILD_TYPE=$(BUILD_TYPE) --build-arg BASE_IMAGE=$(BASE_IMAGE) -t local-ai-backend:transformers -f backend/Dockerfile.python --build-arg BACKEND=transformers .
-
-docker-build-diffusers:
-	docker build --progress=plain --build-arg BUILD_TYPE=$(BUILD_TYPE) --build-arg BASE_IMAGE=$(BASE_IMAGE) -t local-ai-backend:diffusers -f backend/Dockerfile.python --build-arg BACKEND=diffusers ./backend
-
-docker-save-diffusers: backend-images
-	docker save local-ai-backend:diffusers -o backend-images/diffusers.tar
-
-docker-build-whisper:
-	docker build --build-arg BUILD_TYPE=$(BUILD_TYPE) --build-arg BASE_IMAGE=$(BASE_IMAGE) --build-arg CUDA_MAJOR_VERSION=$(CUDA_MAJOR_VERSION) --build-arg CUDA_MINOR_VERSION=$(CUDA_MINOR_VERSION) -t local-ai-backend:whisper -f backend/Dockerfile.golang --build-arg BACKEND=whisper  .
-
-docker-save-whisper: backend-images
-	docker save local-ai-backend:whisper -o backend-images/whisper.tar
-
-docker-build-faster-whisper:
-	docker build --build-arg BUILD_TYPE=$(BUILD_TYPE) --build-arg BASE_IMAGE=$(BASE_IMAGE) -t local-ai-backend:faster-whisper -f backend/Dockerfile.python --build-arg BACKEND=faster-whisper .
-
-docker-build-coqui:
-	docker build --build-arg BUILD_TYPE=$(BUILD_TYPE) --build-arg BASE_IMAGE=$(BASE_IMAGE) -t local-ai-backend:coqui -f backend/Dockerfile.python --build-arg BACKEND=coqui .
-
-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-vibevoice:
-	docker build --progress=plain --build-arg BUILD_TYPE=$(BUILD_TYPE) --build-arg BASE_IMAGE=$(BASE_IMAGE) -t local-ai-backend:vibevoice -f backend/Dockerfile.python --build-arg BACKEND=vibevoice ./backend
-
-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 .
-
-docker-build-backends: docker-build-llama-cpp docker-build-rerankers docker-build-vllm docker-build-transformers docker-build-diffusers docker-build-kokoro docker-build-faster-whisper docker-build-coqui docker-build-bark docker-build-chatterbox docker-build-vibevoice docker-build-exllama2
+clean-mock-backend:
+	rm -f tests/e2e/mock-backend/mock-backend
 
 ########################################################
 ### END Backends

README.md

@@ -51,34 +51,16 @@
 **LocalAI** is the free, Open Source OpenAI alternative. LocalAI act as a drop-in replacement REST API that's compatible with OpenAI (Elevenlabs, Anthropic... ) API specifications for local AI inferencing. It allows you to run LLMs, generate images, audio (and not only) locally or on-prem with consumer grade hardware, supporting multiple model families. Does not require GPU. It is created and maintained by [Ettore Di Giacinto](https://github.com/mudler).
 
 
-## 📚🆕 Local Stack Family
+## Local Stack Family
 
-🆕 LocalAI is now part of a comprehensive suite of AI tools designed to work together:
+Liking LocalAI? LocalAI is part of an integrated suite of AI infrastructure tools, you might also like:
+
+- **[LocalAGI](https://github.com/mudler/LocalAGI)** - AI agent orchestration platform with OpenAI Responses API compatibility and advanced agentic capabilities
+- **[LocalRecall](https://github.com/mudler/LocalRecall)** - MCP/REST API knowledge base system providing persistent memory and storage for AI agents
+- 🆕 **[Cogito](https://github.com/mudler/cogito)** - Go library for building intelligent, co-operative agentic software and LLM-powered workflows, focusing on improving results for small, open source language models that scales to any LLM. Powers LocalAGI and LocalAI MCP/Agentic capabilities
+- 🆕 **[Wiz](https://github.com/mudler/wiz)** - Terminal-based AI agent accessible via Ctrl+Space keybinding. Portable, local-LLM friendly shell assistant with TUI/CLI modes, tool execution with approval, MCP protocol support, and multi-shell compatibility (zsh, bash, fish)
+- 🆕 **[SkillServer](https://github.com/mudler/skillserver)** - Simple, centralized skills database for AI agents via MCP. Manages skills as Markdown files with MCP server integration, web UI for editing, Git synchronization, and full-text search capabilities
 
-<table>
-  <tr>
-    <td width="50%" valign="top">
-      <a href="https://github.com/mudler/LocalAGI">
-        <img src="https://raw.githubusercontent.com/mudler/LocalAGI/refs/heads/main/webui/react-ui/public/logo_2.png" width="300" alt="LocalAGI Logo">
-      </a>
-    </td>
-    <td width="50%" valign="top">
-      <h3><a href="https://github.com/mudler/LocalAGI">LocalAGI</a></h3>
-      <p>A powerful Local AI agent management platform that serves as a drop-in replacement for OpenAI's Responses API, enhanced with advanced agentic capabilities.</p>
-    </td>
-  </tr>
-  <tr>
-    <td width="50%" valign="top">
-      <a href="https://github.com/mudler/LocalRecall">
-        <img src="https://raw.githubusercontent.com/mudler/LocalRecall/refs/heads/main/static/localrecall_horizontal.png" width="300" alt="LocalRecall Logo">
-      </a>
-    </td>
-    <td width="50%" valign="top">
-      <h3><a href="https://github.com/mudler/LocalRecall">LocalRecall</a></h3>
-      <p>A REST-ful API and knowledge base management system that provides persistent memory and storage capabilities for AI agents.</p>
-    </td>
-  </tr>
-</table>
 
 ## Screenshots / Video
 
@@ -111,6 +93,8 @@
 
 ## 💻 Quickstart
 
+> ⚠️ **Note:** The `install.sh` script is currently experiencing issues due to the heavy changes currently undergoing in LocalAI and may produce broken or misconfigured installations. Please use Docker installation (see below) or manual binary installation until [issue #8032](https://github.com/mudler/LocalAI/issues/8032) is resolved.
+
 Run the installer script:
 
 ```bash
@@ -128,7 +112,7 @@ For more installation options, see [Installer Options](https://localai.io/instal
 
 > Note: the DMGs are not signed by Apple as quarantined. See https://github.com/mudler/LocalAI/issues/6268 for a workaround, fix is tracked here: https://github.com/mudler/LocalAI/issues/6244
 
-Or run with docker:
+### Containers (Docker, podman, ...)
 
 > **💡 Docker Run vs Docker Start**
 > 
@@ -137,55 +121,59 @@ Or run with docker:
 > 
 > If you've already run LocalAI before and want to start it again, use: `docker start -i local-ai`
 
-### CPU only image:
+#### CPU only image:
 
 ```bash
 docker run -ti --name local-ai -p 8080:8080 localai/localai:latest
 ```
 
-### NVIDIA GPU Images:
+#### NVIDIA GPU Images:
 
 ```bash
+# CUDA 13.0
+docker run -ti --name local-ai -p 8080:8080 --gpus all localai/localai:latest-gpu-nvidia-cuda-13
+
 # CUDA 12.0
 docker run -ti --name local-ai -p 8080:8080 --gpus all localai/localai:latest-gpu-nvidia-cuda-12
 
-# CUDA 11.7
-docker run -ti --name local-ai -p 8080:8080 --gpus all localai/localai:latest-gpu-nvidia-cuda-11
-
 # NVIDIA Jetson (L4T) ARM64
+# CUDA 12 (for Nvidia AGX Orin and similar platforms)
 docker run -ti --name local-ai -p 8080:8080 --gpus all localai/localai:latest-nvidia-l4t-arm64
+
+# CUDA 13 (for Nvidia DGX Spark)
+docker run -ti --name local-ai -p 8080:8080 --gpus all localai/localai:latest-nvidia-l4t-arm64-cuda-13
 ```
 
-### AMD GPU Images (ROCm):
+#### AMD GPU Images (ROCm):
 
 ```bash
 docker run -ti --name local-ai -p 8080:8080 --device=/dev/kfd --device=/dev/dri --group-add=video localai/localai:latest-gpu-hipblas
 ```
 
-### Intel GPU Images (oneAPI):
+#### Intel GPU Images (oneAPI):
 
 ```bash
 docker run -ti --name local-ai -p 8080:8080 --device=/dev/dri/card1 --device=/dev/dri/renderD128 localai/localai:latest-gpu-intel
 ```
 
-### Vulkan GPU Images:
+#### Vulkan GPU Images:
 
 ```bash
 docker run -ti --name local-ai -p 8080:8080 localai/localai:latest-gpu-vulkan
 ```
 
-### AIO Images (pre-downloaded models):
+#### AIO Images (pre-downloaded models):
 
 ```bash
 # CPU version
 docker run -ti --name local-ai -p 8080:8080 localai/localai:latest-aio-cpu
 
+# NVIDIA CUDA 13 version
+docker run -ti --name local-ai -p 8080:8080 --gpus all localai/localai:latest-aio-gpu-nvidia-cuda-13
+
 # NVIDIA CUDA 12 version
 docker run -ti --name local-ai -p 8080:8080 --gpus all localai/localai:latest-aio-gpu-nvidia-cuda-12
 
-# NVIDIA CUDA 11 version
-docker run -ti --name local-ai -p 8080:8080 --gpus all localai/localai:latest-aio-gpu-nvidia-cuda-11
-
 # Intel GPU version
 docker run -ti --name local-ai -p 8080:8080 localai/localai:latest-aio-gpu-intel
 
@@ -251,6 +239,7 @@ Roadmap items: [List of issues](https://github.com/mudler/LocalAI/issues?q=is%3A
 - 🔈 [Audio to Text](https://localai.io/features/audio-to-text/) (Audio transcription with `whisper.cpp`)
 - 🎨 [Image generation](https://localai.io/features/image-generation)
 - 🔥 [OpenAI-alike tools API](https://localai.io/features/openai-functions/) 
+- ⚡ [Realtime API](https://localai.io/features/openai-realtime/) (Speech-to-speech) 
 - 🧠 [Embeddings generation for vector databases](https://localai.io/features/embeddings/)
 - ✍️ [Constrained grammars](https://localai.io/features/constrained_grammars/)
 - 🖼️ [Download Models directly from Huggingface ](https://localai.io/models/)
@@ -269,39 +258,39 @@ LocalAI supports a comprehensive range of AI backends with multiple acceleration
 ### 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 |
+| **llama.cpp** | LLM inference in C/C++ | CUDA 12/13, ROCm, Intel SYCL, Vulkan, Metal, CPU |
+| **vLLM** | Fast LLM inference with PagedAttention | CUDA 12/13, ROCm, Intel |
+| **transformers** | HuggingFace transformers framework | CUDA 12/13, ROCm, Intel, CPU |
 | **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 |
+| **whisper.cpp** | OpenAI Whisper in C/C++ | CUDA 12/13, ROCm, Intel SYCL, Vulkan, CPU |
+| **faster-whisper** | Fast Whisper with CTranslate2 | CUDA 12/13, ROCm, Intel, CPU |
+| **coqui** | Advanced TTS with 1100+ languages | CUDA 12/13, ROCm, Intel, CPU |
+| **kokoro** | Lightweight TTS model | CUDA 12/13, ROCm, Intel, CPU |
+| **chatterbox** | Production-grade TTS | CUDA 12/13, CPU |
 | **piper** | Fast neural TTS system | CPU |
 | **kitten-tts** | Kitten TTS models | CPU |
 | **silero-vad** | Voice Activity Detection | CPU |
-| **neutts** | Text-to-speech with voice cloning | CUDA 12, ROCm, CPU |
+| **neutts** | Text-to-speech with voice cloning | CUDA 12/13, ROCm, CPU |
+| **vibevoice** | Real-time TTS with voice cloning | CUDA 12/13, ROCm, Intel, CPU |
+| **pocket-tts** | Lightweight CPU-based TTS | CUDA 12/13, ROCm, Intel, CPU |
+| **qwen-tts** | High-quality TTS with custom voice, voice design, and voice cloning | CUDA 12/13, ROCm, Intel, 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 |
+| **stablediffusion.cpp** | Stable Diffusion in C/C++ | CUDA 12/13, Intel SYCL, Vulkan, CPU |
+| **diffusers** | HuggingFace diffusion models | CUDA 12/13, 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 |
+| **rfdetr** | Real-time object detection | CUDA 12/13, Intel, CPU |
+| **rerankers** | Document reranking API | CUDA 12/13, ROCm, Intel, CPU |
 | **local-store** | Vector database | CPU |
 | **huggingface** | HuggingFace API integration | API-based |
 
@@ -309,13 +298,14 @@ LocalAI supports a comprehensive range of AI backends with multiple acceleration
 
 | 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, neutts | 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+ |
+| **NVIDIA CUDA 13** | All CUDA-compatible backends | Nvidia hardware |
+| **AMD ROCm** | llama.cpp, whisper, vllm, transformers, diffusers, rerankers, coqui, kokoro, neutts, vibevoice, pocket-tts, qwen-tts | AMD Graphics |
+| **Intel oneAPI** | llama.cpp, whisper, stablediffusion, vllm, transformers, diffusers, rfdetr, rerankers, coqui, kokoro, vibevoice, pocket-tts, qwen-tts | Intel Arc, Intel iGPUs |
+| **Apple Metal** | llama.cpp, whisper, diffusers, MLX, MLX-VLM | Apple M1/M2/M3+ |
 | **Vulkan** | llama.cpp, whisper, stablediffusion | Cross-platform GPUs |
-| **NVIDIA Jetson** | llama.cpp, whisper, stablediffusion, diffusers, rfdetr | ARM64 embedded AI |
+| **NVIDIA Jetson (CUDA 12)** | llama.cpp, whisper, stablediffusion, diffusers, rfdetr | ARM64 embedded AI (AGX Orin, etc.) |
+| **NVIDIA Jetson (CUDA 13)** | llama.cpp, whisper, stablediffusion, diffusers, rfdetr | ARM64 embedded AI (DGX Spark) |
 | **CPU Optimized** | All backends | AVX/AVX2/AVX512, quantization support |
 
 ### 🔗 Community and integrations
@@ -334,6 +324,10 @@ Agentic Libraries:
 MCPs:
 - https://github.com/mudler/MCPs
 
+OS Assistant:
+
+- https://github.com/mudler/Keygeist - Keygeist is an AI-powered keyboard operator that listens for key combinations and responds with AI-generated text typed directly into your Linux box.
+
 Model galleries
 - https://github.com/go-skynet/model-gallery
 
@@ -408,6 +402,10 @@ A huge thank you to our generous sponsors who support this project covering CI e
   </a>
 </p>
 
+### Individual sponsors
+
+A special thanks to individual sponsors that contributed to the project, a full list is in [Github](https://github.com/sponsors/mudler) and [buymeacoffee](https://buymeacoffee.com/mudler), a special shout out goes to [drikster80](https://github.com/drikster80) for being generous. Thank you everyone!
+
 ## 🌟 Star history
 
 [![LocalAI Star history Chart](https://api.star-history.com/svg?repos=go-skynet/LocalAI&type=Date)](https://star-history.com/#go-skynet/LocalAI&Date)

backend/Dockerfile.golang

@@ -1,4 +1,4 @@
-ARG BASE_IMAGE=ubuntu:22.04
+ARG BASE_IMAGE=ubuntu:24.04
 
 FROM ${BASE_IMAGE} AS builder
 ARG BACKEND=rerankers
@@ -12,8 +12,8 @@ ENV CUDA_MINOR_VERSION=${CUDA_MINOR_VERSION}
 ENV DEBIAN_FRONTEND=noninteractive
 ARG TARGETARCH
 ARG TARGETVARIANT
-ARG GO_VERSION=1.22.6
-ARG UBUNTU_VERSION=2204
+ARG GO_VERSION=1.25.4
+ARG UBUNTU_VERSION=2404
 
