feat(qwen-tts): add Qwen-tts backend (#8163 )

* feat(qwen-tts): add Qwen-tts backend Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * Update intel deps Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * Drop flash-attn for cuda13 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> --------- Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
chore(model gallery): 🤖 add 1 new models via gallery agent (#8170 )
2026-02-03 03:02:38 -05:00 · 2026-01-23 15:18:41 +01:00 · 2026-01-23 08:19:39 +01:00 · 2026-01-23 08:18:05 +01:00 · 2026-01-23 00:38:28 +01:00 · 2026-01-22 23:32:05 +00:00
212 changed files with 24029 additions and 2200 deletions
--- a/.github/workflows/backend.yml
+++ b/.github/workflows/backend.yml
--- a/.github/workflows/dependabot_auto.yml
+++ b/.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
--- a/.github/workflows/generate_grpc_cache.yaml
+++ b/.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}}
--- a/.github/workflows/generate_intel_image.yaml
+++ b/.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: ${{ matrix.platforms }}
--- a/.github/workflows/image-pr.yml
+++ b/.github/workflows/image-pr.yml
@@ -1,94 +1,95 @@
 ---
-name: 'build container images tests'
-
-on:
-  pull_request:
-
-concurrency:
-  group: ci-${{ 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: "9"
+              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'
+  
--- a/.github/workflows/image.yml
+++ b/.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: "9"
+              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: ${{ secrets.LOCALAI_REGISTRY_PASSWORD }}
+      strategy:
+        matrix:
+          include:
+            - build-type: 'cublas'
+              cuda-major-version: "12"
+              cuda-minor-version: "0"
+              platforms: 'linux/arm64'
+              tag-latest: 'auto'
+              tag-suffix: '-nvidia-l4t-arm64'
+              base-image: "nvcr.io/nvidia/l4t-jetpack:r36.4.0"
+              runs-on: 'ubuntu-24.04-arm'
+              makeflags: "--jobs=4 --output-sync=target"
+              skip-drivers: 'true'
+              ubuntu-version: "2204"
+              ubuntu-codename: 'jammy'
+            - build-type: 'cublas'
+              cuda-major-version: "13"
+              cuda-minor-version: "0"
+              platforms: 'linux/arm64'
+              tag-latest: 'auto'
+              tag-suffix: '-nvidia-l4t-arm64-cuda-13'
+              base-image: "ubuntu:24.04"
+              runs-on: 'ubuntu-24.04-arm'
+              makeflags: "--jobs=4 --output-sync=target"
+              skip-drivers: 'false'
+              ubuntu-version: '2404'
+              ubuntu-codename: 'noble'
+  
--- a/.github/workflows/image_build.yml
+++ b/.github/workflows/image_build.yml
@@ -23,7 +23,7 @@ on:
        type: string
      cuda-minor-version:
        description: 'CUDA minor version'
-        default: "4"
+        default: "9"
        type: string
      platforms:
        description: 'Platforms'
@@ -61,6 +61,11 @@ on:
        required: false
        default: '2204'
        type: string
+      ubuntu-codename:
+        description: 'Ubuntu codename'
+        required: false
+        default: 'noble'
+        type: string
    secrets:
      dockerUsername:
        required: true
@@ -244,6 +249,7 @@ jobs:
            MAKEFLAGS=${{ 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
--- a/.github/workflows/test-extra.yml
+++ b/.github/workflows/test-extra.yml
@@ -247,3 +247,60 @@ 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 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 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 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
--- a/.gitignore
+++ b/.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
--- a/AGENTS.md
+++ b/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 bark for ROCM/hipblas
+
+- The Makefile has targets like `docker-build-bark` 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-bark`
+- 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
+	...
+	$(MAKE) -C backend/python/<backend-name> test
+```
+
+**Step 4d: Add Backend Definition**
+
+Add a backend definition variable in the backend definitions section (around line 428-457). The format depends on the backend type:
+
+**For Python backends with root context** (like `faster-whisper`, `bark`):
+```makefile
+BACKEND_<BACKEND_NAME> = <backend-name>|python|.|false|true
+```
+
+**For Python backends with `./backend` context** (like `chatterbox`, `moonshine`):
+```makefile
+BACKEND_<BACKEND_NAME> = <backend-name>|python|./backend|false|true
+```
+
+**For Go backends**:
+```makefile
+BACKEND_<BACKEND_NAME> = <backend-name>|golang|.|false|true
+```
+
+**Step 4e: Generate Docker Build Target**
+
+Add an eval call to generate the docker-build target (around line 480-501):
+
+```makefile
+$(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 +234,49 @@ 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
--- a/CONTRIBUTING.md
+++ b/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.
--- a/59
+++ b/59
@@ -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 && \
--- a/Dockerfile.aio
+++ b/Dockerfile.aio
@@ -1,4 +1,4 @@
-ARG BASE_IMAGE=ubuntu:22.04
+ARG BASE_IMAGE=ubuntu:24.04

 FROM ${BASE_IMAGE} 

--- a/59
+++ b/59
@@ -1,3 +1,6 @@
+# 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/moonshine backends/pocket-tts backends/qwen-tts
+
 GOCMD=go
 GOTEST=$(GOCMD) test
 GOVET=$(GOCMD) vet
@@ -6,11 +9,14 @@ LAUNCHER_BINARY_NAME=local-ai-launcher

 CUDA_MAJOR_VERSION?=13
 CUDA_MINOR_VERSION?=0
-UBUNTU_VERSION?=2204
+UBUNTU_VERSION?=2404
+UBUNTU_CODENAME?=noble

 GORELEASER?=

 export BUILD_TYPE?=
+export CUDA_MAJOR_VERSION?=12
+export CUDA_MINOR_VERSION?=9

 GO_TAGS?=
 BUILD_ID?=
@@ -164,6 +170,7 @@ docker-build-aio:
 		--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 $(MAKE) docker-aio
@@ -194,6 +201,7 @@ prepare-e2e:
 		--build-arg CUDA_MAJOR_VERSION=$(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 .
@@ -307,6 +315,9 @@ prepare-test-extra: protogen-python
 	$(MAKE) -C backend/python/chatterbox
 	$(MAKE) -C backend/python/vllm
 	$(MAKE) -C backend/python/vibevoice
+	$(MAKE) -C backend/python/moonshine
+	$(MAKE) -C backend/python/pocket-tts
+	$(MAKE) -C backend/python/qwen-tts

 test-extra: prepare-test-extra
 	$(MAKE) -C backend/python/transformers test
@@ -314,11 +325,14 @@ test-extra: prepare-test-extra
 	$(MAKE) -C backend/python/chatterbox test
 	$(MAKE) -C backend/python/vllm 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

 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 \
@@ -330,19 +344,21 @@ docker:
 		--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=$(BUILD_TYPE) \
 		--build-arg UBUNTU_VERSION=$(UBUNTU_VERSION) \
-		-t $(DOCKER_IMAGE)-cuda-11 .
+		--build-arg UBUNTU_CODENAME=$(UBUNTU_CODENAME) \
+		-t $(DOCKER_IMAGE)-cuda-12 .

 docker-aio:
 	@echo "Building AIO image with base $(BASE_IMAGE) as $(DOCKER_AIO_IMAGE)"
@@ -352,6 +368,7 @@ docker-aio:
 		--build-arg CUDA_MAJOR_VERSION=$(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:
@@ -360,7 +377,7 @@ 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="$(DOCKER_MAKEFLAGS)" \
@@ -368,6 +385,7 @@ docker-image-intel:
 		--build-arg CUDA_MAJOR_VERSION=$(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) .

 ########################################################
@@ -433,16 +451,17 @@ BACKEND_FASTER_WHISPER = faster-whisper|python|.|false|true
 BACKEND_COQUI = coqui|python|.|false|true
 BACKEND_BARK = bark|python|.|false|true
 BACKEND_EXLLAMA2 = exllama2|python|.|false|true
-
-# Python backends with ./backend context
-BACKEND_RFDETR = rfdetr|python|./backend|false|true
-BACKEND_KITTEN_TTS = kitten-tts|python|./backend|false|true
-BACKEND_NEUTTS = neutts|python|./backend|false|true
-BACKEND_KOKORO = kokoro|python|./backend|false|true
-BACKEND_VLLM = vllm|python|./backend|false|true
-BACKEND_DIFFUSERS = diffusers|python|./backend|--progress=plain|true
-BACKEND_CHATTERBOX = chatterbox|python|./backend|false|true
-BACKEND_VIBEVOICE = vibevoice|python|./backend|--progress=plain|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_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

 # Helper function to build docker image for a backend
 # Usage: $(call docker-build-backend,BACKEND_NAME,DOCKERFILE_TYPE,BUILD_CONTEXT,PROGRESS_FLAG,NEEDS_BACKEND_ARG)
@@ -453,6 +472,7 @@ define docker-build-backend
 		--build-arg CUDA_MAJOR_VERSION=$(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
@@ -486,12 +506,15 @@ $(eval $(call generate-docker-build-target,$(BACKEND_VLLM)))
 $(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)))

 # Pattern rule for docker-save targets
 docker-save-%: backend-images
 	docker save local-ai-backend:$* -o backend-images/$*.tar

-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
+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 docker-build-moonshine docker-build-pocket-tts docker-build-qwen-tts

 ########################################################
 ### END Backends
--- a/README.md
+++ b/README.md
@@ -111,6 +111,8 @@

 ## 💻 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 +130,7 @@ 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**
 > 
@@ -137,13 +139,13 @@ 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
@@ -152,9 +154,6 @@ docker run -ti --name local-ai -p 8080:8080 --gpus all localai/localai:latest-gp
 # 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
@@ -163,25 +162,25 @@ docker run -ti --name local-ai -p 8080:8080 --gpus all localai/localai:latest-nv
 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
@@ -193,9 +192,6 @@ docker run -ti --name local-ai -p 8080:8080 --gpus all localai/localai:latest-ai
 # 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

