test CI (remove me)

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
chore(deps): switch to ubuntu 24.04
2026-02-03 03:02:38 -05:00 · 2025-07-04 18:49:15 +02:00 · 2025-07-04 18:48:53 +02:00 · 2025-07-04 18:42:01 +02:00 · 2025-07-04 18:39:36 +02:00 · 2025-07-04 18:38:18 +02:00
60 changed files with 1534 additions and 331 deletions
--- a/.github/workflows/backend.yml
+++ b/.github/workflows/backend.yml
@@ -7,7 +7,7 @@ on:
      - master
    tags:
      - '*'
-  #pull_request:
+  pull_request:

 concurrency:
  group: ci-backends-${{ github.head_ref || github.ref }}-${{ github.repository }}
@@ -26,7 +26,6 @@ jobs:
      runs-on: ${{ matrix.runs-on }}
      base-image: ${{ matrix.base-image }}
      backend: ${{ matrix.backend }}
-      latest-image: ${{ matrix.latest-image }}
      dockerfile: $${ matrix.dockerfile }}
      context: $${ matrix.context }}
    secrets:
@@ -47,9 +46,8 @@ jobs:
            tag-latest: 'true'
            tag-suffix: '-gpu-nvidia-cuda-11-rerankers'
            runs-on: 'ubuntu-latest'
-            base-image: "ubuntu:22.04"
+            base-image: "ubuntu:24.04"
            backend: "rerankers"
-            latest-image: 'latest-gpu-nvidia-cuda-11-rerankers'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          - build-type: 'cublas'
@@ -59,9 +57,8 @@ jobs:
            tag-latest: 'true'
            tag-suffix: '-gpu-nvidia-cuda-11-vllm'
            runs-on: 'ubuntu-latest'
-            base-image: "ubuntu:22.04"
+            base-image: "ubuntu:24.04"
            backend: "vllm"
-            latest-image: 'latest-gpu-nvidia-cuda-11-vllm'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          - build-type: 'cublas'
@@ -71,9 +68,8 @@ jobs:
            tag-latest: 'true'
            tag-suffix: '-gpu-nvidia-cuda-11-transformers'
            runs-on: 'ubuntu-latest'
-            base-image: "ubuntu:22.04"
+            base-image: "ubuntu:24.04"
            backend: "transformers"
-            latest-image: 'latest-gpu-nvidia-cuda-11-transformers'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          - build-type: 'cublas'
@@ -83,9 +79,8 @@ jobs:
            tag-latest: 'true'
            tag-suffix: '-gpu-nvidia-cuda-11-diffusers'
            runs-on: 'ubuntu-latest'
-            base-image: "ubuntu:22.04"
+            base-image: "ubuntu:24.04"
            backend: "diffusers"
-            latest-image: 'latest-gpu-nvidia-cuda-11-diffusers'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          # CUDA 11 additional backends
@@ -96,9 +91,8 @@ jobs:
            tag-latest: 'true'
            tag-suffix: '-gpu-nvidia-cuda-11-kokoro'
            runs-on: 'ubuntu-latest'
-            base-image: "ubuntu:22.04"
+            base-image: "ubuntu:24.04"
            backend: "kokoro"
-            latest-image: 'latest-gpu-nvidia-cuda-11-kokoro'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          - build-type: 'cublas'
@@ -108,9 +102,8 @@ jobs:
            tag-latest: 'true'
            tag-suffix: '-gpu-nvidia-cuda-11-faster-whisper'
            runs-on: 'ubuntu-latest'
-            base-image: "ubuntu:22.04"
+            base-image: "ubuntu:24.04"
            backend: "faster-whisper"
-            latest-image: 'latest-gpu-nvidia-cuda-11-faster-whisper'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          - build-type: 'cublas'
@@ -120,9 +113,8 @@ jobs:
            tag-latest: 'true'
            tag-suffix: '-gpu-nvidia-cuda-11-coqui'
            runs-on: 'ubuntu-latest'
-            base-image: "ubuntu:22.04"
+            base-image: "ubuntu:24.04"
            backend: "coqui"
-            latest-image: 'latest-gpu-nvidia-cuda-11-coqui'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          - build-type: 'cublas'
@@ -132,9 +124,8 @@ jobs:
            tag-latest: 'true'
            tag-suffix: '-gpu-nvidia-cuda-11-bark'
            runs-on: 'ubuntu-latest'
-            base-image: "ubuntu:22.04"
+            base-image: "ubuntu:24.04"
            backend: "bark"
-            latest-image: 'latest-gpu-nvidia-cuda-11-bark'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          - build-type: 'cublas'
@@ -144,9 +135,8 @@ jobs:
            tag-latest: 'true'
            tag-suffix: '-gpu-nvidia-cuda-11-chatterbox'
            runs-on: 'ubuntu-latest'
-            base-image: "ubuntu:22.04"
+            base-image: "ubuntu:24.04"
            backend: "chatterbox"
-            latest-image: 'latest-gpu-nvidia-cuda-11-chatterbox'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          # CUDA 12 builds
@@ -157,9 +147,8 @@ jobs:
            tag-latest: 'true'
            tag-suffix: '-gpu-nvidia-cuda-12-rerankers'
            runs-on: 'ubuntu-latest'
-            base-image: "ubuntu:22.04"
+            base-image: "ubuntu:24.04"
            backend: "rerankers"
-            latest-image: 'latest-gpu-nvidia-cuda-12-rerankers'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          - build-type: 'cublas'
@@ -169,9 +158,8 @@ jobs:
            tag-latest: 'true'
            tag-suffix: '-gpu-nvidia-cuda-12-vllm'
            runs-on: 'ubuntu-latest'
-            base-image: "ubuntu:22.04"
+            base-image: "ubuntu:24.04"
            backend: "vllm"
-            latest-image: 'latest-gpu-nvidia-cuda-12-vllm'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          - build-type: 'cublas'
@@ -181,9 +169,8 @@ jobs:
            tag-latest: 'true'
            tag-suffix: '-gpu-nvidia-cuda-12-transformers'
            runs-on: 'ubuntu-latest'
-            base-image: "ubuntu:22.04"
+            base-image: "ubuntu:24.04"
            backend: "transformers"
-            latest-image: 'latest-gpu-nvidia-cuda-12-transformers'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          - build-type: 'cublas'
@@ -193,9 +180,8 @@ jobs:
            tag-latest: 'true'
            tag-suffix: '-gpu-nvidia-cuda-12-diffusers'
            runs-on: 'ubuntu-latest'
-            base-image: "ubuntu:22.04"
-            backend: "diffusers"
-            latest-image: 'latest-gpu-nvidia-cuda-12-diffusers'
+            base-image: "ubuntu:24.04"
+            backend: "diffusers"            
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          # CUDA 12 additional backends
@@ -206,9 +192,8 @@ jobs:
            tag-latest: 'true'
            tag-suffix: '-gpu-nvidia-cuda-12-kokoro'
            runs-on: 'ubuntu-latest'
-            base-image: "ubuntu:22.04"
+            base-image: "ubuntu:24.04"
            backend: "kokoro"
-            latest-image: 'latest-gpu-nvidia-cuda-12-kokoro'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          - build-type: 'cublas'
@@ -218,9 +203,8 @@ jobs:
            tag-latest: 'true'
            tag-suffix: '-gpu-nvidia-cuda-12-faster-whisper'
            runs-on: 'ubuntu-latest'
-            base-image: "ubuntu:22.04"
+            base-image: "ubuntu:24.04"
            backend: "faster-whisper"
-            latest-image: 'latest-gpu-nvidia-cuda-12-faster-whisper'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          - build-type: 'cublas'
@@ -230,9 +214,8 @@ jobs:
            tag-latest: 'true'
            tag-suffix: '-gpu-nvidia-cuda-12-coqui'
            runs-on: 'ubuntu-latest'
-            base-image: "ubuntu:22.04"
+            base-image: "ubuntu:24.04"
            backend: "coqui"
-            latest-image: 'latest-gpu-nvidia-cuda-12-coqui'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          - build-type: 'cublas'
@@ -242,9 +225,8 @@ jobs:
            tag-latest: 'true'
            tag-suffix: '-gpu-nvidia-cuda-12-bark'
            runs-on: 'ubuntu-latest'
-            base-image: "ubuntu:22.04"
+            base-image: "ubuntu:24.04"
            backend: "bark"
-            latest-image: 'latest-gpu-nvidia-cuda-12-bark'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          - build-type: 'cublas'
@@ -254,9 +236,8 @@ jobs:
            tag-latest: 'true'
            tag-suffix: '-gpu-nvidia-cuda-12-chatterbox'
            runs-on: 'ubuntu-latest'
-            base-image: "ubuntu:22.04"
+            base-image: "ubuntu:24.04"
            backend: "chatterbox"
-            latest-image: 'latest-gpu-nvidia-cuda-12-chatterbox'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          # hipblas builds
@@ -267,9 +248,8 @@ jobs:
            tag-latest: 'true'
            tag-suffix: '-gpu-rocm-hipblas-rerankers'
            runs-on: 'ubuntu-latest'
-            base-image: "rocm/dev-ubuntu-22.04:6.1"
+            base-image: "rocm/dev-ubuntu-24.04:6.4.1"
            backend: "rerankers"
-            latest-image: 'latest-gpu-rocm-hipblas-rerankers'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          - build-type: 'hipblas'
@@ -279,9 +259,8 @@ jobs:
            tag-latest: 'true'
            tag-suffix: '-gpu-rocm-hipblas-vllm'
            runs-on: 'ubuntu-latest'
-            base-image: "rocm/dev-ubuntu-22.04:6.1"
+            base-image: "rocm/dev-ubuntu-24.04:6.4.1"
            backend: "vllm"
-            latest-image: 'latest-gpu-rocm-hipblas-vllm'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          - build-type: 'hipblas'
@@ -291,9 +270,8 @@ jobs:
            tag-latest: 'true'
            tag-suffix: '-gpu-rocm-hipblas-transformers'
            runs-on: 'ubuntu-latest'
-            base-image: "rocm/dev-ubuntu-22.04:6.1"
+            base-image: "rocm/dev-ubuntu-24.04:6.4.1"
            backend: "transformers"
-            latest-image: 'latest-gpu-rocm-hipblas-transformers'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          - build-type: 'hipblas'
@@ -303,9 +281,8 @@ jobs:
            tag-latest: 'true'
            tag-suffix: '-gpu-rocm-hipblas-diffusers'
            runs-on: 'ubuntu-latest'
-            base-image: "rocm/dev-ubuntu-22.04:6.1"
+            base-image: "rocm/dev-ubuntu-24.04:6.4.1"
            backend: "diffusers"
-            latest-image: 'latest-gpu-rocm-hipblas-diffusers'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          # ROCm additional backends
@@ -316,9 +293,8 @@ jobs:
            tag-latest: 'true'
            tag-suffix: '-gpu-rocm-hipblas-kokoro'
            runs-on: 'ubuntu-latest'
-            base-image: "rocm/dev-ubuntu-22.04:6.1"
+            base-image: "rocm/dev-ubuntu-24.04:6.4.1"
            backend: "kokoro"
-            latest-image: 'latest-gpu-rocm-hipblas-kokoro'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          - build-type: 'hipblas'
@@ -328,9 +304,8 @@ jobs:
            tag-latest: 'true'
            tag-suffix: '-gpu-rocm-hipblas-faster-whisper'
            runs-on: 'ubuntu-latest'
-            base-image: "rocm/dev-ubuntu-22.04:6.1"
+            base-image: "rocm/dev-ubuntu-24.04:6.4.1"
            backend: "faster-whisper"
-            latest-image: 'latest-gpu-rocm-hipblas-faster-whisper'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          - build-type: 'hipblas'
@@ -340,9 +315,8 @@ jobs:
            tag-latest: 'true'
            tag-suffix: '-gpu-rocm-hipblas-coqui'
            runs-on: 'ubuntu-latest'
-            base-image: "rocm/dev-ubuntu-22.04:6.1"
+            base-image: "rocm/dev-ubuntu-24.04:6.4.1"
            backend: "coqui"
-            latest-image: 'latest-gpu-rocm-hipblas-coqui'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          - build-type: 'hipblas'
@@ -352,9 +326,8 @@ jobs:
            tag-latest: 'true'
            tag-suffix: '-gpu-rocm-hipblas-bark'
            runs-on: 'ubuntu-latest'
-            base-image: "rocm/dev-ubuntu-22.04:6.1"
+            base-image: "rocm/dev-ubuntu-24.04:6.4.1"
            backend: "bark"
-            latest-image: 'latest-gpu-rocm-hipblas-bark'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          # sycl builds
@@ -367,7 +340,6 @@ jobs:
            runs-on: 'ubuntu-latest'
            base-image: "quay.io/go-skynet/intel-oneapi-base:latest"
            backend: "rerankers"
-            latest-image: 'latest-gpu-intel-sycl-f32-rerankers'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          - build-type: 'sycl_f16'
@@ -379,7 +351,6 @@ jobs:
            runs-on: 'ubuntu-latest'
            base-image: "quay.io/go-skynet/intel-oneapi-base:latest"
            backend: "rerankers"
-            latest-image: 'latest-gpu-intel-sycl-f16-rerankers'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          - build-type: 'sycl_f32'
@@ -391,7 +362,6 @@ jobs:
            runs-on: 'ubuntu-latest'
            base-image: "quay.io/go-skynet/intel-oneapi-base:latest"
            backend: "vllm"
-            latest-image: 'latest-gpu-intel-sycl-f32-vllm'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          - build-type: 'sycl_f16'
@@ -403,7 +373,6 @@ jobs:
            runs-on: 'ubuntu-latest'
            base-image: "quay.io/go-skynet/intel-oneapi-base:latest"
            backend: "vllm"
-            latest-image: 'latest-gpu-intel-sycl-f16-vllm'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          - build-type: 'sycl_f32'
@@ -415,7 +384,6 @@ jobs:
            runs-on: 'ubuntu-latest'
            base-image: "quay.io/go-skynet/intel-oneapi-base:latest"
            backend: "transformers"
-            latest-image: 'latest-gpu-intel-sycl-f32-transformers'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          - build-type: 'sycl_f16'
@@ -427,7 +395,6 @@ jobs:
            runs-on: 'ubuntu-latest'
            base-image: "quay.io/go-skynet/intel-oneapi-base:latest"
            backend: "transformers"
-            latest-image: 'latest-gpu-intel-sycl-f16-transformers'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          - build-type: 'sycl_f32'
@@ -439,7 +406,6 @@ jobs:
            runs-on: 'ubuntu-latest'
            base-image: "quay.io/go-skynet/intel-oneapi-base:latest"
            backend: "diffusers"
-            latest-image: 'latest-gpu-intel-sycl-f32-diffusers'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          # SYCL additional backends
@@ -452,7 +418,6 @@ jobs:
            runs-on: 'ubuntu-latest'
            base-image: "quay.io/go-skynet/intel-oneapi-base:latest"
            backend: "kokoro"
-            latest-image: 'latest-gpu-intel-sycl-f32-kokoro'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          - build-type: 'sycl_f16'
@@ -464,7 +429,6 @@ jobs:
            runs-on: 'ubuntu-latest'
            base-image: "quay.io/go-skynet/intel-oneapi-base:latest"
            backend: "kokoro"
-            latest-image: 'latest-gpu-intel-sycl-f16-kokoro'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          - build-type: 'sycl_f32'
@@ -476,7 +440,6 @@ jobs:
            runs-on: 'ubuntu-latest'
            base-image: "quay.io/go-skynet/intel-oneapi-base:latest"
            backend: "faster-whisper"
-            latest-image: 'latest-gpu-intel-sycl-f32-faster-whisper'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          - build-type: 'sycl_f16'
@@ -488,7 +451,6 @@ jobs:
            runs-on: 'ubuntu-latest'
            base-image: "quay.io/go-skynet/intel-oneapi-base:latest"
            backend: "faster-whisper"
-            latest-image: 'latest-gpu-intel-sycl-f16-faster-whisper'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          - build-type: 'sycl_f32'
@@ -500,7 +462,6 @@ jobs:
            runs-on: 'ubuntu-latest'
            base-image: "quay.io/go-skynet/intel-oneapi-base:latest"
            backend: "coqui"
-            latest-image: 'latest-gpu-intel-sycl-f32-coqui'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          - build-type: 'sycl_f16'
@@ -512,7 +473,6 @@ jobs:
            runs-on: 'ubuntu-latest'
            base-image: "quay.io/go-skynet/intel-oneapi-base:latest"
            backend: "coqui"
-            latest-image: 'latest-gpu-intel-sycl-f16-coqui'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          - build-type: 'sycl_f32'
@@ -524,7 +484,6 @@ jobs:
            runs-on: 'ubuntu-latest'
            base-image: "quay.io/go-skynet/intel-oneapi-base:latest"
            backend: "bark"
-            latest-image: 'latest-gpu-intel-sycl-f32-bark'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          - build-type: 'sycl_f16'
@@ -536,7 +495,6 @@ jobs:
            runs-on: 'ubuntu-latest'
            base-image: "quay.io/go-skynet/intel-oneapi-base:latest"
            backend: "bark"
-            latest-image: 'latest-gpu-intel-sycl-f16-bark'
            dockerfile: "./backend/Dockerfile.python"
            context: "./backend"
          # bark-cpp
@@ -547,8 +505,7 @@ jobs:
            tag-latest: 'true'
            tag-suffix: '-bark-cpp'
            runs-on: 'ubuntu-latest'
-            base-image: "ubuntu:22.04"
+            base-image: "ubuntu:24.04"
            backend: "bark"
-            latest-image: 'latest-bark-cpp'
            dockerfile: "./backend/Dockerfile.go"
            context: "./"
--- a/.github/workflows/backend_build.yml
+++ b/.github/workflows/backend_build.yml
@@ -28,10 +28,6 @@ on:
        description: 'Tag latest'
        default: ''
        type: string
-      latest-image:
-        description: 'Tag latest'
-        default: ''
-        type: string
      tag-suffix:
        description: 'Tag suffix'
        default: ''
@@ -153,7 +149,7 @@ jobs:
            type=sha
          flavor: |
            latest=${{ inputs.tag-latest }}
-            suffix=${{ inputs.tag-suffix }}
+            suffix=${{ inputs.tag-suffix }},onlatest=true

      - name: Docker meta for PR
        id: meta_pull_request
@@ -168,7 +164,7 @@ jobs:
            type=sha,suffix=${{ github.event.number }}-${{ inputs.backend }}-${{ inputs.build-type }}-${{ inputs.cuda-major-version }}-${{ inputs.cuda-minor-version }}
          flavor: |
            latest=${{ inputs.tag-latest }}
-            suffix=${{ inputs.tag-suffix }}
+            suffix=${{ inputs.tag-suffix }},onlatest=true
 ## End testing image
      - name: Set up QEMU
        uses: docker/setup-qemu-action@master
@@ -210,7 +206,6 @@ jobs:
          cache-from: type=gha
          platforms: ${{ inputs.platforms }}
          push: ${{ github.event_name != 'pull_request' }}
-          load: ${{ github.event_name != 'pull_request' && inputs.latest-image != '' && github.ref_type == 'tag' }}
          tags: ${{ steps.meta.outputs.tags }}
          labels: ${{ steps.meta.outputs.labels }}

@@ -233,18 +228,7 @@ jobs:
          tags: ${{ steps.meta_pull_request.outputs.tags }}
          labels: ${{ steps.meta_pull_request.outputs.labels }}

-      - name: Cleanup
-        run: |
-          docker builder prune -f
-          docker system prune --force --volumes --all

-      - name: Latest tag
-        if: github.event_name != 'pull_request' && inputs.latest-image != '' && github.ref_type == 'tag'
-        run: |
-          docker tag localai/localai-backends:${{ steps.meta.outputs.version }} localai/localai-backends:${{ inputs.latest-image }}
-          docker push localai/localai-backends:${{ inputs.latest-image }}
-          docker tag quay.io/go-skynet/local-ai-backends:${{ steps.meta.outputs.version }} quay.io/go-skynet/local-ai-backends:${{ inputs.latest-image }}
-          docker push quay.io/go-skynet/local-ai-backends:${{ inputs.latest-image }}