 RUN apt-get update && \
     apt-get install -y --no-install-recommends \
@@ -40,11 +40,45 @@ RUN <<EOT bash
         apt-get update && \
         apt-get install -y  --no-install-recommends \
             software-properties-common pciutils wget gpg-agent && \
-        wget -qO - https://packages.lunarg.com/lunarg-signing-key-pub.asc | apt-key add - && \
-        wget -qO /etc/apt/sources.list.d/lunarg-vulkan-jammy.list https://packages.lunarg.com/vulkan/lunarg-vulkan-jammy.list && \
-        apt-get update && \
-        apt-get install -y \
-            vulkan-sdk && \
+        apt-get install -y libglm-dev cmake libxcb-dri3-0 libxcb-present0 libpciaccess0 \
+            libpng-dev libxcb-keysyms1-dev libxcb-dri3-dev libx11-dev g++ gcc \
+            libwayland-dev libxrandr-dev libxcb-randr0-dev libxcb-ewmh-dev \
+            git python-is-python3 bison libx11-xcb-dev liblz4-dev libzstd-dev \
+            ocaml-core ninja-build pkg-config libxml2-dev wayland-protocols python3-jsonschema \
+            clang-format qtbase5-dev qt6-base-dev libxcb-glx0-dev sudo xz-utils
+        if [ "amd64" = "$TARGETARCH" ]; then
+            wget "https://sdk.lunarg.com/sdk/download/1.4.335.0/linux/vulkansdk-linux-x86_64-1.4.335.0.tar.xz" && \
+            tar -xf vulkansdk-linux-x86_64-1.4.335.0.tar.xz && \
+            rm vulkansdk-linux-x86_64-1.4.335.0.tar.xz && \
+            mkdir -p /opt/vulkan-sdk && \
+            mv 1.4.335.0 /opt/vulkan-sdk/ && \
+            cd /opt/vulkan-sdk/1.4.335.0 && \
+            ./vulkansdk --no-deps --maxjobs \
+                vulkan-loader \
+                vulkan-validationlayers \
+                vulkan-extensionlayer \
+                vulkan-tools \
+                shaderc && \
+            cp -rfv /opt/vulkan-sdk/1.4.335.0/x86_64/bin/* /usr/bin/ && \
+            cp -rfv /opt/vulkan-sdk/1.4.335.0/x86_64/lib/* /usr/lib/x86_64-linux-gnu/ && \
+            cp -rfv /opt/vulkan-sdk/1.4.335.0/x86_64/include/* /usr/include/ && \
+            cp -rfv /opt/vulkan-sdk/1.4.335.0/x86_64/share/* /usr/share/ && \
+            rm -rf /opt/vulkan-sdk
+        fi
+        if [ "arm64" = "$TARGETARCH" ]; then
+            mkdir vulkan && cd vulkan && \
+            curl -L -o vulkan-sdk.tar.xz https://github.com/mudler/vulkan-sdk-arm/releases/download/1.4.335.0/vulkansdk-ubuntu-24.04-arm-1.4.335.0.tar.xz && \
+            tar -xvf vulkan-sdk.tar.xz && \
+            rm vulkan-sdk.tar.xz && \
+            cd 1.4.335.0 && \
+            cp -rfv aarch64/bin/* /usr/bin/ && \
+            cp -rfv aarch64/lib/* /usr/lib/aarch64-linux-gnu/ && \
+            cp -rfv aarch64/include/* /usr/include/ && \
+            cp -rfv aarch64/share/* /usr/share/ && \
+            cd ../.. && \
+            rm -rf vulkan
+        fi
+        ldconfig && \
         apt-get clean && \
         rm -rf /var/lib/apt/lists/*
     fi
@@ -148,6 +182,8 @@ EOT
 
 COPY . /LocalAI
 
+RUN git config --global --add safe.directory /LocalAI
+
 RUN cd /LocalAI && make protogen-go && make -C /LocalAI/backend/go/${BACKEND} build
 
 FROM scratch

backend/Dockerfile.llama-cpp

@@ -1,4 +1,4 @@
-ARG BASE_IMAGE=ubuntu:22.04
+ARG BASE_IMAGE=ubuntu:24.04
 ARG GRPC_BASE_IMAGE=${BASE_IMAGE}
 
 
@@ -10,7 +10,8 @@ FROM ${GRPC_BASE_IMAGE} AS grpc
 ARG GRPC_MAKEFLAGS="-j4 -Otarget"
 ARG GRPC_VERSION=v1.65.0
 ARG CMAKE_FROM_SOURCE=false
-ARG CMAKE_VERSION=3.26.4
+# CUDA Toolkit 13.x compatibility: CMake 3.31.9+ fixes toolchain detection/arch table issues
+ARG CMAKE_VERSION=3.31.10
 
 ENV MAKEFLAGS=${GRPC_MAKEFLAGS}
 
@@ -26,7 +27,7 @@ RUN apt-get update && \
 
 # Install CMake (the version in 22.04 is too old)
 RUN <<EOT bash
-    if [ "${CMAKE_FROM_SOURCE}}" = "true" ]; then
+    if [ "${CMAKE_FROM_SOURCE}" = "true" ]; then
         curl -L -s https://github.com/Kitware/CMake/releases/download/v${CMAKE_VERSION}/cmake-${CMAKE_VERSION}.tar.gz -o cmake.tar.gz && tar xvf cmake.tar.gz && cd cmake-${CMAKE_VERSION} && ./configure && make && make install
     else
         apt-get update && \
@@ -50,6 +51,13 @@ RUN git clone --recurse-submodules --jobs 4 -b ${GRPC_VERSION} --depth 1 --shall
     rm -rf /build
 
 FROM ${BASE_IMAGE} AS builder
+ARG CMAKE_FROM_SOURCE=false
+ARG CMAKE_VERSION=3.31.10
+# We can target specific CUDA ARCHITECTURES like --build-arg CUDA_DOCKER_ARCH='75;86;89;120'
+ARG CUDA_DOCKER_ARCH
+ENV CUDA_DOCKER_ARCH=${CUDA_DOCKER_ARCH}
+ARG CMAKE_ARGS
+ENV CMAKE_ARGS=${CMAKE_ARGS}
 ARG BACKEND=rerankers
 ARG BUILD_TYPE
 ENV BUILD_TYPE=${BUILD_TYPE}
@@ -61,8 +69,8 @@ ENV CUDA_MINOR_VERSION=${CUDA_MINOR_VERSION}
 ENV DEBIAN_FRONTEND=noninteractive
 ARG TARGETARCH
 ARG TARGETVARIANT
-ARG GO_VERSION=1.22.6
-ARG UBUNTU_VERSION=2204
+ARG GO_VERSION=1.25.4
+ARG UBUNTU_VERSION=2404
 
 RUN apt-get update && \
     apt-get install -y --no-install-recommends \
@@ -70,6 +78,7 @@ RUN apt-get update && \
         ccache git \
         ca-certificates \
         make \
+        pkg-config libcurl4-openssl-dev \
         curl unzip \
         libssl-dev wget && \
     apt-get clean && \
@@ -88,11 +97,45 @@ RUN <<EOT bash
         apt-get update && \
         apt-get install -y  --no-install-recommends \
             software-properties-common pciutils wget gpg-agent && \
-        wget -qO - https://packages.lunarg.com/lunarg-signing-key-pub.asc | apt-key add - && \
-        wget -qO /etc/apt/sources.list.d/lunarg-vulkan-jammy.list https://packages.lunarg.com/vulkan/lunarg-vulkan-jammy.list && \
-        apt-get update && \
-        apt-get install -y \
-            vulkan-sdk && \
+        apt-get install -y libglm-dev cmake libxcb-dri3-0 libxcb-present0 libpciaccess0 \
+            libpng-dev libxcb-keysyms1-dev libxcb-dri3-dev libx11-dev g++ gcc \
+            libwayland-dev libxrandr-dev libxcb-randr0-dev libxcb-ewmh-dev \
+            git python-is-python3 bison libx11-xcb-dev liblz4-dev libzstd-dev \
+            ocaml-core ninja-build pkg-config libxml2-dev wayland-protocols python3-jsonschema \
+            clang-format qtbase5-dev qt6-base-dev libxcb-glx0-dev sudo xz-utils
+        if [ "amd64" = "$TARGETARCH" ]; then
+            wget "https://sdk.lunarg.com/sdk/download/1.4.335.0/linux/vulkansdk-linux-x86_64-1.4.335.0.tar.xz" && \
+            tar -xf vulkansdk-linux-x86_64-1.4.335.0.tar.xz && \
+            rm vulkansdk-linux-x86_64-1.4.335.0.tar.xz && \
+            mkdir -p /opt/vulkan-sdk && \
+            mv 1.4.335.0 /opt/vulkan-sdk/ && \
+            cd /opt/vulkan-sdk/1.4.335.0 && \
+            ./vulkansdk --no-deps --maxjobs \
+                vulkan-loader \
+                vulkan-validationlayers \
+                vulkan-extensionlayer \
+                vulkan-tools \
+                shaderc && \
+            cp -rfv /opt/vulkan-sdk/1.4.335.0/x86_64/bin/* /usr/bin/ && \
+            cp -rfv /opt/vulkan-sdk/1.4.335.0/x86_64/lib/* /usr/lib/x86_64-linux-gnu/ && \
+            cp -rfv /opt/vulkan-sdk/1.4.335.0/x86_64/include/* /usr/include/ && \
+            cp -rfv /opt/vulkan-sdk/1.4.335.0/x86_64/share/* /usr/share/ && \
+            rm -rf /opt/vulkan-sdk
+        fi
+        if [ "arm64" = "$TARGETARCH" ]; then
+            mkdir vulkan && cd vulkan && \
+            curl -L -o vulkan-sdk.tar.xz https://github.com/mudler/vulkan-sdk-arm/releases/download/1.4.335.0/vulkansdk-ubuntu-24.04-arm-1.4.335.0.tar.xz && \
+            tar -xvf vulkan-sdk.tar.xz && \
+            rm vulkan-sdk.tar.xz && \
+            cd 1.4.335.0 && \
+            cp -rfv aarch64/bin/* /usr/bin/ && \
+            cp -rfv aarch64/lib/* /usr/lib/aarch64-linux-gnu/ && \
+            cp -rfv aarch64/include/* /usr/include/ && \
+            cp -rfv aarch64/share/* /usr/share/ && \
+            cd ../.. && \
+            rm -rf vulkan
+        fi
+        ldconfig && \
         apt-get clean && \
         rm -rf /var/lib/apt/lists/*
     fi
@@ -189,7 +232,7 @@ EOT
 
 # Install CMake (the version in 22.04 is too old)
 RUN <<EOT bash
-    if [ "${CMAKE_FROM_SOURCE}}" = "true" ]; then
+    if [ "${CMAKE_FROM_SOURCE}" = "true" ]; then
         curl -L -s https://github.com/Kitware/CMake/releases/download/v${CMAKE_VERSION}/cmake-${CMAKE_VERSION}.tar.gz -o cmake.tar.gz && tar xvf cmake.tar.gz && cd cmake-${CMAKE_VERSION} && ./configure && make && make install
     else
         apt-get update && \
@@ -205,19 +248,30 @@ COPY --from=grpc /opt/grpc /usr/local
 
 COPY . /LocalAI
 
-## Otherwise just run the normal build
-RUN <<EOT bash
-if [ "${TARGETARCH}" = "arm64" ] || [ "${BUILD_TYPE}" = "hipblas" ]; then \
-        cd /LocalAI/backend/cpp/llama-cpp && make llama-cpp-fallback && \
-        make llama-cpp-grpc && make llama-cpp-rpc-server; \
-    else \
-        cd /LocalAI/backend/cpp/llama-cpp && make llama-cpp-avx && \
-        make llama-cpp-avx2 && \
-        make llama-cpp-avx512 && \
-        make llama-cpp-fallback && \
-        make llama-cpp-grpc && \
-        make llama-cpp-rpc-server; \
-    fi  
+RUN <<'EOT' bash
+set -euxo pipefail
+
+if [[ -n "${CUDA_DOCKER_ARCH:-}" ]]; then
+  CUDA_ARCH_ESC="${CUDA_DOCKER_ARCH//;/\\;}"
+  export CMAKE_ARGS="${CMAKE_ARGS:-} -DCMAKE_CUDA_ARCHITECTURES=${CUDA_ARCH_ESC}"
+  echo "CMAKE_ARGS(env) = ${CMAKE_ARGS}"
+  rm -rf /LocalAI/backend/cpp/llama-cpp-*-build
+fi
+
+if [ "${TARGETARCH}" = "arm64" ] || [ "${BUILD_TYPE}" = "hipblas" ]; then
+  cd /LocalAI/backend/cpp/llama-cpp
+  make llama-cpp-fallback
+  make llama-cpp-grpc
+  make llama-cpp-rpc-server
+else
+  cd /LocalAI/backend/cpp/llama-cpp
+  make llama-cpp-avx
+  make llama-cpp-avx2
+  make llama-cpp-avx512
+  make llama-cpp-fallback
+  make llama-cpp-grpc
+  make llama-cpp-rpc-server
+fi
 EOT
 
 

backend/Dockerfile.python

@@ -1,4 +1,4 @@
-ARG BASE_IMAGE=ubuntu:22.04
+ARG BASE_IMAGE=ubuntu:24.04
 
 FROM ${BASE_IMAGE} AS builder
 ARG BACKEND=rerankers
@@ -12,7 +12,7 @@ ENV CUDA_MINOR_VERSION=${CUDA_MINOR_VERSION}
 ENV DEBIAN_FRONTEND=noninteractive
 ARG TARGETARCH
 ARG TARGETVARIANT
-ARG UBUNTU_VERSION=2204
+ARG UBUNTU_VERSION=2404
 
 RUN apt-get update && \
     apt-get install -y --no-install-recommends \
@@ -54,11 +54,45 @@ RUN <<EOT bash
         apt-get update && \
         apt-get install -y  --no-install-recommends \
             software-properties-common pciutils wget gpg-agent && \
-        wget -qO - https://packages.lunarg.com/lunarg-signing-key-pub.asc | apt-key add - && \
-        wget -qO /etc/apt/sources.list.d/lunarg-vulkan-jammy.list https://packages.lunarg.com/vulkan/lunarg-vulkan-jammy.list && \
-        apt-get update && \
-        apt-get install -y \
-            vulkan-sdk && \
+        apt-get install -y libglm-dev cmake libxcb-dri3-0 libxcb-present0 libpciaccess0 \
+            libpng-dev libxcb-keysyms1-dev libxcb-dri3-dev libx11-dev g++ gcc \
+            libwayland-dev libxrandr-dev libxcb-randr0-dev libxcb-ewmh-dev \
+            git python-is-python3 bison libx11-xcb-dev liblz4-dev libzstd-dev \
+            ocaml-core ninja-build pkg-config libxml2-dev wayland-protocols python3-jsonschema \
+            clang-format qtbase5-dev qt6-base-dev libxcb-glx0-dev sudo xz-utils
+        if [ "amd64" = "$TARGETARCH" ]; then
+            wget "https://sdk.lunarg.com/sdk/download/1.4.335.0/linux/vulkansdk-linux-x86_64-1.4.335.0.tar.xz" && \
+            tar -xf vulkansdk-linux-x86_64-1.4.335.0.tar.xz && \
+            rm vulkansdk-linux-x86_64-1.4.335.0.tar.xz && \
+            mkdir -p /opt/vulkan-sdk && \
+            mv 1.4.335.0 /opt/vulkan-sdk/ && \
+            cd /opt/vulkan-sdk/1.4.335.0 && \
+            ./vulkansdk --no-deps --maxjobs \
+                vulkan-loader \
+                vulkan-validationlayers \
+                vulkan-extensionlayer \
+                vulkan-tools \
+                shaderc && \
+            cp -rfv /opt/vulkan-sdk/1.4.335.0/x86_64/bin/* /usr/bin/ && \
+            cp -rfv /opt/vulkan-sdk/1.4.335.0/x86_64/lib/* /usr/lib/x86_64-linux-gnu/ && \
+            cp -rfv /opt/vulkan-sdk/1.4.335.0/x86_64/include/* /usr/include/ && \
+            cp -rfv /opt/vulkan-sdk/1.4.335.0/x86_64/share/* /usr/share/ && \
+            rm -rf /opt/vulkan-sdk
+        fi
+        if [ "arm64" = "$TARGETARCH" ]; then
+            mkdir vulkan && cd vulkan && \
+            curl -L -o vulkan-sdk.tar.xz https://github.com/mudler/vulkan-sdk-arm/releases/download/1.4.335.0/vulkansdk-ubuntu-24.04-arm-1.4.335.0.tar.xz && \
+            tar -xvf vulkan-sdk.tar.xz && \
+            rm vulkan-sdk.tar.xz && \
+            cd 1.4.335.0 && \
+            cp -rfv aarch64/bin/* /usr/bin/ && \
+            cp -rfv aarch64/lib/* /usr/lib/aarch64-linux-gnu/ && \
+            cp -rfv aarch64/include/* /usr/include/ && \
+            cp -rfv aarch64/share/* /usr/share/ && \
+            cd ../.. && \
+            rm -rf vulkan
+        fi
+        ldconfig && \
         apt-get clean && \
         rm -rf /var/lib/apt/lists/*
     fi
@@ -142,7 +176,8 @@ RUN if [ "${BUILD_TYPE}" = "hipblas" ]; then \
 # Install uv as a system package
 RUN curl -LsSf https://astral.sh/uv/install.sh | UV_INSTALL_DIR=/usr/bin sh
 ENV PATH="/root/.cargo/bin:${PATH}"
-
+# Increase timeout for uv installs behind slow networks
+ENV UV_HTTP_TIMEOUT=180
 RUN curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y
 
 # Install grpcio-tools (the version in 22.04 is too old)
@@ -155,12 +190,18 @@ RUN <<EOT bash
 EOT
 
 
-COPY python/${BACKEND} /${BACKEND}
-COPY backend.proto /${BACKEND}/backend.proto
-COPY python/common/ /${BACKEND}/common
+COPY backend/python/${BACKEND} /${BACKEND}
+COPY backend/backend.proto /${BACKEND}/backend.proto
+COPY backend/python/common/ /${BACKEND}/common
+COPY scripts/build/package-gpu-libs.sh /package-gpu-libs.sh
 