@@ -279,9 +275,9 @@ 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 11/12/13, ROCm, Intel SYCL, Vulkan, Metal, CPU |
+| **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 11/12/13, ROCm, Intel, CPU |
+| **transformers** | HuggingFace transformers framework | CUDA 12/13, ROCm, Intel, CPU |
 | **exllama2** | GPTQ inference library | CUDA 12/13 |
 | **MLX** | Apple Silicon LLM inference | Metal (M1/M2/M3+) |
 | **MLX-VLM** | Apple Silicon Vision-Language Models | Metal (M1/M2/M3+) |
@@ -295,24 +291,26 @@ LocalAI supports a comprehensive range of AI backends with multiple acceleration
 | **bark-cpp** | C++ implementation of Bark | CUDA, Metal, 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 11/12/13, 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 12/13, 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 12/13, Intel SYCL, Vulkan, CPU |
-| **diffusers** | HuggingFace diffusion models | CUDA 11/12/13, ROCm, Intel, Metal, 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 12/13, Intel, CPU |
-| **rerankers** | Document reranking API | CUDA 11/12/13, ROCm, Intel, CPU |
+| **rerankers** | Document reranking API | CUDA 12/13, ROCm, Intel, CPU |
 | **local-store** | Vector database | CPU |
 | **huggingface** | HuggingFace API integration | API-based |

@@ -320,11 +318,10 @@ 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 |
 | **NVIDIA CUDA 13** | All CUDA-compatible backends | Nvidia hardware |
-| **AMD ROCm** | llama.cpp, whisper, vllm, transformers, diffusers, rerankers, coqui, kokoro, bark, neutts, vibevoice | AMD Graphics |
-| **Intel oneAPI** | llama.cpp, whisper, stablediffusion, vllm, transformers, diffusers, rfdetr, rerankers, exllama2, coqui, kokoro, bark, vibevoice | Intel Arc, Intel iGPUs |
+| **AMD ROCm** | llama.cpp, whisper, vllm, transformers, diffusers, rerankers, coqui, kokoro, bark, neutts, vibevoice, pocket-tts, qwen-tts | AMD Graphics |
+| **Intel oneAPI** | llama.cpp, whisper, stablediffusion, vllm, transformers, diffusers, rfdetr, rerankers, exllama2, coqui, kokoro, bark, vibevoice, pocket-tts, qwen-tts | Intel Arc, Intel iGPUs |
 | **Apple Metal** | llama.cpp, whisper, diffusers, MLX, MLX-VLM, bark-cpp | Apple M1/M2/M3+ |
 | **Vulkan** | llama.cpp, whisper, stablediffusion | Cross-platform GPUs |
 | **NVIDIA Jetson (CUDA 12)** | llama.cpp, whisper, stablediffusion, diffusers, rfdetr | ARM64 embedded AI (AGX Orin, etc.) |
--- a/backend/Dockerfile.golang
+++ b/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
--- a/backend/Dockerfile.llama-cpp
+++ b/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


--- a/backend/Dockerfile.python
+++ b/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}/ /
--- a/backend/README.md
+++ b/backend/README.md
@@ -65,7 +65,7 @@ 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 +132,7 @@ 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 +209,4 @@ 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
--- a/backend/backend.proto
+++ b/backend/backend.proto
@@ -32,6 +32,8 @@ service Backend {
  rpc GetMetrics(MetricsRequest) returns (MetricsResponse);

  rpc VAD(VADRequest) returns (VADResponse) {}
+
+  rpc ModelMetadata(ModelOptions) returns (ModelMetadataResponse) {}
 }

 // Define the empty request
@@ -410,3 +412,8 @@ 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)
+}
--- a/backend/cpp/llama-cpp/CMakeLists.txt
+++ b/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()
--- a/backend/cpp/llama-cpp/Makefile
+++ b/backend/cpp/llama-cpp/Makefile
@@ -1,5 +1,5 @@

-LLAMA_VERSION?=e57f52334b2e8436a94f7e332462dfc63a08f995
+LLAMA_VERSION?=a5eaa1d6a3732bc0f460b02b61c95680bba5a012
 LLAMA_REPO?=https://github.com/ggerganov/llama.cpp

 CMAKE_ARGS?=
@@ -7,7 +7,7 @@ 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
@@ -107,39 +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})
-ifeq ($(OS),Darwin)
-	CMAKE_ARGS="$(CMAKE_ARGS) -DGGML_AVX=on -DGGML_AVX2=off -DGGML_AVX512=off -DGGML_FMA=off -DGGML_F16C=off" $(MAKE) VARIANT="llama-cpp-avx-build" build-llama-cpp-grpc-server
-else ifeq ($(ARCH),$(filter $(ARCH),aarch64 arm64))
-	CMAKE_ARGS="$(CMAKE_ARGS) -DGGML_AVX=on -DGGML_AVX2=off -DGGML_AVX512=off -DGGML_FMA=off -DGGML_F16C=off" $(MAKE) VARIANT="llama-cpp-avx-build" build-llama-cpp-grpc-server
-else
-	CMAKE_ARGS="$(CMAKE_ARGS) -DGGML_AVX=on -DGGML_AVX2=off -DGGML_AVX512=off -DGGML_FMA=off -DGGML_F16C=off -DCMAKE_C_FLAGS=-mno-bmi2 -DCMAKE_CXX_FLAGS=-mno-bmi2" $(MAKE) VARIANT="llama-cpp-avx-build" build-llama-cpp-grpc-server
-endif
+	CMAKE_ARGS="$(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})
-ifeq ($(OS),Darwin)
-	CMAKE_ARGS="$(CMAKE_ARGS) -DGGML_AVX=off -DGGML_AVX2=off -DGGML_AVX512=off -DGGML_FMA=off -DGGML_F16C=off" $(MAKE) VARIANT="llama-cpp-fallback-build" build-llama-cpp-grpc-server
-else ifeq ($(ARCH),$(filter $(ARCH),aarch64 arm64))
-	CMAKE_ARGS="$(CMAKE_ARGS) -DGGML_AVX=off -DGGML_AVX2=off -DGGML_AVX512=off -DGGML_FMA=off -DGGML_F16C=off" $(MAKE) VARIANT="llama-cpp-fallback-build" build-llama-cpp-grpc-server
-else
-	CMAKE_ARGS="$(CMAKE_ARGS) -DGGML_AVX=off -DGGML_AVX2=off -DGGML_AVX512=off -DGGML_FMA=off -DGGML_F16C=off -DCMAKE_C_FLAGS='-mno-bmi -mno-bmi2' -DCMAKE_CXX_FLAGS='-mno-bmi -mno-bmi2'" $(MAKE) VARIANT="llama-cpp-fallback-build" build-llama-cpp-grpc-server
-endif
+	CMAKE_ARGS="$(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})
-ifeq ($(OS),Darwin)
-	CMAKE_ARGS="$(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" $(MAKE) VARIANT="llama-cpp-grpc-build" build-llama-cpp-grpc-server
-else ifeq ($(ARCH),$(filter $(ARCH),aarch64 arm64))
-	CMAKE_ARGS="$(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" $(MAKE) VARIANT="llama-cpp-grpc-build" build-llama-cpp-grpc-server
-else
-	CMAKE_ARGS="$(CMAKE_ARGS) -DGGML_RPC=ON -DGGML_AVX=off -DGGML_AVX2=off -DGGML_AVX512=off -DGGML_FMA=off -DGGML_F16C=off -DCMAKE_C_FLAGS='-mno-bmi -mno-bmi2' -DCMAKE_CXX_FLAGS='-mno-bmi -mno-bmi2'" TARGET="--target grpc-server --target rpc-server" $(MAKE) VARIANT="llama-cpp-grpc-build" build-llama-cpp-grpc-server
-endif
+	CMAKE_ARGS="$(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
--- a/backend/cpp/llama-cpp/grpc-server.cpp
+++ b/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
@@ -293,8 +294,6 @@ json parse_options(bool streaming, const backend::PredictOptions* predict, const
    return data;
 }

-// Sanitize tools JSON to remove null values from tool.parameters.properties
-// This prevents Jinja template errors when processing tools with malformed parameter schemas

 const std::vector<ggml_type> kv_cache_types = {
    GGML_TYPE_F32,
@@ -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.has_speculative() && !params.speculative.model.path.empty()) {
+                error_msg += " (with draft model: " + params.speculative.model.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;
    }
 };
--- a/backend/cpp/llama-cpp/package.sh
+++ b/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 $CURDIR/package/lib/
--- a/backend/go/stablediffusion-ggml/Makefile
+++ b/backend/go/stablediffusion-ggml/Makefile
@@ -8,7 +8,7 @@ JOBS?=$(shell nproc --ignore=1)

 # stablediffusion.cpp (ggml)
 STABLEDIFFUSION_GGML_REPO?=https://github.com/leejet/stable-diffusion.cpp
-STABLEDIFFUSION_GGML_VERSION?=4ff2c8c74bd17c2cfffe3a01be77743fb3efba2f
+STABLEDIFFUSION_GGML_VERSION?=5e4579c11d0678f9765463582d024e58270faa9c

 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)
--- a/backend/go/stablediffusion-ggml/package.sh
+++ b/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 $CURDIR/package/lib/
--- a/backend/go/whisper/Makefile
+++ b/backend/go/whisper/Makefile
@@ -8,7 +8,7 @@ JOBS?=$(shell nproc --ignore=1)

 # whisper.cpp version
 WHISPER_REPO?=https://github.com/ggml-org/whisper.cpp
-WHISPER_CPP_VERSION?=e9898ddfb908ffaa7026c66852a023889a5a7202
+WHISPER_CPP_VERSION?=7aa8818647303b567c3a21fe4220b2681988e220
 SO_TARGET?=libgowhisper.so

