feather-a10-runtime / overlay /scripts /benchmark_checkpoint_report.py
Jackoatmon's picture
Update Feather training runtime image
951f760 verified
#!/usr/bin/env python3
from __future__ import annotations
import json
def build_checkpoint_report(files: list[str]) -> dict[str, object]:
by_job: dict[str, dict[str, object]] = {}
for path in files:
parts = path.split("/")
if len(parts) < 3 or parts[0] != "jobs":
continue
job_id = parts[1]
filename = parts[-1]
if filename not in {"best_bpb.pt", "pretrain_final.pt", "latest.pt"}:
continue
row = by_job.setdefault(job_id, {"job_id": job_id, "paths": []})
row["paths"].append(path)
candidates = []
for job_id, row in by_job.items():
paths = list(row["paths"])
preferred = None
for suffix in ("pretrain_final.pt", "best_bpb.pt", "latest.pt"):
for path in paths:
if path.endswith(suffix):
preferred = path
break
if preferred is not None:
break
candidates.append({
"job_id": job_id,
"preferred_path": preferred,
"available_paths": sorted(paths),
})
candidates.sort(key=lambda row: row["job_id"], reverse=True)
return {
"n_candidates": len(candidates),
"candidates": candidates,
}
def main() -> int:
print(json.dumps(build_checkpoint_report([]), indent=2, sort_keys=True))
return 0
if __name__ == "__main__":
raise SystemExit(main())