 RUN cd /${BACKEND} && PORTABLE_PYTHON=true make
 
+# Package GPU libraries into the backend's lib directory
+RUN mkdir -p /${BACKEND}/lib && \
+    TARGET_LIB_DIR="/${BACKEND}/lib" BUILD_TYPE="${BUILD_TYPE}" CUDA_MAJOR_VERSION="${CUDA_MAJOR_VERSION}" \
+    bash /package-gpu-libs.sh "/${BACKEND}/lib"
+
 FROM scratch
 ARG BACKEND=rerankers
 COPY --from=builder /${BACKEND}/ /

backend/README.md

@@ -46,7 +46,7 @@ The backend system provides language-specific Dockerfiles that handle the build
 - **vllm**: High-performance LLM inference
 - **mlx**: Apple Silicon optimization
 - **diffusers**: Stable Diffusion models
-- **Audio**: bark, coqui, faster-whisper, kitten-tts
+- **Audio**: coqui, faster-whisper, kitten-tts
 - **Vision**: mlx-vlm, rfdetr
 - **Specialized**: rerankers, chatterbox, kokoro
 
@@ -55,7 +55,6 @@ The backend system provides language-specific Dockerfiles that handle the build
 - **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/`)
@@ -65,7 +64,7 @@ The backend system provides language-specific Dockerfiles that handle the build
 ## Hardware Acceleration Support
 
 ### CUDA (NVIDIA)
-- **Versions**: CUDA 11.x, 12.x
+- **Versions**: CUDA 12.x, 13.x
 - **Features**: cuBLAS, cuDNN, TensorRT optimization
 - **Targets**: x86_64, ARM64 (Jetson)
 
@@ -132,8 +131,7 @@ For ARM64/Mac builds, docker can't be used, and the makefile in the respective b
 ### Build Types
 
 - **`cpu`**: CPU-only optimization
-- **`cublas11`**: CUDA 11.x with cuBLAS
-- **`cublas12`**: CUDA 12.x with cuBLAS
+- **`cublas12`**, **`cublas13`**: CUDA 12.x, 13.x with cuBLAS
 - **`hipblas`**: ROCm with rocBLAS
 - **`intel`**: Intel oneAPI optimization
 - **`vulkan`**: Vulkan-based acceleration
@@ -210,4 +208,4 @@ When contributing to the backend system:
 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
+5. **Validate**: Test across different hardware targets

backend/backend.proto

@@ -17,6 +17,7 @@ service Backend {
   rpc GenerateVideo(GenerateVideoRequest) returns (Result) {}
   rpc AudioTranscription(TranscriptRequest) returns (TranscriptResult) {}
   rpc TTS(TTSRequest) returns (Result) {}
+  rpc TTSStream(TTSRequest) returns (stream Reply) {}
   rpc SoundGeneration(SoundGenerationRequest) returns (Result) {}
   rpc TokenizeString(PredictOptions) returns (TokenizationResponse) {}
   rpc Status(HealthMessage) returns (StatusResponse) {}
@@ -32,6 +33,8 @@ service Backend {
   rpc GetMetrics(MetricsRequest) returns (MetricsResponse);
 
   rpc VAD(VADRequest) returns (VADResponse) {}
+
+  rpc ModelMetadata(ModelOptions) returns (ModelMetadataResponse) {}
 }
 
 // Define the empty request
@@ -296,12 +299,12 @@ message TranscriptSegment {
   int64 end = 3;
   string text = 4;
   repeated int32 tokens = 5;
+  string speaker = 6;
 }
 
 message GenerateImageRequest {
   int32 height = 1;
   int32 width = 2;
-  int32 mode = 3;
   int32 step = 4;
   int32 seed = 5;
   string positive_prompt = 6;
@@ -411,3 +414,8 @@ message Detection {
 message DetectResponse {
   repeated Detection Detections = 1;
 }
+
+message ModelMetadataResponse {
+  bool supports_thinking = 1;
+  string rendered_template = 2;  // The rendered chat template with enable_thinking=true (empty if not applicable)
+}

backend/cpp/llama-cpp/CMakeLists.txt

@@ -70,4 +70,4 @@ target_link_libraries(${TARGET} PRIVATE common llama mtmd ${CMAKE_THREAD_LIBS_IN
 target_compile_features(${TARGET} PRIVATE cxx_std_11)
 if(TARGET BUILD_INFO)
   add_dependencies(${TARGET} BUILD_INFO)
-endif()
+endif()

backend/cpp/llama-cpp/Makefile

@@ -1,5 +1,5 @@
 
-LLAMA_VERSION?=5b6c9bc0f3c8f55598b9999b65aff7ce4119bc15
+LLAMA_VERSION?=2634ed207a17db1a54bd8df0555bd8499a6ab691
 LLAMA_REPO?=https://github.com/ggerganov/llama.cpp
 
 CMAKE_ARGS?=
@@ -7,7 +7,8 @@ BUILD_TYPE?=
 NATIVE?=false
 ONEAPI_VARS?=/opt/intel/oneapi/setvars.sh
 TARGET?=--target grpc-server
-JOBS?=$(shell nproc)
+JOBS?=$(shell nproc 2>/dev/null || sysctl -n hw.ncpu 2>/dev/null || echo 1)
+ARCH?=$(shell uname -m)
 
 # Disable Shared libs as we are linking on static gRPC and we can't mix shared and static
 CMAKE_ARGS+=-DBUILD_SHARED_LIBS=OFF -DLLAMA_CURL=OFF
@@ -106,21 +107,21 @@ llama-cpp-avx: llama.cpp
 	cp -rf $(CURRENT_MAKEFILE_DIR)/../llama-cpp $(CURRENT_MAKEFILE_DIR)/../llama-cpp-avx-build
 	$(MAKE) -C $(CURRENT_MAKEFILE_DIR)/../llama-cpp-avx-build purge
 	$(info ${GREEN}I llama-cpp build info:avx${RESET})
-	CMAKE_ARGS="$(CMAKE_ARGS) -DGGML_AVX=on -DGGML_AVX2=off -DGGML_AVX512=off -DGGML_FMA=off -DGGML_F16C=off" $(MAKE) VARIANT="llama-cpp-avx-build" build-llama-cpp-grpc-server
+	CMAKE_ARGS="$(CMAKE_ARGS) -DGGML_AVX=on -DGGML_AVX2=off -DGGML_AVX512=off -DGGML_FMA=off -DGGML_F16C=off -DGGML_BMI2=off" $(MAKE) VARIANT="llama-cpp-avx-build" build-llama-cpp-grpc-server
 	cp -rfv $(CURRENT_MAKEFILE_DIR)/../llama-cpp-avx-build/grpc-server llama-cpp-avx
 
 llama-cpp-fallback: llama.cpp
 	cp -rf $(CURRENT_MAKEFILE_DIR)/../llama-cpp $(CURRENT_MAKEFILE_DIR)/../llama-cpp-fallback-build
 	$(MAKE) -C $(CURRENT_MAKEFILE_DIR)/../llama-cpp-fallback-build purge
 	$(info ${GREEN}I llama-cpp build info:fallback${RESET})
-	CMAKE_ARGS="$(CMAKE_ARGS) -DGGML_AVX=off -DGGML_AVX2=off -DGGML_AVX512=off -DGGML_FMA=off -DGGML_F16C=off" $(MAKE) VARIANT="llama-cpp-fallback-build" build-llama-cpp-grpc-server
+	CMAKE_ARGS="$(CMAKE_ARGS) -DGGML_AVX=off -DGGML_AVX2=off -DGGML_AVX512=off -DGGML_FMA=off -DGGML_F16C=off -DGGML_BMI2=off" $(MAKE) VARIANT="llama-cpp-fallback-build" build-llama-cpp-grpc-server
 	cp -rfv $(CURRENT_MAKEFILE_DIR)/../llama-cpp-fallback-build/grpc-server llama-cpp-fallback
 
 llama-cpp-grpc: llama.cpp
 	cp -rf $(CURRENT_MAKEFILE_DIR)/../llama-cpp $(CURRENT_MAKEFILE_DIR)/../llama-cpp-grpc-build
 	$(MAKE) -C $(CURRENT_MAKEFILE_DIR)/../llama-cpp-grpc-build purge
 	$(info ${GREEN}I llama-cpp build info:grpc${RESET})
-	CMAKE_ARGS="$(CMAKE_ARGS) -DGGML_RPC=ON -DGGML_AVX=off -DGGML_AVX2=off -DGGML_AVX512=off -DGGML_FMA=off -DGGML_F16C=off" TARGET="--target grpc-server --target rpc-server" $(MAKE) VARIANT="llama-cpp-grpc-build" build-llama-cpp-grpc-server
+	CMAKE_ARGS="$(CMAKE_ARGS) -DGGML_RPC=ON -DGGML_AVX=off -DGGML_AVX2=off -DGGML_AVX512=off -DGGML_FMA=off -DGGML_F16C=off -DGGML_BMI2=off" TARGET="--target grpc-server --target rpc-server" $(MAKE) VARIANT="llama-cpp-grpc-build" build-llama-cpp-grpc-server
 	cp -rfv $(CURRENT_MAKEFILE_DIR)/../llama-cpp-grpc-build/grpc-server llama-cpp-grpc
 
 llama-cpp-rpc-server: llama-cpp-grpc

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

@@ -23,6 +23,7 @@
 #include <grpcpp/health_check_service_interface.h>
 #include <regex>
 #include <atomic>
+#include <mutex>
 #include <signal.h>
 #include <thread>
 
@@ -82,8 +83,8 @@ static void start_llama_server(server_context& ctx_server) {
 
     // 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.impl->chat_templates.get()),
-    //     common_chat_format_example(ctx_server.impl->chat_templates.get(), ctx_server.impl->params_base.use_jinja).c_str(), ctx_server.impl->params_base.default_template_kwargs);
+    //     common_chat_templates_source(ctx_server.impl->chat_params.tmpls.get()),
+    //     common_chat_format_example(ctx_server.impl->chat_params.tmpls.get(), ctx_server.impl->params_base.use_jinja).c_str(), ctx_server.impl->params_base.default_template_kwargs);
 
     // Keep the chat templates initialized in load_model() so they can be used when UseTokenizerTemplate is enabled
     // Templates will only be used conditionally in Predict/PredictStream when UseTokenizerTemplate is true and Messages are provided
@@ -358,9 +359,7 @@ static void params_parse(server_context& /*ctx_server*/, const backend::ModelOpt
 
     params.model.path = request->modelfile();
     if (!request->mmproj().empty()) {
-    // get the directory of modelfile
-      std::string model_dir = params.model.path.substr(0, params.model.path.find_last_of("/\\"));
-      params.mmproj.path = model_dir + "/"+ request->mmproj();
+      params.mmproj.path = request->mmproj();
     }
     //  params.model_alias ??
     params.model_alias =  request->modelfile();
@@ -392,8 +391,9 @@ static void params_parse(server_context& /*ctx_server*/, const backend::ModelOpt
     // Initialize fit_params options (can be overridden by options)
     // fit_params: whether to auto-adjust params to fit device memory (default: true as in llama.cpp)
     params.fit_params = true;
-    // fit_params_target: target margin per device in bytes (default: 1GB)
-    params.fit_params_target = 1024 * 1024 * 1024;
+    // fit_params_target: target margin per device in bytes (default: 1GB per device)
+    // Initialize as vector with default value for all devices
+    params.fit_params_target = std::vector<size_t>(llama_max_devices(), 1024 * 1024 * 1024);
     // fit_params_min_ctx: minimum context size for fit (default: 4096)
     params.fit_params_min_ctx = 4096;
 
@@ -470,10 +470,28 @@ static void params_parse(server_context& /*ctx_server*/, const backend::ModelOpt
         } else if (!strcmp(optname, "fit_params_target") || !strcmp(optname, "fit_target")) {
             if (optval != NULL) {
                 try {
-                    // Value is in MiB, convert to bytes
-                    params.fit_params_target = static_cast<size_t>(std::stoi(optval_str)) * 1024 * 1024;
+                    // Value is in MiB, can be comma-separated list for multiple devices
+                    // Single value is broadcast across all devices
+                    std::string arg_next = optval_str;
+                    const std::regex regex{ R"([,/]+)" };
+                    std::sregex_token_iterator it{ arg_next.begin(), arg_next.end(), regex, -1 };
+                    std::vector<std::string> split_arg{ it, {} };
+                    if (split_arg.size() >= llama_max_devices()) {
+                        // Too many values provided
+                        continue;
+                    }
+                    if (split_arg.size() == 1) {
+                        // Single value: broadcast to all devices
+                        size_t value_mib = std::stoul(split_arg[0]);
+                        std::fill(params.fit_params_target.begin(), params.fit_params_target.end(), value_mib * 1024 * 1024);
+                    } else {
+                        // Multiple values: set per device
+                        for (size_t i = 0; i < split_arg.size() && i < params.fit_params_target.size(); i++) {
+                            params.fit_params_target[i] = std::stoul(split_arg[i]) * 1024 * 1024;
+                        }
+                    }
                 } catch (const std::exception& e) {
-                    // If conversion fails, keep default value (1GB)
+                    // If conversion fails, keep default value (1GB per device)
                 }
             }
         } else if (!strcmp(optname, "fit_params_min_ctx") || !strcmp(optname, "fit_ctx")) {
@@ -688,13 +706,13 @@ private:
 public:
     BackendServiceImpl(server_context& ctx) : ctx_server(ctx) {}
 
-    grpc::Status Health(ServerContext* /*context*/, const backend::HealthMessage* /*request*/, backend::Reply* reply) {
+    grpc::Status Health(ServerContext* /*context*/, const backend::HealthMessage* /*request*/, backend::Reply* reply) override {
         // Implement Health RPC
         reply->set_message("OK");
         return Status::OK;
     }
 
-    grpc::Status LoadModel(ServerContext* /*context*/, const backend::ModelOptions* request, backend::Result* result) {
+    grpc::Status LoadModel(ServerContext* /*context*/, const backend::ModelOptions* request, backend::Result* result) override {
         // Implement LoadModel RPC
         common_params params;
         params_parse(ctx_server, request, params);
@@ -711,11 +729,72 @@ public:
         LOG_INF("\n");
         LOG_INF("%s\n", common_params_get_system_info(params).c_str());
         LOG_INF("\n");
+        
+        // Capture error messages during model loading
+        struct error_capture {
+            std::string captured_error;
+            std::mutex error_mutex;
+            ggml_log_callback original_callback;
+            void* original_user_data;
+        } error_capture_data;
+        
+        // Get original log callback
+        llama_log_get(&error_capture_data.original_callback, &error_capture_data.original_user_data);
+        
+        // Set custom callback to capture errors
+        llama_log_set([](ggml_log_level level, const char * text, void * user_data) {
+            auto* capture = static_cast<error_capture*>(user_data);
+            
+            // Capture error messages
+            if (level == GGML_LOG_LEVEL_ERROR) {
+                std::lock_guard<std::mutex> lock(capture->error_mutex);
+                // Append error message, removing trailing newlines
+                std::string msg(text);
+                while (!msg.empty() && (msg.back() == '\n' || msg.back() == '\r')) {
+                    msg.pop_back();
+                }
+                if (!msg.empty()) {
+                    if (!capture->captured_error.empty()) {
+                        capture->captured_error.append("; ");
+                    }
+                    capture->captured_error.append(msg);
+                }
+            }
+            
+            // Also call original callback to preserve logging
+            if (capture->original_callback) {
+                capture->original_callback(level, text, capture->original_user_data);
+            }
+        }, &error_capture_data);
+        
         // load the model
-        if (!ctx_server.load_model(params)) {
-            result->set_message("Failed loading model");
+        bool load_success = ctx_server.load_model(params);
+        
+        // Restore original log callback
+        llama_log_set(error_capture_data.original_callback, error_capture_data.original_user_data);
+        
+        if (!load_success) {
+            std::string error_msg = "Failed to load model: " + params.model.path;
+            if (!params.mmproj.path.empty()) {
+                error_msg += " (with mmproj: " + params.mmproj.path + ")";
+            }
+            if (params.speculative.has_dft() && !params.speculative.mparams_dft.path.empty()) {
+                error_msg += " (with draft model: " + params.speculative.mparams_dft.path + ")";
+            }
+            
+            // Add captured error details if available
+            {
+                std::lock_guard<std::mutex> lock(error_capture_data.error_mutex);
+                if (!error_capture_data.captured_error.empty()) {
+                    error_msg += ". Error: " + error_capture_data.captured_error;
+                } else {
+                    error_msg += ". Model file may not exist or be invalid.";
+                }
+            }
+            
+            result->set_message(error_msg);
             result->set_success(false);
-            return Status::CANCELLED;
+            return grpc::Status(grpc::StatusCode::INTERNAL, error_msg);
         }
 