 CMAKE_ARGS+=-DBUILD_SHARED_LIBS=OFF
--- a/backend/go/whisper/package.sh
+++ b/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/
--- a/backend/index.yaml
+++ b/backend/index.yaml
@@ -275,6 +275,24 @@
    amd: "rocm-faster-whisper"
    nvidia-cuda-13: "cuda13-faster-whisper"
    nvidia-cuda-12: "cuda12-faster-whisper"
+- &moonshine
+  description: |
+    Moonshine is a fast, accurate, and efficient speech-to-text transcription model using ONNX Runtime.
+    It provides real-time transcription capabilities with support for multiple model sizes and GPU acceleration.
+  urls:
+    - https://github.com/moonshine-ai/moonshine
+  tags:
+    - speech-to-text
+    - transcription
+    - ONNX
+  license: MIT
+  name: "moonshine"
+  alias: "moonshine"
+  capabilities:
+    nvidia: "cuda12-moonshine"
+    default: "cpu-moonshine"
+    nvidia-cuda-13: "cuda13-moonshine"
+    nvidia-cuda-12: "cuda12-moonshine"
 - &kokoro
  icon: https://avatars.githubusercontent.com/u/166769057?v=4
  description: |
@@ -410,6 +428,50 @@
    nvidia-l4t-cuda-12: "nvidia-l4t-vibevoice"
    nvidia-l4t-cuda-13: "cuda13-nvidia-l4t-arm64-vibevoice"
  icon: https://avatars.githubusercontent.com/u/6154722?s=200&v=4
+- &qwen-tts
+  urls:
+    - https://github.com/QwenLM/Qwen3-TTS
+  description: |
+    Qwen3-TTS is a high-quality text-to-speech model supporting custom voice, voice design, and voice cloning.
+  tags:
+    - text-to-speech
+    - TTS
+  license: apache-2.0
+  name: "qwen-tts"
+  alias: "qwen-tts"
+  capabilities:
+    nvidia: "cuda12-qwen-tts"
+    intel: "intel-qwen-tts"
+    amd: "rocm-qwen-tts"
+    nvidia-l4t: "nvidia-l4t-qwen-tts"
+    default: "cpu-qwen-tts"
+    nvidia-cuda-13: "cuda13-qwen-tts"
+    nvidia-cuda-12: "cuda12-qwen-tts"
+    nvidia-l4t-cuda-12: "nvidia-l4t-qwen-tts"
+    nvidia-l4t-cuda-13: "cuda13-nvidia-l4t-arm64-qwen-tts"
+  icon: https://avatars.githubusercontent.com/u/6154722?s=200&v=4
+- &pocket-tts
+  urls:
+    - https://github.com/kyutai-labs/pocket-tts
+  description: |
+    Pocket TTS is a lightweight text-to-speech model designed to run efficiently on CPUs.
+  tags:
+    - text-to-speech
+    - TTS
+  license: mit
+  name: "pocket-tts"
+  alias: "pocket-tts"
+  capabilities:
+    nvidia: "cuda12-pocket-tts"
+    intel: "intel-pocket-tts"
+    amd: "rocm-pocket-tts"
+    nvidia-l4t: "nvidia-l4t-pocket-tts"
+    default: "cpu-pocket-tts"
+    nvidia-cuda-13: "cuda13-pocket-tts"
+    nvidia-cuda-12: "cuda12-pocket-tts"
+    nvidia-l4t-cuda-12: "nvidia-l4t-pocket-tts"
+    nvidia-l4t-cuda-13: "cuda13-nvidia-l4t-arm64-pocket-tts"
+  icon: https://avatars.githubusercontent.com/u/6154722?s=200&v=4
 - &piper
  name: "piper"
  uri: "quay.io/go-skynet/local-ai-backends:latest-piper"
@@ -497,18 +559,14 @@
    default: "cpu-neutts"
    nvidia: "cuda12-neutts"
    amd: "rocm-neutts"
-    nvidia-l4t: "nvidia-l4t-neutts"
    nvidia-cuda-12: "cuda12-neutts"
-    nvidia-l4t-cuda-12: "nvidia-l4t-arm64-neutts"
 - !!merge <<: *neutts
  name: "neutts-development"
  capabilities:
    default: "cpu-neutts-development"
    nvidia: "cuda12-neutts-development"
    amd: "rocm-neutts-development"
-    nvidia-l4t: "nvidia-l4t-neutts-development"
    nvidia-cuda-12: "cuda12-neutts-development"
-    nvidia-l4t-cuda-12: "nvidia-l4t-arm64-neutts-development"
 - !!merge <<: *llamacpp
  name: "llama-cpp-development"
  capabilities:
@@ -538,11 +596,6 @@
  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-rocm-hipblas-neutts"
  mirrors:
    - localai/localai-backends:latest-gpu-rocm-hipblas-neutts
- !!merge <<: *neutts
-  name: "nvidia-l4t-arm64-neutts"
-  uri: "quay.io/go-skynet/local-ai-backends:latest-nvidia-l4t-arm64-neutts"
-  mirrors:
-    - localai/localai-backends:latest-nvidia-l4t-arm64-neutts
 - !!merge <<: *neutts
  name: "cpu-neutts-development"
  uri: "quay.io/go-skynet/local-ai-backends:master-cpu-neutts"
@@ -558,11 +611,6 @@
  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-rocm-hipblas-neutts"
  mirrors:
    - localai/localai-backends:master-gpu-rocm-hipblas-neutts
- !!merge <<: *neutts
-  name: "nvidia-l4t-arm64-neutts-development"
-  uri: "quay.io/go-skynet/local-ai-backends:master-nvidia-l4t-arm64-neutts"
-  mirrors:
-    - localai/localai-backends:master-nvidia-l4t-arm64-neutts
 - !!merge <<: *mlx
  name: "mlx-development"
  uri: "quay.io/go-skynet/local-ai-backends:master-metal-darwin-arm64-mlx"
@@ -634,11 +682,6 @@
  uri: "quay.io/go-skynet/local-ai-backends:master-cpu-llama-cpp"
  mirrors:
    - localai/localai-backends:master-cpu-llama-cpp
- !!merge <<: *llamacpp
-  name: "cuda11-llama-cpp"
-  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-11-llama-cpp"
-  mirrors:
-    - localai/localai-backends:latest-gpu-nvidia-cuda-11-llama-cpp
 - !!merge <<: *llamacpp
  name: "cuda12-llama-cpp"
  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-12-llama-cpp"
@@ -679,11 +722,6 @@
  uri: "quay.io/go-skynet/local-ai-backends:master-metal-darwin-arm64-llama-cpp"
  mirrors:
    - localai/localai-backends:master-metal-darwin-arm64-llama-cpp
- !!merge <<: *llamacpp
-  name: "cuda11-llama-cpp-development"
-  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-11-llama-cpp"
-  mirrors:
-    - localai/localai-backends:master-gpu-nvidia-cuda-11-llama-cpp
 - !!merge <<: *llamacpp
  name: "cuda12-llama-cpp-development"
  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-12-llama-cpp"
@@ -755,11 +793,6 @@
  uri: "quay.io/go-skynet/local-ai-backends:master-cpu-whisper"
  mirrors:
    - localai/localai-backends:master-cpu-whisper
- !!merge <<: *whispercpp
-  name: "cuda11-whisper"
-  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-11-whisper"
-  mirrors:
-    - localai/localai-backends:latest-gpu-nvidia-cuda-11-whisper
 - !!merge <<: *whispercpp
  name: "cuda12-whisper"
  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-12-whisper"
@@ -800,11 +833,6 @@
  uri: "quay.io/go-skynet/local-ai-backends:master-metal-darwin-arm64-whisper"
  mirrors:
    - localai/localai-backends:master-metal-darwin-arm64-whisper
- !!merge <<: *whispercpp
-  name: "cuda11-whisper-development"
-  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-11-whisper"
-  mirrors:
-    - localai/localai-backends:master-gpu-nvidia-cuda-11-whisper
 - !!merge <<: *whispercpp
  name: "cuda12-whisper-development"
  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-12-whisper"
@@ -879,11 +907,6 @@
  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-intel-sycl-f16-stablediffusion-ggml"
  mirrors:
    - localai/localai-backends:latest-gpu-intel-sycl-f16-stablediffusion-ggml
- !!merge <<: *stablediffusionggml
-  name: "cuda11-stablediffusion-ggml"
-  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-11-stablediffusion-ggml"
-  mirrors:
-    - localai/localai-backends:latest-gpu-nvidia-cuda-11-stablediffusion-ggml
 - !!merge <<: *stablediffusionggml
  name: "cuda12-stablediffusion-ggml-development"
  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-12-stablediffusion-ggml"
@@ -899,11 +922,6 @@
  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-intel-sycl-f16-stablediffusion-ggml"
  mirrors:
    - localai/localai-backends:master-gpu-intel-sycl-f16-stablediffusion-ggml