      - name: job summary
        run: |
--- 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:2025.1.0-0-devel-ubuntu22.04
+          - base-image: intel/oneapi-basekit:2025.1.3-0-devel-ubuntu24.04
            runs-on: 'ubuntu-latest'
            platforms: 'linux/amd64'
    runs-on: ${{matrix.runs-on}}
--- a/.github/workflows/image-pr.yml
+++ b/.github/workflows/image-pr.yml
@@ -43,22 +43,22 @@ jobs:
            tag-suffix: '-gpu-nvidia-cuda12-ffmpeg'
            ffmpeg: 'true'
            runs-on: 'ubuntu-latest'
-            base-image: "ubuntu:22.04"
+            base-image: "ubuntu:24.04"
            makeflags: "--jobs=3 --output-sync=target"
          - build-type: 'hipblas'
            platforms: 'linux/amd64'
            tag-latest: 'false'
            tag-suffix: '-hipblas'
            ffmpeg: 'false'
-            base-image: "rocm/dev-ubuntu-22.04:6.1"
-            grpc-base-image: "ubuntu:22.04"
+            base-image: "rocm/dev-ubuntu-24.04:6.4.1"
+            grpc-base-image: "ubuntu:24.04"
            runs-on: 'ubuntu-latest'
            makeflags: "--jobs=3 --output-sync=target"
          - build-type: 'sycl_f16'
            platforms: 'linux/amd64'
            tag-latest: 'false'
            base-image: "quay.io/go-skynet/intel-oneapi-base:latest"
-            grpc-base-image: "ubuntu:22.04"
+            grpc-base-image: "ubuntu:24.04"
            tag-suffix: 'sycl-f16-ffmpeg'
            ffmpeg: 'true'
            runs-on: 'ubuntu-latest'
@@ -69,5 +69,5 @@ jobs:
            tag-suffix: '-vulkan-ffmpeg-core'
            ffmpeg: 'true'
            runs-on: 'ubuntu-latest'
-            base-image: "ubuntu:22.04"
+            base-image: "ubuntu:24.04"
            makeflags: "--jobs=4 --output-sync=target"
--- a/.github/workflows/image.yml
+++ b/.github/workflows/image.yml
@@ -38,11 +38,11 @@ jobs:
        include:
          - build-type: 'hipblas'
            platforms: 'linux/amd64'
-            tag-latest: 'false'
-            tag-suffix: '-hipblas'
+            tag-latest: 'auto'
+            tag-suffix: '-gpu-hipblas'
            ffmpeg: 'true'
-            base-image: "rocm/dev-ubuntu-22.04:6.1"
-            grpc-base-image: "ubuntu:22.04"
+            base-image: "rocm/dev-ubuntu-24.04:6.4.1"
+            grpc-base-image: "ubuntu:24.04"
            runs-on: 'ubuntu-latest'
            makeflags: "--jobs=3 --output-sync=target"
            aio: "-aio-gpu-hipblas"
@@ -77,7 +77,7 @@ jobs:
            tag-latest: 'auto'
            tag-suffix: ''
            ffmpeg: 'true'
-            base-image: "ubuntu:22.04"
+            base-image: "ubuntu:24.04"
            runs-on: 'ubuntu-latest'
            aio: "-aio-cpu"
            makeflags: "--jobs=4 --output-sync=target"
@@ -86,11 +86,11 @@ jobs:
            cuda-major-version: "11"
            cuda-minor-version: "7"
            platforms: 'linux/amd64'
-            tag-latest: 'false'
+            tag-latest: 'auto'
            tag-suffix: '-gpu-nvidia-cuda11'
            ffmpeg: 'true'
            runs-on: 'ubuntu-latest'
-            base-image: "ubuntu:22.04"
+            base-image: "ubuntu:24.04"
            makeflags: "--jobs=4 --output-sync=target"
            skip-drivers: 'false'
            aio: "-aio-gpu-nvidia-cuda-11"
@@ -98,29 +98,29 @@ jobs:
            cuda-major-version: "12"
            cuda-minor-version: "0"
            platforms: 'linux/amd64'
-            tag-latest: 'false'
+            tag-latest: 'auto'
            tag-suffix: '-gpu-nvidia-cuda12'
            ffmpeg: 'true'
            runs-on: 'ubuntu-latest'
-            base-image: "ubuntu:22.04"
+            base-image: "ubuntu:24.04"
            skip-drivers: 'false'
            makeflags: "--jobs=4 --output-sync=target"
            aio: "-aio-gpu-nvidia-cuda-12"
          - build-type: 'vulkan'
            platforms: 'linux/amd64'
-            tag-latest: 'false'
+            tag-latest: 'auto'
            tag-suffix: '-vulkan'
            ffmpeg: 'true'
            runs-on: 'ubuntu-latest'
-            base-image: "ubuntu:22.04"
+            base-image: "ubuntu:24.04"
            skip-drivers: 'false'
            makeflags: "--jobs=4 --output-sync=target"
            aio: "-aio-gpu-vulkan"
          - build-type: 'sycl_f16'
            platforms: 'linux/amd64'
-            tag-latest: 'false'
+            tag-latest: 'auto'
            base-image: "quay.io/go-skynet/intel-oneapi-base:latest"
-            grpc-base-image: "ubuntu:22.04"
+            grpc-base-image: "ubuntu:24.04"
            tag-suffix: '-gpu-intel-f16'
            ffmpeg: 'true'
            runs-on: 'ubuntu-latest'
@@ -128,9 +128,9 @@ jobs:
            aio: "-aio-gpu-intel-f16"
          - build-type: 'sycl_f32'
            platforms: 'linux/amd64'
-            tag-latest: 'false'
+            tag-latest: 'auto'
            base-image: "quay.io/go-skynet/intel-oneapi-base:latest"
-            grpc-base-image: "ubuntu:22.04"
+            grpc-base-image: "ubuntu:24.04"
            tag-suffix: '-gpu-intel-f32'
            ffmpeg: 'true'
            runs-on: 'ubuntu-latest'
@@ -165,7 +165,7 @@ jobs:
            cuda-major-version: "12"
            cuda-minor-version: "0"
            platforms: 'linux/arm64'
-            tag-latest: 'false'
+            tag-latest: 'auto'
            tag-suffix: '-nvidia-l4t-arm64'
            ffmpeg: 'true'
            base-image: "nvcr.io/nvidia/l4t-jetpack:r36.4.0"
--- a/11
+++ b/11
@@ -1,4 +1,4 @@
-ARG BASE_IMAGE=ubuntu:22.04
+ARG BASE_IMAGE=ubuntu:24.04
 ARG GRPC_BASE_IMAGE=${BASE_IMAGE}
 ARG INTEL_BASE_IMAGE=${BASE_IMAGE}

@@ -24,6 +24,8 @@ ARG TARGETARCH
 ARG TARGETVARIANT
 ENV BUILD_TYPE=${BUILD_TYPE}

+RUN mkdir -p /run/localai
+
 # Vulkan requirements
 RUN <<EOT bash
    if [ "${BUILD_TYPE}" = "vulkan" ] && [ "${SKIP_DRIVERS}" = "false" ]; then
@@ -36,7 +38,8 @@ RUN <<EOT bash
        apt-get install -y \
            vulkan-sdk && \
        apt-get clean && \
-        rm -rf /var/lib/apt/lists/*
+        rm -rf /var/lib/apt/lists/* && \
+        echo "vulkan" > /run/localai/capability
    fi
 EOT

@@ -63,7 +66,8 @@ RUN <<EOT bash
            libcusparse-dev-${CUDA_MAJOR_VERSION}-${CUDA_MINOR_VERSION} \
            libcusolver-dev-${CUDA_MAJOR_VERSION}-${CUDA_MINOR_VERSION} && \
        apt-get clean && \
-        rm -rf /var/lib/apt/lists/*
+        rm -rf /var/lib/apt/lists/* && \
+        echo "nvidia" > /run/localai/capability
    fi
 EOT

@@ -83,6 +87,7 @@ RUN if [ "${BUILD_TYPE}" = "hipblas" ] && [ "${SKIP_DRIVERS}" = "false" ]; then
            rocblas-dev && \
        apt-get clean && \
        rm -rf /var/lib/apt/lists/* && \
+        echo "amd" > /run/localai/capability && \
        # I have no idea why, but the ROCM lib packages don't trigger ldconfig after they install, which results in local-ai and others not being able
        # to locate the libraries. We run ldconfig ourselves to work around this packaging deficiency
        ldconfig \
--- 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/24
+++ b/24
@@ -6,11 +6,11 @@ BINARY_NAME=local-ai
 DETECT_LIBS?=true

 # llama.cpp versions
-CPPLLAMA_VERSION?=8846aace4934ad29651ea61b8c7e3f6b0556e3d2
+CPPLLAMA_VERSION?=bee28421be25fd447f61cb6db64d556cbfce32ec

 # whisper.cpp version
 WHISPER_REPO?=https://github.com/ggml-org/whisper.cpp
-WHISPER_CPP_VERSION?=32cf4e2aba799aff069011f37ca025401433cf9f
+WHISPER_CPP_VERSION?=d9999d54c868b8bfcd376aa26067e787d53e679e

 # go-piper version
 PIPER_REPO?=https://github.com/mudler/go-piper
@@ -18,7 +18,7 @@ PIPER_VERSION?=e10ca041a885d4a8f3871d52924b47792d5e5aa0

 # bark.cpp
 BARKCPP_REPO?=https://github.com/PABannier/bark.cpp.git
-BARKCPP_VERSION?=v1.0.0
+BARKCPP_VERSION?=5d5be84f089ab9ea53b7a793f088d3fbf7247495

 # stablediffusion.cpp (ggml)
 STABLEDIFFUSION_GGML_REPO?=https://github.com/richiejp/stable-diffusion.cpp
@@ -265,8 +265,8 @@ sources/bark.cpp/build/libbark.a: sources/bark.cpp
 	cmake $(CMAKE_ARGS) .. && \
 	cmake --build . --config Release

-backend/go/bark/libbark.a: sources/bark.cpp/build/libbark.a
-	$(MAKE) -C backend/go/bark libbark.a
+backend/go/bark-cpp/libbark.a: sources/bark.cpp/build/libbark.a
+	$(MAKE) -C backend/go/bark-cpp libbark.a

 ## go-piper
 sources/go-piper:
@@ -355,7 +355,7 @@ clean: ## Remove build related file
 	rm -rf release/
 	rm -rf backend-assets/*
 	$(MAKE) -C backend/cpp/grpc clean
-	$(MAKE) -C backend/go/bark clean
+	$(MAKE) -C backend/go/bark-cpp clean
 	$(MAKE) -C backend/cpp/llama clean
 	$(MAKE) -C backend/go/image/stablediffusion-ggml clean
 	rm -rf backend/cpp/llama-* || true
@@ -778,9 +778,9 @@ backend-assets/util/llama-cpp-rpc-server: backend-assets/grpc/llama-cpp-grpc
 	mkdir -p backend-assets/util/
 	cp -rf backend/cpp/llama-grpc/llama.cpp/build/bin/rpc-server backend-assets/util/llama-cpp-rpc-server

-backend-assets/grpc/bark-cpp: backend/go/bark/libbark.a backend-assets/grpc
-	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/backend/go/bark/ LIBRARY_PATH=$(CURDIR)/backend/go/bark/ \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/bark-cpp ./backend/go/bark/
+backend-assets/grpc/bark-cpp: backend/go/bark-cpp/libbark.a backend-assets/grpc
+	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(CURDIR)/backend/go/bark-cpp/ LIBRARY_PATH=$(CURDIR)/backend/go/bark-cpp/ \
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/bark-cpp ./backend/go/bark-cpp/
 ifneq ($(UPX),)
 	$(UPX) backend-assets/grpc/bark-cpp
 endif
@@ -817,7 +817,7 @@ grpcs: prepare $(GRPC_BACKENDS)
 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 \
@@ -852,7 +852,7 @@ docker-aio-all:

 docker-image-intel:
 	docker build \
-		--build-arg BASE_IMAGE=intel/oneapi-basekit:2025.1.0-0-devel-ubuntu24.04 \
+		--build-arg BASE_IMAGE=intel/oneapi-basekit:2025.1.3-0-devel-ubuntu24.04 \
 		--build-arg IMAGE_TYPE=$(IMAGE_TYPE) \
 		--build-arg GO_TAGS="$(GO_TAGS)" \
 		--build-arg MAKEFLAGS="$(DOCKER_MAKEFLAGS)" \
@@ -860,7 +860,7 @@ docker-image-intel:

 docker-image-intel-xpu:
 	docker build \
-		--build-arg BASE_IMAGE=intel/oneapi-basekit:2025.1.0-0-devel-ubuntu22.04 \
+		--build-arg BASE_IMAGE=intel/oneapi-basekit:2025.1.3-0-devel-ubuntu24.04 \
 		--build-arg IMAGE_TYPE=$(IMAGE_TYPE) \
 		--build-arg GO_TAGS="$(GO_TAGS)" \
 		--build-arg MAKEFLAGS="$(DOCKER_MAKEFLAGS)" \
--- a/README.md
+++ b/README.md
@@ -141,10 +141,10 @@ docker run -ti --name local-ai -p 8080:8080 --device=/dev/kfd --device=/dev/dri

 ```bash
 # Intel GPU with FP16 support
-docker run -ti --name local-ai -p 8080:8080 localai/localai:latest-gpu-intel-f16
+docker run -ti --name local-ai -p 8080:8080 --device=/dev/dri/card1 --device=/dev/dri/renderD128 localai/localai:latest-gpu-intel-f16

 # Intel GPU with FP32 support
-docker run -ti --name local-ai -p 8080:8080 localai/localai:latest-gpu-intel-f32
+docker run -ti --name local-ai -p 8080:8080 --device=/dev/dri/card1 --device=/dev/dri/renderD128 localai/localai:latest-gpu-intel-f32
 ```

 ### Vulkan GPU Images:
@@ -215,6 +215,7 @@ Roadmap items: [List of issues](https://github.com/mudler/LocalAI/issues?q=is%3A

 ## 🚀 [Features](https://localai.io/features/)

+- 🧩 [Backend Gallery](https://localai.io/backends/): Install/remove backends on the fly, powered by OCI images — fully customizable and API-driven.
 - 📖 [Text generation with GPTs](https://localai.io/features/text-generation/) (`llama.cpp`, `transformers`, `vllm` ... [:book: and more](https://localai.io/model-compatibility/index.html#model-compatibility-table))
 - 🗣 [Text to Audio](https://localai.io/features/text-to-audio/)
 - 🔈 [Audio to Text](https://localai.io/features/audio-to-text/) (Audio transcription with `whisper.cpp`)
--- a/backend/Dockerfile.go
+++ b/backend/Dockerfile.go
@@ -1,4 +1,4 @@
-ARG BASE_IMAGE=ubuntu:22.04
+ARG BASE_IMAGE=ubuntu:24.04

 FROM ${BASE_IMAGE} AS builder
 ARG BACKEND=rerankers
@@ -123,9 +123,9 @@ EOT

 COPY . /LocalAI

-RUN cd /LocalAI && make backend-assets/grpc/bark-cpp
+RUN cd /LocalAI && make backend-assets/grpc/${BACKEND}

 FROM scratch

-COPY --from=builder /LocalAI/backend-assets/grpc/bark-cpp ./
-COPY --from=builder /LocalAI/backend/go/bark/run.sh ./
+COPY --from=builder /LocalAI/backend-assets/grpc/${BACKEND} ./
+COPY --from=builder /LocalAI/backend/go/${BACKEND}/run.sh ./
--- 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
--- a/backend/backend.proto
+++ b/backend/backend.proto
@@ -258,6 +258,8 @@ message ModelOptions {
  repeated GrammarTrigger GrammarTriggers = 65;

  bool Reranking = 71;
+
+  repeated string Overrides = 72;
 }

 message Result {
--- a/backend/cpp/llama/grpc-server.cpp
+++ b/backend/cpp/llama/grpc-server.cpp
@@ -12,6 +12,7 @@

 #include "backend.pb.h"
 #include "backend.grpc.pb.h"
+#include "common.h"
 #include <getopt.h>
 #include <grpcpp/ext/proto_server_reflection_plugin.h>
 #include <grpcpp/grpcpp.h>
@@ -260,6 +261,13 @@ static void params_parse(const backend::ModelOptions* request,
        }
    }

+    // Add kv_overrides
+    if (request->overrides_size() > 0) {
+        for (int i = 0; i < request->overrides_size(); i++) {
+            string_parse_kv_override(request->overrides(i).c_str(), params.kv_overrides);
+        }
+    }
+
    // TODO: Add yarn

    if (!request->tensorsplit().empty()) {
--- a/backend/go/bark-cpp/Makefile
+++ b/backend/go/bark-cpp/Makefile
--- a/backend/go/bark-cpp/gobark.cpp
+++ b/backend/go/bark-cpp/gobark.cpp
--- a/backend/go/bark-cpp/gobark.go
+++ b/backend/go/bark-cpp/gobark.go
--- a/backend/go/bark-cpp/gobark.h
+++ b/backend/go/bark-cpp/gobark.h
--- a/backend/go/bark-cpp/main.go
+++ b/backend/go/bark-cpp/main.go
--- a/backend/go/bark-cpp/run.sh
+++ b/backend/go/bark-cpp/run.sh
--- a/backend/index.yaml
+++ b/backend/index.yaml
@@ -1,8 +1,7 @@
 ---
 ## vLLM
 - &vllm
-  name: "cuda11-vllm"
-  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-11-vllm"
+  name: "vllm"
  license: apache-2.0
  urls:
    - https://github.com/vllm-project/vllm
@@ -29,6 +28,19 @@
        Speculative decoding
        Chunked prefill
  alias: "vllm"
+  capabilities:
+    nvidia: "cuda12-vllm"
+    amd: "rocm-vllm"
+    intel: "intel-sycl-f16-vllm"
+- !!merge <<: *vllm
+  name: "vllm-development"
+  capabilities:
+    nvidia: "cuda12-vllm-development"
+    amd: "rocm-vllm-development"
+    intel: "intel-sycl-f16-vllm-development"
+- !!merge <<: *vllm
+  name: "cuda11-vllm"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-11-vllm"
 - !!merge <<: *vllm
  name: "cuda12-vllm"
  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-12-vllm"
@@ -57,43 +69,52 @@
  name: "intel-sycl-f16-vllm-development"
  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-intel-sycl-f16-vllm"
 ## Rerankers
- name: "cuda11-rerankers"
+- &rerankers
+  name: "rerankers"
+  alias: "rerankers"
+  capabilities:
+    nvidia: "cuda12-rerankers"
+    intel: "intel-sycl-f16-rerankers"
+    amd: "rocm-rerankers"
+- !!merge <<: *rerankers
+  name: "rerankers-development"
+  capabilities:
+    nvidia: "cuda12-rerankers-development"
+    intel: "intel-sycl-f16-rerankers-development"
+    amd: "rocm-rerankers-development"
+- !!merge <<: *rerankers
+  name: "cuda11-rerankers"
  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-11-rerankers"
-  alias: "cuda11-rerankers"
- name: "cuda12-rerankers"
+- !!merge <<: *rerankers
+  name: "cuda12-rerankers"
  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-12-rerankers"
-  alias: "cuda12-rerankers"
- name: "intel-sycl-f32-rerankers"
+- !!merge <<: *rerankers
+  name: "intel-sycl-f32-rerankers"
  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-intel-sycl-f32-rerankers"
-  alias: "intel-sycl-f32-rerankers"
- name: "intel-sycl-f16-rerankers"
+- !!merge <<: *rerankers
+  name: "intel-sycl-f16-rerankers"
  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-intel-sycl-f16-rerankers"
-  alias: "intel-sycl-f16-rerankers"
- name: "rocm-rerankers"
+- !!merge <<: *rerankers
+  name: "rocm-rerankers"
  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-rocm-hipblas-rerankers"
-  alias: "rocm-rerankers"
- name: "cuda11-rerankers-development"
+- !!merge <<: *rerankers
+  name: "cuda11-rerankers-development"
  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-11-rerankers"
-  alias: "rerankers"
-
- name: "cuda12-rerankers-development"
+- !!merge <<: *rerankers
+  name: "cuda12-rerankers-development"
  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-12-rerankers"
-  alias: "rerankers"
- name: "rocm-rerankers-development"
+- !!merge <<: *rerankers
+  name: "rocm-rerankers-development"
  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-rocm-hipblas-rerankers"
-  alias: "rerankers"
-
- name: "intel-sycl-f32-rerankers-development"
+- !!merge <<: *rerankers
+  name: "intel-sycl-f32-rerankers-development"
  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-intel-sycl-f32-rerankers"
-  alias: "rerankers"
-
- name: "intel-sycl-f16-rerankers-development"
+- !!merge <<: *rerankers
+  name: "intel-sycl-f16-rerankers-development"
  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-intel-sycl-f16-rerankers"
-  alias: "rerankers"
 ## Transformers
 - &transformers
-  name: "cuda12-transformers"
-  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-12-transformers"
+  name: "transformers"
  icon: https://camo.githubusercontent.com/26569a27b8a30a488dd345024b71dbc05da7ff1b2ba97bb6080c9f1ee0f26cc7/68747470733a2f2f68756767696e67666163652e636f2f64617461736574732f68756767696e67666163652f646f63756d656e746174696f6e2d696d616765732f7265736f6c76652f6d61696e2f7472616e73666f726d6572732f7472616e73666f726d6572735f61735f615f6d6f64656c5f646566696e6974696f6e2e706e67
  alias: "transformers"
  license: apache-2.0
@@ -105,6 +126,19 @@
  tags:
    - text-to-text
    - multimodal
+  capabilities:
+    nvidia: "cuda12-transformers"
+    intel: "intel-sycl-f16-transformers"
+    amd: "rocm-transformers"
+- !!merge <<: *transformers
+  name: "transformers-development"
+  capabilities:
+    nvidia: "cuda12-transformers-development"
+    intel: "intel-sycl-f16-transformers-development"
+    amd: "rocm-transformers-development"
+- !!merge <<: *transformers
+  name: "cuda12-transformers"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-12-transformers"
 - !!merge <<: *transformers
  name: "rocm-transformers"
  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-rocm-hipblas-transformers"
@@ -143,10 +177,21 @@
    - image-generation
    - video-generation
    - diffusion-models
-  name: "cuda12-diffusers"
  license: apache-2.0
-  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-12-diffusers"
  alias: "diffusers"
+  capabilities:
+    nvidia: "cuda12-diffusers"
+    intel: "intel-sycl-f32-diffusers"
+    amd: "rocm-diffusers"
+- !!merge <<: *diffusers
+  name: "diffusers-development"
+  capabilities:
+    nvidia: "cuda12-diffusers-development"
+    intel: "intel-sycl-f32-diffusers-development"
+    amd: "rocm-diffusers-development"
+- !!merge <<: *diffusers
+  name: "cuda12-diffusers"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-12-diffusers"
 - !!merge <<: *diffusers
  name: "rocm-diffusers"
  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-rocm-hipblas-diffusers"
@@ -170,6 +215,7 @@
  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-intel-sycl-f32-diffusers"
  ## exllama2
 - &exllama2
+  name: "exllama2"
  urls:
    - https://github.com/turboderp-org/exllamav2
  tags:
@@ -179,9 +225,20 @@
  license: MIT
  description: |
    ExLlamaV2 is an inference library for running local LLMs on modern consumer GPUs.
+  alias: "exllama2"
+  capabilities:
+    nvidia: "cuda12-exllama2"
+    intel: "intel-sycl-f32-exllama2"
+    amd: "rocm-exllama2"
+- !!merge <<: *exllama2
+  name: "exllama2-development"
+  capabilities:
+    nvidia: "cuda12-exllama2-development"
+    intel: "intel-sycl-f32-exllama2-development"
+    amd: "rocm-exllama2-development"
+- !!merge <<: *exllama2
  name: "cuda11-exllama2"
  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-11-exllama2"
-  alias: "exllama2"
 - !!merge <<: *exllama2
  name: "cuda12-exllama2"
  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-12-exllama2"
@@ -204,9 +261,21 @@
    - TTS
    - LLM
  license: apache-2.0
+  alias: "kokoro"
+  name: "kokoro"
+  capabilities:
+    nvidia: "cuda12-kokoro"
+    intel: "intel-sycl-f32-kokoro"
+    amd: "rocm-kokoro"
+- !!merge <<: *kokoro
+  name: "kokoro-development"
+  capabilities:
+    nvidia: "cuda12-kokoro-development"
+    intel: "intel-sycl-f32-kokoro-development"
+    amd: "rocm-kokoro-development"
+- !!merge <<: *kokoro
  name: "cuda11-kokoro-development"
  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-11-kokoro"
-  alias: "kokoro"
 - !!merge <<: *kokoro
  name: "cuda12-kokoro-development"
  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-12-kokoro"
@@ -225,6 +294,15 @@
 - !!merge <<: *kokoro
  name: "sycl-f32-kokoro-development"
  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-intel-sycl-f32-kokoro"
+- !!merge <<: *kokoro
+  name: "cuda11-kokoro"
+  uri: "quay.io/go-skynet/local-ai-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"
+- !!merge <<: *kokoro
+  name: "rocm-kokoro"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-rocm-hipblas-kokoro"
 ## faster-whisper
 - &faster-whisper
  icon: https://avatars.githubusercontent.com/u/1520500?s=200&v=4
@@ -237,9 +315,20 @@
    - speech-to-text
    - Whisper
  license: MIT
-  name: "cuda11-faster-whisper-development"
-  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-11-faster-whisper"
-  alias: "faster-whisper"
+  name: "faster-whisper"
+  capabilities:
+    nvidia: "cuda12-faster-whisper"
+    intel: "intel-sycl-f32-faster-whisper"
+    amd: "rocm-faster-whisper"
+- !!merge <<: *faster-whisper
+  name: "faster-whisper-development"
+  capabilities:
+    nvidia: "cuda12-faster-whisper-development"
+    intel: "intel-sycl-f32-faster-whisper-development"
+    amd: "rocm-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"
 - !!merge <<: *faster-whisper
  name: "cuda12-faster-whisper-development"
  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-12-faster-whisper"
@@ -274,10 +363,28 @@
    - text-to-speech
    - TTS
  license: mpl-2.0
+  name: "coqui"
+  alias: "coqui"
+  capabilities:
+    nvidia: "cuda12-coqui"
+    intel: "intel-sycl-f32-coqui"
+    amd: "rocm-coqui"
+  icon: https://avatars.githubusercontent.com/u/1338804?s=200&v=4
+- !!merge <<: *coqui
+  name: "coqui-development"
+  capabilities:
+    nvidia: "cuda12-coqui-development"
+    intel: "intel-sycl-f32-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"
+- !!merge <<: *coqui
+  name: "cuda12-coqui"
+  uri: "quay.io/go-skynet/local-ai-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"
-  alias: "coqui"
-  icon: https://avatars.githubusercontent.com/u/1338804?s=200&v=4
 - !!merge <<: *coqui
  name: "cuda12-coqui-development"
  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-12-coqui"
@@ -296,6 +403,9 @@
 - !!merge <<: *coqui
  name: "sycl-f16-coqui-development"
  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-intel-sycl-f16-coqui"
+- !!merge <<: *coqui
+  name: "rocm-coqui"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-rocm-hipblas-coqui"
 ## bark
 - &bark
  urls:
@@ -306,13 +416,25 @@
    - text-to-speech
    - TTS
  license: MIT
-  name: "cuda11-bark-development"
-  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-11-bark"
+  name: "bark"
  alias: "bark"
+  capabilities:
+    cuda: "cuda12-bark"
+    intel: "intel-sycl-f32-bark"
+    rocm: "rocm-bark"
  icon: https://avatars.githubusercontent.com/u/99442120?s=200&v=4
 - !!merge <<: *bark
-  name: "cuda12-bark-development"
-  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-12-bark"
+  name: "bark-development"
+  capabilities:
+    nvidia: "cuda12-bark-development"
+    intel: "intel-sycl-f32-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"
+- !!merge <<: *bark
+  name: "cuda11-bark"
+  uri: "quay.io/go-skynet/local-ai-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"
@@ -328,6 +450,15 @@
 - !!merge <<: *bark
  name: "sycl-f16-bark-development"
  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-intel-sycl-f16-bark"
+- !!merge <<: *bark
+  name: "cuda12-bark"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-nvidia-cuda-12-bark"
+- !!merge <<: *bark
+  name: "rocm-bark"
+  uri: "quay.io/go-skynet/local-ai-backends:latest-gpu-rocm-hipblas-bark"
+- !!merge <<: *bark
+  name: "cuda12-bark-development"
+  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-12-bark"
 - &barkcpp
  urls:
    - https://github.com/PABannier/bark.cpp
@@ -369,15 +500,22 @@
    - TTS
  license: MIT
  icon: https://private-user-images.githubusercontent.com/660224/448166653-bd8c5f03-e91d-4ee5-b680-57355da204d1.png?jwt=eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJpc3MiOiJnaXRodWIuY29tIiwiYXVkIjoicmF3LmdpdGh1YnVzZXJjb250ZW50LmNvbSIsImtleSI6ImtleTUiLCJleHAiOjE3NTAxOTE0MDAsIm5iZiI6MTc1MDE5MTEwMCwicGF0aCI6Ii82NjAyMjQvNDQ4MTY2NjUzLWJkOGM1ZjAzLWU5MWQtNGVlNS1iNjgwLTU3MzU1ZGEyMDRkMS5wbmc_WC1BbXotQWxnb3JpdGhtPUFXUzQtSE1BQy1TSEEyNTYmWC1BbXotQ3JlZGVudGlhbD1BS0lBVkNPRFlMU0E1M1BRSzRaQSUyRjIwMjUwNjE3JTJGdXMtZWFzdC0xJTJGczMlMkZhd3M0X3JlcXVlc3QmWC1BbXotRGF0ZT0yMDI1MDYxN1QyMDExNDBaJlgtQW16LUV4cGlyZXM9MzAwJlgtQW16LVNpZ25hdHVyZT1hMmI1NGY3OGFiZTlhNGFkNTVlYTY4NTIwMWEzODRiZGE4YzdhNGQ5MGNhNzE3MDYyYTA2NDIxYTkyYzhiODkwJlgtQW16LVNpZ25lZEhlYWRlcnM9aG9zdCJ9.mR9kM9xX0TdzPuSpuspCllHYQiq79dFQ2rtuNvjrl6w
-  name: "cuda11-chatterbox-development"
-  uri: "quay.io/go-skynet/local-ai-backends:master-gpu-nvidia-cuda-11-chatterbox"
-  alias: "chatterbox"
+  name: "chatterbox"
+  capabilities:
+    nvidia: "cuda12-chatterbox"
+- !!merge <<: *chatterbox
+  name: "chatterbox-development"
+  capabilities:
+    nvidia: "cuda12-chatterbox-development"
 - !!merge <<: *chatterbox
  name: "cuda12-chatterbox-development"
  uri: "quay.io/go-skynet/local-ai-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"
+- !!merge <<: *chatterbox
+  name: "cuda11-chatterbox-development"
+  uri: "quay.io/go-skynet/local-ai-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"
--- a/backend/python/diffusers/backend.py
+++ b/backend/python/diffusers/backend.py
@@ -38,9 +38,7 @@ DISABLE_CPU_OFFLOAD = os.environ.get("DISABLE_CPU_OFFLOAD", "0") == "1"
 FRAMES = os.environ.get("FRAMES", "64")