         // Process grammar triggers now that vocab is available
@@ -803,7 +882,7 @@ public:
             std::string prompt_str;
             std::vector<raw_buffer> files; // Declare files early so it's accessible in both branches
             // Handle chat templates when UseTokenizerTemplate is enabled and Messages are provided
-            if (request->usetokenizertemplate() && request->messages_size() > 0 && ctx_server.impl->chat_templates != nullptr) {
+            if (request->usetokenizertemplate() && request->messages_size() > 0 && ctx_server.impl->chat_params.tmpls != nullptr) {
                 // Convert proto Messages to JSON format compatible with oaicompat_chat_params_parse
                 json body_json;
                 json messages_json = json::array();
@@ -1182,12 +1261,7 @@ public:
                 // Use the same approach as server.cpp: call oaicompat_chat_params_parse
                 // This handles all template application, grammar merging, etc. automatically
                 // Files extracted from multimodal content in messages will be added to the files vector
-                // Create parser options with current chat_templates to ensure tmpls is not null
-                oaicompat_parser_options parser_opt = ctx_server.impl->oai_parser_opt;
-                parser_opt.tmpls = ctx_server.impl->chat_templates.get(); // Ensure tmpls is set to current chat_templates
-                // Update allow_image and allow_audio based on current mctx state
-                parser_opt.allow_image = ctx_server.impl->mctx ? mtmd_support_vision(ctx_server.impl->mctx) : false;
-                parser_opt.allow_audio = ctx_server.impl->mctx ? mtmd_support_audio(ctx_server.impl->mctx) : false;
+                // chat_params already contains tmpls, allow_image, and allow_audio set during model loading
 
                 // Debug: Log tools before template processing
                 if (body_json.contains("tools")) {
@@ -1233,7 +1307,7 @@ public:
                     }
                 }
 
-                json parsed_data = oaicompat_chat_params_parse(body_json, parser_opt, files);
+                json parsed_data = oaicompat_chat_params_parse(body_json, ctx_server.impl->chat_params, files);
 
                 // Debug: Log tools after template processing
                 if (parsed_data.contains("tools")) {
@@ -1286,7 +1360,7 @@ public:
 
             // If not using chat templates, extract files from image_data/audio_data fields
             // (If using chat templates, files were already extracted by oaicompat_chat_params_parse)
-            if (!request->usetokenizertemplate() || request->messages_size() == 0 || ctx_server.impl->chat_templates == nullptr) {
+            if (!request->usetokenizertemplate() || request->messages_size() == 0 || ctx_server.impl->chat_params.tmpls == nullptr) {
                 const auto &images_data = data.find("image_data");
                 if (images_data != data.end() && images_data->is_array())
                 {
@@ -1494,7 +1568,7 @@ public:
         return grpc::Status::OK;
     }
 
-    grpc::Status Predict(ServerContext* context, const backend::PredictOptions* request, backend::Reply* reply) {
+    grpc::Status Predict(ServerContext* context, const backend::PredictOptions* request, backend::Reply* reply) override {
          if (params_base.model.path.empty()) {
              return grpc::Status(grpc::StatusCode::FAILED_PRECONDITION, "Model not loaded");
          }
@@ -1514,7 +1588,7 @@ public:
             std::string prompt_str;
             std::vector<raw_buffer> files; // Declare files early so it's accessible in both branches
             // Handle chat templates when UseTokenizerTemplate is enabled and Messages are provided
-            if (request->usetokenizertemplate() && request->messages_size() > 0 && ctx_server.impl->chat_templates != nullptr) {
+            if (request->usetokenizertemplate() && request->messages_size() > 0 && ctx_server.impl->chat_params.tmpls != nullptr) {
                 // Convert proto Messages to JSON format compatible with oaicompat_chat_params_parse
                 json body_json;
                 json messages_json = json::array();
@@ -1918,12 +1992,7 @@ public:
                 // Use the same approach as server.cpp: call oaicompat_chat_params_parse
                 // This handles all template application, grammar merging, etc. automatically
                 // Files extracted from multimodal content in messages will be added to the files vector
-                // Create parser options with current chat_templates to ensure tmpls is not null
-                oaicompat_parser_options parser_opt = ctx_server.impl->oai_parser_opt;
-                parser_opt.tmpls = ctx_server.impl->chat_templates.get(); // Ensure tmpls is set to current chat_templates
-                // Update allow_image and allow_audio based on current mctx state
-                parser_opt.allow_image = ctx_server.impl->mctx ? mtmd_support_vision(ctx_server.impl->mctx) : false;
-                parser_opt.allow_audio = ctx_server.impl->mctx ? mtmd_support_audio(ctx_server.impl->mctx) : false;
+                // chat_params already contains tmpls, allow_image, and allow_audio set during model loading
 
                 // Debug: Log tools before template processing
                 if (body_json.contains("tools")) {
@@ -1969,7 +2038,7 @@ public:
                     }
                 }
 
-                json parsed_data = oaicompat_chat_params_parse(body_json, parser_opt, files);
+                json parsed_data = oaicompat_chat_params_parse(body_json, ctx_server.impl->chat_params, files);
 
                 // Debug: Log tools after template processing
                 if (parsed_data.contains("tools")) {
@@ -2022,7 +2091,7 @@ public:
 
             // If not using chat templates, extract files from image_data/audio_data fields
             // (If using chat templates, files were already extracted by oaicompat_chat_params_parse)
-            if (!request->usetokenizertemplate() || request->messages_size() == 0 || ctx_server.impl->chat_templates == nullptr) {
+            if (!request->usetokenizertemplate() || request->messages_size() == 0 || ctx_server.impl->chat_params.tmpls == nullptr) {
                 const auto &images_data = data.find("image_data");
                 if (images_data != data.end() && images_data->is_array())
                 {
@@ -2165,7 +2234,7 @@ public:
         return grpc::Status::OK;
     }
 
-    grpc::Status Embedding(ServerContext* context, const backend::PredictOptions* request, backend::EmbeddingResult* embeddingResult) {
+    grpc::Status Embedding(ServerContext* context, const backend::PredictOptions* request, backend::EmbeddingResult* embeddingResult) override {
         if (params_base.model.path.empty()) {
             return grpc::Status(grpc::StatusCode::FAILED_PRECONDITION, "Model not loaded");
         }
@@ -2260,7 +2329,7 @@ public:
         return grpc::Status::OK;
     }
 
-    grpc::Status Rerank(ServerContext* context, const backend::RerankRequest* request, backend::RerankResult* rerankResult) {
+    grpc::Status Rerank(ServerContext* context, const backend::RerankRequest* request, backend::RerankResult* rerankResult) override {
         if (!params_base.embedding || params_base.pooling_type != LLAMA_POOLING_TYPE_RANK) {
             return grpc::Status(grpc::StatusCode::UNIMPLEMENTED, "This server does not support reranking. Start it with `--reranking` and without `--embedding`");
         }
@@ -2346,7 +2415,7 @@ public:
         return grpc::Status::OK;
     }
 
-    grpc::Status TokenizeString(ServerContext* /*context*/, const backend::PredictOptions* request, backend::TokenizationResponse* response) {
+    grpc::Status TokenizeString(ServerContext* /*context*/, const backend::PredictOptions* request, backend::TokenizationResponse* response) override {
         if (params_base.model.path.empty()) {
             return grpc::Status(grpc::StatusCode::FAILED_PRECONDITION, "Model not loaded");
         }
@@ -2369,7 +2438,7 @@ public:
         return grpc::Status::OK;
     }
 
-    grpc::Status GetMetrics(ServerContext* /*context*/, const backend::MetricsRequest* /*request*/, backend::MetricsResponse* response) {
+    grpc::Status GetMetrics(ServerContext* /*context*/, const backend::MetricsRequest* /*request*/, backend::MetricsResponse* response) override {
 
 // request slots data using task queue
         auto rd = ctx_server.get_response_reader();
@@ -2407,6 +2476,47 @@ public:
         response->set_prompt_tokens_processed(res_metrics->n_prompt_tokens_processed_total);
 
 
+        return grpc::Status::OK;
+    }
+
+    grpc::Status ModelMetadata(ServerContext* /*context*/, const backend::ModelOptions* /*request*/, backend::ModelMetadataResponse* response) override {
+        // Check if model is loaded
+        if (params_base.model.path.empty()) {
+            return grpc::Status(grpc::StatusCode::FAILED_PRECONDITION, "Model not loaded");
+        }
+
+        // Check if chat templates are initialized
+        if (ctx_server.impl->chat_params.tmpls == nullptr) {
+            // If templates are not initialized, we can't detect thinking support
+            // Return false as default
+            response->set_supports_thinking(false);
+            response->set_rendered_template("");
+            return grpc::Status::OK;
+        }
+
+        // Detect thinking support using llama.cpp's function
+        bool supports_thinking = common_chat_templates_support_enable_thinking(ctx_server.impl->chat_params.tmpls.get());
+        response->set_supports_thinking(supports_thinking);
+
+        // Render the template with enable_thinking=true so Go code can detect thinking tokens
+        // This allows reusing existing detection functions in Go
+        std::string rendered_template = "";
+        if (params_base.use_jinja) {
+            // Render the template with enable_thinking=true to see what the actual prompt looks like
+            common_chat_templates_inputs dummy_inputs;
+            common_chat_msg msg;
+            msg.role = "user";
+            msg.content = "test";
+            dummy_inputs.messages = {msg};
+            dummy_inputs.enable_thinking = true;
+            dummy_inputs.use_jinja = params_base.use_jinja;
+            
+            const auto rendered = common_chat_templates_apply(ctx_server.impl->chat_params.tmpls.get(), dummy_inputs);
+            rendered_template = rendered.prompt;
+        }
+        
+        response->set_rendered_template(rendered_template);
+
         return grpc::Status::OK;
     }
 };

backend/cpp/llama-cpp/package.sh

@@ -6,6 +6,7 @@
 set -e
 
 CURDIR=$(dirname "$(realpath $0)")
+REPO_ROOT="${CURDIR}/../../.."
 
 # Create lib directory
 mkdir -p $CURDIR/package/lib
@@ -37,6 +38,15 @@ else
     exit 1
 fi
 
+# Package GPU libraries based on BUILD_TYPE
+# The GPU library packaging script will detect BUILD_TYPE and copy appropriate GPU libraries
+GPU_LIB_SCRIPT="${REPO_ROOT}/scripts/build/package-gpu-libs.sh"
+if [ -f "$GPU_LIB_SCRIPT" ]; then
+    echo "Packaging GPU libraries for BUILD_TYPE=${BUILD_TYPE:-cpu}..."
+    source "$GPU_LIB_SCRIPT" "$CURDIR/package/lib"
+    package_gpu_libs
+fi
+
 echo "Packaging completed successfully" 
 ls -liah $CURDIR/package/
 ls -liah $CURDIR/package/lib/

backend/go/bark-cpp/Makefile

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

backend/go/bark-cpp/gobark.cpp

@@ -1,85 +0,0 @@
-#include <iostream>
-#include <tuple>
-
-#include "bark.h"
-#include "gobark.h"
-#include "common.h"
-#include "ggml.h"
-
-struct bark_context *c;
-
-void bark_print_progress_callback(struct bark_context *bctx, enum bark_encoding_step step, int progress, void *user_data) {
-    if (step == bark_encoding_step::SEMANTIC) {
-        printf("\rGenerating semantic tokens... %d%%", progress);
-    } else if (step == bark_encoding_step::COARSE) {
-        printf("\rGenerating coarse tokens... %d%%", progress);
-    } else if (step == bark_encoding_step::FINE) {
-        printf("\rGenerating fine tokens... %d%%", progress);
-    }
-    fflush(stdout);
-}
-
-int load_model(char *model) {
-    // initialize bark context
-    struct bark_context_params ctx_params = bark_context_default_params();
-    bark_params params;
-
-    params.model_path = model;
-
-   // ctx_params.verbosity = verbosity;
-    ctx_params.progress_callback = bark_print_progress_callback;
-    ctx_params.progress_callback_user_data = nullptr;
-
-    struct bark_context *bctx = bark_load_model(params.model_path.c_str(), ctx_params, params.seed);
-    if (!bctx) {
-        fprintf(stderr, "%s: Could not load model\n", __func__);
-        return 1;
-    }
-
-    c = bctx;
-
-    return 0;
-}
-
-int tts(char *text,int  threads, char *dst ) {
-
-    ggml_time_init();
-    const int64_t t_main_start_us = ggml_time_us();
-
-    // generate audio
-    if (!bark_generate_audio(c, text, threads)) {
-        fprintf(stderr, "%s: An error occurred. If the problem persists, feel free to open an issue to report it.\n", __func__);
-        return 1;
-    }
-
-    const float *audio_data = bark_get_audio_data(c);
-    if (audio_data == NULL) {
-        fprintf(stderr, "%s: Could not get audio data\n", __func__);
-        return 1;
-    }
-
-    const int audio_arr_size = bark_get_audio_data_size(c);
-
-    std::vector<float> audio_arr(audio_data, audio_data + audio_arr_size);
-
-    write_wav_on_disk(audio_arr, dst);
-
-    // report timing
-    {
-        const int64_t t_main_end_us = ggml_time_us();
-        const int64_t t_load_us = bark_get_load_time(c);
-        const int64_t t_eval_us = bark_get_eval_time(c);
-
-        printf("\n\n");
-        printf("%s:     load time = %8.2f ms\n", __func__, t_load_us / 1000.0f);
-        printf("%s:     eval time = %8.2f ms\n", __func__, t_eval_us / 1000.0f);
-        printf("%s:    total time = %8.2f ms\n", __func__, (t_main_end_us - t_main_start_us) / 1000.0f);
-    }
-    
-    return 0;
-}
-
-int unload() {
-    bark_free(c);
-}
-

backend/go/bark-cpp/gobark.go

@@ -1,52 +0,0 @@
-package main
-
-// #cgo CXXFLAGS: -I${SRCDIR}/sources/bark.cpp/ -I${SRCDIR}/sources/bark.cpp/encodec.cpp -I${SRCDIR}/sources/bark.cpp/encodec.cpp/ggml/include -I${SRCDIR}/sources/bark.cpp/examples -I${SRCDIR}/sources/bark.cpp/spm-headers
-// #cgo LDFLAGS: -L${SRCDIR}/ -L${SRCDIR}/sources/bark.cpp/build/examples -L${SRCDIR}/sources/bark.cpp/build/encodec.cpp/ggml/src/ -L${SRCDIR}/sources/bark.cpp/build/encodec.cpp/ -lbark -lencodec -lcommon -lggml -lgomp
-// #include <gobark.h>
-// #include <stdlib.h>
-import "C"
-
-import (
-	"fmt"
-	"unsafe"
-
-	"github.com/mudler/LocalAI/pkg/grpc/base"
-	pb "github.com/mudler/LocalAI/pkg/grpc/proto"
-)
-
-type Bark struct {
-	base.SingleThread
-	threads int
-}
-
-func (sd *Bark) Load(opts *pb.ModelOptions) error {
-
-	sd.threads = int(opts.Threads)
-
-	modelFile := C.CString(opts.ModelFile)
-	defer C.free(unsafe.Pointer(modelFile))
-
-	ret := C.load_model(modelFile)
-	if ret != 0 {
-		return fmt.Errorf("inference failed")
-	}
-
-	return nil
-}
-
-func (sd *Bark) TTS(opts *pb.TTSRequest) error {
-	t := C.CString(opts.Text)
-	defer C.free(unsafe.Pointer(t))
-
-	dst := C.CString(opts.Dst)
-	defer C.free(unsafe.Pointer(dst))
-
-	threads := C.int(sd.threads)
-
-	ret := C.tts(t, threads, dst)
-	if ret != 0 {
-		return fmt.Errorf("inference failed")
-	}
-
-	return nil
-}

backend/go/bark-cpp/gobark.h

@@ -1,8 +0,0 @@
-#ifdef __cplusplus
-extern "C" {
-#endif
-int load_model(char *model);
-int tts(char *text,int  threads, char *dst );
-#ifdef __cplusplus
-}
-#endif

backend/go/bark-cpp/main.go

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

backend/go/bark-cpp/package.sh

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

backend/go/bark-cpp/run.sh

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

backend/go/stablediffusion-ggml/Makefile

@@ -8,7 +8,7 @@ JOBS?=$(shell nproc --ignore=1)
 
 # stablediffusion.cpp (ggml)
 STABLEDIFFUSION_GGML_REPO?=https://github.com/leejet/stable-diffusion.cpp
-STABLEDIFFUSION_GGML_VERSION?=bda7fab9f208dff4b67179a68f694b6ddec13326
+STABLEDIFFUSION_GGML_VERSION?=e411520407663e1ddf8ff2e5ed4ff3a116fbbc97
 