- !!merge <<: *stablediffusionggml
-  name: "cuda11-stablediffusion-ggml-development"
-  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-11-stablediffusion-ggml"
-  mirrors:
-    - localai/localai-backends:master-gpu-nvidia-cuda-11-stablediffusion-ggml
 - !!merge <<: *stablediffusionggml
  name: "nvidia-l4t-arm64-stablediffusion-ggml-development"
  uri: "quay.io/go-skynet/local-ai-backends:master-nvidia-l4t-arm64-stablediffusion-ggml"
@@ -1054,11 +1072,6 @@
    intel: "intel-rerankers-development"
    amd: "rocm-rerankers-development"
    nvidia-cuda-13: "cuda13-rerankers-development"
- !!merge <<: *rerankers
-  name: "cuda11-rerankers"
-  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-11-rerankers"
-  mirrors:
-    - localai/localai-backends:latest-gpu-nvidia-cuda-11-rerankers
 - !!merge <<: *rerankers
  name: "cuda12-rerankers"
  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-12-rerankers"
@@ -1074,11 +1087,6 @@
  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-rocm-hipblas-rerankers"
  mirrors:
    - localai/localai-backends:latest-gpu-rocm-hipblas-rerankers
- !!merge <<: *rerankers
-  name: "cuda11-rerankers-development"
-  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-11-rerankers"
-  mirrors:
-    - localai/localai-backends:master-gpu-nvidia-cuda-11-rerankers
 - !!merge <<: *rerankers
  name: "cuda12-rerankers-development"
  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-12-rerankers"
@@ -1127,16 +1135,6 @@
  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-intel-transformers"
  mirrors:
    - localai/localai-backends:latest-gpu-intel-transformers
- !!merge <<: *transformers
-  name: "cuda11-transformers-development"
-  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-11-transformers"
-  mirrors:
-    - localai/localai-backends:master-gpu-nvidia-cuda-11-transformers
- !!merge <<: *transformers
-  name: "cuda11-transformers"
-  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-11-transformers"
-  mirrors:
-    - localai/localai-backends:latest-gpu-nvidia-cuda-11-transformers
 - !!merge <<: *transformers
  name: "cuda12-transformers-development"
  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-12-transformers"
@@ -1213,21 +1211,11 @@
  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-rocm-hipblas-diffusers"
  mirrors:
    - localai/localai-backends:latest-gpu-rocm-hipblas-diffusers
- !!merge <<: *diffusers
-  name: "cuda11-diffusers"
-  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-11-diffusers"
-  mirrors:
-    - localai/localai-backends:latest-gpu-nvidia-cuda-11-diffusers
 - !!merge <<: *diffusers
  name: "intel-diffusers"
  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-intel-diffusers"
  mirrors:
    - localai/localai-backends:latest-gpu-intel-diffusers
- !!merge <<: *diffusers
-  name: "cuda11-diffusers-development"
-  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-11-diffusers"
-  mirrors:
-    - localai/localai-backends:master-gpu-nvidia-cuda-11-diffusers
 - !!merge <<: *diffusers
  name: "cuda12-diffusers-development"
  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-12-diffusers"
@@ -1269,21 +1257,11 @@
  capabilities:
    nvidia: "cuda12-exllama2-development"
    intel: "intel-exllama2-development"
- !!merge <<: *exllama2
-  name: "cuda11-exllama2"
-  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-11-exllama2"
-  mirrors:
-    - localai/localai-backends:latest-gpu-nvidia-cuda-11-exllama2
 - !!merge <<: *exllama2
  name: "cuda12-exllama2"
  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-12-exllama2"
  mirrors:
    - localai/localai-backends:latest-gpu-nvidia-cuda-12-exllama2
- !!merge <<: *exllama2
-  name: "cuda11-exllama2-development"
-  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-11-exllama2"
-  mirrors:
-    - localai/localai-backends:master-gpu-nvidia-cuda-11-exllama2
 - !!merge <<: *exllama2
  name: "cuda12-exllama2-development"
  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-12-exllama2"
@@ -1297,11 +1275,6 @@
    intel: "intel-kokoro-development"
    amd: "rocm-kokoro-development"
    nvidia-l4t: "nvidia-l4t-kokoro-development"
- !!merge <<: *kokoro
-  name: "cuda11-kokoro-development"
-  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-11-kokoro"
-  mirrors:
-    - localai/localai-backends:master-gpu-nvidia-cuda-11-kokoro
 - !!merge <<: *kokoro
  name: "cuda12-kokoro-development"
  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-12-kokoro"
@@ -1332,11 +1305,6 @@
  uri: "quay.io/go-skynet/local-ai-backends:master-nvidia-l4t-kokoro"
  mirrors:
    - localai/localai-backends:master-nvidia-l4t-kokoro
- !!merge <<: *kokoro
-  name: "cuda11-kokoro"
-  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-11-kokoro"
-  mirrors:
-    - localai/localai-backends:latest-gpu-nvidia-cuda-11-kokoro
 - !!merge <<: *kokoro
  name: "cuda12-kokoro"
  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-12-kokoro"
@@ -1365,11 +1333,6 @@
    intel: "intel-faster-whisper-development"
    amd: "rocm-faster-whisper-development"
    nvidia-cuda-13: "cuda13-faster-whisper-development"
- !!merge <<: *faster-whisper
-  name: "cuda11-faster-whisper"
-  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-11-faster-whisper"
-  mirrors:
-    - localai/localai-backends:latest-gpu-nvidia-cuda-11-faster-whisper
 - !!merge <<: *faster-whisper
  name: "cuda12-faster-whisper-development"
  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-12-faster-whisper"
@@ -1400,6 +1363,44 @@
  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-13-faster-whisper"
  mirrors:
    - localai/localai-backends:master-gpu-nvidia-cuda-13-faster-whisper
+## moonshine
+- !!merge <<: *moonshine
+  name: "moonshine-development"
+  capabilities:
+    nvidia: "cuda12-moonshine-development"
+    default: "cpu-moonshine-development"
+    nvidia-cuda-13: "cuda13-moonshine-development"
+    nvidia-cuda-12: "cuda12-moonshine-development"
+- !!merge <<: *moonshine
+  name: "cpu-moonshine"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-cpu-moonshine"
+  mirrors:
+    - localai/localai-backends:latest-cpu-moonshine
+- !!merge <<: *moonshine
+  name: "cpu-moonshine-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-cpu-moonshine"
+  mirrors:
+    - localai/localai-backends:master-cpu-moonshine
+- !!merge <<: *moonshine
+  name: "cuda12-moonshine"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-12-moonshine"
+  mirrors:
+    - localai/localai-backends:latest-gpu-nvidia-cuda-12-moonshine
+- !!merge <<: *moonshine
+  name: "cuda12-moonshine-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-12-moonshine"
+  mirrors:
+    - localai/localai-backends:master-gpu-nvidia-cuda-12-moonshine
+- !!merge <<: *moonshine
+  name: "cuda13-moonshine"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-13-moonshine"
+  mirrors:
+    - localai/localai-backends:latest-gpu-nvidia-cuda-13-moonshine
+- !!merge <<: *moonshine
+  name: "cuda13-moonshine-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-13-moonshine"
+  mirrors:
+    - localai/localai-backends:master-gpu-nvidia-cuda-13-moonshine
 ## coqui