 if XPU:
-    import intel_extension_for_pytorch as ipex
-
-    print(ipex.xpu.get_device_name(0))
+    print(torch.xpu.get_device_name(0))

 # If MAX_WORKERS are specified in the environment use it, otherwise default to 1
 MAX_WORKERS = int(os.environ.get('PYTHON_GRPC_MAX_WORKERS', '1'))
@@ -336,6 +334,8 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
                request.LoraAdapter = os.path.join(request.ModelPath, request.LoraAdapter)

            device = "cpu" if not request.CUDA else "cuda"
+            if XPU:
+                device = "xpu"
            self.device = device
            if request.LoraAdapter:
                # Check if its a local file and not a directory ( we load lora differently for a safetensor file )
@@ -359,12 +359,11 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):

                self.pipe.set_adapters(adapters_name, adapter_weights=adapters_weights)

-            if request.CUDA:
-                self.pipe.to('cuda')
+            if device != "cpu":
+                self.pipe.to(device)
                if self.controlnet:
-                    self.controlnet.to('cuda')
-            if XPU:
-                self.pipe = self.pipe.to("xpu")
+                    self.controlnet.to(device)
+
        except Exception as err:
            return backend_pb2.Result(success=False, message=f"Unexpected {err=}, {type(err)=}")
        # Implement your logic here for the LoadModel service
--- a/backend/python/diffusers/run.sh
+++ b/backend/python/diffusers/run.sh
@@ -6,4 +6,10 @@ else
    source $backend_dir/../common/libbackend.sh
 fi

-startBackend $@
+if [ -d "/opt/intel" ]; then
+    # Assumes we are using the Intel oneAPI container image
+    # https://github.com/intel/intel-extension-for-pytorch/issues/538
+    export XPU=1
+fi
+
+startBackend $@
--- a/core/application/startup.go
+++ b/core/application/startup.go
@@ -57,7 +57,7 @@ func New(opts ...config.AppOption) (*Application, error) {
 		}
 	}

-	if err := pkgStartup.InstallModels(options.Galleries, options.ModelPath, options.EnforcePredownloadScans, nil, options.ModelsURL...); err != nil {
+	if err := pkgStartup.InstallModels(options.Galleries, options.BackendGalleries, options.ModelPath, options.BackendsPath, options.EnforcePredownloadScans, options.AutoloadBackendGalleries, nil, options.ModelsURL...); err != nil {
 		log.Error().Err(err).Msg("error installing models")
 	}

@@ -86,13 +86,13 @@ func New(opts ...config.AppOption) (*Application, error) {
 	}

 	if options.PreloadJSONModels != "" {
-		if err := services.ApplyGalleryFromString(options.ModelPath, options.PreloadJSONModels, options.EnforcePredownloadScans, options.Galleries); err != nil {
+		if err := services.ApplyGalleryFromString(options.ModelPath, options.BackendsPath, options.EnforcePredownloadScans, options.AutoloadBackendGalleries, options.Galleries, options.BackendGalleries, options.PreloadJSONModels); err != nil {
 			return nil, err
 		}
 	}

 	if options.PreloadModelsFromPath != "" {
-		if err := services.ApplyGalleryFromFile(options.ModelPath, options.PreloadModelsFromPath, options.EnforcePredownloadScans, options.Galleries); err != nil {
+		if err := services.ApplyGalleryFromFile(options.ModelPath, options.BackendsPath, options.EnforcePredownloadScans, options.AutoloadBackendGalleries, options.Galleries, options.BackendGalleries, options.PreloadModelsFromPath); err != nil {
 			return nil, err
 		}
 	}
--- a/core/backend/llm.go
+++ b/core/backend/llm.go
@@ -4,8 +4,8 @@ import (
 	"context"
 	"encoding/json"
 	"fmt"
-	"os"
 	"regexp"
+	"slices"
 	"strings"
 	"sync"
 	"unicode/utf8"
@@ -14,6 +14,7 @@ import (

 	"github.com/mudler/LocalAI/core/config"
 	"github.com/mudler/LocalAI/core/schema"
+	"github.com/mudler/LocalAI/core/services"

 	"github.com/mudler/LocalAI/core/gallery"
 	"github.com/mudler/LocalAI/pkg/grpc/proto"
@@ -34,17 +35,22 @@ type TokenUsage struct {
 	TimingTokenGeneration  float64
 }

-func ModelInference(ctx context.Context, s string, messages []schema.Message, images, videos, audios []string, loader *model.ModelLoader, c *config.BackendConfig, o *config.ApplicationConfig, tokenCallback func(string, TokenUsage) bool) (func() (LLMResponse, error), error) {
+func ModelInference(ctx context.Context, s string, messages []schema.Message, images, videos, audios []string, loader *model.ModelLoader, c *config.BackendConfig, cl *config.BackendConfigLoader, o *config.ApplicationConfig, tokenCallback func(string, TokenUsage) bool) (func() (LLMResponse, error), error) {
 	modelFile := c.Model

 	// Check if the modelFile exists, if it doesn't try to load it from the gallery
 	if o.AutoloadGalleries { // experimental
-		if _, err := os.Stat(modelFile); os.IsNotExist(err) {
+		modelNames, err := services.ListModels(cl, loader, nil, services.SKIP_ALWAYS)
+		if err != nil {
+			return nil, err
+		}
+		if !slices.Contains(modelNames, c.Name) {
 			utils.ResetDownloadTimers()
 			// if we failed to load the model, we try to download it
-			err := gallery.InstallModelFromGallery(o.Galleries, modelFile, loader.ModelPath, gallery.GalleryModel{}, utils.DisplayDownloadFunction, o.EnforcePredownloadScans)
+			err := gallery.InstallModelFromGallery(o.Galleries, o.BackendGalleries, c.Name, loader.ModelPath, o.BackendsPath, gallery.GalleryModel{}, utils.DisplayDownloadFunction, o.EnforcePredownloadScans, o.AutoloadBackendGalleries)
 			if err != nil {
-				return nil, err
+				log.Error().Err(err).Msgf("failed to install model %q from gallery", modelFile)
+				//return nil, err
 			}
 		}
 	}
--- a/core/backend/options.go
+++ b/core/backend/options.go
@@ -143,6 +143,7 @@ func grpcModelOpts(c config.BackendConfig) *pb.ModelOptions {
 		CLIPModel:            c.Diffusers.ClipModel,
 		CLIPSubfolder:        c.Diffusers.ClipSubFolder,
 		Options:              c.Options,
+		Overrides:            c.Overrides,
 		CLIPSkip:             int32(c.Diffusers.ClipSkip),
 		ControlNet:           c.Diffusers.ControlNet,
 		ContextSize:          int32(ctxSize),
--- a/core/cli/backends.go
+++ b/core/cli/backends.go
@@ -0,0 +1,156 @@
+package cli
+
+import (
+	"encoding/json"
+	"fmt"
+
+	cliContext "github.com/mudler/LocalAI/core/cli/context"
+	"github.com/mudler/LocalAI/core/config"
+
+	"github.com/mudler/LocalAI/core/gallery"
+	"github.com/mudler/LocalAI/pkg/downloader"
+	"github.com/mudler/LocalAI/pkg/startup"
+	"github.com/rs/zerolog/log"
+	"github.com/schollz/progressbar/v3"
+)
+
+type BackendsCMDFlags struct {
+	BackendGalleries string `env:"LOCALAI_BACKEND_GALLERIES,BACKEND_GALLERIES" help:"JSON list of backend galleries" group:"backends" default:"${backends}"`
+	BackendsPath     string `env:"LOCALAI_BACKENDS_PATH,BACKENDS_PATH" type:"path" default:"${basepath}/backends" help:"Path containing backends used for inferencing" group:"storage"`
+}
+
+type BackendsList struct {
+	BackendsCMDFlags `embed:""`
+}
+
+type BackendsInstallSingle struct {
+	InstallArgs []string `arg:"" optional:"" name:"backend" help:"Backend images to install"`
+
+	BackendsCMDFlags `embed:""`
+}
+
+type BackendsInstall struct {
+	BackendArgs []string `arg:"" optional:"" name:"backends" help:"Backend configuration URLs to load"`
+
+	BackendsCMDFlags `embed:""`
+}
+
+type BackendsUninstall struct {
+	BackendArgs []string `arg:"" name:"backends" help:"Backend names to uninstall"`
+
+	BackendsCMDFlags `embed:""`
+}
+
+type BackendsCMD struct {
+	List          BackendsList          `cmd:"" help:"List the backends available in your galleries" default:"withargs"`
+	Install       BackendsInstall       `cmd:"" help:"Install a backend from the gallery"`
+	InstallSingle BackendsInstallSingle `cmd:"" help:"Install a single backend from the gallery"`
+	Uninstall     BackendsUninstall     `cmd:"" help:"Uninstall a backend"`
+}
+
+func (bi *BackendsInstallSingle) Run(ctx *cliContext.Context) error {
+	for _, backend := range bi.InstallArgs {
+		progressBar := progressbar.NewOptions(
+			1000,
+			progressbar.OptionSetDescription(fmt.Sprintf("downloading backend %s", backend)),
+			progressbar.OptionShowBytes(false),
+			progressbar.OptionClearOnFinish(),
+		)
+		progressCallback := func(fileName string, current string, total string, percentage float64) {
+			v := int(percentage * 10)
+			err := progressBar.Set(v)
+			if err != nil {
+				log.Error().Err(err).Str("filename", fileName).Int("value", v).Msg("error while updating progress bar")
+			}
+		}
+
+		if err := gallery.InstallBackend(bi.BackendsPath, &gallery.GalleryBackend{
+			URI: backend,
+		}, progressCallback); err != nil {
+			return err
+		}
+	}
+
+	return nil
+}
+
+func (bl *BackendsList) Run(ctx *cliContext.Context) error {
+	var galleries []config.Gallery
+	if err := json.Unmarshal([]byte(bl.BackendGalleries), &galleries); err != nil {
+		log.Error().Err(err).Msg("unable to load galleries")
+	}
+
+	backends, err := gallery.AvailableBackends(galleries, bl.BackendsPath)
+	if err != nil {
+		return err
+	}
+	for _, backend := range backends {
+		if backend.Installed {
+			fmt.Printf(" * %s@%s (installed)\n", backend.Gallery.Name, backend.Name)
+		} else {
+			fmt.Printf(" - %s@%s\n", backend.Gallery.Name, backend.Name)
+		}
+	}
+	return nil
+}
+
+func (bi *BackendsInstall) Run(ctx *cliContext.Context) error {
+	var galleries []config.Gallery
+	if err := json.Unmarshal([]byte(bi.BackendGalleries), &galleries); err != nil {
+		log.Error().Err(err).Msg("unable to load galleries")
+	}
+
+	for _, backendName := range bi.BackendArgs {
+
+		progressBar := progressbar.NewOptions(
+			1000,
+			progressbar.OptionSetDescription(fmt.Sprintf("downloading backend %s", backendName)),
+			progressbar.OptionShowBytes(false),
+			progressbar.OptionClearOnFinish(),
+		)
+		progressCallback := func(fileName string, current string, total string, percentage float64) {
+			v := int(percentage * 10)
+			err := progressBar.Set(v)
+			if err != nil {
+				log.Error().Err(err).Str("filename", fileName).Int("value", v).Msg("error while updating progress bar")
+			}
+		}
+
+		backendURI := downloader.URI(backendName)
+
+		if !backendURI.LooksLikeOCI() {
+			backends, err := gallery.AvailableBackends(galleries, bi.BackendsPath)
+			if err != nil {
+				return err
+			}
+
+			backend := gallery.FindGalleryElement(backends, backendName, bi.BackendsPath)
+			if backend == nil {
+				log.Error().Str("backend", backendName).Msg("backend not found")
+				return fmt.Errorf("backend not found: %s", backendName)
+			}
+
+			log.Info().Str("backend", backendName).Str("license", backend.License).Msg("installing backend")
+		}
+
+		err := startup.InstallExternalBackends(galleries, bi.BackendsPath, progressCallback, backendName)
+		if err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+func (bu *BackendsUninstall) Run(ctx *cliContext.Context) error {
+	for _, backendName := range bu.BackendArgs {
+		log.Info().Str("backend", backendName).Msg("uninstalling backend")
+
+		err := gallery.DeleteBackendFromSystem(bu.BackendsPath, backendName)
+		if err != nil {
+			return err
+		}
+
+		fmt.Printf("Backend %s uninstalled successfully\n", backendName)
+	}
+	return nil
+}
--- a/core/cli/cli.go
+++ b/core/cli/cli.go
@@ -11,6 +11,7 @@ var CLI struct {
 	Run             RunCMD             `cmd:"" help:"Run LocalAI, this the default command if no other command is specified. Run 'local-ai run --help' for more information" default:"withargs"`
 	Federated       FederatedCLI       `cmd:"" help:"Run LocalAI in federated mode"`
 	Models          ModelsCMD          `cmd:"" help:"Manage LocalAI models and definitions"`
+	Backends        BackendsCMD        `cmd:"" help:"Manage LocalAI backends and definitions"`
 	TTS             TTSCMD             `cmd:"" help:"Convert text to speech"`
 	SoundGeneration SoundGenerationCMD `cmd:"" help:"Generates audio files from text or audio"`
 	Transcript      TranscriptCMD      `cmd:"" help:"Convert audio to text"`
--- a/core/cli/models.go
+++ b/core/cli/models.go
@@ -16,8 +16,10 @@ import (
 )

 type ModelsCMDFlags struct {
-	Galleries  string `env:"LOCALAI_GALLERIES,GALLERIES" help:"JSON list of galleries" group:"models" default:"${galleries}"`
-	ModelsPath string `env:"LOCALAI_MODELS_PATH,MODELS_PATH" type:"path" default:"${basepath}/models" help:"Path containing models used for inferencing" group:"storage"`
+	Galleries        string `env:"LOCALAI_GALLERIES,GALLERIES" help:"JSON list of galleries" group:"models" default:"${galleries}"`
+	BackendGalleries string `env:"LOCALAI_BACKEND_GALLERIES,BACKEND_GALLERIES" help:"JSON list of backend galleries" group:"backends" default:"${backends}"`
+	ModelsPath       string `env:"LOCALAI_MODELS_PATH,MODELS_PATH" type:"path" default:"${basepath}/models" help:"Path containing models used for inferencing" group:"storage"`
+	BackendsPath     string `env:"LOCALAI_BACKENDS_PATH,BACKENDS_PATH" type:"path" default:"${basepath}/backends" help:"Path containing backends used for inferencing" group:"storage"`
 }

 type ModelsList struct {
@@ -25,8 +27,9 @@ type ModelsList struct {
 }

 type ModelsInstall struct {
-	DisablePredownloadScan bool     `env:"LOCALAI_DISABLE_PREDOWNLOAD_SCAN" help:"If true, disables the best-effort security scanner before downloading any files." group:"hardening" default:"false"`
-	ModelArgs              []string `arg:"" optional:"" name:"models" help:"Model configuration URLs to load"`
+	DisablePredownloadScan   bool     `env:"LOCALAI_DISABLE_PREDOWNLOAD_SCAN" help:"If true, disables the best-effort security scanner before downloading any files." group:"hardening" default:"false"`
+	AutoloadBackendGalleries bool     `env:"LOCALAI_AUTOLOAD_BACKEND_GALLERIES" help:"If true, automatically loads backend galleries" group:"backends" default:"true"`
+	ModelArgs                []string `arg:"" optional:"" name:"models" help:"Model configuration URLs to load"`

 	ModelsCMDFlags `embed:""`
 }
@@ -62,6 +65,11 @@ func (mi *ModelsInstall) Run(ctx *cliContext.Context) error {
 		log.Error().Err(err).Msg("unable to load galleries")
 	}

+	var backendGalleries []config.Gallery
+	if err := json.Unmarshal([]byte(mi.BackendGalleries), &backendGalleries); err != nil {
+		log.Error().Err(err).Msg("unable to load backend galleries")
+	}
+
 	for _, modelName := range mi.ModelArgs {

 		progressBar := progressbar.NewOptions(
@@ -100,7 +108,7 @@ func (mi *ModelsInstall) Run(ctx *cliContext.Context) error {
 			log.Info().Str("model", modelName).Str("license", model.License).Msg("installing model")
 		}

-		err = startup.InstallModels(galleries, mi.ModelsPath, !mi.DisablePredownloadScan, progressCallback, modelName)
+		err = startup.InstallModels(galleries, backendGalleries, mi.ModelsPath, mi.BackendsPath, !mi.DisablePredownloadScan, mi.AutoloadBackendGalleries, progressCallback, modelName)
 		if err != nil {
 			return err
 		}
--- a/core/cli/run.go
+++ b/core/cli/run.go
@@ -30,13 +30,14 @@ type RunCMD struct {
 	LocalaiConfigDir             string        `env:"LOCALAI_CONFIG_DIR" type:"path" default:"${basepath}/configuration" help:"Directory for dynamic loading of certain configuration files (currently api_keys.json and external_backends.json)" group:"storage"`
 	LocalaiConfigDirPollInterval time.Duration `env:"LOCALAI_CONFIG_DIR_POLL_INTERVAL" help:"Typically the config path picks up changes automatically, but if your system has broken fsnotify events, set this to an interval to poll the LocalAI Config Dir (example: 1m)" group:"storage"`
 	// The alias on this option is there to preserve functionality with the old `--config-file` parameter
-	ModelsConfigFile    string   `env:"LOCALAI_MODELS_CONFIG_FILE,CONFIG_FILE" aliases:"config-file" help:"YAML file containing a list of model backend configs" group:"storage"`
-	BackendGalleries    string   `env:"LOCALAI_BACKEND_GALLERIES,BACKEND_GALLERIES" help:"JSON list of backend galleries" group:"backends" default:"${backends}"`
-	Galleries           string   `env:"LOCALAI_GALLERIES,GALLERIES" help:"JSON list of galleries" group:"models" default:"${galleries}"`
-	AutoloadGalleries   bool     `env:"LOCALAI_AUTOLOAD_GALLERIES,AUTOLOAD_GALLERIES" group:"models"`
-	PreloadModels       string   `env:"LOCALAI_PRELOAD_MODELS,PRELOAD_MODELS" help:"A List of models to apply in JSON at start" group:"models"`
-	Models              []string `env:"LOCALAI_MODELS,MODELS" help:"A List of model configuration URLs to load" group:"models"`
-	PreloadModelsConfig string   `env:"LOCALAI_PRELOAD_MODELS_CONFIG,PRELOAD_MODELS_CONFIG" help:"A List of models to apply at startup. Path to a YAML config file" group:"models"`
+	ModelsConfigFile         string   `env:"LOCALAI_MODELS_CONFIG_FILE,CONFIG_FILE" aliases:"config-file" help:"YAML file containing a list of model backend configs" group:"storage"`
+	BackendGalleries         string   `env:"LOCALAI_BACKEND_GALLERIES,BACKEND_GALLERIES" help:"JSON list of backend galleries" group:"backends" default:"${backends}"`
+	Galleries                string   `env:"LOCALAI_GALLERIES,GALLERIES" help:"JSON list of galleries" group:"models" default:"${galleries}"`
+	AutoloadGalleries        bool     `env:"LOCALAI_AUTOLOAD_GALLERIES,AUTOLOAD_GALLERIES" group:"models" default:"true"`
+	AutoloadBackendGalleries bool     `env:"LOCALAI_AUTOLOAD_BACKEND_GALLERIES,AUTOLOAD_BACKEND_GALLERIES" group:"backends" default:"true"`
+	PreloadModels            string   `env:"LOCALAI_PRELOAD_MODELS,PRELOAD_MODELS" help:"A List of models to apply in JSON at start" group:"models"`
+	Models                   []string `env:"LOCALAI_MODELS,MODELS" help:"A List of model configuration URLs to load" group:"models"`
+	PreloadModelsConfig      string   `env:"LOCALAI_PRELOAD_MODELS_CONFIG,PRELOAD_MODELS_CONFIG" help:"A List of models to apply at startup. Path to a YAML config file" group:"models"`