 CMAKE_ARGS+=-DGGML_MAX_NAME=128
 
@@ -28,7 +28,12 @@ else ifeq ($(BUILD_TYPE),clblas)
 	CMAKE_ARGS+=-DGGML_CLBLAST=ON -DCLBlast_DIR=/some/path
 # If it's hipblas we do have also to set CC=/opt/rocm/llvm/bin/clang CXX=/opt/rocm/llvm/bin/clang++
 else ifeq ($(BUILD_TYPE),hipblas)
-	CMAKE_ARGS+=-DSD_HIPBLAS=ON -DGGML_HIPBLAS=ON
+	ROCM_HOME ?= /opt/rocm
+	ROCM_PATH ?= /opt/rocm
+	export CXX=$(ROCM_HOME)/llvm/bin/clang++
+	export CC=$(ROCM_HOME)/llvm/bin/clang
+	AMDGPU_TARGETS?=gfx803,gfx900,gfx906,gfx908,gfx90a,gfx942,gfx1010,gfx1030,gfx1032,gfx1100,gfx1101,gfx1102,gfx1200,gfx1201
+	CMAKE_ARGS+=-DSD_HIPBLAS=ON -DGGML_HIPBLAS=ON -DAMDGPU_TARGETS=$(AMDGPU_TARGETS)
 else ifeq ($(BUILD_TYPE),vulkan)
 	CMAKE_ARGS+=-DSD_VULKAN=ON -DGGML_VULKAN=ON
 else ifeq ($(OS),Darwin)

backend/go/stablediffusion-ggml/gosd.cpp

@@ -55,6 +55,7 @@ const char* schedulers[] = {
     "sgm_uniform",
     "simple",
     "smoothstep",
+    "kl_optimal",
     "lcm",
 };
 
@@ -147,26 +148,26 @@ static std::string lora_dir_path;
 static void build_embedding_vec(const char* embedding_dir) {
     embedding_vec.clear();
     embedding_strings.clear();
-    
+
     if (!embedding_dir || strlen(embedding_dir) == 0) {
         return;
     }
-    
+
     if (!std::filesystem::exists(embedding_dir) || !std::filesystem::is_directory(embedding_dir)) {
         fprintf(stderr, "Embedding directory does not exist or is not a directory: %s\n", embedding_dir);
         return;
     }
-    
+
     static const std::vector<std::string> valid_ext = {".pt", ".safetensors", ".gguf"};
-    
+
     for (const auto& entry : std::filesystem::directory_iterator(embedding_dir)) {
         if (!entry.is_regular_file()) {
             continue;
         }
-        
+
         auto path = entry.path();
         std::string ext = path.extension().string();
-        
+
         bool valid = false;
         for (const auto& e : valid_ext) {
             if (ext == e) {
@@ -177,51 +178,51 @@ static void build_embedding_vec(const char* embedding_dir) {
         if (!valid) {
             continue;
         }
-        
+
         std::string name = path.stem().string();
         std::string full_path = path.string();
-        
+
         // Store strings in persistent storage
         embedding_strings.push_back(name);
         embedding_strings.push_back(full_path);
-        
+
         sd_embedding_t item;
         item.name = embedding_strings[embedding_strings.size() - 2].c_str();
         item.path = embedding_strings[embedding_strings.size() - 1].c_str();
-        
+
         embedding_vec.push_back(item);
         fprintf(stderr, "Found embedding: %s -> %s\n", item.name, item.path);
     }
-    
+
     fprintf(stderr, "Loaded %zu embeddings from %s\n", embedding_vec.size(), embedding_dir);
 }
 
 // Discover LoRA files in directory and build a map of name -> path
 static std::map<std::string, std::string> discover_lora_files(const char* lora_dir) {
     std::map<std::string, std::string> lora_map;
-    
+
     if (!lora_dir || strlen(lora_dir) == 0) {
         fprintf(stderr, "LoRA directory not specified\n");
         return lora_map;
     }
-    
+
     if (!std::filesystem::exists(lora_dir) || !std::filesystem::is_directory(lora_dir)) {
         fprintf(stderr, "LoRA directory does not exist or is not a directory: %s\n", lora_dir);
         return lora_map;
     }
-    
+
     static const std::vector<std::string> valid_ext = {".safetensors", ".ckpt", ".pt", ".gguf"};
-    
+
     fprintf(stderr, "Discovering LoRA files in: %s\n", lora_dir);
-    
+
     for (const auto& entry : std::filesystem::directory_iterator(lora_dir)) {
         if (!entry.is_regular_file()) {
             continue;
         }
-        
+
         auto path = entry.path();
         std::string ext = path.extension().string();
-        
+
         bool valid = false;
         for (const auto& e : valid_ext) {
             if (ext == e) {
@@ -232,17 +233,17 @@ static std::map<std::string, std::string> discover_lora_files(const char* lora_d
         if (!valid) {
             continue;
         }
-        
+
         std::string name = path.stem().string();  // stem() already removes extension
         std::string full_path = path.string();
-        
+
         // Store the name (without extension) -> full path mapping
         // This allows users to specify just the name in <lora:name:strength>
         lora_map[name] = full_path;
-        
+
         fprintf(stderr, "Found LoRA file: %s -> %s\n", name.c_str(), full_path.c_str());
     }
-    
+
     fprintf(stderr, "Discovered %zu LoRA files in %s\n", lora_map.size(), lora_dir);
     return lora_map;
 }
@@ -264,31 +265,31 @@ static bool is_absolute_path(const std::string& p) {
 static std::pair<std::vector<sd_lora_t>, std::string> parse_loras_from_prompt(const std::string& prompt, const char* lora_dir) {
     std::vector<sd_lora_t> loras;
     std::string cleaned_prompt = prompt;
-    
+
     if (!lora_dir || strlen(lora_dir) == 0) {
         fprintf(stderr, "LoRA directory not set, cannot parse LoRAs from prompt\n");
         return {loras, cleaned_prompt};
     }
-    
+
     // Discover LoRA files for name-based lookup
     std::map<std::string, std::string> discovered_lora_map = discover_lora_files(lora_dir);
-    
+
     // Map to accumulate multipliers for the same LoRA (matches upstream)
     std::map<std::string, float> lora_map;
     std::map<std::string, float> high_noise_lora_map;
-    
+
     static const std::regex re(R"(<lora:([^:>]+):([^>]+)>)");
     static const std::vector<std::string> valid_ext = {".pt", ".safetensors", ".gguf"};
     std::smatch m;
-    
+
     std::string tmp = prompt;
-    
+
     fprintf(stderr, "Parsing LoRAs from prompt: %s\n", prompt.c_str());
-    
+
     while (std::regex_search(tmp, m, re)) {
         std::string raw_path = m[1].str();
         const std::string raw_mul = m[2].str();
-        
+
         float mul = 0.f;
         try {
             mul = std::stof(raw_mul);
@@ -298,14 +299,14 @@ static std::pair<std::vector<sd_lora_t>, std::string> parse_loras_from_prompt(co
             fprintf(stderr, "Invalid LoRA multiplier '%s', skipping\n", raw_mul.c_str());
             continue;
         }
-        
+
         bool is_high_noise = false;
         static const std::string prefix = "|high_noise|";
         if (raw_path.rfind(prefix, 0) == 0) {
             raw_path.erase(0, prefix.size());
             is_high_noise = true;
         }
-        
+
         std::filesystem::path final_path;
         if (is_absolute_path(raw_path)) {
             final_path = raw_path;
@@ -334,7 +335,7 @@ static std::pair<std::vector<sd_lora_t>, std::string> parse_loras_from_prompt(co
                 }
             }
         }
-        
+
         // Try adding extensions if file doesn't exist
         if (!std::filesystem::exists(final_path)) {
             bool found = false;
@@ -354,24 +355,24 @@ static std::pair<std::vector<sd_lora_t>, std::string> parse_loras_from_prompt(co
                 continue;
             }
         }
-        
+
         // Normalize path (matches upstream)
         const std::string key = final_path.lexically_normal().string();
-        
+
         // Accumulate multiplier if same LoRA appears multiple times (matches upstream)
         if (is_high_noise) {
             high_noise_lora_map[key] += mul;
         } else {
             lora_map[key] += mul;
         }
-        
-        fprintf(stderr, "Parsed LoRA: path='%s', multiplier=%.2f, is_high_noise=%s\n", 
+
+        fprintf(stderr, "Parsed LoRA: path='%s', multiplier=%.2f, is_high_noise=%s\n",
                 key.c_str(), mul, is_high_noise ? "true" : "false");
-        
+
         cleaned_prompt = std::regex_replace(cleaned_prompt, re, "", std::regex_constants::format_first_only);
         tmp = m.suffix().str();
     }
-    
+
     // Build final LoRA vector from accumulated maps (matches upstream)
     // Store all path strings first to ensure they persist
     for (const auto& kv : lora_map) {
@@ -380,7 +381,7 @@ static std::pair<std::vector<sd_lora_t>, std::string> parse_loras_from_prompt(co
     for (const auto& kv : high_noise_lora_map) {
         lora_strings.push_back(kv.first);
     }
-    
+
     // Now build the LoRA vector with pointers to the stored strings
     size_t string_idx = 0;
     for (const auto& kv : lora_map) {
@@ -391,7 +392,7 @@ static std::pair<std::vector<sd_lora_t>, std::string> parse_loras_from_prompt(co
         loras.push_back(item);
         string_idx++;
     }
-    
+
     for (const auto& kv : high_noise_lora_map) {
         sd_lora_t item;
         item.is_high_noise = true;
@@ -400,7 +401,7 @@ static std::pair<std::vector<sd_lora_t>, std::string> parse_loras_from_prompt(co
         loras.push_back(item);
         string_idx++;
     }
-    
+
     // Clean up extra spaces
     std::regex space_regex(R"(\s+)");
     cleaned_prompt = std::regex_replace(cleaned_prompt, space_regex, " ");
@@ -413,9 +414,9 @@ static std::pair<std::vector<sd_lora_t>, std::string> parse_loras_from_prompt(co
     if (last != std::string::npos) {
         cleaned_prompt.erase(last + 1);
     }
-    
+
     fprintf(stderr, "Parsed %zu LoRA(s) from prompt. Cleaned prompt: %s\n", loras.size(), cleaned_prompt.c_str());
-    
+
     return {loras, cleaned_prompt};
 }
 
@@ -752,7 +753,7 @@ int load_model(const char *model, char *model_path, char* options[], int threads
         }
     }
     if (scheduler == SCHEDULER_COUNT) {
-      scheduler = sd_get_default_scheduler(sd_ctx);
+      scheduler = sd_get_default_scheduler(sd_ctx, sample_method);
       fprintf(stderr, "Invalid scheduler, using default: %s\n", schedulers[scheduler]);
     }
 
@@ -787,7 +788,7 @@ sd_img_gen_params_t* sd_img_gen_params_new(void) {
     sd_img_gen_params_t *params = (sd_img_gen_params_t *)std::malloc(sizeof(sd_img_gen_params_t));
     sd_img_gen_params_init(params);
     sd_sample_params_init(&params->sample_params);
-    sd_easycache_params_init(&params->easycache);
+    sd_cache_params_init(&params->cache);
     params->control_strength = 0.9f;
     return params;
 }
@@ -819,18 +820,18 @@ void sd_img_gen_params_set_prompts(sd_img_gen_params_t *params, const char *prom
         fprintf(stderr, "Note: Found %zu LoRAs in negative prompt (may not be supported)\n", neg_loras.size());
     }
     cleaned_negative_prompt_storage = cleaned_negative;
-    
+
     // Set the cleaned prompts
     params->prompt = cleaned_prompt_storage.c_str();
     params->negative_prompt = cleaned_negative_prompt_storage.c_str();
-    
+
     // Set LoRAs in params
     params->loras = lora_vec.empty() ? nullptr : lora_vec.data();
     params->lora_count = static_cast<uint32_t>(lora_vec.size());
-    
+
     fprintf(stderr, "Set prompts with %zu LoRAs. Original prompt: %s\n", lora_vec.size(), prompt ? prompt : "(null)");
     fprintf(stderr, "Cleaned prompt: %s\n", cleaned_prompt_storage.c_str());
-    
+
     // Debug: Verify LoRAs are set correctly
     if (params->loras && params->lora_count > 0) {
         fprintf(stderr, "DEBUG: LoRAs set in params structure:\n");
@@ -1042,7 +1043,7 @@ int gen_image(sd_img_gen_params_t *p, int steps, char *dst, float cfg_scale, cha
         fprintf(stderr, "Using %u LoRA(s) in generation:\n", p->lora_count);
         for (uint32_t i = 0; i < p->lora_count; i++) {
             fprintf(stderr, "  LoRA[%u]: path='%s', multiplier=%.2f, is_high_noise=%s\n",
-                    i, 
+                    i,
                     p->loras[i].path ? p->loras[i].path : "(null)",
                     p->loras[i].multiplier,
                     p->loras[i].is_high_noise ? "true" : "false");

backend/go/stablediffusion-ggml/package.sh

@@ -6,6 +6,7 @@
 set -e
 
 CURDIR=$(dirname "$(realpath $0)")
+REPO_ROOT="${CURDIR}/../../.."
 
 # Create lib directory
 mkdir -p $CURDIR/package/lib
@@ -50,6 +51,15 @@ else
     exit 1
 fi
 
+# Package GPU libraries based on BUILD_TYPE
+# The GPU library packaging script will detect BUILD_TYPE and copy appropriate GPU libraries
+GPU_LIB_SCRIPT="${REPO_ROOT}/scripts/build/package-gpu-libs.sh"
+if [ -f "$GPU_LIB_SCRIPT" ]; then
+    echo "Packaging GPU libraries for BUILD_TYPE=${BUILD_TYPE:-cpu}..."
+    source "$GPU_LIB_SCRIPT" "$CURDIR/package/lib"
+    package_gpu_libs
+fi
+
 echo "Packaging completed successfully"
 ls -liah $CURDIR/package/
 ls -liah $CURDIR/package/lib/

backend/go/whisper/Makefile

@@ -8,7 +8,7 @@ JOBS?=$(shell nproc --ignore=1)
 
 # whisper.cpp version
 WHISPER_REPO?=https://github.com/ggml-org/whisper.cpp
-WHISPER_CPP_VERSION?=6c22e792cb0ee155b6587ce71a8410c3aeb06949
+WHISPER_CPP_VERSION?=aa1bc0d1a6dfd70dbb9f60c11df12441e03a9075
 SO_TARGET?=libgowhisper.so
 
 CMAKE_ARGS+=-DBUILD_SHARED_LIBS=OFF

backend/go/whisper/gowhisper.go

@@ -130,8 +130,9 @@ func (w *Whisper) AudioTranscription(opts *pb.TranscriptRequest) (pb.TranscriptR
 	segments := []*pb.TranscriptSegment{}
 	text := ""
 	for i := range int(segsLen) {
-		s := CppGetSegmentStart(i)
-		t := CppGetSegmentEnd(i)
+		// segment start/end conversion factor taken from https://github.com/ggml-org/whisper.cpp/blob/master/examples/cli/cli.cpp#L895
+		s := CppGetSegmentStart(i) * (10000000)
+		t := CppGetSegmentEnd(i) * (10000000)
 		txt := strings.Clone(CppGetSegmentText(i))
 		tokens := make([]int32, CppNTokens(i))
 

backend/go/whisper/package.sh

@@ -6,6 +6,7 @@
 set -e
 
 CURDIR=$(dirname "$(realpath $0)")
+REPO_ROOT="${CURDIR}/../../.."
 