 - !!merge <<: *coqui
@@ -1408,21 +1409,11 @@
    nvidia: "cuda12-coqui-development"
    intel: "intel-coqui-development"
    amd: "rocm-coqui-development"
- !!merge <<: *coqui
-  name: "cuda11-coqui"
-  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-11-coqui"
-  mirrors:
-    - localai/localai-backends:latest-gpu-nvidia-cuda-11-coqui
 - !!merge <<: *coqui
  name: "cuda12-coqui"
  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-12-coqui"
  mirrors:
    - localai/localai-backends:latest-gpu-nvidia-cuda-12-coqui
- !!merge <<: *coqui
-  name: "cuda11-coqui-development"
-  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-11-coqui"
-  mirrors:
-    - localai/localai-backends:master-gpu-nvidia-cuda-11-coqui
 - !!merge <<: *coqui
  name: "cuda12-coqui-development"
  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-12-coqui"
@@ -1455,16 +1446,6 @@
    nvidia: "cuda12-bark-development"
    intel: "intel-bark-development"
    amd: "rocm-bark-development"
- !!merge <<: *bark
-  name: "cuda11-bark-development"
-  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-11-bark"
-  mirrors:
-    - localai/localai-backends:master-gpu-nvidia-cuda-11-bark
- !!merge <<: *bark
-  name: "cuda11-bark"
-  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-11-bark"
-  mirrors:
-    - localai/localai-backends:latest-gpu-nvidia-cuda-11-bark
 - !!merge <<: *bark
  name: "rocm-bark-development"
  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-rocm-hipblas-bark"
@@ -1546,16 +1527,6 @@
  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-12-chatterbox"
  mirrors:
    - localai/localai-backends:master-gpu-nvidia-cuda-12-chatterbox
- !!merge <<: *chatterbox
-  name: "cuda11-chatterbox"
-  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-11-chatterbox"
-  mirrors:
-    - localai/localai-backends:latest-gpu-nvidia-cuda-11-chatterbox
- !!merge <<: *chatterbox
-  name: "cuda11-chatterbox-development"
-  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-11-chatterbox"
-  mirrors:
-    - localai/localai-backends:master-gpu-nvidia-cuda-11-chatterbox
 - !!merge <<: *chatterbox
  name: "cuda12-chatterbox"
  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-12-chatterbox"
@@ -1664,3 +1635,169 @@
  uri: "quay.io/go-skynet/local-ai-backends:master-nvidia-l4t-cuda-13-arm64-vibevoice"
  mirrors:
    - localai/localai-backends:master-nvidia-l4t-cuda-13-arm64-vibevoice
+## qwen-tts
+- !!merge <<: *qwen-tts
+  name: "qwen-tts-development"
+  capabilities:
+    nvidia: "cuda12-qwen-tts-development"
+    intel: "intel-qwen-tts-development"
+    amd: "rocm-qwen-tts-development"
+    nvidia-l4t: "nvidia-l4t-qwen-tts-development"
+    default: "cpu-qwen-tts-development"
+    nvidia-cuda-13: "cuda13-qwen-tts-development"
+    nvidia-cuda-12: "cuda12-qwen-tts-development"
+    nvidia-l4t-cuda-12: "nvidia-l4t-qwen-tts-development"
+    nvidia-l4t-cuda-13: "cuda13-nvidia-l4t-arm64-qwen-tts-development"
+- !!merge <<: *qwen-tts
+  name: "cpu-qwen-tts"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-cpu-qwen-tts"
+  mirrors:
+    - localai/localai-backends:latest-cpu-qwen-tts
+- !!merge <<: *qwen-tts
+  name: "cpu-qwen-tts-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-cpu-qwen-tts"
+  mirrors:
+    - localai/localai-backends:master-cpu-qwen-tts
+- !!merge <<: *qwen-tts
+  name: "cuda12-qwen-tts"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-12-qwen-tts"
+  mirrors:
+    - localai/localai-backends:latest-gpu-nvidia-cuda-12-qwen-tts
+- !!merge <<: *qwen-tts
+  name: "cuda12-qwen-tts-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-12-qwen-tts"
+  mirrors:
+    - localai/localai-backends:master-gpu-nvidia-cuda-12-qwen-tts
+- !!merge <<: *qwen-tts
+  name: "cuda13-qwen-tts"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-13-qwen-tts"
+  mirrors:
+    - localai/localai-backends:latest-gpu-nvidia-cuda-13-qwen-tts
+- !!merge <<: *qwen-tts
+  name: "cuda13-qwen-tts-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-13-qwen-tts"
+  mirrors:
+    - localai/localai-backends:master-gpu-nvidia-cuda-13-qwen-tts
+- !!merge <<: *qwen-tts
+  name: "intel-qwen-tts"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-intel-qwen-tts"
+  mirrors:
+    - localai/localai-backends:latest-gpu-intel-qwen-tts
+- !!merge <<: *qwen-tts
+  name: "intel-qwen-tts-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-intel-qwen-tts"
+  mirrors:
+    - localai/localai-backends:master-gpu-intel-qwen-tts
+- !!merge <<: *qwen-tts
+  name: "rocm-qwen-tts"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-rocm-hipblas-qwen-tts"
+  mirrors:
+    - localai/localai-backends:latest-gpu-rocm-hipblas-qwen-tts
+- !!merge <<: *qwen-tts
+  name: "rocm-qwen-tts-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-rocm-hipblas-qwen-tts"
+  mirrors:
+    - localai/localai-backends:master-gpu-rocm-hipblas-qwen-tts
+- !!merge <<: *qwen-tts
+  name: "nvidia-l4t-qwen-tts"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-nvidia-l4t-qwen-tts"
+  mirrors:
+    - localai/localai-backends:latest-nvidia-l4t-qwen-tts
+- !!merge <<: *qwen-tts
+  name: "nvidia-l4t-qwen-tts-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-nvidia-l4t-qwen-tts"
+  mirrors:
+    - localai/localai-backends:master-nvidia-l4t-qwen-tts
+- !!merge <<: *qwen-tts
+  name: "cuda13-nvidia-l4t-arm64-qwen-tts"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-nvidia-l4t-cuda-13-arm64-qwen-tts"
+  mirrors:
+    - localai/localai-backends:latest-nvidia-l4t-cuda-13-arm64-qwen-tts
+- !!merge <<: *qwen-tts
+  name: "cuda13-nvidia-l4t-arm64-qwen-tts-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-nvidia-l4t-cuda-13-arm64-qwen-tts"
+  mirrors:
+    - localai/localai-backends:master-nvidia-l4t-cuda-13-arm64-qwen-tts
+## pocket-tts
+- !!merge <<: *pocket-tts
+  name: "pocket-tts-development"
+  capabilities:
+    nvidia: "cuda12-pocket-tts-development"
+    intel: "intel-pocket-tts-development"
+    amd: "rocm-pocket-tts-development"
+    nvidia-l4t: "nvidia-l4t-pocket-tts-development"
+    default: "cpu-pocket-tts-development"
+    nvidia-cuda-13: "cuda13-pocket-tts-development"
+    nvidia-cuda-12: "cuda12-pocket-tts-development"
+    nvidia-l4t-cuda-12: "nvidia-l4t-pocket-tts-development"
+    nvidia-l4t-cuda-13: "cuda13-nvidia-l4t-arm64-pocket-tts-development"
+- !!merge <<: *pocket-tts
+  name: "cpu-pocket-tts"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-cpu-pocket-tts"
+  mirrors:
+    - localai/localai-backends:latest-cpu-pocket-tts
+- !!merge <<: *pocket-tts
+  name: "cpu-pocket-tts-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-cpu-pocket-tts"
+  mirrors:
+    - localai/localai-backends:master-cpu-pocket-tts
+- !!merge <<: *pocket-tts
+  name: "cuda12-pocket-tts"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-12-pocket-tts"
+  mirrors:
+    - localai/localai-backends:latest-gpu-nvidia-cuda-12-pocket-tts
+- !!merge <<: *pocket-tts
+  name: "cuda12-pocket-tts-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-12-pocket-tts"
+  mirrors:
+    - localai/localai-backends:master-gpu-nvidia-cuda-12-pocket-tts
+- !!merge <<: *pocket-tts
+  name: "cuda13-pocket-tts"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-13-pocket-tts"
+  mirrors:
+    - localai/localai-backends:latest-gpu-nvidia-cuda-13-pocket-tts
+- !!merge <<: *pocket-tts
+  name: "cuda13-pocket-tts-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-13-pocket-tts"
+  mirrors:
+    - localai/localai-backends:master-gpu-nvidia-cuda-13-pocket-tts
+- !!merge <<: *pocket-tts
+  name: "intel-pocket-tts"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-intel-pocket-tts"
+  mirrors:
+    - localai/localai-backends:latest-gpu-intel-pocket-tts
+- !!merge <<: *pocket-tts
+  name: "intel-pocket-tts-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-intel-pocket-tts"
+  mirrors:
+    - localai/localai-backends:master-gpu-intel-pocket-tts
+- !!merge <<: *pocket-tts
+  name: "rocm-pocket-tts"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-rocm-hipblas-pocket-tts"
+  mirrors:
+    - localai/localai-backends:latest-gpu-rocm-hipblas-pocket-tts
+- !!merge <<: *pocket-tts
+  name: "rocm-pocket-tts-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-rocm-hipblas-pocket-tts"
+  mirrors:
+    - localai/localai-backends:master-gpu-rocm-hipblas-pocket-tts
+- !!merge <<: *pocket-tts
+  name: "nvidia-l4t-pocket-tts"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-nvidia-l4t-pocket-tts"
+  mirrors:
+    - localai/localai-backends:latest-nvidia-l4t-pocket-tts
+- !!merge <<: *pocket-tts
+  name: "nvidia-l4t-pocket-tts-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-nvidia-l4t-pocket-tts"
+  mirrors:
+    - localai/localai-backends:master-nvidia-l4t-pocket-tts
+- !!merge <<: *pocket-tts
+  name: "cuda13-nvidia-l4t-arm64-pocket-tts"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-nvidia-l4t-cuda-13-arm64-pocket-tts"
+  mirrors:
+    - localai/localai-backends:latest-nvidia-l4t-cuda-13-arm64-pocket-tts
+- !!merge <<: *pocket-tts
+  name: "cuda13-nvidia-l4t-arm64-pocket-tts-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-nvidia-l4t-cuda-13-arm64-pocket-tts"
+  mirrors:
+    - localai/localai-backends:master-nvidia-l4t-cuda-13-arm64-pocket-tts
--- a/backend/python/README.md
+++ b/backend/python/README.md
@@ -85,7 +85,7 @@ 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 +95,8 @@ 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