 	F16         bool `name:"f16" env:"LOCALAI_F16,F16" help:"Enable GPU acceleration" group:"performance"`
 	Threads     int  `env:"LOCALAI_THREADS,THREADS" short:"t" help:"Number of threads used for parallel computation. Usage of the number of physical cores in the system is suggested" group:"performance"`
@@ -192,6 +193,10 @@ func (r *RunCMD) Run(ctx *cliContext.Context) error {
 		opts = append(opts, config.EnableGalleriesAutoload)
 	}

+	if r.AutoloadBackendGalleries {
+		opts = append(opts, config.EnableBackendGalleriesAutoload)
+	}
+
 	if r.PreloadBackendOnly {
 		_, err := application.New(opts...)
 		return err
--- a/core/config/application_config.go
+++ b/core/config/application_config.go
@@ -55,7 +55,7 @@ type ApplicationConfig struct {

 	ExternalGRPCBackends map[string]string

-	AutoloadGalleries bool
+	AutoloadGalleries, AutoloadBackendGalleries bool

 	SingleBackend           bool
 	ParallelBackendRequests bool
@@ -192,6 +192,10 @@ var EnableGalleriesAutoload = func(o *ApplicationConfig) {
 	o.AutoloadGalleries = true
 }

+var EnableBackendGalleriesAutoload = func(o *ApplicationConfig) {
+	o.AutoloadBackendGalleries = true
+}
+
 func WithExternalBackend(name string, uri string) AppOption {
 	return func(o *ApplicationConfig) {
 		if o.ExternalGRPCBackends == nil {
--- a/core/config/backend_config.go
+++ b/core/config/backend_config.go
@@ -70,7 +70,8 @@ type BackendConfig struct {
 	Description string `yaml:"description"`
 	Usage       string `yaml:"usage"`

-	Options []string `yaml:"options"`
+	Options   []string `yaml:"options"`
+	Overrides []string `yaml:"overrides"`
 }

 // Pipeline defines other models to use for audio-to-audio
--- a/core/gallery/backend_types.go
+++ b/core/gallery/backend_types.go
@@ -30,7 +30,7 @@ func (m *GalleryBackend) SetGallery(gallery config.Gallery) {
 }

 func (m *GalleryBackend) IsMeta() bool {
-	return len(m.CapabilitiesMap) > 0
+	return len(m.CapabilitiesMap) > 0 && m.URI == ""
 }

 func (m *GalleryBackend) SetInstalled(installed bool) {
--- a/core/gallery/backends.go
+++ b/core/gallery/backends.go
@@ -62,7 +62,7 @@ func findBestBackendFromMeta(backend *GalleryBackend, systemState *system.System
 		return nil
 	}

-	realBackend := backend.CapabilitiesMap[systemState.GPUVendor]
+	realBackend := backend.CapabilitiesMap[systemState.Capability()]
 	if realBackend == "" {
 		return nil
 	}
@@ -71,7 +71,22 @@ func findBestBackendFromMeta(backend *GalleryBackend, systemState *system.System
 }

 // Installs a model from the gallery
-func InstallBackendFromGallery(galleries []config.Gallery, systemState *system.SystemState, name string, basePath string, downloadStatus func(string, string, string, float64)) error {
+func InstallBackendFromGallery(galleries []config.Gallery, systemState *system.SystemState, name string, basePath string, downloadStatus func(string, string, string, float64), force bool) error {
+	if !force {
+		// check if we already have the backend installed
+		backends, err := ListSystemBackends(basePath)
+		if err != nil {
+			return err
+		}
+		if _, ok := backends[name]; ok {
+			return nil
+		}
+	}
+
+	if name == "" {
+		return fmt.Errorf("backend name is empty")
+	}
+
 	log.Debug().Interface("galleries", galleries).Str("name", name).Msg("Installing backend from gallery")

 	backends, err := AvailableBackends(galleries, basePath)
@@ -147,7 +162,7 @@ func InstallBackend(basePath string, config *GalleryBackend, downloadStatus func
 		return fmt.Errorf("failed to create backend path %q: %v", backendPath, err)
 	}

-	if err := oci.ExtractOCIImage(img, backendPath, downloadStatus); err != nil {
+	if err := oci.ExtractOCIImage(img, config.URI, backendPath, downloadStatus); err != nil {
 		return fmt.Errorf("failed to extract image %q: %v", config.URI, err)
 	}

@@ -231,14 +246,32 @@ func ListSystemBackends(basePath string) (map[string]string, error) {
 	for _, backend := range backends {
 		if backend.IsDir() {
 			runFile := filepath.Join(basePath, backend.Name(), runFile)
-			backendsNames[backend.Name()] = runFile
+			// Skip if metadata file don't exist
+			metadataFilePath := filepath.Join(basePath, backend.Name(), metadataFile)
+			if _, err := os.Stat(metadataFilePath); os.IsNotExist(err) {
+				continue
+			}

 			// Check for alias in metadata
 			metadata, err := readBackendMetadata(filepath.Join(basePath, backend.Name()))
 			if err != nil {
 				return nil, err
 			}
-			if metadata != nil && metadata.Alias != "" {
+
+			if metadata == nil {
+				continue
+			}
+
+			if _, exists := backendsNames[backend.Name()]; !exists {
+				// We don't want to override aliases if already set, and if we are meta backend
+				if _, err := os.Stat(runFile); err == nil {
+					backendsNames[backend.Name()] = runFile
+				} else {
+					backendsNames[backend.Name()] = ""
+				}
+			}
+
+			if metadata.Alias != "" {
 				backendsNames[metadata.Alias] = runFile
 			}
 		}
--- a/core/gallery/backends_test.go
+++ b/core/gallery/backends_test.go
@@ -42,13 +42,13 @@ var _ = Describe("Gallery Backends", func() {

 	Describe("InstallBackendFromGallery", func() {
 		It("should return error when backend is not found", func() {
-			err := InstallBackendFromGallery(galleries, nil, "non-existent", tempDir, nil)
+			err := InstallBackendFromGallery(galleries, nil, "non-existent", tempDir, nil, true)
 			Expect(err).To(HaveOccurred())
 			Expect(err.Error()).To(ContainSubstring("no backend found with name \"non-existent\""))
 		})

 		It("should install backend from gallery", func() {
-			err := InstallBackendFromGallery(galleries, nil, "test-backend", tempDir, nil)
+			err := InstallBackendFromGallery(galleries, nil, "test-backend", tempDir, nil, true)
 			Expect(err).ToNot(HaveOccurred())
 			Expect(filepath.Join(tempDir, "test-backend", "run.sh")).To(BeARegularFile())
 		})
@@ -181,7 +181,7 @@ var _ = Describe("Gallery Backends", func() {

 			// Test with NVIDIA system state
 			nvidiaSystemState := &system.SystemState{GPUVendor: "nvidia"}
-			err = InstallBackendFromGallery([]config.Gallery{gallery}, nvidiaSystemState, "meta-backend", tempDir, nil)
+			err = InstallBackendFromGallery([]config.Gallery{gallery}, nvidiaSystemState, "meta-backend", tempDir, nil, true)
 			Expect(err).NotTo(HaveOccurred())

 			metaBackendPath := filepath.Join(tempDir, "meta-backend")
@@ -206,6 +206,144 @@ var _ = Describe("Gallery Backends", func() {
 			Expect(concreteBackendPath).NotTo(BeADirectory())
 		})

+		It("should handle meta backend deletion correctly with aliases", func() {
+			metaBackend := &GalleryBackend{
+				Metadata: Metadata{
+					Name: "meta-backend",
+				},
+				Alias: "backend-alias",
+				CapabilitiesMap: map[string]string{
+					"nvidia": "nvidia-backend",
+					"amd":    "amd-backend",
+					"intel":  "intel-backend",
+				},
+			}
+
+			nvidiaBackend := &GalleryBackend{
+				Metadata: Metadata{
+					Name: "nvidia-backend",
+				},
+				Alias: "backend-alias",
+				URI:   testImage,
+			}
+
+			amdBackend := &GalleryBackend{
+				Metadata: Metadata{
+					Name: "amd-backend",
+				},
+				Alias: "backend-alias",
+				URI:   testImage,
+			}
+
+			gallery := config.Gallery{
+				Name: "test-gallery",
+				URL:  "file://" + filepath.Join(tempDir, "backend-gallery.yaml"),
+			}
+
+			galleryBackend := GalleryBackends{amdBackend, nvidiaBackend, metaBackend}
+
+			dat, err := yaml.Marshal(galleryBackend)
+			Expect(err).NotTo(HaveOccurred())
+			err = os.WriteFile(filepath.Join(tempDir, "backend-gallery.yaml"), dat, 0644)
+			Expect(err).NotTo(HaveOccurred())
+
+			// Test with NVIDIA system state
+			nvidiaSystemState := &system.SystemState{GPUVendor: "nvidia"}
+			err = InstallBackendFromGallery([]config.Gallery{gallery}, nvidiaSystemState, "meta-backend", tempDir, nil, true)
+			Expect(err).NotTo(HaveOccurred())
+
+			metaBackendPath := filepath.Join(tempDir, "meta-backend")
+			Expect(metaBackendPath).To(BeADirectory())
+
+			concreteBackendPath := filepath.Join(tempDir, "nvidia-backend")
+			Expect(concreteBackendPath).To(BeADirectory())
+
+			allBackends, err := ListSystemBackends(tempDir)
+			Expect(err).NotTo(HaveOccurred())
+			Expect(allBackends).To(HaveKey("meta-backend"))
+			Expect(allBackends).To(HaveKey("nvidia-backend"))
+			Expect(allBackends["meta-backend"]).To(BeEmpty())
+
+			// Delete meta backend by name
+			err = DeleteBackendFromSystem(tempDir, "meta-backend")
+			Expect(err).NotTo(HaveOccurred())
+
+			// Verify meta backend directory is deleted
+			Expect(metaBackendPath).NotTo(BeADirectory())
+
+			// Verify concrete backend directory is deleted
+			Expect(concreteBackendPath).NotTo(BeADirectory())
+		})
+
+		It("should handle meta backend deletion correctly with aliases pointing to the same backend", func() {
+			metaBackend := &GalleryBackend{
+				Metadata: Metadata{
+					Name: "meta-backend",
+				},
+				Alias: "meta-backend",
+				CapabilitiesMap: map[string]string{
+					"nvidia": "nvidia-backend",
+					"amd":    "amd-backend",
+					"intel":  "intel-backend",
+				},
+			}
+
+			nvidiaBackend := &GalleryBackend{
+				Metadata: Metadata{
+					Name: "nvidia-backend",
+				},
+				Alias: "meta-backend",
+				URI:   testImage,
+			}
+
+			amdBackend := &GalleryBackend{
+				Metadata: Metadata{
+					Name: "amd-backend",
+				},
+				Alias: "meta-backend",
+				URI:   testImage,
+			}
+
+			gallery := config.Gallery{
+				Name: "test-gallery",
+				URL:  "file://" + filepath.Join(tempDir, "backend-gallery.yaml"),
+			}
+
+			galleryBackend := GalleryBackends{amdBackend, nvidiaBackend, metaBackend}
+
+			dat, err := yaml.Marshal(galleryBackend)
+			Expect(err).NotTo(HaveOccurred())
+			err = os.WriteFile(filepath.Join(tempDir, "backend-gallery.yaml"), dat, 0644)
+			Expect(err).NotTo(HaveOccurred())
+
+			// Test with NVIDIA system state
+			nvidiaSystemState := &system.SystemState{GPUVendor: "nvidia"}
+			err = InstallBackendFromGallery([]config.Gallery{gallery}, nvidiaSystemState, "meta-backend", tempDir, nil, true)
+			Expect(err).NotTo(HaveOccurred())
+
+			metaBackendPath := filepath.Join(tempDir, "meta-backend")
+			Expect(metaBackendPath).To(BeADirectory())
+
+			concreteBackendPath := filepath.Join(tempDir, "nvidia-backend")
+			Expect(concreteBackendPath).To(BeADirectory())
+
+			allBackends, err := ListSystemBackends(tempDir)
+			Expect(err).NotTo(HaveOccurred())
+			Expect(allBackends).To(HaveKey("meta-backend"))
+			Expect(allBackends).To(HaveKey("nvidia-backend"))
+			Expect(allBackends["meta-backend"]).To(Equal(filepath.Join(tempDir, "nvidia-backend", "run.sh")))
+
+			// Delete meta backend by name
+			err = DeleteBackendFromSystem(tempDir, "meta-backend")
+			Expect(err).NotTo(HaveOccurred())
+
+			// Verify meta backend directory is deleted
+			Expect(metaBackendPath).NotTo(BeADirectory())
+
+			// Verify concrete backend directory is deleted
+			Expect(concreteBackendPath).NotTo(BeADirectory())
+		})
+
 		It("should list meta backends correctly in system backends", func() {
 			// Create a meta backend directory with metadata
 			metaBackendPath := filepath.Join(tempDir, "meta-backend")
@@ -227,7 +365,9 @@ var _ = Describe("Gallery Backends", func() {
 			concreteBackendPath := filepath.Join(tempDir, "concrete-backend")
 			err = os.MkdirAll(concreteBackendPath, 0750)
 			Expect(err).NotTo(HaveOccurred())
-			err = os.WriteFile(filepath.Join(concreteBackendPath, "run.sh"), []byte("#!/bin/bash"), 0755)
+			err = os.WriteFile(filepath.Join(concreteBackendPath, "metadata.json"), []byte("{}"), 0755)
+			Expect(err).NotTo(HaveOccurred())
+			err = os.WriteFile(filepath.Join(concreteBackendPath, "run.sh"), []byte(""), 0755)
 			Expect(err).NotTo(HaveOccurred())

 			// List system backends
@@ -238,8 +378,8 @@ var _ = Describe("Gallery Backends", func() {
 			Expect(backends).To(HaveKey("meta-backend"))
 			Expect(backends).To(HaveKey("concrete-backend"))

-			// meta-backend should point to its own run.sh
-			Expect(backends["meta-backend"]).To(Equal(filepath.Join(tempDir, "meta-backend", "run.sh")))
+			// meta-backend should be empty
+			Expect(backends["meta-backend"]).To(BeEmpty())
 			// concrete-backend should point to its own run.sh
 			Expect(backends["concrete-backend"]).To(Equal(filepath.Join(tempDir, "concrete-backend", "run.sh")))
 		})
@@ -319,6 +459,10 @@ var _ = Describe("Gallery Backends", func() {
 			for _, name := range backendNames {
 				err := os.MkdirAll(filepath.Join(tempDir, name), 0750)
 				Expect(err).NotTo(HaveOccurred())
+				err = os.WriteFile(filepath.Join(tempDir, name, "metadata.json"), []byte("{}"), 0755)
+				Expect(err).NotTo(HaveOccurred())
+				err = os.WriteFile(filepath.Join(tempDir, name, "run.sh"), []byte(""), 0755)
+				Expect(err).NotTo(HaveOccurred())
 			}

 			backends, err := ListSystemBackends(tempDir)
@@ -349,6 +493,8 @@ var _ = Describe("Gallery Backends", func() {
 			Expect(err).NotTo(HaveOccurred())
 			err = os.WriteFile(filepath.Join(backendPath, "metadata.json"), metadataData, 0644)
 			Expect(err).NotTo(HaveOccurred())
+			err = os.WriteFile(filepath.Join(backendPath, "run.sh"), []byte(""), 0755)
+			Expect(err).NotTo(HaveOccurred())

 			backends, err := ListSystemBackends(tempDir)
 			Expect(err).NotTo(HaveOccurred())
--- a/core/gallery/gallery.go
+++ b/core/gallery/gallery.go
@@ -121,7 +121,12 @@ func AvailableGalleryModels(galleries []config.Gallery, basePath string) (Galler

 	// Get models from galleries
 	for _, gallery := range galleries {
-		galleryModels, err := getGalleryElements[*GalleryModel](gallery, basePath)
+		galleryModels, err := getGalleryElements[*GalleryModel](gallery, basePath, func(model *GalleryModel) bool {
+			if _, err := os.Stat(filepath.Join(basePath, fmt.Sprintf("%s.yaml", model.GetName()))); err == nil {
+				return true
+			}
+			return false
+		})
 		if err != nil {
 			return nil, err
 		}
@@ -137,7 +142,14 @@ func AvailableBackends(galleries []config.Gallery, basePath string) (GalleryElem

 	// Get models from galleries
 	for _, gallery := range galleries {
-		galleryModels, err := getGalleryElements[*GalleryBackend](gallery, basePath)
+		galleryModels, err := getGalleryElements[*GalleryBackend](gallery, basePath, func(backend *GalleryBackend) bool {
+			backends, err := ListSystemBackends(basePath)
+			if err != nil {
+				return false
+			}
+			_, exists := backends[backend.GetName()]
+			return exists
+		})
 		if err != nil {
 			return nil, err
 		}
@@ -162,7 +174,7 @@ func findGalleryURLFromReferenceURL(url string, basePath string) (string, error)
 	return refFile, err
 }

-func getGalleryElements[T GalleryElement](gallery config.Gallery, basePath string) ([]T, error) {
+func getGalleryElements[T GalleryElement](gallery config.Gallery, basePath string, isInstalledCallback func(T) bool) ([]T, error) {
 	var models []T = []T{}

 	if strings.HasSuffix(gallery.URL, ".ref") {
@@ -187,15 +199,7 @@ func getGalleryElements[T GalleryElement](gallery config.Gallery, basePath strin
 	// Add gallery to models
 	for _, model := range models {
 		model.SetGallery(gallery)
-		// we check if the model was already installed by checking if the config file exists
-		// TODO: (what to do if the model doesn't install a config file?)
-		// TODO: This is sub-optimal now that the gallery handles both backends and models - we need to abstract this away
-		if _, err := os.Stat(filepath.Join(basePath, fmt.Sprintf("%s.yaml", model.GetName()))); err == nil {
-			model.SetInstalled(true)
-		}
-		if _, err := os.Stat(filepath.Join(basePath, model.GetName())); err == nil {
-			model.SetInstalled(true)
-		}
+		model.SetInstalled(isInstalledCallback(model))
 	}
 	return models, nil
 }
--- a/core/gallery/models.go
+++ b/core/gallery/models.go
@@ -10,6 +10,7 @@ import (
 	"dario.cat/mergo"
 	"github.com/mudler/LocalAI/core/config"
 	lconfig "github.com/mudler/LocalAI/core/config"
+	"github.com/mudler/LocalAI/core/system"
 	"github.com/mudler/LocalAI/pkg/downloader"
 	"github.com/mudler/LocalAI/pkg/utils"

@@ -69,7 +70,9 @@ type PromptTemplate struct {
 }

 // Installs a model from the gallery
-func InstallModelFromGallery(galleries []config.Gallery, name string, basePath string, req GalleryModel, downloadStatus func(string, string, string, float64), enforceScan bool) error {
+func InstallModelFromGallery(
+	modelGalleries, backendGalleries []config.Gallery,
+	name string, basePath, backendBasePath string, req GalleryModel, downloadStatus func(string, string, string, float64), enforceScan, automaticallyInstallBackend bool) error {

 	applyModel := func(model *GalleryModel) error {
 		name = strings.ReplaceAll(name, string(os.PathSeparator), "__")
@@ -119,14 +122,26 @@ func InstallModelFromGallery(galleries []config.Gallery, name string, basePath s
 			return err
 		}

-		if err := InstallModel(basePath, installName, &config, model.Overrides, downloadStatus, enforceScan); err != nil {
+		installedModel, err := InstallModel(basePath, installName, &config, model.Overrides, downloadStatus, enforceScan)
+		if err != nil {
 			return err
 		}

+		if automaticallyInstallBackend && installedModel.Backend != "" {
+			systemState, err := system.GetSystemState()
+			if err != nil {
+				return err
+			}
+
+			if err := InstallBackendFromGallery(backendGalleries, systemState, installedModel.Backend, backendBasePath, downloadStatus, false); err != nil {
+				return err
+			}
+		}
+
 		return nil
 	}

-	models, err := AvailableGalleryModels(galleries, basePath)
+	models, err := AvailableGalleryModels(modelGalleries, basePath)
 	if err != nil {
 		return err
 	}
@@ -139,11 +154,11 @@ func InstallModelFromGallery(galleries []config.Gallery, name string, basePath s
 	return applyModel(model)
 }

-func InstallModel(basePath, nameOverride string, config *ModelConfig, configOverrides map[string]interface{}, downloadStatus func(string, string, string, float64), enforceScan bool) error {
+func InstallModel(basePath, nameOverride string, config *ModelConfig, configOverrides map[string]interface{}, downloadStatus func(string, string, string, float64), enforceScan bool) (*lconfig.BackendConfig, error) {
 	// Create base path if it doesn't exist
 	err := os.MkdirAll(basePath, 0750)
 	if err != nil {
-		return fmt.Errorf("failed to create base path: %v", err)
+		return nil, fmt.Errorf("failed to create base path: %v", err)
 	}

 	if len(configOverrides) > 0 {
@@ -155,7 +170,7 @@ func InstallModel(basePath, nameOverride string, config *ModelConfig, configOver
 		log.Debug().Msgf("Checking %q exists and matches SHA", file.Filename)

 		if err := utils.VerifyPath(file.Filename, basePath); err != nil {
-			return err
+			return nil, err
 		}