 # Create lib directory
 mkdir -p $CURDIR/package/lib
@@ -50,6 +51,15 @@ else
     exit 1
 fi
 
+# Package GPU libraries based on BUILD_TYPE
+# The GPU library packaging script will detect BUILD_TYPE and copy appropriate GPU libraries
+GPU_LIB_SCRIPT="${REPO_ROOT}/scripts/build/package-gpu-libs.sh"
+if [ -f "$GPU_LIB_SCRIPT" ]; then
+    echo "Packaging GPU libraries for BUILD_TYPE=${BUILD_TYPE:-cpu}..."
+    source "$GPU_LIB_SCRIPT" "$CURDIR/package/lib"
+    package_gpu_libs
+fi
+
 echo "Packaging completed successfully"
 ls -liah $CURDIR/package/
 ls -liah $CURDIR/package/lib/

backend/python/README.md

@@ -16,10 +16,8 @@ The Python backends use a unified build system based on `libbackend.sh` that pro
 - **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
@@ -85,7 +83,7 @@ runUnittests
 The build system automatically detects and configures for different hardware:
 
 - **CPU** - Standard CPU-only builds
-- **CUDA** - NVIDIA GPU acceleration (supports CUDA 11/12)
+- **CUDA** - NVIDIA GPU acceleration (supports CUDA 12/13)
 - **Intel** - Intel XPU/GPU optimization
 - **MLX** - Apple Silicon (M1/M2/M3) optimization
 - **HIP** - AMD GPU acceleration
@@ -95,8 +93,8 @@ The build system automatically detects and configures for different hardware:
 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-cublas13.txt` - CUDA 13 packages
 - `requirements-intel.txt` - Intel-optimized packages
 - `requirements-mps.txt` - Apple Silicon packages
 

backend/python/bark/README.md

@@ -1,16 +0,0 @@
-# Creating a separate environment for ttsbark project
-
-```
-make ttsbark
-```
-
-# Testing the gRPC server
-
-```
-<The path of your python interpreter> -m unittest test_ttsbark.py
-```
-
-For example
-```
-/opt/conda/envs/bark/bin/python -m unittest extra/grpc/bark/test_ttsbark.py
-``````

backend/python/bark/requirements-cpu.txt

@@ -1,4 +0,0 @@
-transformers
-accelerate
-torch==2.4.1
-torchaudio==2.4.1

backend/python/bark/requirements-cublas11.txt

@@ -1,5 +0,0 @@
---extra-index-url https://download.pytorch.org/whl/cu118
-torch==2.4.1+cu118
-torchaudio==2.4.1+cu118
-transformers
-accelerate

backend/python/bark/requirements-cublas12.txt

@@ -1,4 +0,0 @@
-torch==2.4.1
-torchaudio==2.4.1
-transformers
-accelerate

backend/python/bark/requirements-hipblas.txt

@@ -1,5 +0,0 @@
---extra-index-url https://download.pytorch.org/whl/rocm6.0
-torch==2.4.1+rocm6.0
-torchaudio==2.4.1+rocm6.0
-transformers
-accelerate

backend/python/bark/requirements-intel.txt

@@ -1,9 +0,0 @@
---extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
-intel-extension-for-pytorch==2.8.10+xpu
-torch==2.3.1+cxx11.abi
-torchaudio==2.3.1+cxx11.abi
-oneccl_bind_pt==2.3.100+xpu
-optimum[openvino]
-setuptools
-transformers
-accelerate

backend/python/bark/requirements.txt

@@ -1,4 +0,0 @@
-bark==0.1.5
-grpcio==1.76.0
-protobuf
-certifi

backend/python/chatterbox/install.sh

@@ -17,4 +17,9 @@ if [ "x${BUILD_PROFILE}" == "xintel" ]; then
 fi
 EXTRA_PIP_INSTALL_FLAGS+=" --no-build-isolation"
 
+if [ "x${BUILD_PROFILE}" == "xl4t12" ]; then
+    USE_PIP=true
+fi
+
+
 installRequirements

backend/python/chatterbox/requirements-cublas11.txt

@@ -1,8 +0,0 @@
---extra-index-url https://download.pytorch.org/whl/cu118
-torch==2.6.0+cu118
-torchaudio==2.6.0+cu118
-transformers==4.46.3
-numpy>=1.24.0,<1.26.0
-# https://github.com/mudler/LocalAI/pull/6240#issuecomment-3329518289
-chatterbox-tts@git+https://git@github.com/mudler/chatterbox.git@faster
-accelerate

backend/python/chatterbox/requirements-hipblas.txt

@@ -1,6 +1,6 @@
---extra-index-url https://download.pytorch.org/whl/rocm6.0
-torch==2.6.0+rocm6.1
-torchaudio==2.6.0+rocm6.1
+--extra-index-url https://download.pytorch.org/whl/rocm6.4
+torch==2.9.1+rocm6.4
+torchaudio==2.9.1+rocm6.4
 transformers
 numpy>=1.24.0,<1.26.0
 # https://github.com/mudler/LocalAI/pull/6240#issuecomment-3329518289

backend/python/chatterbox/requirements-install.txt

@@ -0,0 +1,5 @@
+# Build dependencies needed for packages installed from source (e.g., git dependencies)
+# When using --no-build-isolation, these must be installed in the venv first
+wheel
+setuptools
+packaging

backend/python/chatterbox/requirements-intel.txt

@@ -1,7 +1,6 @@
---extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
-intel-extension-for-pytorch==2.3.110+xpu
-torch==2.3.1+cxx11.abi
-torchaudio==2.3.1+cxx11.abi
+--extra-index-url https://download.pytorch.org/whl/xpu
+torch
+torchaudio
 transformers
 numpy>=1.24.0,<1.26.0
 # https://github.com/mudler/LocalAI/pull/6240#issuecomment-3329518289

backend/python/common/libbackend.sh

@@ -1,7 +1,7 @@
 #!/usr/bin/env bash
 set -euo pipefail
 
-# 
+#
 # use the library by adding the following line to a script:
 # source $(dirname $0)/../common/libbackend.sh
 #
@@ -206,8 +206,8 @@ function init() {
 
 # getBuildProfile will inspect the system to determine which build profile is appropriate:
 # returns one of the following:
-# - cublas11
 # - cublas12
+# - cublas13
 # - hipblas
 # - intel
 function getBuildProfile() {
@@ -392,13 +392,13 @@ 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 more specific version of BUILD_TYPE, ex: cuda-12 or cuda-13
 # 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.
 # you may want to add the following line to a requirements-intel.txt if you use one:
 #
-# --index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
+# --index-url https://download.pytorch.org/whl/xpu
 #
 # If you need to add extra flags into the pip install command you can do so by setting the variable EXTRA_PIP_INSTALL_FLAGS
 # before calling installRequirements.  For example:
@@ -465,6 +465,14 @@ function startBackend() {
     if [ "x${PORTABLE_PYTHON}" == "xtrue" ] || [ -x "$(_portable_python)" ]; then
         _makeVenvPortable --update-pyvenv-cfg
     fi
+
+    # Set up GPU library paths if a lib directory exists
+    # This allows backends to include their own GPU libraries (CUDA, ROCm, etc.)
+    if [ -d "${EDIR}/lib" ]; then
+        export LD_LIBRARY_PATH="${EDIR}/lib:${LD_LIBRARY_PATH:-}"
+        echo "Added ${EDIR}/lib to LD_LIBRARY_PATH for GPU libraries"
+    fi
+
     if [ ! -z "${BACKEND_FILE:-}" ]; then
         exec "${EDIR}/venv/bin/python" "${BACKEND_FILE}" "$@"
     elif [ -e "${MY_DIR}/server.py" ]; then

backend/python/common/template/requirements-hipblas.txt

@@ -1,2 +1,2 @@
---extra-index-url https://download.pytorch.org/whl/rocm6.0
+--extra-index-url https://download.pytorch.org/whl/rocm6.4
 torch

backend/python/common/template/requirements-intel.txt

@@ -1,5 +1,4 @@
---extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
-intel-extension-for-pytorch==2.8.10+xpu
+--extra-index-url https://download.pytorch.org/whl/xpu
 torch==2.8.0
 oneccl_bind_pt==2.8.0+xpu
 optimum[openvino]

