ollama37/.github/workflows/inference.yml

name: Inference Tests

on:
  workflow_dispatch: # Manual trigger
    inputs:
      use_existing_container:
        description: "Use existing running container"
        required: false
        default: "false"
        type: choice
        options:
          - "true"
          - "false"
      judge_mode:
        description: "Test judge mode"
        required: false
        default: "dual"
        type: choice
        options:
          - "simple"
          - "llm"
          - "dual"
      judge_model:
        description: "LLM model for judging (if llm/dual mode)"
        required: false
        default: "gemma3:12b"
        type: string
  workflow_call: # Called by other workflows
    inputs:
      use_existing_container:
        description: "Container is already running"
        required: false
        default: false
        type: boolean
      judge_mode:
        description: "Test judge mode (simple, llm, dual)"
        required: false
        default: "dual"
        type: string
      judge_model:
        description: "LLM model for judging"
        required: false
        default: "gemma3:12b"
        type: string
    outputs:
      result:
        description: "Inference test result"
        value: ${{ jobs.inference.outputs.result }}

env:
  OLLAMA_HOST: http://localhost:11434

jobs:
  inference:
    name: Inference Tests
    runs-on: self-hosted
    outputs:
      result: ${{ steps.inference-tests.outcome }}

    steps:
      - name: Checkout
        uses: actions/checkout@v4

      - name: Setup Node.js
        uses: actions/setup-node@v4
        with:
          node-version: "20"

      - name: Install test runner dependencies
        run: cd tests && npm ci

      - name: Start container (if needed)
        if: ${{ inputs.use_existing_container != 'true' && inputs.use_existing_container != true }}
        run: |
          cd docker
          docker compose down 2>/dev/null || true
          docker compose up -d
          sleep 10

      - name: Run inference tests
        id: inference-tests
        run: |
          cd tests

          # Build judge flags based on input
          JUDGE_FLAGS=""
          if [ "${{ inputs.judge_mode }}" = "simple" ]; then
            JUDGE_FLAGS="--no-llm"
          elif [ "${{ inputs.judge_mode }}" = "llm" ]; then
            JUDGE_FLAGS="--judge-model ${{ inputs.judge_model || 'gemma3:12b' }}"
          else
            # dual mode (default)
            JUDGE_FLAGS="--dual-judge --judge-model ${{ inputs.judge_model || 'gemma3:12b' }}"
          fi

          echo "Judge mode: ${{ inputs.judge_mode || 'dual' }}"
          echo "Judge flags: $JUDGE_FLAGS"

          # Progress goes to stderr (visible), JSON results go to file
          npm run --silent dev -- run --suite inference $JUDGE_FLAGS --output json > /tmp/inference-results.json || true

          echo "--- JSON Results ---"
          cat /tmp/inference-results.json

      - name: Check test results
        run: |
          FAILED=$(jq '.summary.failed' /tmp/inference-results.json)
          echo "Failed tests: $FAILED"
          if [ "$FAILED" -gt 0 ]; then
            echo "::error::$FAILED inference test(s) failed"
            exit 1
          fi

      - name: Unload test model from VRAM
        if: always()
        run: |
          echo "Unloading gemma3:4b from VRAM..."
          curl -s http://localhost:11434/api/generate -d '{"model":"gemma3:4b","keep_alive":0}' || true
          echo "Model unloaded"

      - name: Upload inference results
        uses: actions/upload-artifact@v4
        if: always()
        with:
          name: inference-test-results
          path: /tmp/inference-results.json

      - name: Stop container (if we started it)
        if: ${{ always() && inputs.use_existing_container != 'true' && inputs.use_existing_container != true }}
        run: |
          cd docker
          docker compose down || true
          echo "Container stopped"