 		// Create file path
@@ -165,19 +180,19 @@ func InstallModel(basePath, nameOverride string, config *ModelConfig, configOver
 			scanResults, err := downloader.HuggingFaceScan(downloader.URI(file.URI))
 			if err != nil && errors.Is(err, downloader.ErrUnsafeFilesFound) {
 				log.Error().Str("model", config.Name).Strs("clamAV", scanResults.ClamAVInfectedFiles).Strs("pickles", scanResults.DangerousPickles).Msg("Contains unsafe file(s)!")
-				return err
+				return nil, err
 			}
 		}
 		uri := downloader.URI(file.URI)
 		if err := uri.DownloadFile(filePath, file.SHA256, i, len(config.Files), downloadStatus); err != nil {
-			return err
+			return nil, err
 		}
 	}

 	// Write prompt template contents to separate files
 	for _, template := range config.PromptTemplates {
 		if err := utils.VerifyPath(template.Name+".tmpl", basePath); err != nil {
-			return err
+			return nil, err
 		}
 		// Create file path
 		filePath := filepath.Join(basePath, template.Name+".tmpl")
@@ -185,12 +200,12 @@ func InstallModel(basePath, nameOverride string, config *ModelConfig, configOver
 		// Create parent directory
 		err := os.MkdirAll(filepath.Dir(filePath), 0750)
 		if err != nil {
-			return fmt.Errorf("failed to create parent directory for prompt template %q: %v", template.Name, err)
+			return nil, fmt.Errorf("failed to create parent directory for prompt template %q: %v", template.Name, err)
 		}
 		// Create and write file content
 		err = os.WriteFile(filePath, []byte(template.Content), 0600)
 		if err != nil {
-			return fmt.Errorf("failed to write prompt template %q: %v", template.Name, err)
+			return nil, fmt.Errorf("failed to write prompt template %q: %v", template.Name, err)
 		}

 		log.Debug().Msgf("Prompt template %q written", template.Name)
@@ -202,9 +217,11 @@ func InstallModel(basePath, nameOverride string, config *ModelConfig, configOver
 	}

 	if err := utils.VerifyPath(name+".yaml", basePath); err != nil {
-		return err
+		return nil, err
 	}

+	backendConfig := lconfig.BackendConfig{}
+
 	// write config file
 	if len(configOverrides) != 0 || len(config.ConfigFile) != 0 {
 		configFilePath := filepath.Join(basePath, name+".yaml")
@@ -213,33 +230,33 @@ func InstallModel(basePath, nameOverride string, config *ModelConfig, configOver
 		configMap := make(map[string]interface{})
 		err = yaml.Unmarshal([]byte(config.ConfigFile), &configMap)
 		if err != nil {
-			return fmt.Errorf("failed to unmarshal config YAML: %v", err)
+			return nil, fmt.Errorf("failed to unmarshal config YAML: %v", err)
 		}

 		configMap["name"] = name

 		if err := mergo.Merge(&configMap, configOverrides, mergo.WithOverride); err != nil {
-			return err
+			return nil, err
 		}

 		// Write updated config file
 		updatedConfigYAML, err := yaml.Marshal(configMap)
 		if err != nil {
-			return fmt.Errorf("failed to marshal updated config YAML: %v", err)
+			return nil, fmt.Errorf("failed to marshal updated config YAML: %v", err)
 		}

-		backendConfig := lconfig.BackendConfig{}
 		err = yaml.Unmarshal(updatedConfigYAML, &backendConfig)
 		if err != nil {
-			return fmt.Errorf("failed to unmarshal updated config YAML: %v", err)
+			return nil, fmt.Errorf("failed to unmarshal updated config YAML: %v", err)
 		}
+
 		if !backendConfig.Validate() {
-			return fmt.Errorf("failed to validate updated config YAML")
+			return nil, fmt.Errorf("failed to validate updated config YAML")
 		}

 		err = os.WriteFile(configFilePath, updatedConfigYAML, 0600)
 		if err != nil {
-			return fmt.Errorf("failed to write updated config file: %v", err)
+			return nil, fmt.Errorf("failed to write updated config file: %v", err)
 		}

 		log.Debug().Msgf("Written config file %s", configFilePath)
@@ -249,14 +266,12 @@ func InstallModel(basePath, nameOverride string, config *ModelConfig, configOver
 	modelFile := filepath.Join(basePath, galleryFileName(name))
 	data, err := yaml.Marshal(config)
 	if err != nil {
-		return err
+		return nil, err
 	}

 	log.Debug().Msgf("Written gallery file %s", modelFile)

-	return os.WriteFile(modelFile, data, 0600)
-
-	//return nil
+	return &backendConfig, os.WriteFile(modelFile, data, 0600)
 }

 func galleryFileName(name string) string {
--- a/core/gallery/models_test.go
+++ b/core/gallery/models_test.go
@@ -29,7 +29,7 @@ var _ = Describe("Model test", func() {
 			defer os.RemoveAll(tempdir)
 			c, err := ReadConfigFile[ModelConfig](filepath.Join(os.Getenv("FIXTURES"), "gallery_simple.yaml"))
 			Expect(err).ToNot(HaveOccurred())
-			err = InstallModel(tempdir, "", c, map[string]interface{}{}, func(string, string, string, float64) {}, true)
+			_, err = InstallModel(tempdir, "", c, map[string]interface{}{}, func(string, string, string, float64) {}, true)
 			Expect(err).ToNot(HaveOccurred())

 			for _, f := range []string{"cerebras", "cerebras-completion.tmpl", "cerebras-chat.tmpl", "cerebras.yaml"} {
@@ -79,7 +79,7 @@ var _ = Describe("Model test", func() {
 			Expect(models[0].URL).To(Equal(bertEmbeddingsURL))
 			Expect(models[0].Installed).To(BeFalse())

-			err = InstallModelFromGallery(galleries, "test@bert", tempdir, GalleryModel{}, func(s1, s2, s3 string, f float64) {}, true)
+			err = InstallModelFromGallery(galleries, []config.Gallery{}, "test@bert", tempdir, "", GalleryModel{}, func(s1, s2, s3 string, f float64) {}, true, true)
 			Expect(err).ToNot(HaveOccurred())

 			dat, err := os.ReadFile(filepath.Join(tempdir, "bert.yaml"))
@@ -116,7 +116,7 @@ var _ = Describe("Model test", func() {
 			c, err := ReadConfigFile[ModelConfig](filepath.Join(os.Getenv("FIXTURES"), "gallery_simple.yaml"))
 			Expect(err).ToNot(HaveOccurred())

-			err = InstallModel(tempdir, "foo", c, map[string]interface{}{}, func(string, string, string, float64) {}, true)
+			_, err = InstallModel(tempdir, "foo", c, map[string]interface{}{}, func(string, string, string, float64) {}, true)
 			Expect(err).ToNot(HaveOccurred())

 			for _, f := range []string{"cerebras", "cerebras-completion.tmpl", "cerebras-chat.tmpl", "foo.yaml"} {
@@ -132,7 +132,7 @@ var _ = Describe("Model test", func() {
 			c, err := ReadConfigFile[ModelConfig](filepath.Join(os.Getenv("FIXTURES"), "gallery_simple.yaml"))
 			Expect(err).ToNot(HaveOccurred())

-			err = InstallModel(tempdir, "foo", c, map[string]interface{}{"backend": "foo"}, func(string, string, string, float64) {}, true)
+			_, err = InstallModel(tempdir, "foo", c, map[string]interface{}{"backend": "foo"}, func(string, string, string, float64) {}, true)
 			Expect(err).ToNot(HaveOccurred())

 			for _, f := range []string{"cerebras", "cerebras-completion.tmpl", "cerebras-chat.tmpl", "foo.yaml"} {
@@ -158,7 +158,7 @@ var _ = Describe("Model test", func() {
 			c, err := ReadConfigFile[ModelConfig](filepath.Join(os.Getenv("FIXTURES"), "gallery_simple.yaml"))
 			Expect(err).ToNot(HaveOccurred())

-			err = InstallModel(tempdir, "../../../foo", c, map[string]interface{}{}, func(string, string, string, float64) {}, true)
+			_, err = InstallModel(tempdir, "../../../foo", c, map[string]interface{}{}, func(string, string, string, float64) {}, true)
 			Expect(err).To(HaveOccurred())
 		})
 	})
--- a/core/http/endpoints/openai/chat.go
+++ b/core/http/endpoints/openai/chat.go
@@ -41,7 +41,7 @@ func ChatEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, evaluat
 		}
 		responses <- initialMessage

-		ComputeChoices(req, s, config, startupOptions, loader, func(s string, c *[]schema.Choice) {}, func(s string, tokenUsage backend.TokenUsage) bool {
+		ComputeChoices(req, s, config, cl, startupOptions, loader, func(s string, c *[]schema.Choice) {}, func(s string, tokenUsage backend.TokenUsage) bool {
 			usage := schema.OpenAIUsage{
 				PromptTokens:     tokenUsage.Prompt,
 				CompletionTokens: tokenUsage.Completion,
@@ -68,7 +68,7 @@ func ChatEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, evaluat
 	}
 	processTools := func(noAction string, prompt string, req *schema.OpenAIRequest, config *config.BackendConfig, loader *model.ModelLoader, responses chan schema.OpenAIResponse, extraUsage bool) {
 		result := ""
-		_, tokenUsage, _ := ComputeChoices(req, prompt, config, startupOptions, loader, func(s string, c *[]schema.Choice) {}, func(s string, usage backend.TokenUsage) bool {
+		_, tokenUsage, _ := ComputeChoices(req, prompt, config, cl, startupOptions, loader, func(s string, c *[]schema.Choice) {}, func(s string, usage backend.TokenUsage) bool {
 			result += s
 			// TODO: Change generated BNF grammar to be compliant with the schema so we can
 			// stream the result token by token here.
@@ -92,7 +92,7 @@ func ChatEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, evaluat
 			}
 			responses <- initialMessage

-			result, err := handleQuestion(config, req, ml, startupOptions, functionResults, result, prompt)
+			result, err := handleQuestion(config, cl, req, ml, startupOptions, functionResults, result, prompt)
 			if err != nil {
 				log.Error().Err(err).Msg("error handling question")
 				return
@@ -383,7 +383,8 @@ func ChatEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, evaluat

 		// no streaming mode
 		default:
-			result, tokenUsage, err := ComputeChoices(input, predInput, config, startupOptions, ml, func(s string, c *[]schema.Choice) {
+
+			tokenCallback := func(s string, c *[]schema.Choice) {
 				if !shouldUseFn {
 					// no function is called, just reply and use stop as finish reason
 					*c = append(*c, schema.Choice{FinishReason: "stop", Index: 0, Message: &schema.Message{Role: "assistant", Content: &s}})
@@ -403,7 +404,7 @@ func ChatEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, evaluat

 				switch {
 				case noActionsToRun:
-					result, err := handleQuestion(config, input, ml, startupOptions, results, s, predInput)
+					result, err := handleQuestion(config, cl, input, ml, startupOptions, results, s, predInput)
 					if err != nil {
 						log.Error().Err(err).Msg("error handling question")
 						return
@@ -458,7 +459,18 @@ func ChatEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, evaluat
 					}
 				}

-			}, nil)
+			}
+
+			result, tokenUsage, err := ComputeChoices(
+				input,
+				predInput,
+				config,
+				cl,
+				startupOptions,
+				ml,
+				tokenCallback,
+				nil,
+			)
 			if err != nil {
 				return err
 			}
@@ -489,7 +501,7 @@ func ChatEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, evaluat
 	}
 }

-func handleQuestion(config *config.BackendConfig, input *schema.OpenAIRequest, ml *model.ModelLoader, o *config.ApplicationConfig, funcResults []functions.FuncCallResults, result, prompt string) (string, error) {
+func handleQuestion(config *config.BackendConfig, cl *config.BackendConfigLoader, input *schema.OpenAIRequest, ml *model.ModelLoader, o *config.ApplicationConfig, funcResults []functions.FuncCallResults, result, prompt string) (string, error) {

 	if len(funcResults) == 0 && result != "" {
 		log.Debug().Msgf("nothing function results but we had a message from the LLM")
@@ -538,7 +550,7 @@ func handleQuestion(config *config.BackendConfig, input *schema.OpenAIRequest, m
 		audios = append(audios, m.StringAudios...)
 	}

-	predFunc, err := backend.ModelInference(input.Context, prompt, input.Messages, images, videos, audios, ml, config, o, nil)
+	predFunc, err := backend.ModelInference(input.Context, prompt, input.Messages, images, videos, audios, ml, config, cl, o, nil)
 	if err != nil {
 		log.Error().Err(err).Msg("model inference failed")
 		return "", err
--- a/core/http/endpoints/openai/completion.go
+++ b/core/http/endpoints/openai/completion.go
@@ -31,7 +31,7 @@ func CompletionEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, e
 	created := int(time.Now().Unix())

 	process := func(id string, s string, req *schema.OpenAIRequest, config *config.BackendConfig, loader *model.ModelLoader, responses chan schema.OpenAIResponse, extraUsage bool) {
-		ComputeChoices(req, s, config, appConfig, loader, func(s string, c *[]schema.Choice) {}, func(s string, tokenUsage backend.TokenUsage) bool {
+		tokenCallback := func(s string, tokenUsage backend.TokenUsage) bool {
 			usage := schema.OpenAIUsage{
 				PromptTokens:     tokenUsage.Prompt,
 				CompletionTokens: tokenUsage.Completion,
@@ -58,7 +58,8 @@ func CompletionEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, e

 			responses <- resp
 			return true
-		})
+		}
+		ComputeChoices(req, s, config, cl, appConfig, loader, func(s string, c *[]schema.Choice) {}, tokenCallback)
 		close(responses)
 	}

@@ -168,7 +169,7 @@ func CompletionEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, e
 			}

 			r, tokenUsage, err := ComputeChoices(
-				input, i, config, appConfig, ml, func(s string, c *[]schema.Choice) {
+				input, i, config, cl, appConfig, ml, func(s string, c *[]schema.Choice) {
 					*c = append(*c, schema.Choice{Text: s, FinishReason: "stop", Index: k})
 				}, nil)
 			if err != nil {
--- a/core/http/endpoints/openai/edit.go
+++ b/core/http/endpoints/openai/edit.go
@@ -56,7 +56,7 @@ func EditEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, evaluat
 				log.Debug().Msgf("Template found, input modified to: %s", i)
 			}

-			r, tokenUsage, err := ComputeChoices(input, i, config, appConfig, ml, func(s string, c *[]schema.Choice) {
+			r, tokenUsage, err := ComputeChoices(input, i, config, cl, appConfig, ml, func(s string, c *[]schema.Choice) {
 				*c = append(*c, schema.Choice{Text: s})
 			}, nil)
 			if err != nil {
--- a/core/http/endpoints/openai/inference.go
+++ b/core/http/endpoints/openai/inference.go
@@ -12,6 +12,7 @@ func ComputeChoices(
 	req *schema.OpenAIRequest,
 	predInput string,
 	config *config.BackendConfig,
+	bcl *config.BackendConfigLoader,
 	o *config.ApplicationConfig,
 	loader *model.ModelLoader,
 	cb func(string, *[]schema.Choice),
@@ -37,7 +38,7 @@ func ComputeChoices(
 	}

 	// get the model function to call for the result
-	predFunc, err := backend.ModelInference(req.Context, predInput, req.Messages, images, videos, audios, loader, config, o, tokenCallback)
+	predFunc, err := backend.ModelInference(req.Context, predInput, req.Messages, images, videos, audios, loader, config, bcl, o, tokenCallback)
 	if err != nil {
 		return result, backend.TokenUsage{}, err
 	}
--- a/core/http/routes/ui_backend_gallery.go
+++ b/core/http/routes/ui_backend_gallery.go
@@ -223,7 +223,7 @@ func registerBackendGalleryRoutes(app *fiber.App, appConfig *config.ApplicationC
 			return c.SendString(elements.ProgressBar("0"))
 		}

-		if status.Progress == 100 {
+		if status.Progress == 100 && status.Processed && status.Message == "completed" {
 			c.Set("HX-Trigger", "done") // this triggers /browse/backend/job/:uid
 			return c.SendString(elements.ProgressBar("100"))
 		}
--- a/core/http/routes/ui_gallery.go
+++ b/core/http/routes/ui_gallery.go
@@ -243,7 +243,7 @@ func registerGalleryRoutes(app *fiber.App, cl *config.BackendConfigLoader, appCo
 			return c.SendString(elements.ProgressBar("0"))
 		}

-		if status.Progress == 100 {
+		if status.Progress == 100 && status.Processed && status.Message == "completed" {
 			c.Set("HX-Trigger", "done") // this triggers /browse/job/:uid (which is when the job is done)
 			return c.SendString(elements.ProgressBar("100"))
 		}
--- a/core/services/backends.go
+++ b/core/services/backends.go
@@ -24,7 +24,7 @@ func (g *GalleryService) backendHandler(op *GalleryOp[gallery.GalleryBackend], s
 		g.modelLoader.DeleteExternalBackend(op.GalleryElementName)
 	} else {
 		log.Warn().Msgf("installing backend %s", op.GalleryElementName)
-		err = gallery.InstallBackendFromGallery(g.appConfig.BackendGalleries, systemState, op.GalleryElementName, g.appConfig.BackendsPath, progressCallback)
+		err = gallery.InstallBackendFromGallery(g.appConfig.BackendGalleries, systemState, op.GalleryElementName, g.appConfig.BackendsPath, progressCallback, true)
 		if err == nil {
 			err = gallery.RegisterBackends(g.appConfig.BackendsPath, g.modelLoader)
 		}
--- a/core/services/models.go
+++ b/core/services/models.go
@@ -7,6 +7,7 @@ import (

 	"github.com/mudler/LocalAI/core/config"
 	"github.com/mudler/LocalAI/core/gallery"
+	"github.com/mudler/LocalAI/core/system"
 	"github.com/mudler/LocalAI/pkg/utils"
 	"gopkg.in/yaml.v2"
 )
@@ -22,7 +23,7 @@ func (g *GalleryService) modelHandler(op *GalleryOp[gallery.GalleryModel], cl *c
 		utils.DisplayDownloadFunction(fileName, current, total, percentage)
 	}

-	err := processModelOperation(op, g.appConfig.ModelPath, g.appConfig.EnforcePredownloadScans, progressCallback)
+	err := processModelOperation(op, g.appConfig.ModelPath, g.appConfig.BackendsPath, g.appConfig.EnforcePredownloadScans, g.appConfig.AutoloadBackendGalleries, progressCallback)
 	if err != nil {
 		return err
 	}
@@ -49,7 +50,7 @@ func (g *GalleryService) modelHandler(op *GalleryOp[gallery.GalleryModel], cl *c
 	return nil
 }

-func prepareModel(modelPath string, req gallery.GalleryModel, downloadStatus func(string, string, string, float64), enforceScan bool) error {
+func installModelFromRemoteConfig(modelPath string, req gallery.GalleryModel, downloadStatus func(string, string, string, float64), enforceScan, automaticallyInstallBackend bool, backendGalleries []config.Gallery, backendBasePath string) error {
 	config, err := gallery.GetGalleryConfigFromURL[gallery.ModelConfig](req.URL, modelPath)
 	if err != nil {
 		return err
@@ -57,7 +58,23 @@ func prepareModel(modelPath string, req gallery.GalleryModel, downloadStatus fun

 	config.Files = append(config.Files, req.AdditionalFiles...)

-	return gallery.InstallModel(modelPath, req.Name, &config, req.Overrides, downloadStatus, enforceScan)
+	installedModel, err := gallery.InstallModel(modelPath, req.Name, &config, req.Overrides, downloadStatus, enforceScan)
+	if err != nil {
+		return err
+	}
+
+	if automaticallyInstallBackend && installedModel.Backend != "" {
+		systemState, err := system.GetSystemState()
+		if err != nil {
+			return err
+		}
+
+		if err := gallery.InstallBackendFromGallery(backendGalleries, systemState, installedModel.Backend, backendBasePath, downloadStatus, false); err != nil {
+			return err
+		}
+	}
+
+	return nil
 }

 type galleryModel struct {
@@ -65,22 +82,22 @@ type galleryModel struct {
 	ID                   string           `json:"id"`
 }

-func processRequests(modelPath string, enforceScan bool, galleries []config.Gallery, requests []galleryModel) error {
+func processRequests(modelPath, backendBasePath string, enforceScan, automaticallyInstallBackend bool, galleries []config.Gallery, backendGalleries []config.Gallery, requests []galleryModel) error {
 	var err error
 	for _, r := range requests {
 		utils.ResetDownloadTimers()
 		if r.ID == "" {
-			err = prepareModel(modelPath, r.GalleryModel, utils.DisplayDownloadFunction, enforceScan)
+			err = installModelFromRemoteConfig(modelPath, r.GalleryModel, utils.DisplayDownloadFunction, enforceScan, automaticallyInstallBackend, backendGalleries, backendBasePath)

 		} else {
 			err = gallery.InstallModelFromGallery(
-				galleries, r.ID, modelPath, r.GalleryModel, utils.DisplayDownloadFunction, enforceScan)
+				galleries, backendGalleries, r.ID, modelPath, backendBasePath, r.GalleryModel, utils.DisplayDownloadFunction, enforceScan, automaticallyInstallBackend)
 		}
 	}
 	return err
 }

-func ApplyGalleryFromFile(modelPath, s string, enforceScan bool, galleries []config.Gallery) error {
+func ApplyGalleryFromFile(modelPath, backendBasePath string, enforceScan, automaticallyInstallBackend bool, galleries []config.Gallery, backendGalleries []config.Gallery, s string) error {
 	dat, err := os.ReadFile(s)
 	if err != nil {
 		return err
@@ -91,24 +108,26 @@ func ApplyGalleryFromFile(modelPath, s string, enforceScan bool, galleries []con
 		return err
 	}

-	return processRequests(modelPath, enforceScan, galleries, requests)
+	return processRequests(modelPath, backendBasePath, enforceScan, automaticallyInstallBackend, galleries, backendGalleries, requests)
 }

-func ApplyGalleryFromString(modelPath, s string, enforceScan bool, galleries []config.Gallery) error {
+func ApplyGalleryFromString(modelPath, backendBasePath string, enforceScan, automaticallyInstallBackend bool, galleries []config.Gallery, backendGalleries []config.Gallery, s string) error {
 	var requests []galleryModel
 	err := json.Unmarshal([]byte(s), &requests)
 	if err != nil {
 		return err
 	}

-	return processRequests(modelPath, enforceScan, galleries, requests)
+	return processRequests(modelPath, backendBasePath, enforceScan, automaticallyInstallBackend, galleries, backendGalleries, requests)
 }

 // processModelOperation handles the installation or deletion of a model
 func processModelOperation(
 	op *GalleryOp[gallery.GalleryModel],
 	modelPath string,
+	backendBasePath string,
 	enforcePredownloadScans bool,
+	automaticallyInstallBackend bool,
 	progressCallback func(string, string, string, float64),
 ) error {
 	// delete a model
@@ -140,7 +159,7 @@ func processModelOperation(