backend/python/coqui/README.md

@@ -1,4 +1,4 @@
-# Creating a separate environment for ttsbark project
+# Creating a separate environment for coqui project
 
 ```
 make coqui

backend/python/coqui/backend.py

@@ -1,6 +1,6 @@
 #!/usr/bin/env python3
 """
-This is an extra gRPC server of LocalAI for Bark TTS
+This is an extra gRPC server of LocalAI for Coqui TTS
 """
 from concurrent import futures
 import time

backend/python/coqui/requirements-cublas11.txt

@@ -1,6 +0,0 @@
---extra-index-url https://download.pytorch.org/whl/cu118
-torch==2.4.1+cu118
-torchaudio==2.4.1+cu118
-transformers==4.48.3
-accelerate
-coqui-tts

backend/python/coqui/requirements-hipblas.txt

@@ -1,6 +1,6 @@
---extra-index-url https://download.pytorch.org/whl/rocm6.0
-torch==2.4.1+rocm6.0
-torchaudio==2.4.1+rocm6.0
+--extra-index-url https://download.pytorch.org/whl/rocm6.4
+torch==2.8.0+rocm6.4
+torchaudio==2.8.0+rocm6.4
 transformers==4.48.3
 accelerate
 coqui-tts

backend/python/coqui/requirements-intel.txt

@@ -1,8 +1,6 @@
---extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
-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
+--extra-index-url https://download.pytorch.org/whl/xpu
+torch==2.8.0+xpu
+torchaudio==2.8.0+xpu
 optimum[openvino]
 setuptools
 transformers==4.48.3

backend/python/diffusers/backend.py

@@ -41,6 +41,10 @@ from optimum.quanto import freeze, qfloat8, quantize
 from transformers import T5EncoderModel
 from safetensors.torch import load_file
 
+# Import LTX-2 specific utilities
+from diffusers.pipelines.ltx2.export_utils import encode_video as ltx2_encode_video
+from diffusers import LTX2VideoTransformer3DModel, GGUFQuantizationConfig
+
 _ONE_DAY_IN_SECONDS = 60 * 60 * 24
 COMPEL = os.environ.get("COMPEL", "0") == "1"
 XPU = os.environ.get("XPU", "0") == "1"
@@ -290,6 +294,104 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
                 pipe.enable_model_cpu_offload()
             return pipe
 
+        # LTX2ImageToVideoPipeline - needs img2vid flag, CPU offload, and special handling
+        if pipeline_type == "LTX2ImageToVideoPipeline":
+            self.img2vid = True
+            self.ltx2_pipeline = True
+            
+            # Check if loading from single file (GGUF)
+            if fromSingleFile and LTX2VideoTransformer3DModel is not None:
+                _, single_file_ext = os.path.splitext(modelFile)
+                if single_file_ext == ".gguf":
+                    # Load transformer from single GGUF file with quantization
+                    transformer_kwargs = {}
+                    quantization_config = GGUFQuantizationConfig(compute_dtype=torchType)
+                    transformer_kwargs["quantization_config"] = quantization_config
+                    
+                    transformer = LTX2VideoTransformer3DModel.from_single_file(
+                        modelFile,
+                        config=request.Model,  # Use request.Model as the config/model_id
+                        subfolder="transformer",
+                        **transformer_kwargs,
+                    )
+                    
+                    # Load pipeline with custom transformer
+                    pipe = load_diffusers_pipeline(
+                        class_name="LTX2ImageToVideoPipeline",
+                        model_id=request.Model,
+                        transformer=transformer,
+                        torch_dtype=torchType,
+                    )
+                else:
+                    # Single file but not GGUF - use standard single file loading
+                    pipe = load_diffusers_pipeline(
+                        class_name="LTX2ImageToVideoPipeline",
+                        model_id=modelFile,
+                        from_single_file=True,
+                        torch_dtype=torchType,
+                    )
+            else:
+                # Standard loading from pretrained
+                pipe = load_diffusers_pipeline(
+                    class_name="LTX2ImageToVideoPipeline",
+                    model_id=request.Model,
+                    torch_dtype=torchType,
+                    variant=variant
+                )
+            
+            if not DISABLE_CPU_OFFLOAD:
+                pipe.enable_model_cpu_offload()
+            return pipe
+
+        # LTX2Pipeline - text-to-video pipeline, needs txt2vid flag, CPU offload, and special handling
+        if pipeline_type == "LTX2Pipeline":
+            self.txt2vid = True
+            self.ltx2_pipeline = True
+            
+            # Check if loading from single file (GGUF)
+            if fromSingleFile and LTX2VideoTransformer3DModel is not None:
+                _, single_file_ext = os.path.splitext(modelFile)
+                if single_file_ext == ".gguf":
+                    # Load transformer from single GGUF file with quantization
+                    transformer_kwargs = {}
+                    quantization_config = GGUFQuantizationConfig(compute_dtype=torchType)
+                    transformer_kwargs["quantization_config"] = quantization_config
+                    
+                    transformer = LTX2VideoTransformer3DModel.from_single_file(
+                        modelFile,
+                        config=request.Model,  # Use request.Model as the config/model_id
+                        subfolder="transformer",
+                        **transformer_kwargs,
+                    )
+                    
+                    # Load pipeline with custom transformer
+                    pipe = load_diffusers_pipeline(
+                        class_name="LTX2Pipeline",
+                        model_id=request.Model,
+                        transformer=transformer,
+                        torch_dtype=torchType,
+                    )
+                else:
+                    # Single file but not GGUF - use standard single file loading
+                    pipe = load_diffusers_pipeline(
+                        class_name="LTX2Pipeline",
+                        model_id=modelFile,
+                        from_single_file=True,
+                        torch_dtype=torchType,
+                    )
+            else:
+                # Standard loading from pretrained
+                pipe = load_diffusers_pipeline(
+                    class_name="LTX2Pipeline",
+                    model_id=request.Model,
+                    torch_dtype=torchType,
+                    variant=variant
+                )
+            
+            if not DISABLE_CPU_OFFLOAD:
+                pipe.enable_model_cpu_offload()
+            return pipe
+
         # ================================================================
         # Dynamic pipeline loading - the default path for most pipelines
         # Uses the dynamic loader to instantiate any pipeline by class name
@@ -404,6 +506,9 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
             fromSingleFile = request.Model.startswith("http") or request.Model.startswith("/") or local
             self.img2vid = False
             self.txt2vid = False
+            self.ltx2_pipeline = False
+
+            print(f"LoadModel: PipelineType from request: {request.PipelineType}", file=sys.stderr)
 
             # Load pipeline using dynamic loader
             # Special cases that require custom initialization are handled first
@@ -414,6 +519,8 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
                 torchType=torchType,
                 variant=variant
             )
+            
+            print(f"LoadModel: After loading - ltx2_pipeline: {self.ltx2_pipeline}, img2vid: {self.img2vid}, txt2vid: {self.txt2vid}, PipelineType: {self.PipelineType}", file=sys.stderr)
 
             if CLIPSKIP and request.CLIPSkip != 0:
                 self.clip_skip = request.CLIPSkip
@@ -651,14 +758,20 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
         try:
             prompt = request.prompt
             if not prompt:
+                print(f"GenerateVideo: No prompt provided for video generation.", file=sys.stderr)
                 return backend_pb2.Result(success=False, message="No prompt provided for video generation")
 
+            # Debug: Print raw request values
+            print(f"GenerateVideo: Raw request values - num_frames: {request.num_frames}, fps: {request.fps}, cfg_scale: {request.cfg_scale}, step: {request.step}", file=sys.stderr)
+
             # 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
             
+            print(f"GenerateVideo: Using values - num_frames: {num_frames}, fps: {fps}, cfg_scale: {cfg_scale}, num_inference_steps: {num_inference_steps}", file=sys.stderr)
+            
             # Prepare generation parameters
             kwargs = {
                 "prompt": prompt,
@@ -684,9 +797,86 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
                 kwargs["end_image"] = load_image(request.end_image)
 
             print(f"Generating video with {kwargs=}", file=sys.stderr)
+            print(f"GenerateVideo: Pipeline type: {self.PipelineType}, ltx2_pipeline flag: {self.ltx2_pipeline}", file=sys.stderr)
 
             # Generate video frames based on pipeline type
-            if self.PipelineType == "WanPipeline":
+            if self.ltx2_pipeline or self.PipelineType in ["LTX2Pipeline", "LTX2ImageToVideoPipeline"]:
+                # LTX-2 generation with audio (supports both text-to-video and image-to-video)
+                # Determine if this is text-to-video (no image) or image-to-video (has image)
+                has_image = bool(request.start_image)
+                
+                # Remove image-related parameters that might have been added earlier
+                kwargs.pop("start_image", None)
+                kwargs.pop("end_image", None)
+                
+                # LTX2ImageToVideoPipeline uses 'image' parameter for image-to-video
+                # LTX2Pipeline (text-to-video) doesn't need an image parameter
+                if has_image:
+                    # Image-to-video: use 'image' parameter
+                    if self.PipelineType == "LTX2ImageToVideoPipeline":
+                        image = load_image(request.start_image)
+                        kwargs["image"] = image
+                        print(f"LTX-2: Using image-to-video mode with image", file=sys.stderr)
+                    else:
+                        # If pipeline type is LTX2Pipeline but we have an image, we can't do image-to-video
+                        return backend_pb2.Result(success=False, message="LTX2Pipeline does not support image-to-video. Use LTX2ImageToVideoPipeline for image-to-video generation.")
+                else:
+                    # Text-to-video: no image parameter needed
+                    # Ensure no image-related kwargs are present
+                    kwargs.pop("image", None)
+                    print(f"LTX-2: Using text-to-video mode (no image)", file=sys.stderr)
+                
+                # LTX-2 uses 'frame_rate' instead of 'fps'
+                frame_rate = float(fps)
+                kwargs["frame_rate"] = frame_rate
+                
+                # LTX-2 requires output_type="np" and return_dict=False
+                kwargs["output_type"] = "np"
+                kwargs["return_dict"] = False
+                
+                # Generate video and audio
+                print(f"LTX-2: Generating with kwargs: {kwargs}", file=sys.stderr)
+                try:
+                    video, audio = self.pipe(**kwargs)
+                    print(f"LTX-2: Generated video shape: {video.shape}, audio shape: {audio.shape}", file=sys.stderr)
+                except Exception as e:
+                    print(f"LTX-2: Error during pipe() call: {e}", file=sys.stderr)
+                    traceback.print_exc()
+                    return backend_pb2.Result(success=False, message=f"Error generating video with LTX-2 pipeline: {e}")
+                
+                # Convert video to uint8 format
+                video = (video * 255).round().astype("uint8")
+                video = torch.from_numpy(video)
+                
+                print(f"LTX-2: Converting video, shape after conversion: {video.shape}", file=sys.stderr)
+                print(f"LTX-2: Audio sample rate: {self.pipe.vocoder.config.output_sampling_rate}", file=sys.stderr)
+                print(f"LTX-2: Output path: {request.dst}", file=sys.stderr)
+                
+                # Use LTX-2's encode_video function which handles audio
+                try:
+                    ltx2_encode_video(
+                        video[0],
+                        fps=frame_rate,
+                        audio=audio[0].float().cpu(),
+                        audio_sample_rate=self.pipe.vocoder.config.output_sampling_rate,
+                        output_path=request.dst,
+                    )
+                    # Verify file was created and has content
+                    import os
+                    if os.path.exists(request.dst):
+                        file_size = os.path.getsize(request.dst)
+                        print(f"LTX-2: Video file created successfully, size: {file_size} bytes", file=sys.stderr)
+                        if file_size == 0:
+                            return backend_pb2.Result(success=False, message=f"Video file was created but is empty (0 bytes). Check LTX-2 encode_video function.")
+                    else:
+                        return backend_pb2.Result(success=False, message=f"Video file was not created at {request.dst}")
+                except Exception as e:
+                    print(f"LTX-2: Error encoding video: {e}", file=sys.stderr)
+                    traceback.print_exc()
+                    return backend_pb2.Result(success=False, message=f"Error encoding video: {e}")
+                
+                return backend_pb2.Result(message="Video generated successfully", success=True)
+            elif self.PipelineType == "WanPipeline":
                 # WAN2.2 text-to-video generation
                 output = self.pipe(**kwargs)
                 frames = output.frames[0]  # WAN2.2 returns frames in this format
@@ -725,11 +915,23 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
                 output = self.pipe(**kwargs)
                 frames = output.frames[0]
             else:
+                print(f"GenerateVideo: Pipeline {self.PipelineType} does not match any known video pipeline handler", file=sys.stderr)
                 return backend_pb2.Result(success=False, message=f"Pipeline {self.PipelineType} does not support video generation")
 
-            # Export video
+            # Export video (for non-LTX-2 pipelines)
+            print(f"GenerateVideo: Exporting video to {request.dst} with fps={fps}", file=sys.stderr)
             export_to_video(frames, request.dst, fps=fps)
             
+            # Verify file was created
+            import os
+            if os.path.exists(request.dst):
+                file_size = os.path.getsize(request.dst)
+                print(f"GenerateVideo: Video file created, size: {file_size} bytes", file=sys.stderr)
+                if file_size == 0:
+                    return backend_pb2.Result(success=False, message=f"Video file was created but is empty (0 bytes)")
+            else:
+                return backend_pb2.Result(success=False, message=f"Video file was not created at {request.dst}")
+            
             return backend_pb2.Result(message="Video generated successfully", success=True)
 
         except Exception as err:

backend/python/diffusers/install.sh

@@ -16,8 +16,12 @@ if [ "x${BUILD_PROFILE}" == "xintel" ]; then
     EXTRA_PIP_INSTALL_FLAGS+=" --upgrade --index-strategy=unsafe-first-match"
 fi
 
+if [ "x${BUILD_PROFILE}" == "xl4t12" ]; then
+    USE_PIP=true
+fi
+
 # Use python 3.12 for l4t
-if [ "x${BUILD_PROFILE}" == "xl4t12" ] || [ "x${BUILD_PROFILE}" == "xl4t13" ]; then
+if [ "x${BUILD_PROFILE}" == "xl4t13" ]; then
   PYTHON_VERSION="3.12"
   PYTHON_PATCH="12"
   PY_STANDALONE_TAG="20251120"

backend/python/diffusers/requirements-cublas11.txt

@@ -1,12 +0,0 @@
---extra-index-url https://download.pytorch.org/whl/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

backend/python/diffusers/requirements-hipblas.txt

@@ -1,6 +1,6 @@
---extra-index-url https://download.pytorch.org/whl/rocm6.3
-torch==2.7.1+rocm6.3
-torchvision==0.22.1+rocm6.3
+--extra-index-url https://download.pytorch.org/whl/rocm6.4
+torch==2.8.0+rocm6.4
+torchvision==0.23.0+rocm6.4
 git+https://github.com/huggingface/diffusers
 opencv-python
 transformers

backend/python/diffusers/requirements-intel.txt

@@ -1,8 +1,6 @@
---extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
-intel-extension-for-pytorch==2.3.110+xpu
-torch==2.5.1+cxx11.abi
-torchvision==0.20.1+cxx11.abi
-oneccl_bind_pt==2.8.0+xpu
+--extra-index-url https://download.pytorch.org/whl/xpu
+torch
+torchvision
 optimum[openvino]
 setuptools
 git+https://github.com/huggingface/diffusers

backend/python/diffusers/requirements.txt

@@ -3,3 +3,4 @@ grpcio==1.76.0
 pillow
 protobuf
 certifi
+av

backend/python/exllama2/.gitignore

@@ -1 +0,0 @@
-source

backend/python/exllama2/Makefile

@@ -1,17 +0,0 @@
-.PHONY: exllama2
-exllama2:
-	bash install.sh
-
-.PHONY: run
-run: exllama2
-	@echo "Running exllama2..."
-	bash run.sh
-	@echo "exllama2 run."
-
-.PHONY: protogen-clean
-protogen-clean:
-	$(RM) backend_pb2_grpc.py backend_pb2.py
-
-.PHONY: clean
-clean: protogen-clean
-	$(RM) -r venv source __pycache__

backend/python/exllama2/backend.py

@@ -1,143 +0,0 @@
-#!/usr/bin/env python3
-import grpc
-from concurrent import futures
-import time
-import backend_pb2
-import backend_pb2_grpc
-import argparse
-import signal
-import sys
-import os
-import glob
-
-from pathlib import Path
-import torch
-import torch.nn.functional as F
-from torch import version as torch_version
-
-
-from exllamav2.generator import (
-    ExLlamaV2BaseGenerator,
-    ExLlamaV2Sampler
-)
-
-
-from exllamav2 import (
-    ExLlamaV2,
-    ExLlamaV2Config,
-    ExLlamaV2Cache,
-    ExLlamaV2Cache_8bit,
-    ExLlamaV2Tokenizer,
-    model_init,
-)
-
-
-_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):
-    def Health(self, request, context):
-        return backend_pb2.Reply(message=bytes("OK", 'utf-8'))
-
-    def LoadModel(self, request, context):
-        try:
-            model_directory = request.ModelFile
-
-            config = ExLlamaV2Config()
-            config.model_dir = model_directory
-            config.prepare()
-
-            model = ExLlamaV2(config)
-
-            cache = ExLlamaV2Cache(model, lazy=True)
-            model.load_autosplit(cache)
-
-            tokenizer = ExLlamaV2Tokenizer(config)
-
-            # Initialize generator
-
-            generator = ExLlamaV2BaseGenerator(model, cache, tokenizer)
-
-            self.generator = generator
-
-            generator.warmup()
-            self.model = model
-            self.tokenizer = tokenizer
-            self.cache = cache
-        except Exception as err:
-            return backend_pb2.Result(success=False, message=f"Unexpected {err=}, {type(err)=}")
-        return backend_pb2.Result(message="Model loaded successfully", success=True)
-
-    def Predict(self, request, context):
-
-        penalty = 1.15
-        if request.Penalty != 0.0:
-            penalty = request.Penalty
-
-        settings = ExLlamaV2Sampler.Settings()
-        settings.temperature = request.Temperature
-        settings.top_k = request.TopK
-        settings.top_p = request.TopP
-        settings.token_repetition_penalty = penalty
-        settings.disallow_tokens(self.tokenizer, [self.tokenizer.eos_token_id])
-        tokens = 512
-
-        if request.Tokens != 0:
-            tokens = request.Tokens
-        output = self.generator.generate_simple(
-            request.Prompt, settings, tokens)
-
-        # Remove prompt from response if present
-        if request.Prompt in output:
-            output = output.replace(request.Prompt, "")
-
-        return backend_pb2.Result(message=bytes(output, encoding='utf-8'))
-
-    def PredictStream(self, request, context):
-        # Implement PredictStream RPC
-        # for reply in some_data_generator():
-        #    yield reply
-        # Not implemented yet
-        return self.Predict(request, context)
-
-
-def serve(address):
-    server = grpc.server(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
-        ])
-    backend_pb2_grpc.add_BackendServicer_to_server(BackendServicer(), server)
-    server.add_insecure_port(address)
-    server.start()
-    print("Server started. Listening on: " + address, file=sys.stderr)
-
-    # Define the signal handler function
-    def signal_handler(sig, frame):
-        print("Received termination signal. Shutting down...")
-        server.stop(0)
-        sys.exit(0)
-
-    # Set the signal handlers for SIGINT and SIGTERM
-    signal.signal(signal.SIGINT, signal_handler)
-    signal.signal(signal.SIGTERM, signal_handler)
-
-    try:
-        while True:
-            time.sleep(_ONE_DAY_IN_SECONDS)
-    except KeyboardInterrupt:
-        server.stop(0)
-
-
-if __name__ == "__main__":
-    parser = argparse.ArgumentParser(description="Run the gRPC server.")
-    parser.add_argument(
-        "--addr", default="localhost:50051", help="The address to bind the server to."
-    )
-    args = parser.parse_args()
-
-    serve(args.addr)

backend/python/exllama2/install.sh

@@ -1,21 +0,0 @@
-#!/bin/bash
-set -e
-
-LIMIT_TARGETS="cublas"
-EXTRA_PIP_INSTALL_FLAGS="--no-build-isolation"
-EXLLAMA2_VERSION=c0ddebaaaf8ffd1b3529c2bb654e650bce2f790f
-
-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
-
-git clone https://github.com/turboderp/exllamav2 $MY_DIR/source
-pushd ${MY_DIR}/source && git checkout -b build ${EXLLAMA2_VERSION} && popd
-
-# This installs exllamav2 in JIT mode so it will compile the appropriate torch extension at runtime
-EXLLAMA_NOCOMPILE= uv pip install ${EXTRA_PIP_INSTALL_FLAGS} ${MY_DIR}/source/

backend/python/exllama2/requirements-cpu.txt

@@ -1,3 +0,0 @@
-transformers
-accelerate
-torch==2.4.1

backend/python/exllama2/requirements-cublas11.txt

@@ -1,4 +0,0 @@
---extra-index-url https://download.pytorch.org/whl/cu118
-torch==2.4.1+cu118
-transformers
-accelerate

backend/python/exllama2/requirements-cublas12.txt

@@ -1,3 +0,0 @@
-torch==2.4.1
-transformers
-accelerate

backend/python/exllama2/requirements-install.txt

@@ -1,4 +0,0 @@
-# This is here to trigger the install script to add --no-build-isolation to the uv pip install commands
-# exllama2 does not specify it's build requirements per PEP517, so we need to provide some things ourselves
-wheel
-setuptools

backend/python/exllama2/requirements.txt

@@ -1,5 +0,0 @@
-grpcio==1.76.0
-protobuf
-certifi
-wheel
-setuptools

backend/python/faster-whisper/backend.py

@@ -1,6 +1,6 @@
 #!/usr/bin/env python3
 """
-This is an extra gRPC server of LocalAI for Bark TTS
+This is an extra gRPC server of LocalAI for Faster Whisper TTS
 """
 from concurrent import futures
 import time
@@ -40,7 +40,7 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
             device = "mps"
         try:
             print("Preparing models, please wait", file=sys.stderr)
-            self.model = WhisperModel(request.Model, device=device, compute_type="float16")
+            self.model = WhisperModel(request.Model, device=device, compute_type="default")
         except Exception as err:
             return backend_pb2.Result(success=False, message=f"Unexpected {err=}, {type(err)=}")
