Buckets:

glennmatlin's picture
download
raw
1.92 kB
"""Read-only probe of HF manifest + working-sample repos to settle the
1.0986B (pool) vs 1.26B (mix) question. Uses the cached HF token."""
from __future__ import annotations
import json
from pathlib import Path
from huggingface_hub import HfApi, hf_hub_download
api = HfApi()
MANIFEST_REPOS = [
"HCAI-Lab/dolma3-6t-corpus-manifest",
"HCAI-Lab/dolma3-olmo3-corpus-manifest",
]
SAMPLE_REPOS = [
"HCAI-Lab/dolma3-6t-sample-10000-docs",
"HCAI-Lab/dolma3_6T_sample_10000_docs",
]
def probe_repo_files(repo: str) -> None:
try:
info = api.repo_info(repo, repo_type="dataset", files_metadata=True)
except Exception as exc: # noqa: BLE001
print(f"[{repo}] repo_info FAILED: {exc}")
return
parquet = [s for s in info.siblings if s.rfilename.endswith(".parquet")]
total_bytes = sum((s.size or 0) for s in parquet)
print(f"[{repo}] parquet_files={len(parquet)} total_bytes={total_bytes/1e9:.1f}GB")
# sample a few filenames
for s in parquet[:3]:
print(f" e.g. {s.rfilename} ({(s.size or 0)/1e6:.1f}MB)")
def probe_sample(repo: str) -> bool:
try:
files = api.list_repo_files(repo, repo_type="dataset")
except Exception as exc: # noqa: BLE001
print(f"[{repo}] list FAILED: {exc}")
return False
print(f"[{repo}] exists; top files: {[f for f in files if not f.startswith('data/')][:8]}")
out = Path("/tmp/hf_probe")
out.mkdir(exist_ok=True)
for fn in ("sample_contract.json", "bin_summary.csv"):
if fn in files:
p = hf_hub_download(repo, fn, repo_type="dataset", local_dir=str(out))
print(f" downloaded {fn} -> {p}")
return True
if __name__ == "__main__":
print("=== MANIFEST REPOS ===")
for r in MANIFEST_REPOS:
probe_repo_files(r)
print("=== SAMPLE REPOS ===")
for r in SAMPLE_REPOS:
if probe_sample(r):
break

Xet Storage Details

Size:
1.92 kB
·
Xet hash:
dafcad2f97b14cd98e2711e7011b1c01ed9757120d99c349fa1ca094511ebc2f

Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.