 	// if the request contains a gallery name, we apply the gallery from the gallery list
 	if op.GalleryElementName != "" {
-		return gallery.InstallModelFromGallery(op.Galleries, op.GalleryElementName, modelPath, op.Req, progressCallback, enforcePredownloadScans)
+		return gallery.InstallModelFromGallery(op.Galleries, op.BackendGalleries, op.GalleryElementName, modelPath, backendBasePath, op.Req, progressCallback, enforcePredownloadScans, automaticallyInstallBackend)
 		// } else if op.ConfigURL != "" {
 		// 	err := startup.InstallModels(op.Galleries, modelPath, enforcePredownloadScans, progressCallback, op.ConfigURL)
 		// 	if err != nil {
@@ -148,6 +167,6 @@ func processModelOperation(
 		// 	}
 		// 	return cl.Preload(modelPath)
 	} else {
-		return prepareModel(modelPath, op.Req, progressCallback, enforcePredownloadScans)
+		return installModelFromRemoteConfig(modelPath, op.Req, progressCallback, enforcePredownloadScans, automaticallyInstallBackend, op.BackendGalleries, backendBasePath)
 	}
 }
--- a/core/services/operation.go
+++ b/core/services/operation.go
@@ -10,8 +10,9 @@ type GalleryOp[T any] struct {
 	GalleryElementName string
 	Delete             bool

-	Req       T
-	Galleries []config.Gallery
+	Req              T
+	Galleries        []config.Gallery
+	BackendGalleries []config.Gallery
 }

 type GalleryOpStatus struct {
--- a/core/system/capabilities.go
+++ b/core/system/capabilities.go
@@ -1,6 +1,7 @@
 package system

 import (
+	"os"
 	"strings"

 	"github.com/mudler/LocalAI/pkg/xsysinfo"
@@ -11,6 +12,29 @@ type SystemState struct {
 	GPUVendor string
 }

+func (s *SystemState) Capability() string {
+	if os.Getenv("LOCALAI_FORCE_META_BACKEND_CAPABILITY") != "" {
+		return os.Getenv("LOCALAI_FORCE_META_BACKEND_CAPABILITY")
+	}
+
+	capabilityRunFile := "/run/localai/capability"
+	if os.Getenv("LOCALAI_FORCE_META_BACKEND_CAPABILITY_RUN_FILE") != "" {
+		capabilityRunFile = os.Getenv("LOCALAI_FORCE_META_BACKEND_CAPABILITY_RUN_FILE")
+	}
+
+	// Check if /run/localai/capability exists and use it
+	// This might be used by e.g. container images to specify which
+	// backends to pull in automatically when installing meta backends.
+	if _, err := os.Stat(capabilityRunFile); err == nil {
+		capability, err := os.ReadFile(capabilityRunFile)
+		if err == nil {
+			return string(capability)
+		}
+	}
+
+	return s.GPUVendor
+}
+
 func GetSystemState() (*SystemState, error) {
 	gpuVendor, _ := detectGPUVendor()
 	log.Debug().Str("gpuVendor", gpuVendor).Msg("GPU vendor")
@@ -30,13 +54,13 @@ func detectGPUVendor() (string, error) {
 		if gpu.DeviceInfo != nil {
 			if gpu.DeviceInfo.Vendor != nil {
 				gpuVendorName := strings.ToUpper(gpu.DeviceInfo.Vendor.Name)
-				if gpuVendorName == "NVIDIA" {
+				if strings.Contains(gpuVendorName, "NVIDIA") {
 					return "nvidia", nil
 				}
-				if gpuVendorName == "AMD" {
+				if strings.Contains(gpuVendorName, "AMD") {
 					return "amd", nil
 				}
-				if gpuVendorName == "INTEL" {
+				if strings.Contains(gpuVendorName, "INTEL") {
 					return "intel", nil
 				}
 				return "nvidia", nil
--- a/docker-compose.yaml
+++ b/docker-compose.yaml
@@ -11,7 +11,7 @@ services:
      dockerfile: Dockerfile
      args:
      - IMAGE_TYPE=core
-      - BASE_IMAGE=ubuntu:22.04
+      - BASE_IMAGE=ubuntu:24.04
    ports:
      - 8080:8080
    env_file:
--- a/docs/content/docs/getting-started/container-images.md
+++ b/docs/content/docs/getting-started/container-images.md
@@ -205,9 +205,9 @@ Standard container images do not have pre-installed models.

 | Description | Quay | Docker Hub                                                  |
 | --- | --- |-------------------------------------------------------------|
-| Latest images from the branch (development) | `quay.io/go-skynet/local-ai:master-hipblas` | `localai/localai:master-hipblas`                      |
+| Latest images from the branch (development) | `quay.io/go-skynet/local-ai:master-gpu-hipblas` | `localai/localai:master-gpu-hipblas`                      |
 | Latest tag | `quay.io/go-skynet/local-ai:latest-gpu-hipblas` | `localai/localai:latest-gpu-hipblas`                      |
-| Versioned image | `quay.io/go-skynet/local-ai:{{< version >}}-hipblas` | `localai/localai:{{< version >}}-hipblas`             |
+| Versioned image | `quay.io/go-skynet/local-ai:{{< version >}}-gpu-hipblas` | `localai/localai:{{< version >}}-gpu-hipblas`             |

