File size: 6,070 Bytes
8f25722 db7db52 8f25722 db7db52 8f25722 655c6db 8f25722 655c6db 8f25722 db7db52 8f25722 db7db52 8f25722 db7db52 8f25722 9c930f3 8f25722 db7db52 8f25722 db7db52 8f25722 db7db52 8f25722 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 | """Sync the working tree to the HuggingFace dataset repo.
Usage:
python scripts/upload_hf.py --dry-run
python scripts/upload_hf.py --include-runs main_ abl_ loao_ de_
Checkpoints (``model.pt``) are only uploaded for the tags listed in
``KEEP_CKPT`` to keep the repo small; metrics/history/curves/predictions for
every run are always uploaded.
"""
from __future__ import annotations
import argparse
import os
from pathlib import Path
from huggingface_hub import HfApi
REPO = "eduzrh/training-model-material"
REPO_TYPE = "dataset"
ROOT = Path(__file__).resolve().parents[1]
# directories uploaded verbatim (relative to repo root)
DIRS = ["code", "docs", "scripts", "data", "results/tables", "results/figures",
"report", "configs", "logs"]
IGNORE = ["**/__pycache__/**", "**/*.pyc", "**/.ipynb_checkpoints/**",
"**/*.pt", "**/predictions.npz", "**/_tables_ro/**", "**/_figures_ro/**"]
KEEP_CKPT: tuple[str, ...] = ("main_pimamba_f0_s0", "main_mamba_f0_s0",
"main_gru_f0_s0", "abl_pimamba_f0_s0",
"next_hymba_f0_s0", "v2_pimamba2_f0_s0")
def _token(explicit: str | None) -> str:
tok = explicit or os.environ.get("HF_TOKEN")
if not tok:
# fall back to the credential saved by `huggingface-cli login`
try:
from huggingface_hub import get_token
tok = get_token()
except Exception: # noqa: BLE001
tok = None
if not tok:
raise SystemExit("no HF token: pass --token, set HF_TOKEN or run hf login")
return tok
def main():
ap = argparse.ArgumentParser()
ap.add_argument("--token", default=None)
ap.add_argument("--repo", default=REPO)
ap.add_argument("--include-runs", nargs="*", default=None,
help="only upload results/runs tags starting with these prefixes")
ap.add_argument("--dry-run", action="store_true")
ap.add_argument("--private", action="store_true")
a = ap.parse_args()
tok = _token(a.token)
api = HfApi(token=tok)
api.create_repo(a.repo, repo_type=REPO_TYPE, exist_ok=True,
private=a.private)
targets: list[Path] = []
for d in DIRS:
p = ROOT / d
if p.exists():
targets.append(p)
# run folders: bulk upload in ONE commit when possible (250+ folders would
# otherwise mean 250+ commits), with stale/debug tags filtered out.
run_root = ROOT / "results" / "runs"
keep_tags = _summary_tags()
extra_ignore = list(IGNORE)
if run_root.exists():
if a.include_runs:
for p in sorted(run_root.iterdir()):
if p.is_dir() and any(p.name.startswith(x) for x in a.include_runs):
targets.append(p)
else:
drop = {d.name for d in run_root.iterdir() if d.is_dir()} - keep_tags
extra_ignore += [f"{t}/**" for t in sorted(drop)]
targets.append(run_root)
print(f"bulk runs: keeping {len(keep_tags)} tags, skipping {len(drop)}")
for f in ["README.md", "requirements.txt", ".gitignore",
"results/RESULTS.md", "results/summary.csv"]:
p = ROOT / f
if p.exists():
targets.append(p)
n_files = 0
total = 0
for p in targets:
rel = p.relative_to(ROOT).as_posix()
files = [p] if p.is_file() else [q for q in p.rglob("*") if q.is_file()]
files = [q for q in files if not _ignored(q.relative_to(ROOT).as_posix())]
if not files:
continue
size = sum(q.stat().st_size for q in files)
n_files += len(files)
total += size
print(f"{rel:45s} {len(files):5d} files {size/1e6:8.2f} MB")
if a.dry_run:
continue
if p.is_file():
api.upload_file(path_or_fileobj=str(p), path_in_repo=rel,
repo_id=a.repo, repo_type=REPO_TYPE,
commit_message=f"sync {rel}")
continue
api.upload_folder(folder_path=str(p), path_in_repo=rel, repo_id=a.repo,
repo_type=REPO_TYPE, ignore_patterns=extra_ignore,
commit_message=f"sync {rel}")
# fold predictions of the seed-0 main runs: needed to reproduce pooled OOF
if not a.dry_run:
for m in ["mlp", "gru", "lstm", "transformer", "mamba", "mamba2", "pimamba"]:
for f in range(5):
pf = run_root / f"main_{m}_f{f}_s0" / "predictions.npz"
if pf.exists():
api.upload_file(path_or_fileobj=str(pf),
path_in_repo=f"results/runs/main_{m}_f{f}_s0/predictions.npz",
repo_id=a.repo, repo_type=REPO_TYPE,
commit_message=f"oof predictions {m} f{f}")
# checkpoints worth keeping
for tag in KEEP_CKPT:
ck = run_root / tag / "model.pt"
if ck.exists():
print(f"checkpoint {tag} ({ck.stat().st_size/1e6:.1f} MB)")
if not a.dry_run:
api.upload_file(path_or_fileobj=str(ck),
path_in_repo=f"results/runs/{tag}/model.pt",
repo_id=a.repo, repo_type=REPO_TYPE,
commit_message=f"checkpoint {tag}")
print(f"\ntotal {n_files} files, {total/1e6:.1f} MB")
def _summary_tags() -> set[str]:
"""Only publish runs that actually appear in results/summary.csv."""
import csv
f = ROOT / "results" / "summary.csv"
if not f.exists():
return set()
with f.open(encoding="utf-8") as fh:
return {r["tag"] for r in csv.DictReader(fh) if r.get("tag")}
def _ignored(rel: str) -> bool:
import fnmatch
for pat in IGNORE:
p = pat.replace("**/", "").replace("/**", "")
if fnmatch.fnmatch(rel, pat) or fnmatch.fnmatch(rel.split("/")[-1], p):
return True
if "/__pycache__/" in rel:
return True
return False
if __name__ == "__main__":
main() |