83 lines
2.6 KiB
YAML
83 lines
2.6 KiB
YAML
name: ai-steward-eval-gate
|
|
|
|
on:
|
|
pull_request:
|
|
paths:
|
|
- 'backend/app/services/adaptive_rag_service.py'
|
|
- 'backend/app/services/project_knowledge_graph.py'
|
|
- 'backend/app/api/rag.py'
|
|
- 'backend/app/api/knowledge_graph.py'
|
|
- 'backend/evals/**'
|
|
- 'scripts/ai_steward_eval.py'
|
|
- 'scripts/rag_followup_smoke.sh'
|
|
push:
|
|
branches: [ main, master ]
|
|
paths:
|
|
- 'backend/evals/**'
|
|
- 'scripts/ai_steward_eval.py'
|
|
- 'scripts/rag_followup_smoke.sh'
|
|
- '.github/workflows/ai-steward-eval-gate.yml'
|
|
|
|
jobs:
|
|
golden-lint:
|
|
runs-on: ubuntu-latest
|
|
steps:
|
|
- uses: actions/checkout@v4
|
|
- uses: actions/setup-python@v5
|
|
with:
|
|
python-version: '3.11'
|
|
- name: Lint golden dataset
|
|
run: |
|
|
python scripts/ai_steward_eval.py --lint-only --input backend/evals/golden_qa_v1.jsonl
|
|
|
|
eval-prereq:
|
|
runs-on: ubuntu-latest
|
|
needs: golden-lint
|
|
steps:
|
|
- name: Check gate configuration
|
|
run: |
|
|
if [ -z "${{ secrets.EVAL_API_TOKEN }}" ] || [ -z "${{ vars.EVAL_API_URL }}" ]; then
|
|
echo "::error::Missing EVAL_API_TOKEN secret or EVAL_API_URL variable. Eval gate must be configured."
|
|
exit 1
|
|
fi
|
|
|
|
eval-gate:
|
|
runs-on: ubuntu-latest
|
|
needs: eval-prereq
|
|
steps:
|
|
- uses: actions/checkout@v4
|
|
- uses: actions/setup-python@v5
|
|
with:
|
|
python-version: '3.11'
|
|
- name: Install deps
|
|
run: |
|
|
python -m pip install --upgrade pip
|
|
python -m pip install httpx
|
|
- name: Run eval gate
|
|
env:
|
|
EVAL_API_TOKEN: ${{ secrets.EVAL_API_TOKEN }}
|
|
EVAL_API_URL: ${{ vars.EVAL_API_URL }}
|
|
run: |
|
|
python scripts/ai_steward_eval.py \
|
|
--input backend/evals/golden_qa_v1.jsonl \
|
|
--api-url "$EVAL_API_URL" \
|
|
--token "$EVAL_API_TOKEN" \
|
|
--output-dir backend/evals/reports \
|
|
--fail-on-threshold \
|
|
--threshold-accuracy 0.80 \
|
|
--threshold-hallucination 0.12 \
|
|
--threshold-followup 0.75 \
|
|
--threshold-query-type 0.80
|
|
- name: Run followup smoke gate
|
|
env:
|
|
EVAL_API_TOKEN: ${{ secrets.EVAL_API_TOKEN }}
|
|
EVAL_API_URL: ${{ vars.EVAL_API_URL }}
|
|
run: |
|
|
BASE_URL="${EVAL_API_URL%/rag/ask}"
|
|
SMOKE_BASE_URL="$BASE_URL" SMOKE_TOKEN="$EVAL_API_TOKEN" bash scripts/rag_followup_smoke.sh
|
|
- name: Upload eval report
|
|
uses: actions/upload-artifact@v4
|
|
with:
|
|
name: ai-steward-eval-report
|
|
path: backend/evals/reports/*.json
|