EMC / scripts /sync.py
mudassiruddin21's picture
Run r3: recipe comparison, keep-the-better rule, extra measurements
a64b4b7 verified
Raw History Blame Contribute Delete
5.68 kB
#!/usr/bin/env python3
"""Push logs, results and (optionally) the draft-model folders to the Hugging Face dataset repo.
The token comes from $HF_TOKEN or $EMC_ROOT/.hf_token; without one nothing is uploaded (the run
never depends on the upload). Paths and host names are scrubbed from text files.
sync.py logs + results + state + generated token data of the current run
sync.py --submissions additionally every runs/*/sub_* folder (weights)
"""
import os
import re
import shutil
import socket
import sys
from pathlib import Path
try:
import fcntl
except ImportError: # not on Linux: no cross-process lock
fcntl = None
ROOT = Path(os.environ.get("EMC_ROOT", str(Path.home() / "emc")))
REPO_ID = os.environ.get("HF_REPO", "mudassiruddin21/EMC")
TEXT = {".log", ".out", ".json", ".jsonl", ".md", ".txt", ".done", ".failed", ""}
SCRUB = [(re.compile(r"hf_[A-Za-z0-9]{16,}"), "hf_***"),
(re.compile(re.escape(str(ROOT.resolve()))), "/workspace/emc"),
(re.compile(re.escape(str(ROOT))), "/workspace/emc"),
(re.compile(re.escape(str(Path.home()))), "/workspace"),
(re.compile(r"\b" + re.escape(socket.gethostname()) + r"\b"), "gpu-node")]
# extra words to keep out of published logs (comma separated), set at launch time
_words = os.environ.get("EMC_SCRUB") or ((ROOT / ".scrub").read_text() if (ROOT / ".scrub").exists() else "")
_words = [w.strip() for w in _words.split(",") if w.strip()]
if _words:
SCRUB.append((re.compile("|".join(map(re.escape, _words)), re.I), "host"))
PATTERNS = (("logs/*", "logs"), ("results/*.json", "results"), ("results/*.md", "results"),
("state/*", "state"), ("work/logs/*.log", "engine_logs"), ("work/results/*.json", "kit_results"),
("runs/*/train_summary.json", "training"), ("data/gen/*.jsonl", "data/gen"))
def token():
if os.environ.get("HF_TOKEN"):
return os.environ["HF_TOKEN"].strip()
p = ROOT / ".hf_token"
return p.read_text().strip() if p.exists() else None
def scrub_copy(src: Path, dst: Path):
dst.parent.mkdir(parents=True, exist_ok=True)
if src.suffix in TEXT and src.stat().st_size < 64 * 2 ** 20 and src.suffix != ".jsonl":
text = src.read_text(errors="replace")
for pat, rep in SCRUB:
text = pat.sub(rep, text)
dst.write_text(text)
else:
shutil.copy(src, dst)
def record_stats():
"""Append a machine snapshot (memory, disk, GPU) to logs/stats.log; never blocks an upload."""
try:
sys.path.insert(0, str(Path(__file__).resolve().parent))
from sysstats import snapshot
(ROOT / "logs").mkdir(parents=True, exist_ok=True)
with open(ROOT / "logs" / "stats.log", "a") as f:
f.write(snapshot(ROOT) + "\n")
except Exception as e: # noqa: BLE001
print(f"machine stats unavailable: {e}")
def main():
record_stats()
tok = token()
if not tok:
print("no Hugging Face token: skipping upload")
return 0
from huggingface_hub import HfApi
api = HfApi(token=tok)
run_id = (ROOT / "state" / "run_id").read_text().strip() if (ROOT / "state" / "run_id").exists() else "run"
with open(ROOT / ".sync.lock", "w") as lock:
if fcntl:
fcntl.flock(lock, fcntl.LOCK_EX)
stage = ROOT / f"export-{os.getpid()}"
shutil.rmtree(stage, ignore_errors=True)
try:
for pattern, sub in PATTERNS:
for p in sorted(ROOT.glob(pattern)):
if p.is_file() and not p.name.startswith("."):
name = f"{p.parent.name}-{p.name}" if sub == "training" else p.name
scrub_copy(p, stage / sub / name)
if stage.exists():
api.upload_folder(folder_path=str(stage), repo_id=REPO_ID, repo_type="dataset",
path_in_repo=f"runs/{run_id}", commit_message=f"Run {run_id}: logs and results")
print(f"uploaded logs/results to {REPO_ID}:runs/{run_id}")
# draft folders: all of them with --submissions, otherwise any whose files are not on
# the Hub yet (so finished weights get saved even if the machine stops early)
subs = sorted(p for p in ROOT.glob("runs/*/sub_*") if p.is_dir())
have = set()
if subs and "--submissions" not in sys.argv:
try:
have = set(api.list_repo_files(REPO_ID, repo_type="dataset"))
except Exception as e: # noqa: BLE001
print(f"could not list the repository: {e}")
for sub in subs:
# a training run records which run it belongs to; folders without that record
# (older runs, or starting weights fetched from the Hub) are not uploaded again
marker = sub.parent / "run_id"
if not marker.exists():
continue
name = f"{sub.parent.name}-{sub.name[4:]}"
dest = f"submissions/{marker.read_text().strip()}/{name}"
missing = [p.name for p in sub.iterdir() if p.is_file() and f"{dest}/{p.name}" not in have]
if "--submissions" in sys.argv or missing:
info = api.upload_folder(folder_path=str(sub), repo_id=REPO_ID, repo_type="dataset",
path_in_repo=dest, commit_message=f"Run {run_id}: draft model {name}")
print(f"uploaded {sub.name} to {REPO_ID}:{dest} ({info})")
finally:
shutil.rmtree(stage, ignore_errors=True)
return 0
if __name__ == "__main__":
sys.exit(main())