--- a/backend/python/bark/requirements-cublas11.txt
+++ b/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
--- a/backend/python/bark/requirements-hipblas.txt
+++ b/backend/python/bark/requirements-hipblas.txt
@@ -1,5 +1,5 @@
--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
 accelerate
--- a/backend/python/bark/requirements-intel.txt
+++ b/backend/python/bark/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==2.3.1+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
+torchaudio
 optimum[openvino]
 setuptools
 transformers
--- a/backend/python/chatterbox/install.sh
+++ b/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
--- a/backend/python/chatterbox/requirements-cublas11.txt
+++ b/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
--- a/backend/python/chatterbox/requirements-hipblas.txt
+++ b/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
--- a/backend/python/chatterbox/requirements-install.txt
+++ b/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
--- a/backend/python/chatterbox/requirements-intel.txt
+++ b/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
--- a/backend/python/common/libbackend.sh
+++ b/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
--- a/backend/python/common/template/requirements-hipblas.txt
+++ b/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
--- a/backend/python/common/template/requirements-intel.txt
+++ b/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]
--- a/backend/python/coqui/requirements-cublas11.txt
+++ b/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
--- a/backend/python/coqui/requirements-hipblas.txt
+++ b/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
--- a/backend/python/coqui/requirements-intel.txt
+++ b/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
--- a/backend/python/diffusers/backend.py
+++ b/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:
--- a/backend/python/diffusers/install.sh
+++ b/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"
--- a/backend/python/diffusers/requirements-cublas11.txt
+++ b/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
--- a/backend/python/diffusers/requirements-hipblas.txt
+++ b/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
--- a/backend/python/diffusers/requirements-intel.txt
+++ b/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
--- a/backend/python/diffusers/requirements.txt
+++ b/backend/python/diffusers/requirements.txt
@@ -3,3 +3,4 @@ grpcio==1.76.0
 pillow
 protobuf
 certifi
+av
--- a/backend/python/exllama2/requirements-cublas11.txt
+++ b/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
--- a/backend/python/faster-whisper/requirements-cublas11.txt
+++ b/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
--- a/backend/python/faster-whisper/requirements-hipblas.txt
+++ b/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
--- a/backend/python/faster-whisper/requirements-intel.txt
+++ b/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
--- a/backend/python/kokoro/install.sh
+++ b/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
--- a/backend/python/kokoro/requirements-cublas11.txt
+++ b/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
--- a/backend/python/kokoro/requirements-hipblas.txt
+++ b/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
--- a/backend/python/kokoro/requirements-intel.txt
+++ b/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
--- a/backend/python/moonshine/Makefile
+++ b/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
--- a/backend/python/moonshine/backend.py
+++ b/backend/python/moonshine/backend.py
@@ -0,0 +1,113 @@
+#!/usr/bin/env python3
+"""
+This is an extra gRPC server of LocalAI for Moonshine transcription
+"""
+from concurrent import futures
+import time
+import argparse
+import signal
+import sys
+import os
+import backend_pb2
+import backend_pb2_grpc
+import moonshine_onnx
+
+import grpc
+
+
+_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):
+        try:
+            print("Preparing models, please wait", file=sys.stderr)
+            # 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)=}")
+        return backend_pb2.Result(message="Model loaded successfully", success=True)
+
+    def AudioTranscription(self, request, context):
+        resultSegments = []
+        text = ""
+        try:
+            # 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:
+                # 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:
+            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),
+        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)
+
--- a/backend/python/moonshine/install.sh
+++ b/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 $backend_dir/../common/libbackend.sh
+fi
+
+installRequirements
+
--- a/backend/python/moonshine/protogen.sh
+++ b/backend/python/moonshine/protogen.sh
@@ -0,0 +1,12 @@
+#!/bin/bash
+set -e
+
+backend_dir=$(dirname $0)
+if [ -d $backend_dir/common ]; then
+    source $backend_dir/common/libbackend.sh
+else
+    source $backend_dir/../common/libbackend.sh
+fi
+
+python3 -m grpc_tools.protoc -I../.. -I./ --python_out=. --grpc_python_out=. backend.proto
+
--- a/backend/python/moonshine/requirements.txt
+++ b/backend/python/moonshine/requirements.txt
@@ -0,0 +1,4 @@
+grpcio==1.71.0
+protobuf
+grpcio-tools
+useful-moonshine-onnx@git+https://git@github.com/moonshine-ai/moonshine.git#subdirectory=moonshine-onnx
--- a/backend/python/moonshine/run.sh
+++ b/backend/python/moonshine/run.sh
@@ -0,0 +1,10 @@
+#!/bin/bash
+backend_dir=$(dirname $0)
+if [ -d $backend_dir/common ]; then
+    source $backend_dir/common/libbackend.sh
+else
+    source $backend_dir/../common/libbackend.sh
+fi
+
+startBackend $@
+
--- a/backend/python/moonshine/test.py
+++ b/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)
+
--- a/backend/python/moonshine/test.sh
+++ b/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
+
--- a/backend/python/neutts/install.sh
+++ b/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 ./
--- a/backend/python/neutts/requirements-hipblas.txt
+++ b/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
--- a/backend/python/pocket-tts/Makefile
+++ b/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__
--- a/backend/python/pocket-tts/backend.py
+++ b/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)
--- a/backend/python/pocket-tts/install.sh
+++ b/backend/python/pocket-tts/install.sh
@@ -0,0 +1,30 @@
+#!/bin/bash
+set -e
+
+backend_dir=$(dirname $0)
+if [ -d $backend_dir/common ]; then
+    source $backend_dir/common/libbackend.sh
+else
+    source $backend_dir/../common/libbackend.sh
+fi
+
+# This is here because the Intel pip index is broken and returns 200 status codes for every package name, it just doesn't return any package links.
+# This makes uv think that the package exists in the Intel pip index, and by default it stops looking at other pip indexes once it finds a match.
+# We need uv to continue falling through to the pypi default index to find optimum[openvino] in the pypi index
+# the --upgrade actually allows us to *downgrade* torch to the version provided in the Intel pip index
+if [ "x${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
--- a/backend/python/pocket-tts/protogen.sh
+++ b/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
--- a/backend/python/pocket-tts/requirements-cpu.txt
+++ b/backend/python/pocket-tts/requirements-cpu.txt
@@ -0,0 +1,4 @@
+--extra-index-url https://download.pytorch.org/whl/cpu
+pocket-tts
+scipy
+torch
--- a/backend/python/pocket-tts/requirements-cublas12.txt
+++ b/backend/python/pocket-tts/requirements-cublas12.txt
@@ -0,0 +1,4 @@
+--extra-index-url https://download.pytorch.org/whl/cu121
+pocket-tts
+scipy
+torch
--- a/backend/python/pocket-tts/requirements-cublas13.txt
+++ b/backend/python/pocket-tts/requirements-cublas13.txt
@@ -0,0 +1,4 @@
+--extra-index-url https://download.pytorch.org/whl/cu130
+pocket-tts
+scipy
+torch
--- a/backend/python/pocket-tts/requirements-hipblas.txt
+++ b/backend/python/pocket-tts/requirements-hipblas.txt
@@ -0,0 +1,4 @@
+--extra-index-url https://download.pytorch.org/whl/rocm6.3
+pocket-tts
+scipy
+torch==2.7.1+rocm6.3
--- a/backend/python/pocket-tts/requirements-intel.txt
+++ b/backend/python/pocket-tts/requirements-intel.txt
@@ -0,0 +1,4 @@
+--extra-index-url https://download.pytorch.org/whl/xpu
+pocket-tts
+scipy
+torch
--- a/backend/python/pocket-tts/requirements-l4t12.txt
+++ b/backend/python/pocket-tts/requirements-l4t12.txt
@@ -0,0 +1,4 @@
+--extra-index-url https://pypi.jetson-ai-lab.io/jp6/cu129/
+pocket-tts
+scipy
+torch
--- a/backend/python/pocket-tts/requirements-l4t13.txt
+++ b/backend/python/pocket-tts/requirements-l4t13.txt
@@ -0,0 +1,4 @@
+--extra-index-url https://download.pytorch.org/whl/cu130
+pocket-tts
+scipy
+torch
--- a/backend/python/pocket-tts/requirements-mps.txt
+++ b/backend/python/pocket-tts/requirements-mps.txt
@@ -0,0 +1,4 @@
+pocket-tts
+scipy
+torch==2.7.1
+torchvision==0.22.1
--- a/backend/python/pocket-tts/requirements.txt
+++ b/backend/python/pocket-tts/requirements.txt
@@ -0,0 +1,4 @@
+grpcio==1.71.0
+protobuf
+certifi
+packaging==24.1
--- a/backend/python/pocket-tts/run.sh
+++ b/backend/python/pocket-tts/run.sh
@@ -0,0 +1,9 @@
+#!/bin/bash
+backend_dir=$(dirname $0)
+if [ -d $backend_dir/common ]; then
+    source $backend_dir/common/libbackend.sh
+else
+    source $backend_dir/../common/libbackend.sh
+fi
+
+startBackend $@
--- a/backend/python/pocket-tts/test.py
+++ b/backend/python/pocket-tts/test.py
@@ -0,0 +1,141 @@
+"""
+A test script to test the gRPC service
+"""
+import unittest
+import subprocess
+import time
+import os
+import tempfile
+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(30)
+
+    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())
+                print(response)
+                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_tts_with_hf_voice(self):
+        """
+        This method tests TTS generation with HuggingFace voice URL
+        """
+        try:
+            self.setUp()
+            with grpc.insecure_channel("localhost:50051") as channel:
+                stub = backend_pb2_grpc.BackendStub(channel)
+                # Load model
+                response = stub.LoadModel(backend_pb2.ModelOptions())
+                self.assertTrue(response.success)
+                
+                # Create temporary output file
+                with tempfile.NamedTemporaryFile(suffix='.wav', delete=False) as tmp_file:
+                    output_path = tmp_file.name
+                
+                # Test TTS with HuggingFace voice URL
+                tts_request = backend_pb2.TTSRequest(
+                    text="Hello world, this is a test.",
+                    dst=output_path,
+                    voice="azelma"
+                )
+                tts_response = stub.TTS(tts_request)
+                self.assertTrue(tts_response.success)
+                
+                # Verify output file exists and is not empty
+                self.assertTrue(os.path.exists(output_path))
+                self.assertGreater(os.path.getsize(output_path), 0)
+                
+                # Cleanup
+                os.unlink(output_path)
+        except Exception as err:
+            print(err)
+            self.fail("TTS service failed")
+        finally:
+            self.tearDown()
+
+    def test_tts_with_default_voice(self):
+        """
+        This method tests TTS generation with default voice (via AudioPath in LoadModel)
+        """
+        try:
+            self.setUp()
+            with grpc.insecure_channel("localhost:50051") as channel:
+                stub = backend_pb2_grpc.BackendStub(channel)
+                # Load model with default voice
+                load_request = backend_pb2.ModelOptions(
+                    Options=["default_voice:azelma"]
+                )
+                response = stub.LoadModel(load_request)
+                self.assertTrue(response.success)