 {{% /tab %}}

--- a/docs/data/version.json
+++ b/docs/data/version.json
@@ -1,3 +1,3 @@
 {
-  "version": "v3.0.0"
+  "version": "v3.1.1"
 }
--- a/docs/layouts/partials/docs/top-header.html
+++ b/docs/layouts/partials/docs/top-header.html
@@ -44,7 +44,7 @@
                {{ end }}
            {{ end -}}
        </div>
-        <div class="d-flex align-items-center m-1">
+        <div class="d-none d-md-flex d-flex align-items-center m-1">
            <h5>Star us on GitHub !&nbsp;</h5>
            <script async defer src="https://buttons.github.io/buttons.js"></script>
            <a class="github-button" href="https://github.com/mudler/LocalAI" data-color-scheme="no-preference: light; light: light; dark: dark;" data-icon="octicon-star" data-size="large" data-show-count="true" aria-label="Star mudler/LocalAI on GitHub">Star</a> 
--- a/docs/static/install.sh
+++ b/docs/static/install.sh
@@ -698,7 +698,7 @@ install_docker() {
            $envs \
            -d -p $PORT:8080 --name local-ai localai/localai:$IMAGE_TAG $STARTCOMMAND
    elif [ "$HAS_AMD" ]; then
-        IMAGE_TAG=${LOCALAI_VERSION}-hipblas
+        IMAGE_TAG=${LOCALAI_VERSION}-gpu-hipblas
        # AIO
        if [ "$USE_AIO" = true ]; then
            IMAGE_TAG=${LOCALAI_VERSION}-aio-gpu-hipblas
--- a/gallery/index.yaml
+++ b/gallery/index.yaml
@@ -1279,6 +1279,337 @@
    - filename: Menlo_Jan-nano-128k-Q4_K_M.gguf
      sha256: a864031a138288da427ca176afd61d7fe2b03fd19a84a656b2691aa1f7a12921
      uri: huggingface://bartowski/Menlo_Jan-nano-128k-GGUF/Menlo_Jan-nano-128k-Q4_K_M.gguf
+- !!merge <<: *qwen3
+  icon: https://huggingface.co/DavidAU/Qwen3-55B-A3B-TOTAL-RECALL-V1.3/resolve/main/qwen3-total-recall.gif
+  name: "qwen3-55b-a3b-total-recall-v1.3-i1"
+  urls:
+    - https://huggingface.co/DavidAU/Qwen3-55B-A3B-TOTAL-RECALL-V1.3
+    - https://huggingface.co/mradermacher/Qwen3-55B-A3B-TOTAL-RECALL-V1.3-i1-GGUF
+  description: |
+    WARNING: MADNESS - UN HINGED and... NSFW. Vivid prose. INTENSE. Visceral Details. Violence. HORROR. GORE. Swearing. UNCENSORED... humor, romance, fun.
+    This repo contains the full precision source code, in "safe tensors" format to generate GGUFs, GPTQ, EXL2, AWQ, HQQ and other formats. The source code can also be used directly.
+
+    This model is for all use cases, but excels in creative use cases specifically.
+
+    This model is based on Qwen3-30B-A3B (MOE, 128 experts, 8 activated), with Brainstorm 40X (by DavidAU - details at bottom of this page.
+
+    This is the refined version -V1.3- from this project (see this repo for all settings, details, system prompts, example generations etc etc):
+
+    https://huggingface.co/DavidAU/Qwen3-55B-A3B-TOTAL-RECALL-Deep-40X-GGUF/
+
+    This version -1.3- is slightly smaller, with further refinements to the Brainstorm adapter.
+
+    This will change generation and reasoning performance within the model.
+  overrides:
+    parameters:
+      model: Qwen3-55B-A3B-TOTAL-RECALL-V1.3.i1-Q4_K_M.gguf
+  files:
+    - filename: Qwen3-55B-A3B-TOTAL-RECALL-V1.3.i1-Q4_K_M.gguf
+      sha256: bcf5a1f8a40e9438a19b23dfb40e872561c310296c5ac804f937a0e3c1376def
+      uri: huggingface://mradermacher/Qwen3-55B-A3B-TOTAL-RECALL-V1.3-i1-GGUF/Qwen3-55B-A3B-TOTAL-RECALL-V1.3.i1-Q4_K_M.gguf
+
+- !!merge <<: *qwen3
+  name: "qwen3-55b-a3b-total-recall-deep-40x"
+  icon: https://huggingface.co/DavidAU/Qwen3-55B-A3B-TOTAL-RECALL-V1.3/resolve/main/qwen3-total-recall.gif
+  urls:
+    - https://huggingface.co/DavidAU/Qwen3-55B-A3B-TOTAL-RECALL-Deep-40X-GGUF
+  description: |
+    WARNING: MADNESS - UN HINGED and... NSFW. Vivid prose. INTENSE. Visceral Details. Violence. HORROR. GORE. Swearing. UNCENSORED... humor, romance, fun.
+    Qwen3-55B-A3B-TOTAL-RECALL-Deep-40X-GGUF
+
+    A highly experimental model ("tamer" versions below) based on Qwen3-30B-A3B (MOE, 128 experts, 8 activated), with Brainstorm 40X (by DavidAU - details at bottom of this page).
+
+    These modifications blow the model (V1) out to 87 layers, 1046 tensors and 55B parameters.
+
+    Note that some versions are smaller than this, with fewer layers/tensors and smaller parameter counts.
+
+    The adapter extensively alters performance, reasoning and output generation.
+
+    Exceptional changes in creative, prose and general performance.
+
+    Regens of the same prompt - even with the same settings - will be very different.
+
+    THREE example generations below - creative (generated with Q3_K_M, V1 model).
+
+    ONE example generation (#4) - non creative (generated with Q3_K_M, V1 model).
+
+    You can run this model on CPU and/or GPU due to unique model construction, size of experts and total activated experts at 3B parameters (8 experts), which translates into roughly almost 6B parameters in this version.
+
+    Two quants uploaded for testing: Q3_K_M, Q4_K_M
+
+    V3, V4 and V5 are also available in these two quants.
+
+    V2 and V6 in Q3_k_m only; as are: V 1.3, 1.4, 1.5, 1.7 and V7 (newest)
+
+    NOTE: V2 and up are from source model 2, V1 and 1.3,1.4,1.5,1.7 are from source model 1.
+  overrides:
+    parameters:
+      model: Qwen3-55B-A3B-TOTAL-RECALL-V5-Deep-40X-q4_K_M.gguf
+  files:
+    - filename: Qwen3-55B-A3B-TOTAL-RECALL-V5-Deep-40X-q4_K_M.gguf
+      sha256: 20ef786a8c8e74eb257aa3069e237cbd40f42d25f5502fed6fa016bb8afbdae4
+      uri: huggingface://DavidAU/Qwen3-55B-A3B-TOTAL-RECALL-Deep-40X-GGUF/Qwen3-55B-A3B-TOTAL-RECALL-V5-Deep-40X-q4_K_M.gguf
+- !!merge <<: *qwen3
+  name: "qwen3-42b-a3b-stranger-thoughts-deep20x-abliterated-uncensored-i1"
+  icon: https://huggingface.co/DavidAU/Qwen3-42B-A3B-Stranger-Thoughts-Deep20x-Abliterated-Uncensored/resolve/main/qwen-42b-ablit.jpg
+  urls:
+    - https://huggingface.co/DavidAU/Qwen3-42B-A3B-Stranger-Thoughts-Deep20x-Abliterated-Uncensored
+    - https://huggingface.co/mradermacher/Qwen3-42B-A3B-Stranger-Thoughts-Deep20x-Abliterated-Uncensored-i1-GGUF
+  description: |
+    WARNING: NSFW. Vivid prose. INTENSE. Visceral Details. Violence. HORROR. GORE. Swearing. UNCENSORED... humor, romance, fun.
+    Qwen3-42B-A3B-Stranger-Thoughts-Deep20x-Abliterated-Uncensored
+
+    This repo contains the full precision source code, in "safe tensors" format to generate GGUFs, GPTQ, EXL2, AWQ, HQQ and other formats. The source code can also be used directly.
+
+    ABOUT:
+
+    Qwen's excellent "Qwen3-30B-A3B", abliterated by "huihui-ai" then combined Brainstorm 20x (tech notes at bottom of the page) in a MOE (128 experts) at 42B parameters (up from 30B).
+
+    This pushes Qwen's abliterated/uncensored model to the absolute limit for creative use cases.
+
+    Prose (all), reasoning, thinking ... all will be very different from reg "Qwen 3s".
+
+    This model will generate horror, fiction, erotica, - you name it - in vivid, stark detail.
+
+    It will NOT hold back.
+
+    Likewise, regen(s) of the same prompt - even at the same settings - will create very different version(s) too.
+
+    See FOUR examples below.
+
+    Model retains full reasoning, and output generation of a Qwen3 MOE ; but has not been tested for "non-creative" use cases.
+
+    Model is set with Qwen's default config:
+
+        40 k context
+        8 of 128 experts activated.
+        Chatml OR Jinja Template (embedded)
+
+    IMPORTANT:
+
+    See usage guide / repo below to get the most out of this model, as settings are very specific.
+
+    USAGE GUIDE:
+
+    Please refer to this model card for
+
+        Specific usage, suggested settings, changing ACTIVE EXPERTS, templates, settings and the like:
+        How to maximize this model in "uncensored" form, with specific notes on "abliterated" models.
+        Rep pen / temp settings specific to getting the model to perform strongly.
+
+    https://huggingface.co/DavidAU/Qwen3-18B-A3B-Stranger-Thoughts-Abliterated-Uncensored-GGUF
+
+    GGUF / QUANTS / SPECIAL SHOUTOUT:
+
+    Special thanks to team Mradermacher for making the quants!
+
+    https://huggingface.co/mradermacher/Qwen3-42B-A3B-Stranger-Thoughts-Deep20x-Abliterated-Uncensored-GGUF
+
+    KNOWN ISSUES:
+
+        Model may "mis-capitalize" word(s) - lowercase, where uppercase should be - from time to time.
+        Model may add extra space from time to time before a word.
+        Incorrect template and/or settings will result in a drop in performance / poor performance.
+  overrides:
+    parameters:
+      model: Qwen3-42B-A3B-Stranger-Thoughts-Deep20x-Abliterated-Uncensored.i1-Q4_K_M.gguf
+  files:
+    - filename: Qwen3-42B-A3B-Stranger-Thoughts-Deep20x-Abliterated-Uncensored.i1-Q4_K_M.gguf
+      sha256: ef4a601adfc2897b214cda2d16f76dcb8215a1b994bc76c696158d68ec535dd8
+      uri: huggingface://mradermacher/Qwen3-42B-A3B-Stranger-Thoughts-Deep20x-Abliterated-Uncensored-i1-GGUF/Qwen3-42B-A3B-Stranger-Thoughts-Deep20x-Abliterated-Uncensored.i1-Q4_K_M.gguf
+- !!merge <<: *qwen3
+  name: "qwen3-22b-a3b-the-harley-quinn"
+  icon: https://huggingface.co/DavidAU/Qwen3-22B-A3B-The-Harley-Quinn/resolve/main/qwen3-harley-quinn-23b.webp
+  urls:
+    - https://huggingface.co/DavidAU/Qwen3-22B-A3B-The-Harley-Quinn
+    - https://huggingface.co/mradermacher/Qwen3-22B-A3B-The-Harley-Quinn-GGUF
+  description: |
+    WARNING: MADNESS - UN HINGED and... NSFW. Vivid prose. INTENSE. Visceral Details. Violence. HORROR. GORE. Swearing. UNCENSORED... humor, romance, fun.
+    Qwen3-22B-A3B-The-Harley-Quinn
+
+    This repo contains the full precision source code, in "safe tensors" format to generate GGUFs, GPTQ, EXL2, AWQ, HQQ and other formats. The source code can also be used directly.
+
+    ABOUT:
+
+    A stranger, yet radically different version of Kalmaze's "Qwen/Qwen3-16B-A3B" with the experts pruned to 64 (from 128, the Qwen 3 30B-A3B version) and then I added 19 layers expanding (Brainstorm 20x by DavidAU info at bottom of this page) the model to 22B total parameters.
+
+    The goal: slightly alter the model, to address some odd creative thinking and output choices.
+
+    Then... Harley Quinn showed up, and then it was a party!
+
+    A wild, out of control (sometimes) but never boring party.
+
+    Please note that the modifications affect the entire model operation; roughly I adjusted the model to think a little "deeper" and "ponder" a bit - but this is a very rough description.
+
+    That being said, reasoning and output generation will be altered regardless of your use case(s).
+
+    These modifications pushes Qwen's model to the absolute limit for creative use cases.
+
+    Detail, vividiness, and creativity all get a boost.
+
+    Prose (all) will also be very different from "default" Qwen3.
+
+    Likewise, regen(s) of the same prompt - even at the same settings - will create very different version(s) too.
+
+    The Brainstrom 20x has also lightly de-censored the model under some conditions.
+
+    However, this model can be prone to bouts of madness.
+
+    It will not always behave, and it will sometimes go -wildly- off script.
+
+    See 4 examples below.
+
+    Model retains full reasoning, and output generation of a Qwen3 MOE ; but has not been tested for "non-creative" use cases.
+
+    Model is set with Qwen's default config:
+
+        40 k context
+        8 of 64 experts activated.
+        Chatml OR Jinja Template (embedded)
+
+    Four example generations below.
+
+    IMPORTANT:
+
+    See usage guide / repo below to get the most out of this model, as settings are very specific.
+
+    If not set correctly, this model will not work the way it should.
+
+    Critical settings:
+
+        Chatml or Jinja Template (embedded, but updated version at repo below)
+        Rep pen of 1.01 or 1.02 ; higher (1.04, 1.05) will result in "Harley Mode".
+        Temp range of .6 to 1.2. ; higher you may need to prompt the model to "output" after thinking.
+        Experts set at 8-10 ; higher will result in "odder" output BUT it might be better.
+
+    That being said, "Harley Quinn" may make her presence known at any moment.
+
+    USAGE GUIDE:
+
+    Please refer to this model card for
+
+        Specific usage, suggested settings, changing ACTIVE EXPERTS, templates, settings and the like:
+        How to maximize this model in "uncensored" form, with specific notes on "abliterated" models.
+        Rep pen / temp settings specific to getting the model to perform strongly.
+
+    https://huggingface.co/DavidAU/Qwen3-18B-A3B-Stranger-Thoughts-Abliterated-Uncensored-GGUF
+
+    GGUF / QUANTS / SPECIAL SHOUTOUT:
+
+    Special thanks to team Mradermacher for making the quants!
+
+    https://huggingface.co/mradermacher/Qwen3-22B-A3B-The-Harley-Quinn-GGUF
+
+    KNOWN ISSUES:
+
+        Model may "mis-capitalize" word(s) - lowercase, where uppercase should be - from time to time.
+        Model may add extra space from time to time before a word.
+        Incorrect template and/or settings will result in a drop in performance / poor performance.
+        Can rant at the end / repeat. Most of the time it will stop on its own.
+
+    Looking for the Abliterated / Uncensored version?
+
+    https://huggingface.co/DavidAU/Qwen3-23B-A3B-The-Harley-Quinn-PUDDIN-Abliterated-Uncensored
+
+    In some cases this "abliterated/uncensored" version may work better than this version.
+    EXAMPLES
+
+    Standard system prompt, rep pen 1.01-1.02, topk 100, topp .95, minp .05, rep pen range 64.
+
+    Tested in LMStudio, quant Q4KS, GPU (CPU output will differ slightly).
+
+    As this is the mid range quant, expected better results from higher quants and/or with more experts activated to be better.
+
+    NOTE: Some formatting lost on copy/paste.
+
+    WARNING: NSFW. Vivid prose. INTENSE. Visceral Details. Violence. HORROR. GORE. Swearing. UNCENSORED... humor, romance, fun.
+  overrides:
+    parameters:
+      model: Qwen3-22B-A3B-The-Harley-Quinn.Q4_K_M.gguf
+  files:
+    - filename: Qwen3-22B-A3B-The-Harley-Quinn.Q4_K_M.gguf
+      sha256: a3666754efde5d6c054de53cff0f38f1bb4a20117e2502eed7018ae57017b0a2
+      uri: huggingface://mradermacher/Qwen3-22B-A3B-The-Harley-Quinn-GGUF/Qwen3-22B-A3B-The-Harley-Quinn.Q4_K_M.gguf
+- !!merge <<: *qwen3
+  name: "qwen3-33b-a3b-stranger-thoughts-abliterated-uncensored"
+  icon: https://huggingface.co/DavidAU/Qwen3-33B-A3B-Stranger-Thoughts-Abliterated-Uncensored/resolve/main/qwen3-33b-ablit.jpg
+  urls:
+    - https://huggingface.co/DavidAU/Qwen3-33B-A3B-Stranger-Thoughts-Abliterated-Uncensored
+    - https://huggingface.co/mradermacher/Qwen3-33B-A3B-Stranger-Thoughts-Abliterated-Uncensored-GGUF
+  description: |
+    WARNING: NSFW. Vivid prose. INTENSE. Visceral Details. Violence. HORROR. GORE. Swearing. UNCENSORED... humor, romance, fun.
+    Qwen3-33B-A3B-Stranger-Thoughts-Abliterated-Uncensored
+
+    This repo contains the full precision source code, in "safe tensors" format to generate GGUFs, GPTQ, EXL2, AWQ, HQQ and other formats. The source code can also be used directly.
+
+    ABOUT:
+
+    A stranger, yet radically different version of "Qwen/Qwen3-30B-A3B", abliterated by "huihui-ai" , with 4 added layers expanding the model to 33B total parameters.
+
+    The goal: slightly alter the model, to address some odd creative thinking and output choices AND de-censor it.
+
+    Please note that the modifications affect the entire model operation; roughly I adjusted the model to think a little "deeper" and "ponder" a bit - but this is a very rough description.
+
+    I also ran reasoning tests (non-creative) to ensure model was not damaged and roughly matched original model performance.
+
+    That being said, reasoning and output generation will be altered regardless of your use case(s)
+  overrides:
+    parameters:
+      model: Qwen3-33B-A3B-Stranger-Thoughts-Abliterated-Uncensored.Q4_K_M.gguf
+  files:
+    - filename: Qwen3-33B-A3B-Stranger-Thoughts-Abliterated-Uncensored.Q4_K_M.gguf
+      sha256: fc0f028ab04d4643032e5bf65c3b51ba947e97b4f562c4fc25c06b6a20b14616
+      uri: huggingface://mradermacher/Qwen3-33B-A3B-Stranger-Thoughts-Abliterated-Uncensored-GGUF/Qwen3-33B-A3B-Stranger-Thoughts-Abliterated-Uncensored.Q4_K_M.gguf
+- !!merge <<: *qwen3
+  name: "pinkpixel_crystal-think-v2"
+  icon: https://huggingface.co/PinkPixel/Crystal-Think-V2/resolve/main/crystal-think-v2-logo.png
+  urls:
+    - https://huggingface.co/PinkPixel/Crystal-Think-V2
+    - https://huggingface.co/bartowski/PinkPixel_Crystal-Think-V2-GGUF
+  description: |
+    Crystal-Think is a specialized mathematical reasoning model based on Qwen3-4B, fine-tuned using Group Relative Policy Optimization (GRPO) on NVIDIA's OpenMathReasoning dataset. Version 2 introduces the new <think></think> reasoning format for enhanced step-by-step mathematical problem solving, algebraic reasoning, and mathematical code generation.
+  overrides:
+    parameters:
+      model: PinkPixel_Crystal-Think-V2-Q4_K_M.gguf
+  files:
+    - filename: PinkPixel_Crystal-Think-V2-Q4_K_M.gguf
+      sha256: 10f2558089c90bc9ef8036ac0b1142ad8991902ec83840a00710fd654df19aaa
+      uri: huggingface://bartowski/PinkPixel_Crystal-Think-V2-GGUF/PinkPixel_Crystal-Think-V2-Q4_K_M.gguf
+- !!merge <<: *qwen3
+  name: "helpingai_dhanishtha-2.0-preview"
+  urls:
+    - https://huggingface.co/HelpingAI/Dhanishtha-2.0-preview
+    - https://huggingface.co/bartowski/HelpingAI_Dhanishtha-2.0-preview-GGUF
+  description: |
+    What makes Dhanishtha-2.0 special? Imagine an AI that doesn't just answer your questions instantly, but actually thinks through problems step-by-step, shows its work, and can even change its mind when it realizes a better approach. That's Dhanishtha-2.0.
+    Quick Summary:
+        🚀 For Everyone: An AI that shows its thinking process and can reconsider its reasoning
+        👩‍💻 For Developers: First model with intermediate thinking capabilities, 39+ language support
+    Dhanishtha-2.0 is a state-of-the-art (SOTA) model developed by HelpingAI, representing the world's first model to feature Intermediate Thinking capabilities. Unlike traditional models that provide single-pass responses, Dhanishtha-2.0 employs a revolutionary multi-phase thinking process that allows the model to think, reconsider, and refine its reasoning multiple times throughout a single response.
+  overrides:
+    parameters:
+      model: HelpingAI_Dhanishtha-2.0-preview-Q4_K_M.gguf
+  files:
+    - filename: HelpingAI_Dhanishtha-2.0-preview-Q4_K_M.gguf
+      sha256: 026a1f80187c9ecdd0227816a35661f3b6b7abe85971121b4c1c25b6cdd7ab86
+      uri: huggingface://bartowski/HelpingAI_Dhanishtha-2.0-preview-GGUF/HelpingAI_Dhanishtha-2.0-preview-Q4_K_M.gguf
+- !!merge <<: *qwen3
+  name: "agentica-org_deepswe-preview"
+  icon: https://hebbkx1anhila5yf.public.blob.vercel-storage.com/IMG_3783-N75vmFhDaJtJkLR4d8pdBymos68DPo.png
+  urls:
+    - https://huggingface.co/agentica-org/DeepSWE-Preview
+    - https://huggingface.co/bartowski/agentica-org_DeepSWE-Preview-GGUF
+  description: |
+    DeepSWE-Preview is a fully open-sourced, state-of-the-art coding agent trained with only reinforcement learning (RL) to excel at software engineering (SWE) tasks. DeepSWE-Preview demonstrates strong reasoning capabilities in navigating complex codebases and viewing/editing multiple files, and it serves as a foundational model for future coding agents. The model achieves an impressive 59.0% on SWE-Bench-Verified, which is currently #1 in the open-weights category.
+
+    DeepSWE-Preview is trained on top of Qwen3-32B with thinking mode enabled. With just 200 steps of RL training, SWE-Bench-Verified score increases by ~20%.
+  overrides:
+    parameters:
+      model: agentica-org_DeepSWE-Preview-Q4_K_M.gguf
+  files:
+    - filename: agentica-org_DeepSWE-Preview-Q4_K_M.gguf
+      sha256: 196a7128d3b7a59f1647792bb72c17db306f773e78d5a47feeeea92e672d761b
+      uri: huggingface://bartowski/agentica-org_DeepSWE-Preview-GGUF/agentica-org_DeepSWE-Preview-Q4_K_M.gguf
 - &gemma3
  url: "github:mudler/LocalAI/gallery/gemma.yaml@master"
  name: "gemma-3-27b-it"
@@ -2006,6 +2337,35 @@
    - filename: gemma-3n-E4B-it-Q8_0.gguf
      sha256: 9f74079242c765116bd1f33123aa07160b5e93578c2d0032594b7ed97576f9c3
      uri: huggingface://ggml-org/gemma-3n-E4B-it-GGUF/gemma-3n-E4B-it-Q8_0.gguf
+- !!merge <<: *gemma3
+  name: "gemma-3-4b-it-max-horror-uncensored-dbl-x-imatrix"
+  icon: https://huggingface.co/DavidAU/Gemma-3-4b-it-MAX-HORROR-Uncensored-DBL-X-Imatrix-GGUF/resolve/main/gemma4-horror-max2.jpg
+  urls:
+    - https://huggingface.co/DavidAU/Gemma-3-4b-it-MAX-HORROR-Uncensored-DBL-X-Imatrix-GGUF
+  description: |
+    Google's newest Gemma-3 model that has been uncensored by David_AU (maintains instruction following / model performance and adds 4 layers to the model) and re-enforced with a system prompt (optional) - see below.
+    The "Horror Imatrix" was built using Grand Horror 16B (at my repo). This adds a "tint" of horror to the model.
+
+    5 examples provided (NSFW / F-Bombs galore) below with prompts at IQ4XS (56 t/s on mid level card).
+
+    Context: 128k.
+
+    "MAXED"
+
+    This means the embed and output tensor are set at "BF16" (full precision) for all quants. This enhances quality, depth and general performance at the cost of a slightly larger quant.
+
+    "HORROR IMATRIX"
+
+    A strong, in house built, imatrix dataset built by David_AU which results in better overall function, instruction following, output quality and stronger connections to ideas, concepts and the world in general.
+
+    This combines with "MAXing" the quant to improve preformance.
+  overrides:
+    parameters:
+      model: Gemma-3-4b-it-MAX-HORROR-Uncensored-D_AU-Q4_K_M-imat.gguf
+  files:
+    - filename: Gemma-3-4b-it-MAX-HORROR-Uncensored-D_AU-Q4_K_M-imat.gguf
+      sha256: 1c577e4c84311c39b3d54b0cef12857ad46e88755f858143accbfcca7cc9fc6b
+      uri: huggingface://DavidAU/Gemma-3-4b-it-MAX-HORROR-Uncensored-DBL-X-Imatrix-GGUF/Gemma-3-4b-it-MAX-HORROR-Uncensored-D_AU-Q4_K_M-imat.gguf
 - &llama4
  url: "github:mudler/LocalAI/gallery/llama3.1-instruct.yaml@master"
  icon: https://avatars.githubusercontent.com/u/153379578
@@ -3284,6 +3644,38 @@
    - filename: sophosympatheia_StrawberryLemonade-L3-70B-v1.0-Q4_K_M.gguf
      sha256: 354472a2946598e0df376f9ecb91f83d7bc9c1b32db46bf48d3ea76f892f2a97
      uri: huggingface://bartowski/sophosympatheia_StrawberryLemonade-L3-70B-v1.0-GGUF/sophosympatheia_StrawberryLemonade-L3-70B-v1.0-Q4_K_M.gguf
+- !!merge <<: *llama33
+  name: "steelskull_l3.3-shakudo-70b"
+  icon: https://cdn-uploads.huggingface.co/production/uploads/64545af5ec40bbbd01242ca6/Y3_fED_Re3U1rd0jOPnAR.jpeg
+  urls:
+    - https://huggingface.co/Steelskull/L3.3-Shakudo-70b
+    - https://huggingface.co/bartowski/Steelskull_L3.3-Shakudo-70b-GGUF
+  description: |
+    L3.3-Shakudo-70b is the result of a multi-stage merging process by Steelskull, designed to create a powerful and creative roleplaying model with a unique flavor. The creation process involved several advanced merging techniques, including weight twisting, to achieve its distinct characteristics.
+    Stage 1: The Cognitive Foundation & Weight Twisting
+
+    The process began by creating a cognitive and tool-use focused base model, L3.3-Cogmoblated-70B. This was achieved through a `model_stock` merge of several models known for their reasoning and instruction-following capabilities. This base was built upon `nbeerbower/Llama-3.1-Nemotron-lorablated-70B`, a model intentionally "ablated" to skew refusal behaviors. This technique, known as weight twisting, helps the final model adopt more desirable response patterns by building upon a foundation that is already aligned against common refusal patterns.
+    Stage 2: The Twin Hydrargyrum - Flavor and Depth
+
+    Two distinct models were then created from the Cogmoblated base:
+
+        L3.3-M1-Hydrargyrum-70B: This model was merged using `SCE`, a technique that enhances creative writing and prose style, giving the model its unique "flavor." The Top_K for this merge were set at 0.22 .
+        L3.3-M2-Hydrargyrum-70B: This model was created using a `Della_Linear` merge, which focuses on integrating the "depth" of various roleplaying and narrative models. The settings for this merge were set at: (lambda: 1.1) (weight: 0.2) (density: 0.7) (epsilon: 0.2)
+
+    Final Stage: Shakudo
+
+    The final model, L3.3-Shakudo-70b, was created by merging the two Hydrargyrum variants using a 50/50 `nuslerp`. This final step combines the rich, creative prose (flavor) from the SCE merge with the strong roleplaying capabilities (depth) from the Della_Linear merge, resulting in a model with a distinct and refined narrative voice.
+
+    A special thank you to Nectar.ai for their generous support of the open-source community and my projects.
+
+    Additionally, a heartfelt thanks to all the Ko-fi supporters who have contributed—your generosity is deeply appreciated and helps keep this work going and the Pods spinning.
+  overrides:
+    parameters:
+      model: Steelskull_L3.3-Shakudo-70b-Q4_K_M.gguf
+  files:
+    - filename: Steelskull_L3.3-Shakudo-70b-Q4_K_M.gguf
+      sha256: 54590c02226f12c6f48a4af6bfed0e3c90130addd1fb8a2b4fcc1f0ab1674ef7
+      uri: huggingface://bartowski/Steelskull_L3.3-Shakudo-70b-GGUF/Steelskull_L3.3-Shakudo-70b-Q4_K_M.gguf
 - &rwkv
  url: "github:mudler/LocalAI/gallery/rwkv.yaml@master"
  name: "rwkv-6-world-7b"
@@ -10599,6 +10991,31 @@
    - filename: AstroSage-70B.Q4_K_M.gguf
      sha256: 1d98dabfa001d358d9f95d2deba93a94ad8baa8839c75a0129cdb6bcf1507f38
      uri: huggingface://mradermacher/AstroSage-70B-GGUF/AstroSage-70B.Q4_K_M.gguf
+- !!merge <<: *llama31
+  name: "thedrummer_anubis-70b-v1.1"
+  icon: https://cdn-uploads.huggingface.co/production/uploads/65f2fd1c25b848bd061b5c2e/G-NwpVtnbdfdnPusYDzx3.png
+  urls:
+    - https://huggingface.co/TheDrummer/Anubis-70B-v1.1
+    - https://huggingface.co/bartowski/TheDrummer_Anubis-70B-v1.1-GGUF
+  description: |
+    A follow up to Anubis 70B v1.0 but with two main strengths: character adherence and unalignment.
+
+    This is not a minor update to Anubis. It is a totally different beast.
+
+        The model does a fantastic job portraying my various characters without fail, adhering to them in such a refreshing and pleasing degree with their dialogue and mannerisms, while also being able to impart a very nice and fresh style that doesn't feel like any other L3.3 models.
+
+        I do think it's a solid improvement though, like it nails characters.
+
+        It feels fresh. I am quite impressed on how it picked up on and empasized subtle details I have not seen other models do in one of my historically accurate character cards.
+
+        Anubis v1.1 is in my main model rotation now, I really like it! -Tarek
+  overrides:
+    parameters:
+      model: TheDrummer_Anubis-70B-v1.1-Q4_K_M.gguf
+  files:
+    - filename: TheDrummer_Anubis-70B-v1.1-Q4_K_M.gguf
+      sha256: a73bed551c64703737f598f1120aac28d1a62c08b5dbe2208da810936bb2522d
+      uri: huggingface://bartowski/TheDrummer_Anubis-70B-v1.1-GGUF/TheDrummer_Anubis-70B-v1.1-Q4_K_M.gguf
 - &deepseek
  url: "github:mudler/LocalAI/gallery/deepseek.yaml@master" ## Deepseek
  name: "deepseek-coder-v2-lite-instruct"
@@ -12905,6 +13322,94 @@
    - filename: Delta-Vector_Austral-24B-Winton-Q4_K_M.gguf
      sha256: feb76e0158d1ebba1809de89d01671b86037f768ebd5f6fb165885ae6338b1b7
      uri: huggingface://bartowski/Delta-Vector_Austral-24B-Winton-GGUF/Delta-Vector_Austral-24B-Winton-Q4_K_M.gguf
+- !!merge <<: *mistral03
+  name: "mistral-small-3.2-46b-the-brilliant-raconteur-ii-instruct-2506"
+  url: "github:mudler/LocalAI/gallery/chatml.yaml@master"
+  icon: https://huggingface.co/DavidAU/Mistral-Small-3.2-46B-The-Brilliant-Raconteur-II-Instruct-2506/resolve/main/mistral-2506.jpg
+  urls:
+    - https://huggingface.co/DavidAU/Mistral-Small-3.2-46B-The-Brilliant-Raconteur-II-Instruct-2506
+    - https://huggingface.co/mradermacher/Mistral-Small-3.2-46B-The-Brilliant-Raconteur-II-Instruct-2506-GGUF
+  description: |
+    WARNING: MADNESS - UN HINGED and... NSFW. Vivid prose. INTENSE. Visceral Details. Violence. HORROR. GORE. Swearing. UNCENSORED... humor, romance, fun.
+    Mistral-Small-3.2-46B-The-Brilliant-Raconteur-II-Instruct-2506
+
+    This repo contains the full precision source code, in "safe tensors" format to generate GGUFs, GPTQ, EXL2, AWQ, HQQ and other formats. The source code can also be used directly.
+
+    ABOUT:
+
+    A stronger, more creative Mistral (Mistral-Small-3.2-24B-Instruct-2506) extended to 79 layers, 46B parameters with Brainstorm 40x by DavidAU (details at very bottom of the page). This is version II, which has a jump in detail, and raw emotion relative to version 1.
+
+    This model pushes Mistral's Instruct 2506 to the limit:
+
+        Regens will be very different, even with same prompt / settings.
+        Output generation will vary vastly on each generation.
+        Reasoning will be changed, and often shorter.
+        Prose, creativity, word choice, and general "flow" are improved.
+        Several system prompts below help push this model even further.
+        Model is partly de-censored / abliterated. Most Mistrals are more uncensored that most other models too.
+        This model can also be used for coding too; even at low quants.
+        Model can be used for all use cases too.
+
+    As this is an instruct model, this model thrives on instructions - both in the system prompt and/or the prompt itself.
+
+    One example below with 3 generations using Q4_K_S.
+
+    Second example below with 2 generations using Q4_K_S.
+
+    Quick Details:
+
+        Model is 128k context, Jinja template (embedded) OR Chatml Template.
+        Reasoning can be turned on/off (see system prompts below) and is OFF by default.
+        Temp range .1 to 1 suggested, with 1-2 for enhanced creative. Above temp 2, is strong but can be very different.
+        Rep pen range: 1 (off) or very light 1.01, 1.02 to 1.05. (model is sensitive to rep pen - this affects reasoning / generation length.)
+        For creative/brainstorming use: suggest 2-5 generations due to variations caused by Brainstorm.
+
+    Observations:
+
+        Sometimes using Chatml (or Alpaca / others ) template (VS Jinja) will result in stronger creative generation.
+        Model can be operated with NO system prompt; however a system prompt will enhance generation.
+        Longer prompts, that more detailed, with more instructions will result in much stronger generations.
+        For prose directives: You may need to add directions, because the model may follow your instructions too closely. IE: "use short sentences" vs "use short sentences sparsely".
+        Reasoning (on) can lead to better creative generation, however sometimes generation with reasoning off is better.
+        Rep pen of up to 1.05 may be needed on quants Q2k/q3ks for some prompts to address "low bit" issues.
+
+    Detailed settings, system prompts, how to and examples below.
+
+    NOTES:
+
+    Image generation should also be possible with this model, just like the base model. Brainstorm was not applied to the image generation systems of the model... yet.
+
+    This is Version II and subject to change / revision.
+
+    This model is a slightly different version of:
+
+    https://huggingface.co/DavidAU/Mistral-Small-3.2-46B-The-Brilliant-Raconteur-Instruct-2506
+  overrides:
+    parameters:
+      model: Mistral-Small-3.2-46B-The-Brilliant-Raconteur-II-Instruct-2506.Q4_K_M.gguf
+  files:
+    - filename: Mistral-Small-3.2-46B-The-Brilliant-Raconteur-II-Instruct-2506.Q4_K_M.gguf
+      sha256: 5c8b6f21ae4f671880fafe60001f30f4c639a680e257701e474777cfcf00f8f6
+      uri: huggingface://mradermacher/Mistral-Small-3.2-46B-The-Brilliant-Raconteur-II-Instruct-2506-GGUF/Mistral-Small-3.2-46B-The-Brilliant-Raconteur-II-Instruct-2506.Q4_K_M.gguf
+- !!merge <<: *mistral03
+  name: "zerofata_ms3.2-paintedfantasy-visage-33b"
+  icon: https://cdn-uploads.huggingface.co/production/uploads/65b19c6c638328850e12d38c/CQeog2SHdGUdmx8vHqL71.png
+  urls:
+    - https://huggingface.co/zerofata/MS3.2-PaintedFantasy-Visage-33B
+    - https://huggingface.co/bartowski/zerofata_MS3.2-PaintedFantasy-Visage-33B-GGUF
+  description: |
+    Another experimental release. Mistral Small 3.2 24B upscaled by 18 layers to create a 33.6B model. This model then went through pretraining, SFT & DPO.
+
+    Can't guarantee the Mistral 3.2 repetition issues are fixed, but this model seems to be less repetitive than my previous attempt.
+
+    This is an uncensored creative model intended to excel at character driven RP / ERP where characters are portrayed creatively and proactively.
+  overrides:
+    parameters:
+      model: zerofata_MS3.2-PaintedFantasy-Visage-33B-Q4_K_M.gguf
+  files:
+    - filename: zerofata_MS3.2-PaintedFantasy-Visage-33B-Q4_K_M.gguf
+      sha256: bd315ad9a4cf0f47ed24f8d387b0cad1dd127e10f2bbe1c6820ae91f700ada56
+      uri: huggingface://bartowski/zerofata_MS3.2-PaintedFantasy-Visage-33B-GGUF/zerofata_MS3.2-PaintedFantasy-Visage-33B-Q4_K_M.gguf
 - &mudler
  url: "github:mudler/LocalAI/gallery/mudler.yaml@master" ### START mudler's LocalAI specific-models
  name: "LocalAI-llama3-8b-function-call-v0.2"
--- a/pkg/downloader/uri.go
+++ b/pkg/downloader/uri.go
@@ -256,7 +256,7 @@ func (uri URI) DownloadFile(filePath, sha string, fileN, total int, downloadStat
 			return fmt.Errorf("failed to get image %q: %v", url, err)
 		}

-		return oci.ExtractOCIImage(img, filepath.Dir(filePath), downloadStatus)
+		return oci.ExtractOCIImage(img, url, filepath.Dir(filePath), downloadStatus)
 	}

 	// Check if the file already exists
--- a/pkg/oci/image.go
+++ b/pkg/oci/image.go
@@ -6,6 +6,7 @@ import (
 	"fmt"
 	"io"
 	"net/http"
+	"os"
 	"runtime"
 	"strconv"
 	"strings"
@@ -21,6 +22,7 @@ import (
 	"github.com/google/go-containerregistry/pkg/v1/mutate"
 	"github.com/google/go-containerregistry/pkg/v1/remote"
 	"github.com/google/go-containerregistry/pkg/v1/remote/transport"
+	"github.com/google/go-containerregistry/pkg/v1/tarball"
 )

 // ref: https://github.com/mudler/luet/blob/master/pkg/helpers/docker/docker.go#L117
@@ -95,31 +97,28 @@ func (pw *progressWriter) Write(p []byte) (int, error) {
 }

 // ExtractOCIImage will extract a given targetImage into a given targetDestination
-func ExtractOCIImage(img v1.Image, targetDestination string, downloadStatus func(string, string, string, float64)) error {
-	var reader io.Reader
-	reader = mutate.Extract(img)
+func ExtractOCIImage(img v1.Image, imageRef string, targetDestination string, downloadStatus func(string, string, string, float64)) error {
+	// Create a temporary tar file
+	tmpTarFile, err := os.CreateTemp("", "localai-oci-*.tar")
+	if err != nil {
+		return fmt.Errorf("failed to create temporary tar file: %v", err)
+	}
+	defer os.Remove(tmpTarFile.Name())
+	defer tmpTarFile.Close()

-	if downloadStatus != nil {
-		var totalSize int64
-		layers, err := img.Layers()
-		if err != nil {
-			return err
-		}
-		for _, layer := range layers {
-			size, err := layer.Size()
-			if err != nil {
-				return err
-			}
-			totalSize += size
-		}
-		reader = io.TeeReader(reader, &progressWriter{total: totalSize, downloadStatus: downloadStatus})
+	// Download the image as tar with progress tracking
+	err = DownloadOCIImageTar(img, imageRef, tmpTarFile.Name(), downloadStatus)
+	if err != nil {
+		return fmt.Errorf("failed to download image tar: %v", err)
 	}

-	_, err := archive.Apply(context.Background(),
-		targetDestination, reader,
-		archive.WithNoSameOwner())
+	// Extract the tar file to the target destination
+	err = ExtractOCIImageFromTar(tmpTarFile.Name(), imageRef, targetDestination, downloadStatus)
+	if err != nil {
+		return fmt.Errorf("failed to extract image tar: %v", err)
+	}

