empire-eval / scripts /fetch_hf.py
Lilambd's picture
Publish Lilambd Empire Eval
7f9a652 verified
Raw History Blame Contribute Delete
3.9 kB
"""Download the Lilambd dataset files the question generators read, pinned to the
current commit of each repo. Writes <mirror>/<name>/... plus _revision.json and
roots.json (input for build.py / verify.py).
usage: python fetch_hf.py --mirror MIRROR_DIR [--only name,name]
needs: pip install huggingface_hub
"""
import argparse, fnmatch, json, os, re
from huggingface_hub import HfApi, hf_hub_download
# name -> list of glob patterns; {LATEST} = newest dated file/folder for that pattern
WANT = {
'world-signals': ['data/{LATEST}/steam_top.csv', 'data/{LATEST}/github_new_repos.csv', 'data/{LATEST}/hf_trending.csv',
'data/{LATEST}/package_downloads.csv', 'data/{LATEST}/podcast_charts.csv', 'data/{LATEST}/search_trends.csv'],
'world-seeds': ['index.csv', 'tables/*.csv'],
'japan-dpc-hospitals': ['hospitals.csv', 'top100_mdc.csv'],
'japan-prefecture-facts': ['facts.csv'],
'world-biodiversity-cells': ['cells.parquet', 'world_totals.csv'],
'machine-economy-daily': ['x402/hosts/{LATEST}.csv', 'npm_top/{LATEST}.csv', 'hf_trending/{LATEST}.csv', 'mcp/new/{LATEST}.csv'],
'security-daily': ['daily/cve_new/{LATEST}.csv', 'daily/patch_first/{LATEST}.csv', 'daily/summary/{LATEST}.csv', 'snapshot/breaches.csv',
'snapshot/cve_signals/published_2025.csv', 'snapshot/cve_signals/published_2026.csv', 'weekly/vendor/{LATEST}.csv'],
'world-pulse-daily': [f'{t}/{{LATEST}}.csv' for t in ['fx_ecb', 'arxiv', 'stablecoins', 'hackernews_front', 'wikipedia_size', 'docker_official_images',
'stackoverflow_tags', 'ai_sdk_downloads', 'crossref', 'zenodo', 'hf_daily_papers', 'github_new_repos',
'crates_io', 'wikipedia_top', 'appstore_charts']],
'world-markets': ['markets.parquet'],
# these six are also built from local copies; fetch them to rebuild everything from HF
'world-hazard': ['data/*.csv'], 'world-earth': ['data/*.csv'], 'world-ground': ['data/*.csv'], 'world-prices': ['data/*.csv'],
'world-science': ['*.csv'], 'world-marketplaces': ['digital.csv', 'offline.parquet'],
}
DATE = re.compile(r'\d{4}-\d{2}-\d{2}')
def resolve(files, pat):
if '{LATEST}' not in pat:
return [f for f in files if fnmatch.fnmatch(f, pat)]
rx = re.compile('^' + re.escape(pat).replace(re.escape('{LATEST}'), r'(\d{4}-\d{2}-\d{2})') + '$')
hits = sorted((m.group(1), f) for f in files for m in [rx.match(f)] if m)
return [hits[-1][1]] if hits else []
def main():
ap = argparse.ArgumentParser()
ap.add_argument('--mirror', required=True)
ap.add_argument('--only', default='')
a = ap.parse_args()
only = [x for x in a.only.split(',') if x]
api = HfApi()
roots = {}
for name, pats in WANT.items():
if only and name not in only:
continue
repo = f'Lilambd/{name}'
info = api.dataset_info(repo)
files = [s.rfilename for s in info.siblings]
got = []
dest = os.path.join(a.mirror, name)
for p in pats:
for f in resolve(files, p):
hf_hub_download(repo, f, repo_type='dataset', revision=info.sha, local_dir=dest)
got.append(f)
json.dump({'repo': repo, 'revision': f'{info.sha[:12]} ({str(info.last_modified)[:10]})', 'sha': info.sha, 'files': got},
open(os.path.join(dest, '_revision.json'), 'w'), indent=1)
roots[name] = os.path.abspath(dest)
print(f'{name}: {len(got)} files @ {info.sha[:12]}')
rp = os.path.join(a.mirror, 'roots.json')
old = json.load(open(rp)) if os.path.exists(rp) else {}
old.update(roots)
json.dump(old, open(rp, 'w'), indent=1)
print('roots ->', rp)
if __name__ == '__main__':
main()