+                
+                # Create temporary output file
+                with tempfile.NamedTemporaryFile(suffix='.wav', delete=False) as tmp_file:
+                    output_path = tmp_file.name
+                
+                # Test TTS without specifying voice (should use default)
+                tts_request = backend_pb2.TTSRequest(
+                    text="Hello world, this is a test.",
+                    dst=output_path
+                )
+                tts_response = stub.TTS(tts_request)
+                self.assertTrue(tts_response.success)
+                
+                # Verify output file exists and is not empty
+                self.assertTrue(os.path.exists(output_path))
+                self.assertGreater(os.path.getsize(output_path), 0)
+                
+                # Cleanup
+                os.unlink(output_path)
+        except Exception as err:
+            print(err)
+            self.fail("TTS service with default voice failed")
+        finally:
+            self.tearDown()
--- a/backend/python/pocket-tts/test.sh
+++ b/backend/python/pocket-tts/test.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
+
+runUnittests
--- a/backend/python/qwen-tts/Makefile
+++ b/backend/python/qwen-tts/Makefile
@@ -0,0 +1,23 @@
+.PHONY: qwen-tts
+qwen-tts:
+	bash install.sh
+
+.PHONY: run
+run: qwen-tts
+	@echo "Running qwen-tts..."
+	bash run.sh
+	@echo "qwen-tts run."
+
+.PHONY: test
+test: qwen-tts
+	@echo "Testing qwen-tts..."
+	bash test.sh
+	@echo "qwen-tts tested."
+
+.PHONY: protogen-clean
+protogen-clean:
+	$(RM) backend_pb2_grpc.py backend_pb2.py
+
+.PHONY: clean
+clean: protogen-clean
+	rm -rf venv __pycache__
--- a/backend/python/qwen-tts/backend.py
+++ b/backend/python/qwen-tts/backend.py
@@ -0,0 +1,475 @@
+#!/usr/bin/env python3
+"""
+This is an extra gRPC server of LocalAI for Qwen3-TTS
+"""
+from concurrent import futures
+import time
+import argparse
+import signal
+import sys
+import os
+import copy
+import traceback
+from pathlib import Path
+import backend_pb2
+import backend_pb2_grpc
+import torch
+import soundfile as sf
+from qwen_tts import Qwen3TTSModel
+
+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
+        self._torch_device = torch.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
+
+        # Get model path from request
+        model_path = request.Model
+        if not model_path:
+            model_path = "Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice"
+
+        # Determine model type from model path or options
+        self.model_type = self.options.get("model_type", None)
+        if not self.model_type:
+            if "CustomVoice" in model_path:
+                self.model_type = "CustomVoice"
+            elif "VoiceDesign" in model_path:
+                self.model_type = "VoiceDesign"
+            elif "Base" in model_path or "0.6B" in model_path or "1.7B" in model_path:
+                self.model_type = "Base"  # VoiceClone model
+            else:
+                # Default to CustomVoice
+                self.model_type = "CustomVoice"
+
+        # Cache for voice clone prompts
+        self._voice_clone_cache = {}
+
+        # Store AudioPath, ModelFile, and ModelPath from LoadModel request
+        # These are used later in TTS for VoiceClone mode
+        self.audio_path = request.AudioPath if hasattr(request, 'AudioPath') and request.AudioPath else None
+        self.model_file = request.ModelFile if hasattr(request, 'ModelFile') and request.ModelFile else None
+        self.model_path = request.ModelPath if hasattr(request, 'ModelPath') and request.ModelPath else None
+
+        # Decide dtype & attention implementation
+        if self.device == "mps":
+            load_dtype = torch.float32  # MPS requires float32
+            device_map = None
+            attn_impl_primary = "sdpa"  # flash_attention_2 not supported on MPS
+        elif self.device == "cuda":
+            load_dtype = torch.bfloat16
+            device_map = "cuda"
+            attn_impl_primary = "flash_attention_2"
+        else:  # cpu
+            load_dtype = torch.float32
+            device_map = "cpu"
+            attn_impl_primary = "sdpa"
+
+        print(f"Using device: {self.device}, torch_dtype: {load_dtype}, attn_implementation: {attn_impl_primary}, model_type: {self.model_type}", file=sys.stderr)
+        print(f"Loading model from: {model_path}", file=sys.stderr)
+
+        # Load model with device-specific logic
+        # Common parameters for all devices
+        load_kwargs = {
+            "dtype": load_dtype,
+            "attn_implementation": attn_impl_primary,
+            "trust_remote_code": True,  # Required for qwen-tts models
+        }
+        
+        try:
+            if self.device == "mps":
+                load_kwargs["device_map"] = None  # load then move
+                self.model = Qwen3TTSModel.from_pretrained(model_path, **load_kwargs)
+                self.model.to("mps")
+            elif self.device == "cuda":
+                load_kwargs["device_map"] = device_map
+                self.model = Qwen3TTSModel.from_pretrained(model_path, **load_kwargs)
+            else:  # cpu
+                load_kwargs["device_map"] = device_map
+                self.model = Qwen3TTSModel.from_pretrained(model_path, **load_kwargs)
+        except Exception as e:
+            error_msg = str(e)
+            print(f"[ERROR] Loading model: {type(e).__name__}: {error_msg}", file=sys.stderr)
+            print(traceback.format_exc(), file=sys.stderr)
+            
+            # Check if it's a missing feature extractor/tokenizer error
+            if "speech_tokenizer" in error_msg or "preprocessor_config.json" in error_msg or "feature extractor" in error_msg.lower():
+                print("\n[ERROR] Model files appear to be incomplete. This usually means:", file=sys.stderr)
+                print("  1. The model download was interrupted or incomplete", file=sys.stderr)
+                print("  2. The model cache is corrupted", file=sys.stderr)
+                print("\nTo fix this, try:", file=sys.stderr)
+                print(f"  rm -rf ~/.cache/huggingface/hub/models--Qwen--Qwen3-TTS-*", file=sys.stderr)
+                print("  Then re-run to trigger a fresh download.", file=sys.stderr)
+                print("\nAlternatively, try using a different model variant:", file=sys.stderr)
+                print("  - Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice", file=sys.stderr)
+                print("  - Qwen/Qwen3-TTS-12Hz-1.7B-VoiceDesign", file=sys.stderr)
+                print("  - Qwen/Qwen3-TTS-12Hz-1.7B-Base", file=sys.stderr)
+            
+            if attn_impl_primary == 'flash_attention_2':
+                print("\nTrying to use SDPA instead of flash_attention_2...", file=sys.stderr)
+                load_kwargs["attn_implementation"] = 'sdpa'
+                try:
+                    if self.device == "mps":
+                        load_kwargs["device_map"] = None
+                        self.model = Qwen3TTSModel.from_pretrained(model_path, **load_kwargs)
+                        self.model.to("mps")
+                    else:
+                        load_kwargs["device_map"] = (self.device if self.device in ("cuda", "cpu") else None)
+                        self.model = Qwen3TTSModel.from_pretrained(model_path, **load_kwargs)
+                except Exception as e2:
+                    print(f"[ERROR] Failed to load with SDPA: {type(e2).__name__}: {e2}", file=sys.stderr)
+                    print(traceback.format_exc(), file=sys.stderr)
+                    raise e2
+            else:
+                raise e
+
+        print(f"Model loaded successfully: {model_path}", file=sys.stderr)
+
+        return backend_pb2.Result(message="Model loaded successfully", success=True)
+
+    def _detect_mode(self, request):
+        """Detect which mode to use based on request parameters."""
+        # Priority: VoiceClone > VoiceDesign > CustomVoice
+
+        # model_type explicitly set
+        if self.model_type == "CustomVoice":
+            return "CustomVoice"
+        if self.model_type == "VoiceClone":
+            return "VoiceClone"
+        if self.model_type == "VoiceDesign":
+            return "VoiceDesign"
+
+        # VoiceClone: AudioPath is provided (from LoadModel, stored in self.audio_path)
+        if self.audio_path:
+            return "VoiceClone"
+        
+        # VoiceDesign: instruct option is provided
+        if "instruct" in self.options and self.options["instruct"]:
+            return "VoiceDesign"
+        
+        # Default to CustomVoice
+        return "CustomVoice"
+
+    def _get_ref_audio_path(self, request):
+        """Get reference audio path from stored AudioPath (from LoadModel)."""
+        if not self.audio_path:
+            return None
+        
+        # If absolute path, use as-is
+        if os.path.isabs(self.audio_path):
+            return self.audio_path
+        
+        # Try relative to ModelFile
+        if self.model_file:
+            model_file_base = os.path.dirname(self.model_file)
+            ref_path = os.path.join(model_file_base, self.audio_path)
+            if os.path.exists(ref_path):
+                return ref_path
+        
+        # Try relative to ModelPath
+        if self.model_path:
+            ref_path = os.path.join(self.model_path, self.audio_path)
+            if os.path.exists(ref_path):
+                return ref_path
+        
+        # Return as-is (might be URL or base64)
+        return self.audio_path
+
+    def _get_voice_clone_prompt(self, request, ref_audio, ref_text):
+        """Get or create voice clone prompt, with caching."""
+        cache_key = f"{ref_audio}:{ref_text}"
+        
+        if cache_key not in self._voice_clone_cache:
+            print(f"Creating voice clone prompt from {ref_audio}", file=sys.stderr)
+            try:
+                prompt_items = self.model.create_voice_clone_prompt(
+                    ref_audio=ref_audio,
+                    ref_text=ref_text,
+                    x_vector_only_mode=self.options.get("x_vector_only_mode", False),
+                )
+                self._voice_clone_cache[cache_key] = prompt_items
+            except Exception as e:
+                print(f"Error creating voice clone prompt: {e}", file=sys.stderr)
+                print(traceback.format_exc(), file=sys.stderr)
+                return None
+        
+        return self._voice_clone_cache[cache_key]
+
+    def TTS(self, request, context):
+        try:
+            # Check if dst is provided
+            if not request.dst:
+                return backend_pb2.Result(
+                    success=False,
+                    message="dst (output path) is required"
+                )
+            
+            # Prepare text
+            text = request.text.strip()
+            if not text:
+                return backend_pb2.Result(
+                    success=False,
+                    message="Text is empty"
+                )
+
+            # Get language (auto-detect if not provided)
+            language = request.language if hasattr(request, 'language') and request.language else None
+            if not language or language == "":
+                language = "Auto"  # Auto-detect language
+
+            # Detect mode
+            mode = self._detect_mode(request)
+            print(f"Detected mode: {mode}", file=sys.stderr)
+
+            # Get generation parameters from options
+            max_new_tokens = self.options.get("max_new_tokens", None)
+            top_p = self.options.get("top_p", None)
+            temperature = self.options.get("temperature", None)
+            do_sample = self.options.get("do_sample", None)
+
+            # Prepare generation kwargs
+            generation_kwargs = {}
+            if max_new_tokens is not None:
+                generation_kwargs["max_new_tokens"] = max_new_tokens
