Add NVIDIA vLLM image and ModelHub release workflow

.dockerignore

@@ -0,0 +1,2 @@
+**
+!Dockerfile

.gitea/workflows/docker-build-push.yml

@@ -0,0 +1,133 @@
+name: Docker Build and Push
+
+on:
+  push:
+    tags:
+      - "v*"
+
+jobs:
+  docker:
+    runs-on: amd64-ubuntu-24.04
+
+    steps:
+      - name: Clone repository
+        run: |
+          git clone "${{ gitea.server_url }}/${{ gitea.repository }}.git" .
+          git checkout "${{ gitea.ref_name }}"
+
+      - name: Set image metadata
+        run: |
+          IMAGE_NAME="$(echo "${{ gitea.repository }}" | tr '[:upper:]' '[:lower:]' | tr '_' '-')"
+          IMAGE="${DOCKER_REGISTRY}/${DOCKER_USERNAME}/${IMAGE_NAME}:${{ gitea.ref_name }}"
+
+          echo "IMAGE_NAME=${IMAGE_NAME}" >> "$GITEA_ENV"
+          echo "IMAGE=${IMAGE}" >> "$GITEA_ENV"
+
+      - name: Load and Validate Task Info
+        run: |
+          set -a
+          . .gitea/workflows/task_info.env
+          set +a
+
+          for name in FRAMEWORK GPU_TYPE TASK_TYPE; do
+            eval "value=\${${name}:-}"
+            if [ "$name" = "FRAMEWORK" ] && [ -z "$value" ]; then
+              echo "${name} is empty in .gitea/workflows/task_info.env"
+              exit 1
+            fi
+
+            echo "${name}=${value}" >> "$GITEA_ENV"
+          done
+
+      - name: Validate Image Verify Metadata
+        run: |
+          if [ -z "${FIXED_TOKEN:-}" ]; then
+            echo "FIXED_TOKEN is not configured on runner"
+            exit 1
+          fi
+
+          if ! response="$(curl --silent --show-error --location --get 'https://modelhub.org.cn/adminApi/image-verify/validate' \
+            --header "Xc-Token: ${FIXED_TOKEN}" \
+            --data-urlencode "gpuType=${GPU_TYPE:-}" \
+            --data-urlencode "taskType=${TASK_TYPE:-}")"; then
+            echo "failed to call image verify validate API"
+            exit 1
+          fi
+
+          VALIDATE_RESPONSE="$response" python3 - <<'PY'
+          import json
+          import os
+          import sys
+
+          raw = os.environ.get("VALIDATE_RESPONSE", "")
+          try:
+              body = json.loads(raw)
+          except json.JSONDecodeError:
+              print("image verify validate API returned invalid JSON")
+              print(raw)
+              sys.exit(1)
+
+          if body.get("code") == 0 and body.get("data") is True:
+              print("image verify metadata validation passed")
+              sys.exit(0)
+
+          message = body.get("message") or "unknown error"
+          print(f"image verify metadata validation failed: {message}")
+          print(raw)
+          sys.exit(1)
+          PY
+
+      - name: Login to Docker Registry
+        run: |
+          echo "$DOCKER_PASSWORD" | docker login "$DOCKER_REGISTRY" \
+            -u "$DOCKER_USERNAME" \
+            --password-stdin
+
+      - name: Build Docker Image
+        run: |
+          docker build -t "$IMAGE" .
+
+      - name: Push Docker Image
+        run: |
+          for attempt in 1 2 3; do
+            echo "Starting docker push attempt ${attempt}/3 for ${IMAGE}"
+            docker push "$IMAGE" &
+            PUSH_PID=$!
+
+            while kill -0 "$PUSH_PID" 2>/dev/null; do
+              echo "docker push is still running at $(date -u '+%Y-%m-%dT%H:%M:%SZ')"
+              sleep 60
+            done
+
+            if wait "$PUSH_PID"; then
+              echo "docker push completed successfully"
+              exit 0
+            fi
+
+            echo "docker push failed on attempt ${attempt}/3"
+            sleep 30
+          done
+
+          echo "docker push failed after 3 attempts"
+          exit 1
+
+      - name: Notify Image Verify
+        run: |
+          if [ -z "${FIXED_TOKEN:-}" ]; then
+            echo "FIXED_TOKEN is not configured on runner"
+            exit 1
+          fi
+
+          curl --silent --show-error --fail-with-body --location --request POST 'https://modelhub.org.cn//adminApi/image-verify' \
+            --header "Xc-Token: ${FIXED_TOKEN}" \
+            --header 'Content-Type: application/json' \
+            --data-raw "{
+              \"framework\": \"${FRAMEWORK}\",
+              \"gpuType\": \"${GPU_TYPE}\",
+              \"imageUrl\": \"${IMAGE}\",
+              \"taskType\": \"${TASK_TYPE}\",
+              \"createBy\": \"${{ gitea.actor }}\",
+              \"repoUrl\": \"${{ gitea.server_url }}/${{ gitea.repository }}\",
+              \"tag\": \"${{ github.ref_name }}\"
+            }"
+

.gitea/workflows/task_info.env

@@ -0,0 +1,3 @@
+FRAMEWORK=vllm
+GPU_TYPE="NVIDIA H800"
+TASK_TYPE=text-generation

Dockerfile

@@ -0,0 +1,5 @@
+FROM harbor.4pd.io/hardcore-tech/vllm/vllm-openai:v0.25.0-cu129
+
+# Text inference does not require the incompatible TorchCodec video decoder.
+RUN python3 -m pip uninstall -y torchcodec
+RUN python3 -c "import vllm.entrypoints.openai.api_server; print('API server import OK')"

README.md

@@ -0,0 +1,21 @@
+# Step-3.5-Flash NVIDIA inference image
+
+Rebuilds the text-inference environment validated on seven NVIDIA H800 GPUs:
+vLLM 0.25.0, CUDA 12.9, BF16, tensor parallelism 1 and pipeline parallelism 7.
+
+TorchCodec is removed because the original image fails to import it with a
+missing `libnvrtc.so.13` dependency. The build verifies the API server import.
+Model weights are not included. Mount them and supply serving arguments at runtime.
+
+## ModelHub release
+
+The workflow is copied from https://dev.modelhub.org.cn/4pdadmin/cicd_demo.
+Push a new `v*` Git tag to trigger image build, push, and review submission.
+The runner supplies `DOCKER_REGISTRY`, `DOCKER_USERNAME`, `DOCKER_PASSWORD`,
+and `FIXED_TOKEN`. It must be able to pull the Harbor base image.
+ModelHub validates `GPU_TYPE="NVIDIA H800"` and `TASK_TYPE=text-generation`
+before building. Approval is required before selecting the image for evaluation.
+
+The image inherits its base image's entrypoint; the validated deployment overrides
+it with `python3 -m vllm.entrypoints.openai.api_server`. Docker run options and
+host model paths are not embedded into this image.