ci(container): publish SenseVoice runtime to GHCR

Signed-off-by: LauraGPT <18321252+LauraGPT@users.noreply.github.com>

.github/workflows/sensevoice-container.yml

@@ -0,0 +1,78 @@
+name: SenseVoice container
+
+on:
+  pull_request:
+    paths:
+      - Dockerfile
+      - requirements.txt
+      - api.py
+      - model.py
+      - .github/workflows/sensevoice-container.yml
+      - tests/test_container_contract.py
+  push:
+    branches:
+      - main
+    tags:
+      - "v*"
+    paths:
+      - Dockerfile
+      - requirements.txt
+      - api.py
+      - model.py
+      - .github/workflows/sensevoice-container.yml
+      - tests/test_container_contract.py
+  workflow_dispatch:
+
+permissions:
+  contents: read
+  packages: write
+
+concurrency:
+  group: sensevoice-container-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  contract:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - name: Validate container contract
+        run: python -m unittest tests.test_container_contract
+
+  build:
+    needs: contract
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: docker/setup-buildx-action@v3
+
+      - name: Log in to GitHub Container Registry
+        if: github.event_name != 'pull_request'
+        uses: docker/login-action@v3
+        with:
+          registry: ghcr.io
+          username: ${{ github.actor }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+
+      - name: Generate image metadata
+        id: meta
+        uses: docker/metadata-action@v5
+        with:
+          images: ghcr.io/${{ github.repository_owner }}/sensevoice
+          tags: |
+            type=raw,value=latest,enable={{is_default_branch}}
+            type=semver,pattern={{version}}
+            type=sha
+
+      - name: Build and publish
+        uses: docker/build-push-action@v6
+        with:
+          context: .
+          platforms: linux/amd64
+          push: ${{ github.event_name != 'pull_request' }}
+          tags: ${{ steps.meta.outputs.tags }}
+          labels: ${{ steps.meta.outputs.labels }}
+          cache-from: type=gha
+          cache-to: type=gha,mode=max
+          provenance: true
+          sbom: true

Dockerfile

@@ -1,20 +1,26 @@
 # ======================================================
 #   FunASR SenseVoiceSmall Inference Server
 # ======================================================
-FROM pytorch/pytorch:2.3.1-cuda12.1-cudnn8-runtime
+FROM pytorch/pytorch:2.12.1-cuda12.6-cudnn9-runtime@sha256:79c5599719e0b1afdb56ac2d14588b530283752d7ae6ec3c36e18ec9deb8b229
+
+LABEL org.opencontainers.image.source="https://github.com/QwenAudio/SenseVoice" \
+      org.opencontainers.image.description="SenseVoiceSmall FastAPI inference server" \
+      org.opencontainers.image.licenses="Apache-2.0"
 
 # Install system dependencies
-RUN apt-get update && apt-get install -y \
-	ffmpeg libsndfile1 git && \
+RUN apt-get update && apt-get install -y --no-install-recommends \
+	ffmpeg libsndfile1 git python3-venv && \
 	rm -rf /var/lib/apt/lists/*
 
 WORKDIR /app
-RUN ls -la /app
+RUN python -m venv --system-site-packages /opt/sensevoice-venv
+ENV PATH=/opt/sensevoice-venv/bin:$PATH
+
 # Copy only requirements first
 COPY requirements.txt /app/
 
 # Install dependencies (cached if requirements.txt didn't change)
-RUN pip install --no-cache-dir -r requirements.txt
+RUN python -m pip install --no-cache-dir -r requirements.txt
 
 # Now copy the rest of your code
 COPY . /app
@@ -34,5 +40,8 @@ ENV MODELSCOPE_CACHE=/models
 # Create model cache directory (helps reuse between restarts)
 RUN mkdir -p /models
 
+HEALTHCHECK --interval=30s --timeout=5s --start-period=5m --retries=3 \
+    CMD python -c "import urllib.request; urllib.request.urlopen('http://127.0.0.1:50000/', timeout=3)" || exit 1
+
 # Start FastAPI app
 CMD ["uvicorn", "api:app", "--host", "0.0.0.0", "--port", "50000"]

README.md

@@ -303,6 +303,20 @@ pip3 install -e ./
 
 SenseVoice can be built and run using Docker to simplify setup, ensure reproducibility, and support both CPU and GPU inference.
 
+### Official image
+
+The official `linux/amd64` runtime image is published to GitHub Container Registry. Model weights are not embedded in the image; they are downloaded into the mounted `/models` cache on first start.
+
+```bash
+docker pull ghcr.io/qwenaudio/sensevoice:latest
+docker run --rm --gpus all \
+  -p 50000:50000 \
+  -v sensevoice-models:/models \
+  ghcr.io/qwenaudio/sensevoice:latest
+```
+
+Open `http://127.0.0.1:50000/docs` after the container becomes healthy. Release tags and immutable `sha-<commit>` tags are available in the [container package](https://github.com/QwenAudio/SenseVoice/pkgs/container/sensevoice).
+
 ### Build with Docker
 ```bash
 docker build -t sensevoice .
@@ -314,7 +328,7 @@ docker run --gpus all -p 50000:50000 sensevoice
 ```
 ### Run (CPU-only)
 ```bash
-docker run -e SENSEVOICE_DEVICE=cpu -p 50000:50000 sensevoice
+docker run --rm -e SENSEVOICE_DEVICE=cpu -p 50000:50000 -v sensevoice-models:/models ghcr.io/qwenaudio/sensevoice:latest
 ```
 ### Docker Compose
 Docker Compose provides an easier way to run SenseVoice with persistent model caching, networking etc. 