+            if top_p is not None:
+                generation_kwargs["top_p"] = top_p
+            if temperature is not None:
+                generation_kwargs["temperature"] = temperature
+            if do_sample is not None:
+                generation_kwargs["do_sample"] = do_sample
+
+            instruct = self.options.get("instruct", "")
+            if instruct is not None and instruct != "":
+                generation_kwargs["instruct"] = instruct
+
+            # Generate audio based on mode
+            if mode == "VoiceClone":
+                # VoiceClone mode
+                ref_audio = self._get_ref_audio_path(request)
+                if not ref_audio:
+                    return backend_pb2.Result(
+                        success=False,
+                        message="AudioPath is required for VoiceClone mode"
+                    )
+                
+                ref_text = self.options.get("ref_text", None)
+                if not ref_text:
+                    # Try to get from request if available
+                    if hasattr(request, 'ref_text') and request.ref_text:
+                        ref_text = request.ref_text
+                    else:
+                        # x_vector_only_mode doesn't require ref_text
+                        if not self.options.get("x_vector_only_mode", False):
+                            return backend_pb2.Result(
+                                success=False,
+                                message="ref_text is required for VoiceClone mode (or set x_vector_only_mode=true)"
+                            )
+
+                # Check if we should use cached prompt
+                use_cached_prompt = self.options.get("use_cached_prompt", True)
+                voice_clone_prompt = None
+                
+                if use_cached_prompt:
+                    voice_clone_prompt = self._get_voice_clone_prompt(request, ref_audio, ref_text)
+                    if voice_clone_prompt is None:
+                        return backend_pb2.Result(
+                            success=False,
+                            message="Failed to create voice clone prompt"
+                        )
+
+                if voice_clone_prompt:
+                    # Use cached prompt
+                    wavs, sr = self.model.generate_voice_clone(
+                        text=text,
+                        language=language,
+                        voice_clone_prompt=voice_clone_prompt,
+                        **generation_kwargs
+                    )
+                else:
+                    # Create prompt on-the-fly
+                    wavs, sr = self.model.generate_voice_clone(
+                        text=text,
+                        language=language,
+                        ref_audio=ref_audio,
+                        ref_text=ref_text,
+                        x_vector_only_mode=self.options.get("x_vector_only_mode", False),
+                        **generation_kwargs
+                    )
+
+            elif mode == "VoiceDesign":
+                # VoiceDesign mode
+                if not instruct:
+                    return backend_pb2.Result(
+                        success=False,
+                        message="instruct option is required for VoiceDesign mode"
+                    )
+
+                wavs, sr = self.model.generate_voice_design(
+                    text=text,
+                    language=language,
+                    instruct=instruct,
+                    **generation_kwargs
+                )
+
+            else:
+                # CustomVoice mode (default)
+                speaker = request.voice if request.voice else None
+                if not speaker:
+                    # Try to get from options
+                    speaker = self.options.get("speaker", None)
+                    if not speaker:
+                        # Use default speaker
+                        speaker = "Vivian"
+                        print(f"No speaker specified, using default: {speaker}", file=sys.stderr)
+
+                # Validate speaker if model supports it
+                if hasattr(self.model, 'get_supported_speakers'):
+                    try:
+                        supported_speakers = self.model.get_supported_speakers()
+                        if speaker not in supported_speakers:
+                            print(f"Warning: Speaker '{speaker}' not in supported list. Available: {supported_speakers}", file=sys.stderr)
+                            # Try to find a close match (case-insensitive)
+                            speaker_lower = speaker.lower()
+                            for sup_speaker in supported_speakers:
+                                if sup_speaker.lower() == speaker_lower:
+                                    speaker = sup_speaker
+                                    print(f"Using matched speaker: {speaker}", file=sys.stderr)
+                                    break
+                    except Exception as e:
+                        print(f"Warning: Could not get supported speakers: {e}", file=sys.stderr)
+
+                wavs, sr = self.model.generate_custom_voice(
+                    text=text,
+                    language=language,
+                    speaker=speaker,
+                    **generation_kwargs
+                )
+
+            # Save output
+            if wavs is not None and len(wavs) > 0:
+                # wavs is a list, take first element
+                audio_data = wavs[0] if isinstance(wavs, list) else wavs
+                sf.write(request.dst, audio_data, sr)
+                print(f"Saved output to {request.dst}", file=sys.stderr)
+            else:
+                return backend_pb2.Result(
+                    success=False,
+                    message="No audio output generated"
+                )
+
+        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)
--- a/backend/python/qwen-tts/install.sh
+++ b/backend/python/qwen-tts/install.sh
@@ -0,0 +1,13 @@
+#!/bin/bash
+set -e
+
+EXTRA_PIP_INSTALL_FLAGS="--no-build-isolation"
+
+backend_dir=$(dirname $0)
+if [ -d $backend_dir/common ]; then
+    source $backend_dir/common/libbackend.sh
+else
+    source $backend_dir/../common/libbackend.sh
+fi
+
+installRequirements
--- a/backend/python/qwen-tts/requirements-cpu.txt
+++ b/backend/python/qwen-tts/requirements-cpu.txt
@@ -0,0 +1,5 @@
+--extra-index-url https://download.pytorch.org/whl/cpu
+torch
+torchaudio
+qwen-tts
+sox
--- a/backend/python/qwen-tts/requirements-cublas12-after.txt
+++ b/backend/python/qwen-tts/requirements-cublas12-after.txt
--- a/backend/python/qwen-tts/requirements-cublas12.txt
+++ b/backend/python/qwen-tts/requirements-cublas12.txt
@@ -0,0 +1,5 @@
+--extra-index-url https://download.pytorch.org/whl/cu121
+torch
+torchaudio
+qwen-tts
+sox
--- a/backend/python/qwen-tts/requirements-cublas13.txt
+++ b/backend/python/qwen-tts/requirements-cublas13.txt
@@ -0,0 +1,5 @@
+--extra-index-url https://download.pytorch.org/whl/cu130
+torch
+torchaudio
+qwen-tts
+sox
--- a/backend/python/qwen-tts/requirements-hipblas.txt
+++ b/backend/python/qwen-tts/requirements-hipblas.txt
@@ -0,0 +1,5 @@
+--extra-index-url https://download.pytorch.org/whl/rocm6.3
+torch==2.7.1+rocm6.3
+torchaudio==2.7.1+rocm6.3
+qwen-tts
+sox
--- a/backend/python/qwen-tts/requirements-intel-after.txt
+++ b/backend/python/qwen-tts/requirements-intel-after.txt
@@ -0,0 +1 @@
+flash-attn
--- a/backend/python/qwen-tts/requirements-intel.txt
+++ b/backend/python/qwen-tts/requirements-intel.txt
@@ -0,0 +1,5 @@
+--extra-index-url https://download.pytorch.org/whl/xpu
+torch
+torchaudio
+qwen-tts
+sox
--- a/backend/python/qwen-tts/requirements-l4t12.txt
+++ b/backend/python/qwen-tts/requirements-l4t12.txt
@@ -0,0 +1,5 @@
+--extra-index-url https://pypi.jetson-ai-lab.io/jp6/cu129/
+torch
+torchaudio
+qwen-tts
+sox
--- a/backend/python/qwen-tts/requirements-l4t13.txt
+++ b/backend/python/qwen-tts/requirements-l4t13.txt
@@ -0,0 +1,5 @@
+--extra-index-url https://download.pytorch.org/whl/cu130
+torch
+torchaudio
+qwen-tts
+sox
--- a/backend/python/qwen-tts/requirements-mps.txt
+++ b/backend/python/qwen-tts/requirements-mps.txt
@@ -0,0 +1,4 @@
+torch==2.7.1
+torchaudio==0.22.1
+qwen-tts
+sox
--- a/backend/python/qwen-tts/requirements.txt
+++ b/backend/python/qwen-tts/requirements.txt
@@ -0,0 +1,6 @@
+grpcio==1.71.0
+protobuf
+certifi
+packaging==24.1
+soundfile
+setuptools
--- a/backend/python/qwen-tts/run.sh
+++ b/backend/python/qwen-tts/run.sh
@@ -0,0 +1,9 @@
+#!/bin/bash
+backend_dir=$(dirname $0)
+if [ -d $backend_dir/common ]; then
+    source $backend_dir/common/libbackend.sh
+else
+    source $backend_dir/../common/libbackend.sh
+fi
+
+startBackend $@
--- a/backend/python/qwen-tts/test.py
+++ b/backend/python/qwen-tts/test.py
@@ -0,0 +1,98 @@
+"""
+A test script to test the gRPC service
+"""
+import unittest
+import subprocess
+import time
+import os
+import sys
+import tempfile
+import threading
+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"],
+            stdout=subprocess.PIPE,
+            stderr=subprocess.PIPE,
+            text=True
+        )
+        time.sleep(5)
+
+    def tearDown(self) -> None:
+        """
+        This method tears down the gRPC service by terminating the server
+        """
+        self.service.terminate()
+        try:
+            stdout, stderr = self.service.communicate(timeout=5)
+            # Output should already be printed by threads, but print any remaining
+            if stdout:
+                print("=== REMAINING STDOUT ===")
+                print(stdout)
+            if stderr:
+                print("=== REMAINING STDERR ===")
+                print(stderr)
+        except subprocess.TimeoutExpired:
+            self.service.kill()
+            stdout, stderr = self.service.communicate()
+            if stdout:
+                print("=== REMAINING STDOUT ===")
+                print(stdout)
+            if stderr:
+                print("=== REMAINING STDERR ===")
+                print(stderr)
+
+    def test_tts(self):
+        """
+        This method tests if the TTS generation works successfully
+        """
+        try:
+            self.setUp()
+            with grpc.insecure_channel("localhost:50051") as channel:
+                stub = backend_pb2_grpc.BackendStub(channel)
+                # Allow up to 10 minutes for model download on first run
+                response = stub.LoadModel(
+                    backend_pb2.ModelOptions(Model="Qwen/Qwen3-TTS-12Hz-0.6B-CustomVoice"),
+                    timeout=600.0
+                )
+                self.assertTrue(response.success)
+                
+                # Create temporary output file
+                with tempfile.NamedTemporaryFile(suffix='.wav', delete=False) as tmp_file:
+                    output_path = tmp_file.name
+                
+                tts_request = backend_pb2.TTSRequest(
+                    text="Hello, this is a test of the qwen-tts backend.",
+                    voice="Vivian",
+                    dst=output_path
+                )
+                # Allow up to 2 minutes for TTS generation
+                tts_response = stub.TTS(tts_request, timeout=120.0)
+                self.assertIsNotNone(tts_response)
+                self.assertTrue(tts_response.success)
+                
+                # Verify output file exists and is not empty
+                self.assertTrue(os.path.exists(output_path))
+                self.assertGreater(os.path.getsize(output_path), 0)
+                
+                # Cleanup
+                os.unlink(output_path)
+        except Exception as err:
+            print(f"Exception: {err}", file=sys.stderr)
+            # Give threads a moment to flush any remaining output
+            time.sleep(1)
+            self.fail("TTS service failed")
+        finally:
+            self.tearDown()
--- a/Show More
+++ b/Show More