CI/CD for ML Pipeline AI Prompt

Design and implement a CI/CD pipeline for this ML project using GitHub Actions.

1. On every pull request — fast checks (< 5 minutes):
   - Code quality: ruff lint, black format check, mypy type checking
   - Unit tests: test data preprocessing, loss functions, metrics, and model architecture
   - Smoke test: train for 2 epochs on 100 samples, assert loss decreases and model saves
   - No data leakage check: run automated leakage detection tests

2. On merge to main — extended checks (< 30 minutes):
   - Integration test: full training run on a small held-out dataset
   - Model performance gate: assert validation metric > {{min_metric_threshold}}
   - Inference test: run the exported model through the serving stack
   - Benchmark: run throughput/latency benchmark and compare to baseline

3. On new model registration — deployment checks:
   - Champion vs challenger comparison on fixed holdout set
   - Deploy to staging if challenger beats champion by > {{improvement_threshold}}%
   - Run smoke test in staging environment
   - Manual approval gate before production deployment

4. GitHub Actions workflow structure:
   - Separate workflow files for each stage
   - Cache: pip dependencies, pre-downloaded datasets for tests
   - Secrets: model registry credentials, cloud storage keys via GitHub Secrets

5. Failure handling:
   - Notify Slack channel on pipeline failure with the failing step and logs link
   - Auto-revert deployment if post-deployment canary metrics degrade

Return: GitHub Actions YAML files for each pipeline stage and a workflow diagram.