         # Implement your logic here for the LoadModel service
@@ -55,11 +55,12 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
             id = 0
             for segment in segments:
                 print("[%.2fs -> %.2fs] %s" % (segment.start, segment.end, segment.text))
-                resultSegments.append(backend_pb2.TranscriptSegment(id=id, start=segment.start, end=segment.end, text=segment.text))
+                resultSegments.append(backend_pb2.TranscriptSegment(id=id, start=int(segment.start)*1e9, end=int(segment.end)*1e9, text=segment.text))
                 text += segment.text
-                id += 1            
+                id += 1
         except Exception as err:
             print(f"Unexpected {err=}, {type(err)=}", file=sys.stderr)
+            raise err
 
         return backend_pb2.TranscriptResult(segments=resultSegments, text=text)
 

backend/python/faster-whisper/requirements-cublas11.txt

@@ -1,9 +0,0 @@
---extra-index-url https://download.pytorch.org/whl/cu118
-torch==2.4.1+cu118
-faster-whisper
-opencv-python
-accelerate
-compel
-peft
-sentencepiece
-optimum-quanto

backend/python/faster-whisper/requirements-hipblas.txt

@@ -1,3 +1,3 @@
---extra-index-url https://download.pytorch.org/whl/rocm6.0
+--extra-index-url https://download.pytorch.org/whl/rocm6.4
 torch
 faster-whisper

backend/python/faster-whisper/requirements-intel.txt

@@ -1,6 +1,4 @@
---extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
-intel-extension-for-pytorch==2.3.110+xpu
-torch==2.3.1+cxx11.abi
-oneccl_bind_pt==2.3.100+xpu
+--extra-index-url https://download.pytorch.org/whl/xpu
+torch
 optimum[openvino]
 faster-whisper

backend/python/kokoro/install.sh

@@ -16,4 +16,8 @@ if [ "x${BUILD_PROFILE}" == "xintel" ]; then
     EXTRA_PIP_INSTALL_FLAGS+=" --upgrade --index-strategy=unsafe-first-match"
 fi
 
+if [ "x${BUILD_PROFILE}" == "xl4t12" ]; then
+    USE_PIP=true
+fi
+
 installRequirements

backend/python/kokoro/requirements-cublas11.txt

@@ -1,7 +0,0 @@
---extra-index-url https://download.pytorch.org/whl/cu118
-torch==2.7.1+cu118
-torchaudio==2.7.1+cu118
-transformers
-accelerate
-kokoro
-soundfile

backend/python/kokoro/requirements-hipblas.txt

@@ -1,6 +1,6 @@
---extra-index-url https://download.pytorch.org/whl/rocm6.3
-torch==2.7.1+rocm6.3
-torchaudio==2.7.1+rocm6.3
+--extra-index-url https://download.pytorch.org/whl/rocm6.4
+torch==2.8.0+rocm6.4
+torchaudio==2.8.0+rocm6.4
 transformers
 accelerate
 kokoro

backend/python/kokoro/requirements-intel.txt

@@ -1,8 +1,6 @@
---extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
-intel-extension-for-pytorch==2.8.10+xpu
-torch==2.5.1+cxx11.abi
-oneccl_bind_pt==2.8.0+xpu
-torchaudio==2.5.1+cxx11.abi
+--extra-index-url https://download.pytorch.org/whl/xpu
+torch
+torchaudio
 optimum[openvino]
 setuptools
 transformers==4.48.3

backend/python/moonshine/Makefile

@@ -0,0 +1,16 @@
+.DEFAULT_GOAL := install
+
+.PHONY: install
+install:
+	bash install.sh
+
+.PHONY: protogen-clean
+protogen-clean:
+	$(RM) backend_pb2_grpc.py backend_pb2.py
+
+.PHONY: clean
+clean: protogen-clean
+	rm -rf venv __pycache__
+
+test: install
+	bash test.sh

backend/python/moonshine/backend.py

@@ -1,6 +1,6 @@
 #!/usr/bin/env python3
 """
-This is an extra gRPC server of LocalAI for Bark TTS
+This is an extra gRPC server of LocalAI for Moonshine transcription
 """
 from concurrent import futures
 import time
@@ -8,11 +8,9 @@ import argparse
 import signal
 import sys
 import os
-from scipy.io.wavfile import write as write_wav
-
 import backend_pb2
 import backend_pb2_grpc
-from bark import SAMPLE_RATE, generate_audio, preload_models
+import moonshine_onnx
 
 import grpc
 
@@ -29,36 +27,52 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
     """
     def Health(self, request, context):
         return backend_pb2.Reply(message=bytes("OK", 'utf-8'))
+    
     def LoadModel(self, request, context):
-        model_name = request.Model
         try:
             print("Preparing models, please wait", file=sys.stderr)
-            # download and load all models
-            preload_models()
+            # Store the model name for use in transcription
+            # Model name format: e.g., "moonshine/tiny"
+            self.model_name = request.Model
+            print(f"Model name set to: {self.model_name}", file=sys.stderr)
         except Exception as err:
             return backend_pb2.Result(success=False, message=f"Unexpected {err=}, {type(err)=}")
-        # Implement your logic here for the LoadModel service
-        # Replace this with your desired response
         return backend_pb2.Result(message="Model loaded successfully", success=True)
 
-    def TTS(self, request, context):
-        model = request.model
-        print(request, file=sys.stderr)
+    def AudioTranscription(self, request, context):
+        resultSegments = []
+        text = ""
         try:
-            audio_array = None
-            if model != "":
-                audio_array = generate_audio(request.text, history_prompt=model)
+            # moonshine_onnx.transcribe returns a list of strings
+            transcriptions = moonshine_onnx.transcribe(request.dst, self.model_name)
+            
+            # Combine all transcriptions into a single text
+            if isinstance(transcriptions, list):
+                text = " ".join(transcriptions)
+                # Create segments for each transcription in the list
+                for id, trans in enumerate(transcriptions):
+                    # Since moonshine doesn't provide timing info, we'll create a single segment
+                    # with id and text, using approximate timing
+                    resultSegments.append(backend_pb2.TranscriptSegment(
+                        id=id, 
+                        start=0, 
+                        end=0, 
+                        text=trans
+                    ))
             else:
-                audio_array = generate_audio(request.text)
-            print("saving to", request.dst, file=sys.stderr)
-            # save audio to disk
-            write_wav(request.dst, SAMPLE_RATE, audio_array)
-            print("saved to", request.dst, file=sys.stderr)
-            print("tts for", file=sys.stderr)
-            print(request, file=sys.stderr)
+                # Handle case where it's not a list (shouldn't happen, but be safe)
+                text = str(transcriptions)
+                resultSegments.append(backend_pb2.TranscriptSegment(
+                    id=0,
+                    start=0,
+                    end=0,
+                    text=text
+                ))
         except Exception as err:
-            return backend_pb2.Result(success=False, message=f"Unexpected {err=}, {type(err)=}")
-        return backend_pb2.Result(success=True)
+            print(f"Unexpected {err=}, {type(err)=}", file=sys.stderr)
+            return backend_pb2.TranscriptResult(segments=[], text="")
+
+        return backend_pb2.TranscriptResult(segments=resultSegments, text=text)
 
 def serve(address):
     server = grpc.server(futures.ThreadPoolExecutor(max_workers=MAX_WORKERS),
@@ -96,3 +110,4 @@ if __name__ == "__main__":
     args = parser.parse_args()
 
     serve(args.addr)
+

backend/python/moonshine/install.sh

@@ -0,0 +1,12 @@
+#!/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
+
+installRequirements
+

backend/python/moonshine/protogen.sh

@@ -0,0 +1,12 @@
+#!/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
+
+python3 -m grpc_tools.protoc -I../.. -I./ --python_out=. --grpc_python_out=. backend.proto
+

backend/python/moonshine/requirements.txt

@@ -0,0 +1,4 @@
+grpcio==1.71.0
+protobuf
+grpcio-tools
+useful-moonshine-onnx@git+https://git@github.com/moonshine-ai/moonshine.git#subdirectory=moonshine-onnx

backend/python/moonshine/run.sh

@@ -1,6 +1,4 @@
 #!/bin/bash
-LIMIT_TARGETS="cublas"
-
 backend_dir=$(dirname $0)
 if [ -d $backend_dir/common ]; then
     source $backend_dir/common/libbackend.sh
@@ -8,4 +6,5 @@ else
     source $backend_dir/../common/libbackend.sh
 fi
 
-startBackend $@
+startBackend $@
+

backend/python/moonshine/test.py

@@ -0,0 +1,139 @@
+"""
+A test script to test the gRPC service for Moonshine transcription
+"""
+import unittest
+import subprocess
+import time
+import os
+import tempfile
+import shutil
+import backend_pb2
+import backend_pb2_grpc
+
+import grpc
+
+
+class TestBackendServicer(unittest.TestCase):
+    """
+    TestBackendServicer is the class that tests the gRPC service
+    """
+    def setUp(self):
+        """
+        This method sets up the gRPC service by starting the server
+        """
+        self.service = subprocess.Popen(["python3", "backend.py", "--addr", "localhost:50051"])
+        time.sleep(10)
+
+    def tearDown(self) -> None:
+        """
+        This method tears down the gRPC service by terminating the server
+        """
+        self.service.terminate()
+        self.service.wait()
+
+    def test_server_startup(self):
+        """
+        This method tests if the server starts up successfully
+        """
+        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 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="moonshine/tiny"))
+                self.assertTrue(response.success)
+                self.assertEqual(response.message, "Model loaded successfully")
+        except Exception as err:
+            print(err)
+            self.fail("LoadModel service failed")
+        finally:
+            self.tearDown()
+
+    def test_audio_transcription(self):
+        """
+        This method tests if audio transcription works successfully
+        """
+        # Create a temporary directory for the audio file
+        temp_dir = tempfile.mkdtemp()
+        audio_file = os.path.join(temp_dir, 'audio.wav')
+        
+        try:
+            # Download the audio file to the temporary directory
+            print(f"Downloading audio file to {audio_file}...")
+            url = "https://cdn.openai.com/whisper/draft-20220913a/micro-machines.wav"
+            result = subprocess.run(
+                ["wget", "-q", url, "-O", audio_file],
+                capture_output=True,
+                text=True
+            )
+            if result.returncode != 0:
+                self.fail(f"Failed to download audio file: {result.stderr}")
+            
+            # Verify the file was downloaded
+            if not os.path.exists(audio_file):
+                self.fail(f"Audio file was not downloaded to {audio_file}")
+            
+            self.setUp()
+            with grpc.insecure_channel("localhost:50051") as channel:
+                stub = backend_pb2_grpc.BackendStub(channel)
+                # Load the model first
+                load_response = stub.LoadModel(backend_pb2.ModelOptions(Model="moonshine/tiny"))
+                self.assertTrue(load_response.success)
+                
+                # Perform transcription
+                transcript_request = backend_pb2.TranscriptRequest(dst=audio_file)
+                transcript_response = stub.AudioTranscription(transcript_request)
+                
+                # Print the transcribed text for debugging
+                print(f"Transcribed text: {transcript_response.text}")
+                print(f"Number of segments: {len(transcript_response.segments)}")
+                
+                # Verify response structure
+                self.assertIsNotNone(transcript_response)
+                self.assertIsNotNone(transcript_response.text)
+                # Protobuf repeated fields return a sequence, not a list
+                self.assertIsNotNone(transcript_response.segments)
+                # Check if segments is iterable (has length)
+                self.assertGreaterEqual(len(transcript_response.segments), 0)
+                
+                # Verify the transcription contains the expected text
+                expected_text = "This is the micro machine man presenting the most midget miniature"
+                self.assertIn(
+                    expected_text.lower(),
+                    transcript_response.text.lower(),
+                    f"Expected text '{expected_text}' not found in transcription: '{transcript_response.text}'"
+                )
+                
+                # If we got segments, verify they have the expected structure
+                if len(transcript_response.segments) > 0:
+                    segment = transcript_response.segments[0]
+                    self.assertIsNotNone(segment.text)
+                    self.assertIsInstance(segment.id, int)
+                else:
+                    # Even if no segments, we should have text
+                    self.assertIsNotNone(transcript_response.text)
+                    self.assertGreater(len(transcript_response.text), 0)
+        except Exception as err:
+            print(err)
+            self.fail("AudioTranscription service failed")
+        finally:
+            self.tearDown()
+            # Clean up the temporary directory
+            if os.path.exists(temp_dir):
+                shutil.rmtree(temp_dir)
+

backend/python/moonshine/test.sh

@@ -0,0 +1,12 @@
+#!/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
+

backend/python/neutts/install.sh

@@ -26,6 +26,12 @@ fi
 
 EXTRA_PIP_INSTALL_FLAGS+=" --no-build-isolation"
 
+
+if [ "x${BUILD_PROFILE}" == "xl4t12" ]; then
+    USE_PIP=true
+fi
+
+
 git clone https://github.com/neuphonic/neutts-air neutts-air
 
 cp -rfv neutts-air/neuttsair ./

backend/python/neutts/requirements-hipblas.txt

@@ -1,5 +1,5 @@
---extra-index-url https://download.pytorch.org/whl/rocm6.3
-torch==2.8.0+rocm6.3
+--extra-index-url https://download.pytorch.org/whl/rocm6.4
+torch==2.8.0+rocm6.4
 transformers==4.56.1
 accelerate
 librosa==0.11.0

backend/python/pocket-tts/Makefile

@@ -0,0 +1,23 @@
+.PHONY: pocket-tts
+pocket-tts:
+	bash install.sh
+
+.PHONY: run
+run: pocket-tts
+	@echo "Running pocket-tts..."
+	bash run.sh
+	@echo "pocket-tts run."
+
+.PHONY: test
+test: pocket-tts
+	@echo "Testing pocket-tts..."
+	bash test.sh
+	@echo "pocket-tts tested."
+
+.PHONY: protogen-clean
+protogen-clean:
+	$(RM) backend_pb2_grpc.py backend_pb2.py
+
+.PHONY: clean
+clean: protogen-clean
+	rm -rf venv __pycache__

backend/python/pocket-tts/backend.py

@@ -0,0 +1,255 @@
+#!/usr/bin/env python3
+"""
+This is an extra gRPC server of LocalAI for Pocket TTS
+"""
+from concurrent import futures
+import time
+import argparse
+import signal
+import sys
+import os
+import traceback
+import scipy.io.wavfile
+import backend_pb2
+import backend_pb2_grpc
+import torch
+from pocket_tts import TTSModel
+
+import grpc
+
+def is_float(s):
+    """Check if a string can be converted to float."""
+    try:
+        float(s)
+        return True
+    except ValueError:
+        return False
+
+def is_int(s):
+    """Check if a string can be converted to int."""
+    try:
+        int(s)
+        return True
+    except ValueError:
+        return False
+
+_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):
+    """
+    BackendServicer is the class that implements the gRPC service
+    """
+    def Health(self, request, context):
+        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"
+        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")
+
+        # Normalize potential 'mpx' typo to 'mps'
+        if device == "mpx":
+            print("Note: device 'mpx' detected, treating it as 'mps'.", file=sys.stderr)
+            device = "mps"
+        
+        # Validate mps availability if requested
+        if device == "mps" and not torch.backends.mps.is_available():
+            print("Warning: MPS not available. Falling back to CPU.", file=sys.stderr)
+            device = "cpu"
+
+        self.device = device
+
+        options = request.Options
+
+        # empty dict
+        self.options = {}
+
+        # The options are a list of strings in this form optname:optvalue
+        # We are storing all the options in a dict so we can use it later when
+        # generating the audio
+        for opt in options:
+            if ":" not in opt:
+                continue
+            key, value = opt.split(":", 1)  # Split only on first colon
+            # if value is a number, convert it to the appropriate type
+            if is_float(value):
+                value = float(value)
+            elif is_int(value):
+                value = int(value)
+            elif value.lower() in ["true", "false"]:
+                value = value.lower() == "true"
+            self.options[key] = value
+
+        # Default voice for caching
+        self.default_voice_url = self.options.get("default_voice", None)
+        self._voice_cache = {}
+
+        try:
+            print("Loading Pocket TTS model", file=sys.stderr)
+            self.tts_model = TTSModel.load_model()
+            print(f"Model loaded successfully. Sample rate: {self.tts_model.sample_rate}", file=sys.stderr)
+
+            # Pre-load default voice if specified
+            if self.default_voice_url:
+                try:
+                    print(f"Pre-loading default voice: {self.default_voice_url}", file=sys.stderr)
+                    voice_state = self.tts_model.get_state_for_audio_prompt(self.default_voice_url)
+                    self._voice_cache[self.default_voice_url] = voice_state
+                    print("Default voice loaded successfully", file=sys.stderr)
+                except Exception as e:
+                    print(f"Warning: Failed to pre-load default voice: {e}", file=sys.stderr)
+
+        except Exception as err:
+            return backend_pb2.Result(success=False, message=f"Unexpected {err=}, {type(err)=}")
+        
+        return backend_pb2.Result(message="Model loaded successfully", success=True)
+
+    def _get_voice_state(self, voice_input):
+        """
+        Get voice state from cache or load it.
+        voice_input can be:
+        - HuggingFace URL (e.g., hf://kyutai/tts-voices/alba-mackenna/casual.wav)
+        - Local file path
+        - None (use default)
+        """
+        # Use default if no voice specified
+        if not voice_input:
+            voice_input = self.default_voice_url
+
+        if not voice_input:
+            return None
+
+        # Check cache first
+        if voice_input in self._voice_cache:
+            return self._voice_cache[voice_input]
+
+        # Load voice state
+        try:
+            print(f"Loading voice from: {voice_input}", file=sys.stderr)
+            voice_state = self.tts_model.get_state_for_audio_prompt(voice_input)
+            self._voice_cache[voice_input] = voice_state
+            return voice_state
+        except Exception as e:
+            print(f"Error loading voice from {voice_input}: {e}", file=sys.stderr)
+            return None
+
+    def TTS(self, request, context):
+        try:
+            # Determine voice input
+            # Priority: request.voice > AudioPath (from ModelOptions) > default
+            voice_input = None
+            
+            if request.voice:
+                voice_input = request.voice
+            elif hasattr(request, 'AudioPath') and request.AudioPath:
+                # Use AudioPath as voice file
+                if os.path.isabs(request.AudioPath):
+                    voice_input = request.AudioPath
+                elif hasattr(request, 'ModelFile') and request.ModelFile:
+                    model_file_base = os.path.dirname(request.ModelFile)
+                    voice_input = os.path.join(model_file_base, request.AudioPath)
+                elif hasattr(request, 'ModelPath') and request.ModelPath:
+                    voice_input = os.path.join(request.ModelPath, request.AudioPath)
+                else:
+                    voice_input = request.AudioPath
+
+            # Get voice state
+            voice_state = self._get_voice_state(voice_input)
+            if voice_state is None:
+                return backend_pb2.Result(
+                    success=False,
+                    message=f"Voice not found or failed to load: {voice_input}. Please provide a valid voice URL or file path."
+                )
+
+            # Prepare text
+            text = request.text.strip()
+
+            if not text:
+                return backend_pb2.Result(
+                    success=False,
+                    message="Text is empty"
+                )
+
+            print(f"Generating audio for text: {text[:50]}...", file=sys.stderr)
+
+            # Generate audio
+            audio = self.tts_model.generate_audio(voice_state, text)
+
+            # Audio is a 1D torch tensor containing PCM data
+            if audio is None or audio.numel() == 0:
+                return backend_pb2.Result(
+                    success=False,
+                    message="No audio generated"
+                )
+
+            # Save audio to file
+            output_path = request.dst
+            if not output_path:
+                output_path = "/tmp/pocket-tts-output.wav"
+
+            # Ensure output directory exists
+            output_dir = os.path.dirname(output_path)
+            if output_dir and not os.path.exists(output_dir):
+                os.makedirs(output_dir, exist_ok=True)
+
+            # Convert torch tensor to numpy and save
+            audio_numpy = audio.numpy()
+            scipy.io.wavfile.write(output_path, self.tts_model.sample_rate, audio_numpy)
+            print(f"Saved audio to {output_path}", file=sys.stderr)
+
+        except Exception as err:
+            print(f"Error in TTS: {err}", file=sys.stderr)
+            print(traceback.format_exc(), file=sys.stderr)
+            return backend_pb2.Result(success=False, message=f"Unexpected {err=}, {type(err)=}")
+        
+        return backend_pb2.Result(success=True)
+
+def serve(address):
+    server = grpc.server(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
+        ])
+    backend_pb2_grpc.add_BackendServicer_to_server(BackendServicer(), server)
+    server.add_insecure_port(address)
+    server.start()
+    print("Server started. Listening on: " + address, file=sys.stderr)
+
+    # Define the signal handler function
+    def signal_handler(sig, frame):
+        print("Received termination signal. Shutting down...")
+        server.stop(0)
+        sys.exit(0)
+
+    # Set the signal handlers for SIGINT and SIGTERM
+    signal.signal(signal.SIGINT, signal_handler)
+    signal.signal(signal.SIGTERM, signal_handler)
+
+    try:
+        while True:
+            time.sleep(_ONE_DAY_IN_SECONDS)
+    except KeyboardInterrupt:
+        server.stop(0)
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description="Run the gRPC server.")
+    parser.add_argument(
+        "--addr", default="localhost:50051", help="The address to bind the server to."
+    )
+    args = parser.parse_args()
+
+    serve(args.addr)

backend/python/pocket-tts/install.sh

@@ -16,4 +16,15 @@ if [ "x${BUILD_PROFILE}" == "xintel" ]; then
     EXTRA_PIP_INSTALL_FLAGS+=" --upgrade --index-strategy=unsafe-first-match"
 fi
 
+# Use python 3.12 for l4t
+if [ "x${BUILD_PROFILE}" == "xl4t13" ]; then
+  PYTHON_VERSION="3.12"
+  PYTHON_PATCH="12"
+  PY_STANDALONE_TAG="20251120"
+fi
+
+if [ "x${BUILD_PROFILE}" == "xl4t12" ]; then
+    USE_PIP=true
+fi
+
 installRequirements

backend/python/pocket-tts/protogen.sh

@@ -0,0 +1,11 @@
+#!/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
+
+python3 -m grpc_tools.protoc -I../.. -I./ --python_out=. --grpc_python_out=. backend.proto