runpod-deploy#
Config-driven RunPod GPU pod orchestration for reusable Python project deployments. Define the pod, storage, run script, and artifacts as a single YAML; runpod-deploy provisions, stages, runs, and pulls results deterministically.
Get started#
Get started
- Quickstart
- The runpod-deploy run lifecycle
- Phase overview
- 1. Validate
- 2. Provision
- 3. SSH wait
- 4. Setup commands
- 5. Stage + secrets + preflight + launch + monitor
- 6. Artifact pull
- 7. Lifecycle action (cleanup)
- 7b. Cost discipline: cleaning up after forensics
- 8. Manifest write
- Where each YAML section maps
- Failure handling
--dry-runvs--offline-dry-run
- Config Reference
- Python API vs. CLI: when to use which
Recipes#
Recipes
- Recipes
- Recipe: local pre-flight, then
runpod-deploy run - Recipe: pull artifacts, then post-process locally
- Recipe: embed deploy metadata in your own artifacts
- Recipe: multi-config sweep
- Recipe: cost reconciliation across past runs
- Recipe: predictions-only-eval
- Recipe: flash_attention_2 graceful fallback
- Recipe: pod-Python reproducibility
- Recipe: multi-manifest forensics via the Python API
- Recipe: stock-out diagnostic
- Recipe: forensics, then cleanup
- Recipe: weekly stale-pod audit
- Recipe: reuse the staging payload via a network volume
- Recipe: recycle a pod for fast iteration
Examples#
Examples