README_ja.md

@@ -246,6 +246,20 @@ export SENSEVOICE_DEVICE=cuda:0
 fastapi run --port 50000
 ```
 
+### 公式 Docker イメージ
+
+公式の `linux/amd64` ランタイムイメージは GitHub Container Registry で公開されます。モデル重みはイメージに含まれず、初回起動時にマウントした `/models` キャッシュへダウンロードされます。
+
+```bash
+docker pull ghcr.io/qwenaudio/sensevoice:latest
+docker run --rm --gpus all \
+  -p 50000:50000 \
+  -v sensevoice-models:/models \
+  ghcr.io/qwenaudio/sensevoice:latest
+```
+
+コンテナが healthy になったら `http://127.0.0.1:50000/docs` を開いてください。リリースタグと不変の `sha-<commit>` タグは[コンテナパッケージ](https://github.com/QwenAudio/SenseVoice/pkgs/container/sensevoice)で確認できます。CPU モードでは `-e SENSEVOICE_DEVICE=cpu` を追加し、`--gpus all` を削除します。
+
 ## 微調整
 
 ### トレーニング環境のインストール

README_zh.md

@@ -274,6 +274,20 @@ export SENSEVOICE_DEVICE=cuda:0
 fastapi run --port 50000
 ```
 
+### Docker 官方镜像
+
+官方 `linux/amd64` 运行时镜像发布在 GitHub Container Registry。镜像不内置模型权重；首次启动会把权重下载到挂载的 `/models` 缓存。
+
+```bash
+docker pull ghcr.io/qwenaudio/sensevoice:latest
+docker run --rm --gpus all \
+  -p 50000:50000 \
+  -v sensevoice-models:/models \
+  ghcr.io/qwenaudio/sensevoice:latest
+```
+
+容器健康后访问 `http://127.0.0.1:50000/docs`。版本 tag 和不可变的 `sha-<commit>` tag 见 [容器包页面](https://github.com/QwenAudio/SenseVoice/pkgs/container/sensevoice)。CPU 模式增加 `-e SENSEVOICE_DEVICE=cpu` 并移除 `--gpus all`。
+
 ## 微调
 
 ### 安装训练环境

tests/test_container_contract.py

@@ -0,0 +1,54 @@
+from pathlib import Path
+import unittest
+
+
+ROOT = Path(__file__).resolve().parents[1]
+
+
+class ContainerContractTest(unittest.TestCase):
+    def test_container_base_matches_repository_torch_floor(self):
+        dockerfile = (ROOT / "Dockerfile").read_text(encoding="utf-8")
+
+        self.assertIn(
+            "pytorch/pytorch:2.12.1-cuda12.6-cudnn9-runtime@sha256:",
+            dockerfile,
+        )
+        self.assertNotIn("pytorch/pytorch:2.3.1", dockerfile)
+
+    def test_container_installs_dependencies_in_an_isolated_venv(self):
+        dockerfile = (ROOT / "Dockerfile").read_text(encoding="utf-8")
+
+        self.assertIn(
+            "python -m venv --system-site-packages /opt/sensevoice-venv",
+            dockerfile,
+        )
+        self.assertIn("ENV PATH=/opt/sensevoice-venv/bin:$PATH", dockerfile)
+        self.assertNotIn("--break-system-packages", dockerfile)
+
+    def test_container_workflow_builds_prs_and_only_pushes_trusted_refs(self):
+        workflow = (
+            ROOT / ".github" / "workflows" / "sensevoice-container.yml"
+        ).read_text(encoding="utf-8")
+
+        self.assertIn("pull_request:", workflow)
+        self.assertIn("packages: write", workflow)
+        self.assertIn("platforms: linux/amd64", workflow)
+        self.assertIn("push: ${{ github.event_name != 'pull_request' }}", workflow)
+        self.assertIn(
+            "ghcr.io/${{ github.repository_owner }}/sensevoice",
+            workflow,
+        )
+
+    def test_readme_uses_public_ghcr_image_and_real_service_port(self):
+        readme = (ROOT / "README.md").read_text(encoding="utf-8")
+
+        self.assertIn("ghcr.io/qwenaudio/sensevoice", readme.lower())
+        self.assertIn("-p 50000:50000", readme)
+        self.assertNotIn(
+            "registry.cn-hangzhou.aliyuncs.com/funasr/sensevoice",
+            readme,
+        )
+
+
+if __name__ == "__main__":
+    unittest.main()