-	return err
+	return nil
 }

 func ParseImageParts(image string) (tag, repository, dstimage string) {
@@ -205,3 +204,164 @@ func GetOCIImageSize(targetImage, targetPlatform string, auth *registrytypes.Aut

 	return size, nil
 }
+
+// DownloadOCIImageTar downloads the compressed layers of an image and then creates an uncompressed tar
+// This provides accurate size estimation and allows for later extraction
+func DownloadOCIImageTar(img v1.Image, imageRef string, tarFilePath string, downloadStatus func(string, string, string, float64)) error {
+	// Get layers to calculate total compressed size for estimation
+	layers, err := img.Layers()
+	if err != nil {
+		return fmt.Errorf("failed to get layers: %v", err)
+	}
+
+	// Calculate total compressed size for progress tracking
+	var totalCompressedSize int64
+	for _, layer := range layers {
+		size, err := layer.Size()
+		if err != nil {
+			return fmt.Errorf("failed to get layer size: %v", err)
+		}
+		totalCompressedSize += size
+	}
+
+	// Create a temporary directory to store the compressed layers
+	tmpDir, err := os.MkdirTemp("", "localai-oci-layers-*")
+	if err != nil {
+		return fmt.Errorf("failed to create temporary directory: %v", err)
+	}
+	defer os.RemoveAll(tmpDir)
+
+	// Download all compressed layers with progress tracking
+	var downloadedLayers []v1.Layer
+	var downloadedSize int64
+
+	// Extract image name from the reference for display
+	imageName := imageRef
+	for i, layer := range layers {
+		layerSize, err := layer.Size()
+		if err != nil {
+			return fmt.Errorf("failed to get layer size: %v", err)
+		}
+
+		// Create a temporary file for this layer
+		layerFile := fmt.Sprintf("%s/layer-%d.tar.gz", tmpDir, i)
+		file, err := os.Create(layerFile)
+		if err != nil {
+			return fmt.Errorf("failed to create layer file: %v", err)
+		}
+
+		// Create progress writer for this layer
+		var writer io.Writer = file
+		if downloadStatus != nil {
+			writer = io.MultiWriter(file, &progressWriter{
+				total:          totalCompressedSize,
+				fileName:       fmt.Sprintf("Downloading %d/%d %s", i+1, len(layers), imageName),
+				downloadStatus: downloadStatus,
+			})
+		}
+
+		// Download the compressed layer
+		layerReader, err := layer.Compressed()
+		if err != nil {
+			file.Close()
+			return fmt.Errorf("failed to get compressed layer: %v", err)
+		}
+
+		_, err = io.Copy(writer, layerReader)
+		file.Close()
+		if err != nil {
+			return fmt.Errorf("failed to download layer %d: %v", i, err)
+		}
+
+		// Load the downloaded layer
+		downloadedLayer, err := tarball.LayerFromFile(layerFile)
+		if err != nil {
+			return fmt.Errorf("failed to load downloaded layer: %v", err)
+		}
+
+		downloadedLayers = append(downloadedLayers, downloadedLayer)
+		downloadedSize += layerSize
+	}
+
+	// Create a local image from the downloaded layers
+	localImg, err := mutate.AppendLayers(img, downloadedLayers...)
+	if err != nil {
+		return fmt.Errorf("failed to create local image: %v", err)
+	}
+
+	// Now extract the uncompressed tar from the local image
+	tarFile, err := os.Create(tarFilePath)
+	if err != nil {
+		return fmt.Errorf("failed to create tar file: %v", err)
+	}
+	defer tarFile.Close()
+
+	// Extract uncompressed tar from local image
+	extractReader := mutate.Extract(localImg)
+	_, err = io.Copy(tarFile, extractReader)
+	if err != nil {
+		return fmt.Errorf("failed to extract uncompressed tar: %v", err)
+	}
+
+	return nil
+}
+
+// ExtractOCIImageFromTar extracts an image from a previously downloaded tar file
+func ExtractOCIImageFromTar(tarFilePath, imageRef, targetDestination string, downloadStatus func(string, string, string, float64)) error {
+	// Open the tar file
+	tarFile, err := os.Open(tarFilePath)
+	if err != nil {
+		return fmt.Errorf("failed to open tar file: %v", err)
+	}
+	defer tarFile.Close()
+
+	// Get file size for progress tracking
+	fileInfo, err := tarFile.Stat()
+	if err != nil {
+		return fmt.Errorf("failed to get file info: %v", err)
+	}
+
+	var reader io.Reader = tarFile
+	if downloadStatus != nil {
+		reader = io.TeeReader(tarFile, &progressWriter{
+			total:          fileInfo.Size(),
+			fileName:       fmt.Sprintf("Extracting %s", imageRef),
+			downloadStatus: downloadStatus,
+		})
+	}
+
+	// Extract the tar file
+	_, err = archive.Apply(context.Background(),
+		targetDestination, reader,
+		archive.WithNoSameOwner())
+
+	return err
+}
+
+// GetOCIImageUncompressedSize returns the total uncompressed size of an image
+func GetOCIImageUncompressedSize(targetImage, targetPlatform string, auth *registrytypes.AuthConfig, t http.RoundTripper) (int64, error) {
+	var totalSize int64
+	var img v1.Image
+	var err error
+
+	img, err = GetImage(targetImage, targetPlatform, auth, t)
+	if err != nil {
+		return totalSize, err
+	}
+
+	layers, err := img.Layers()
+	if err != nil {
+		return totalSize, err
+	}
+
+	for _, layer := range layers {
+		// Use compressed size as an approximation since uncompressed size is not directly available
+		size, err := layer.Size()
+		if err != nil {
+			return totalSize, err
+		}
+		totalSize += size
+	}
+
+	return totalSize, nil
+}
--- a/pkg/oci/image_test.go
+++ b/pkg/oci/image_test.go
@@ -30,7 +30,7 @@ var _ = Describe("OCI", func() {
 			Expect(err).NotTo(HaveOccurred())
 			defer os.RemoveAll(dir)

-			err = ExtractOCIImage(img, dir, nil)
+			err = ExtractOCIImage(img, imageName, dir, nil)
 			Expect(err).NotTo(HaveOccurred())
 		})
 	})
--- a/pkg/startup/backend_preload.go
+++ b/pkg/startup/backend_preload.go
@@ -27,7 +27,7 @@ func InstallExternalBackends(galleries []config.Gallery, backendPath string, dow
 				errs = errors.Join(err, fmt.Errorf("error installing backend %s", backend))
 			}
 		default:
-			err := gallery.InstallBackendFromGallery(galleries, systemState, backend, backendPath, downloadStatus)
+			err := gallery.InstallBackendFromGallery(galleries, systemState, backend, backendPath, downloadStatus, true)
 			if err != nil {
 				errs = errors.Join(err, fmt.Errorf("error installing backend %s", backend))
 			}
--- a/pkg/startup/model_preload.go
+++ b/pkg/startup/model_preload.go
@@ -17,7 +17,7 @@ import (
 // InstallModels will preload models from the given list of URLs and galleries
 // It will download the model if it is not already present in the model path
 // It will also try to resolve if the model is an embedded model YAML configuration
-func InstallModels(galleries []config.Gallery, modelPath string, enforceScan bool, downloadStatus func(string, string, string, float64), models ...string) error {
+func InstallModels(galleries, backendGalleries []config.Gallery, modelPath, backendBasePath string, enforceScan, autoloadBackendGalleries bool, downloadStatus func(string, string, string, float64), models ...string) error {
 	// create an error that groups all errors
 	var err error

@@ -99,7 +99,7 @@ func InstallModels(galleries []config.Gallery, modelPath string, enforceScan boo
 				}
 			} else {
 				// Check if it's a model gallery, or print a warning
-				e, found := installModel(galleries, url, modelPath, downloadStatus, enforceScan)
+				e, found := installModel(galleries, backendGalleries, url, modelPath, backendBasePath, downloadStatus, enforceScan, autoloadBackendGalleries)
 				if e != nil && found {
 					log.Error().Err(err).Msgf("[startup] failed installing model '%s'", url)
 					err = errors.Join(err, e)
@@ -113,7 +113,7 @@ func InstallModels(galleries []config.Gallery, modelPath string, enforceScan boo
 	return err
 }

-func installModel(galleries []config.Gallery, modelName, modelPath string, downloadStatus func(string, string, string, float64), enforceScan bool) (error, bool) {
+func installModel(galleries, backendGalleries []config.Gallery, modelName, modelPath, backendBasePath string, downloadStatus func(string, string, string, float64), enforceScan, autoloadBackendGalleries bool) (error, bool) {
 	models, err := gallery.AvailableGalleryModels(galleries, modelPath)
 	if err != nil {
 		return err, false
@@ -129,7 +129,7 @@ func installModel(galleries []config.Gallery, modelName, modelPath string, downl
 	}

 	log.Info().Str("model", modelName).Str("license", model.License).Msg("installing model")
-	err = gallery.InstallModelFromGallery(galleries, modelName, modelPath, gallery.GalleryModel{}, downloadStatus, enforceScan)
+	err = gallery.InstallModelFromGallery(galleries, backendGalleries, modelName, modelPath, backendBasePath, gallery.GalleryModel{}, downloadStatus, enforceScan, autoloadBackendGalleries)
 	if err != nil {
 		return err, true
 	}
--- a/pkg/startup/model_preload_test.go
+++ b/pkg/startup/model_preload_test.go
@@ -21,7 +21,7 @@ var _ = Describe("Preload test", func() {
 			url := "https://raw.githubusercontent.com/mudler/LocalAI-examples/main/configurations/phi-2.yaml"
 			fileName := fmt.Sprintf("%s.yaml", "phi-2")

-			InstallModels([]config.Gallery{}, tmpdir, true, nil, url)
+			InstallModels([]config.Gallery{}, []config.Gallery{}, tmpdir, "", true, true, nil, url)

 			resultFile := filepath.Join(tmpdir, fileName)

@@ -36,7 +36,7 @@ var _ = Describe("Preload test", func() {
 			url := "huggingface://TheBloke/TinyLlama-1.1B-Chat-v0.3-GGUF/tinyllama-1.1b-chat-v0.3.Q2_K.gguf"
 			fileName := fmt.Sprintf("%s.gguf", "tinyllama-1.1b-chat-v0.3.Q2_K")

-			err = InstallModels([]config.Gallery{}, tmpdir, false, nil, url)
+			err = InstallModels([]config.Gallery{}, []config.Gallery{}, tmpdir, "", false, true, nil, url)
 			Expect(err).ToNot(HaveOccurred())

 			resultFile := filepath.Join(tmpdir, fileName)
Author	SHA1	Message	Date
Ettore Di Giacinto	057d5c25f1	test CI (remove me) Signed-off-by: Ettore Di Giacinto <mudler@localai.io>	2025-07-04 18:49:15 +02:00
Ettore Di Giacinto	cdf70db0d6	chore(deps): switch to ubuntu 24.04 Signed-off-by: Ettore Di Giacinto <mudler@localai.io>	2025-07-04 18:48:53 +02:00
Ettore Di Giacinto	c35dd0a7b8	chore(model gallery): add zerofata_ms3.2-paintedfantasy-visage-33b (#5793 ) Signed-off-by: Ettore Di Giacinto <mudler@localai.io>	2025-07-04 18:42:01 +02:00
Ettore Di Giacinto	2f5af6b246	chore(model gallery): add agentica-org_deepswe-preview (#5792 ) Signed-off-by: Ettore Di Giacinto <mudler@localai.io>	2025-07-04 18:39:36 +02:00
Ettore Di Giacinto	00cf2e0e0a	chore(model gallery): add helpingai_dhanishtha-2.0-preview (#5791 ) Signed-off-by: Ettore Di Giacinto <mudler@localai.io>	2025-07-04 18:38:18 +02:00
LocalAI [bot]	c7a1d9c089	chore: ⬆️ Update ggml-org/llama.cpp to `bee28421be25fd447f61cb6db64d556cbfce32ec` (#5788 ) ⬆️ Update ggml-org/llama.cpp Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: mudler <2420543+mudler@users.noreply.github.com>	2025-07-04 08:46:56 +02:00
LocalAI [bot]	ad7ba52166	chore: ⬆️ Update PABannier/bark.cpp to `5d5be84f089ab9ea53b7a793f088d3fbf7247495` (#4786 ) ⬆️ Update PABannier/bark.cpp Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: mudler <2420543+mudler@users.noreply.github.com>	2025-07-03 22:44:53 +00:00
Ettore Di Giacinto	c5b9f45166	chore(cli): add backends CLI to manipulate and install backends (#5787 ) Signed-off-by: Ettore Di Giacinto <mudler@localai.io>	2025-07-03 19:31:27 +02:00
Ettore Di Giacinto	61b64a65ab	chore(bark-cpp): generalize and move to bark-cpp (#5786 ) Signed-off-by: Ettore Di Giacinto <mudler@localai.io>	2025-07-03 19:31:10 +02:00
Ettore Di Giacinto	8276952920	feat(system): detect and allow to override capabilities (#5785 ) Signed-off-by: Ettore Di Giacinto <mudler@localai.io>	2025-07-03 19:30:52 +02:00
Ettore Di Giacinto	b7cd5bfaec	feat(backends): add metas in the gallery (#5784 ) * chore(backends): add metas in the gallery Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * chore: correctly handle aliases and metas with same names Signed-off-by: Ettore Di Giacinto <mudler@localai.io> --------- Signed-off-by: Ettore Di Giacinto <mudler@localai.io>	2025-07-03 18:01:55 +02:00
LocalAI [bot]	da4312e4d3	chore: ⬆️ Update ggml-org/llama.cpp to `e75ba4c0434eb759eb7ff74e034ebe729053e575` (#5783 ) ⬆️ Update ggml-org/llama.cpp Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: mudler <2420543+mudler@users.noreply.github.com>	2025-07-03 10:01:17 +02:00
LocalAI [bot]	7d507c54ed	chore: ⬆️ Update ggml-org/whisper.cpp to `d9999d54c868b8bfcd376aa26067e787d53e679e` (#5782 ) ⬆️ Update ggml-org/whisper.cpp Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: mudler <2420543+mudler@users.noreply.github.com>	2025-07-03 09:57:36 +02:00
LocalAI [bot]	df7ed49889	docs: ⬆️ update docs version mudler/LocalAI (#5781 ) ⬆️ Update docs version mudler/LocalAI Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: mudler <2420543+mudler@users.noreply.github.com>	2025-07-02 22:45:21 +00:00
Ettore Di Giacinto	bfdc29d316	fix(gallery): correctly show status for downloading OCI images (#5774 ) We can't use the mutate.Extract written bytes as current status as that will be bigger than the compressed image size. Image manifest don't have any guarantee of the type of artifact (can be compressed or not) when showing the layer size. Split the extraction process in two parts: Downloading and extracting as a flattened system, in this way we can display the status of downloading and extracting accordingly. This change also fixes a small nuance in detecting installed backends, now it's more consistent and looks if a metadata.json and/or a path with a `run.sh` file is present. Signed-off-by: Ettore Di Giacinto <mudler@localai.io>	2025-07-02 08:25:48 +02:00
LocalAI [bot]	7fdc006071	chore: ⬆️ Update ggml-org/llama.cpp to `de569441470332ff922c23fb0413cc957be75b25` (#5777 ) ⬆️ Update ggml-org/llama.cpp Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: mudler <2420543+mudler@users.noreply.github.com>	2025-07-02 08:25:29 +02:00
LocalAI [bot]	615830245b	chore: ⬆️ Update ggml-org/whisper.cpp to `bca021c9740b267c2973fba56555be052006023a` (#5776 ) ⬆️ Update ggml-org/whisper.cpp Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: mudler <2420543+mudler@users.noreply.github.com>	2025-07-02 08:24:58 +02:00
LocalAI [bot]	61376c0fa7	docs: ⬆️ update docs version mudler/LocalAI (#5775 ) ⬆️ Update docs version mudler/LocalAI Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: mudler <2420543+mudler@users.noreply.github.com>	2025-07-01 22:45:24 +00:00
Ettore Di Giacinto	d0fb23514f	Revert "fix(gallery): correctly show status for downloading OCI images" This reverts commit `780d034ac9`.	2025-07-01 21:32:04 +02:00
Ettore Di Giacinto	780d034ac9	fix(gallery): correctly show status for downloading OCI images We can't use the mutate.Extract written bytes as current status as that will be bigger than the compressed image size. Image manifest don't have any guarantee of the type of artifact (can be compressed or not) when showing the layer size. Split the extraction process in two parts: Downloading and extracting as a flattened system, in this way we can display the status of downloading and extracting accordingly. This change also fixes a small nuance in detecting installed backends, now it's more consistent and looks if a metadata.json and/or a path with a `run.sh` file is present. Signed-off-by: Ettore Di Giacinto <mudler@localai.io>	2025-07-01 19:56:28 +02:00
Ettore Di Giacinto	ec2a044c7e	chore(model gallery): add pinkpixel_crystal-think-v2 (#5773 ) Signed-off-by: Ettore Di Giacinto <mudler@localai.io>	2025-07-01 16:18:19 +02:00
Ettore Di Giacinto	ad6fdd21fd	chore(model gallery): add steelskull_l3.3-shakudo-70b (#5772 ) Signed-off-by: Ettore Di Giacinto <mudler@localai.io>	2025-07-01 16:15:22 +02:00
Ettore Di Giacinto	cd94e6b352	chore(model gallery): add thedrummer_anubis-70b-v1.1 (#5771 ) Signed-off-by: Ettore Di Giacinto <mudler@localai.io>	2025-07-01 13:54:29 +02:00
Richard Palethorpe	b37cef3718	fix: Diffusers and XPU fixes (#5737 ) * fix(README): Add device flags for Intel/XPU Signed-off-by: Richard Palethorpe <io@richiejp.com> * fix(diffusers/xpu): Set device to XPU and ignore CUDA request when on Intel Signed-off-by: Richard Palethorpe <io@richiejp.com> --------- Signed-off-by: Richard Palethorpe <io@richiejp.com>	2025-07-01 12:36:17 +02:00
Dedy F. Setyawan	9f957d547d	fix(docs): Improve Header Responsiveness - Hide "Star us on GitHub!" on Mobile (#5770 )	2025-07-01 12:15:16 +02:00
LocalAI [bot]	f0d9f0c5d8	chore: ⬆️ Update ggml-org/llama.cpp to `0a5a3b5cdfd887cf0f8e09d9ff89dee130cfcdde` (#5759 ) ⬆️ Update ggml-org/llama.cpp Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: mudler <2420543+mudler@users.noreply.github.com>	2025-06-30 22:46:55 +00:00
LocalAI [bot]	d33e1c72a3	chore: ⬆️ Update ggml-org/llama.cpp to `caf5681fcb47dfe9bafee94ef9aa8f669ac986c7` (#5758 ) ⬆️ Update ggml-org/llama.cpp Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: mudler <2420543+mudler@users.noreply.github.com>	2025-06-29 22:49:47 +00:00
Ettore Di Giacinto	33f9ee06c9	fix(gallery): automatically install model from name (#5757 ) Signed-off-by: Ettore Di Giacinto <mudler@localai.io>	2025-06-29 17:42:58 +02:00
Ettore Di Giacinto	c54677402d	chore(model gallery): add qwen3-33b-a3b-stranger-thoughts-abliterated-uncensored (#5755 ) Signed-off-by: Ettore Di Giacinto <mudler@localai.io>	2025-06-29 10:04:33 +02:00
LocalAI [bot]	3fe3a7b23d	chore: ⬆️ Update ggml-org/llama.cpp to `27208bf657cfe7262791df473927225e48efe482` (#5753 ) ⬆️ Update ggml-org/llama.cpp Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: mudler <2420543+mudler@users.noreply.github.com>	2025-06-29 09:06:37 +02:00
LocalAI [bot]	f8ff6fa1fd	docs: ⬆️ update docs version mudler/LocalAI (#5752 ) ⬆️ Update docs version mudler/LocalAI Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: mudler <2420543+mudler@users.noreply.github.com>	2025-06-28 22:17:49 +02:00
Ettore Di Giacinto	dfadc3696e	feat(llama.cpp): allow to set kv-overrides (#5745 ) Signed-off-by: Ettore Di Giacinto <mudler@localai.io>	2025-06-28 21:26:07 +02:00
Ettore Di Giacinto	dbcf5fb4fc	chore(model gallery): add gemma-3-4b-it-max-horror-uncensored-dbl-x-imatrix (#5751 ) Signed-off-by: Ettore Di Giacinto <mudler@localai.io>	2025-06-28 18:18:14 +02:00
Ettore Di Giacinto	2633137a17	chore(model gallery): add qwen3-22b-a3b-the-harley-quinn (#5750 ) Signed-off-by: Ettore Di Giacinto <mudler@localai.io>	2025-06-28 18:17:10 +02:00
Ettore Di Giacinto	d9c17dd23b	chore(model gallery): add mistral-small-3.2-46b-the-brilliant-raconteur-ii-instruct-2506 (#5749 ) Signed-off-by: Ettore Di Giacinto <mudler@localai.io>	2025-06-28 18:15:25 +02:00
Ettore Di Giacinto	d8b7bd4860	chore(model gallery): add qwen3-42b-a3b-stranger-thoughts-deep20x-abliterated-uncensored-i1 (#5748 ) Signed-off-by: Ettore Di Giacinto <mudler@localai.io>	2025-06-28 18:12:17 +02:00
Ettore Di Giacinto	a611cbc0f4	chore(model gallery): add qwen3-55b-a3b-total-recall-deep-40x (#5747 ) Signed-off-by: Ettore Di Giacinto <mudler@localai.io>	2025-06-28 17:54:32 +02:00
Ettore Di Giacinto	850b525159	chore(model gallery): add qwen3-55b-a3b-total-recall-v1.3-i1 (#5746 ) Signed-off-by: Ettore Di Giacinto <mudler@localai.io>	2025-06-28 17:47:46 +02:00
Ettore Di Giacinto	35b3426a2a	Update README.md Signed-off-by: Ettore Di Giacinto <mudler@users.noreply.github.com>	2025-06-28 09:16:25 +02:00
LocalAI [bot]	cd2b0c0e7c	chore: ⬆️ Update ggml-org/llama.cpp to `72babea5dea56c8a8e8420ccf731b12a5cf37854` (#5743 ) ⬆️ Update ggml-org/llama.cpp Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: mudler <2420543+mudler@users.noreply.github.com>	2025-06-27 23:46:27 +02:00
LocalAI [bot]	73d80c43a8	chore: ⬆️ Update ggml-org/whisper.cpp to `c88ffbf9baeaae8c2cc0a4f496618314bb2ee9e0` (#5742 ) ⬆️ Update ggml-org/whisper.cpp Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: mudler <2420543+mudler@users.noreply.github.com>	2025-06-27 23:45:57 +02:00
LocalAI [bot]	665562b850	docs: ⬆️ update docs version mudler/LocalAI (#5741 ) ⬆️ Update docs version mudler/LocalAI Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: mudler <2420543+mudler@users.noreply.github.com>	2025-06-27 22:23:43 +02:00
Ettore Di Giacinto	7a78e4f482	fix(backends gallery): meta packages do not have URIs (#5740 ) Signed-off-by: Ettore Di Giacinto <mudler@localai.io>	2025-06-27 22:23:14 +02:00
Ettore Di Giacinto	6f41a6f934	fix(backends gallery): correctly identify gpu vendor (#5739 ) Signed-off-by: Ettore Di Giacinto <mudler@localai.io>	2025-06-27 22:22:58 +02:00
Ettore Di Giacinto	bb54f2da2b	feat(gallery): automatically install missing backends along models (#5736 ) Signed-off-by: Ettore Di Giacinto <mudler@localai.io>	2025-06-27 18:25:44 +02:00
Ettore Di Giacinto	e1cc7ee107	fix(ci): enable tag-latest to auto (#5738 ) Signed-off-by: Ettore Di Giacinto <mudler@localai.io>	2025-06-27 18:17:01 +02:00
Ettore Di Giacinto	cfc9dfa3d5	fix(ci): better handling of latest images for backends (#5735 ) Signed-off-by: Ettore Di Giacinto <mudler@localai.io>	2025-06-27 10:32:58 +02:00