|
Download docs/HOST_NOTES.md from codemaivanngu/job-manager: direct link, hf CLI and curl.
- Browser
- Download file 571 Bytes
-
https://huggingface.co/codemaivanngu/job-manager/resolve/main/docs/HOST_NOTES.md
- Command line
-
hf download hf://codemaivanngu/job-manager/docs/HOST_NOTES.md
-
curl -L -o HOST_NOTES.md https://huggingface.co/codemaivanngu/job-manager/resolve/main/docs/HOST_NOTES.md
571 Bytes
Multi-machine rules (v0.2)
- HOST = hostname -s; khai bao GPU cua may: ec host --gpus 0,1,3
- Lock theo (may, gpu): locks/-gpu-.lock => gpu0 cua may A khong chan gpu0 cua may B.
- 1 job = 1 CELL: (run, ckpt_sha, step, benchmark, eval_seed, gen_seed).
- Claim: mv pending -> claimed; ghi claimed_by=":"; touch heartbeats/ moi 60s.
- ec gc thu hoi theo heartbeat (mac dinh 20m), khong theo mtime file job => dung ca khi lech dong ho.
- Bank key = sha256(ckpt_sha | benchmark | eval_seed | contract_sha | env_fp) => dedup muc ckpt, xuyen may.