Datasets:
File size: 3,305 Bytes
d9a31e2 | 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 | #!/usr/bin/env python3
"""Launch parallel European HPLT filtering workers.
The filter itself is intentionally single-process and shard-safe. This launcher
splits the actual HF train parquet file list across workers and starts each
worker in a detached screen session with its own output directory and log.
"""
from __future__ import annotations
import argparse
import subprocess
from pathlib import Path
from huggingface_hub import list_repo_files
def parse_args() -> argparse.Namespace:
ap = argparse.ArgumentParser()
ap.add_argument("--repo-id", default="ashtok897/european-hplt-v1")
ap.add_argument("--workers", type=int, default=4)
ap.add_argument("--out-dir", type=Path, default=Path("data/european_hplt_pl_parallel"))
ap.add_argument("--log-dir", type=Path, default=Path("logs"))
ap.add_argument("--screen-prefix", default="hplt")
ap.add_argument("--target-total-tokens", type=int, default=500_000_000)
ap.add_argument("--max-records-per-worker", type=int, default=200_000)
ap.add_argument("--shard-records", type=int, default=25_000)
ap.add_argument("--shard-tokens", type=int, default=25_000_000)
ap.add_argument("--progress-every", type=int, default=100_000)
ap.add_argument("--dry-run", action="store_true")
return ap.parse_args()
def main() -> None:
args = parse_args()
if args.workers < 1:
raise SystemExit("--workers must be >= 1")
files = sorted(
path
for path in list_repo_files(args.repo_id, repo_type="dataset")
if path.startswith("data/train/") and path.endswith(".parquet")
)
if not files:
raise SystemExit(f"No train parquet files found in {args.repo_id}")
args.out_dir.mkdir(parents=True, exist_ok=True)
args.log_dir.mkdir(parents=True, exist_ok=True)
target_tokens = max(1, args.target_total_tokens // args.workers)
chunks = [files[idx:: args.workers] for idx in range(args.workers)]
print(f"repo={args.repo_id} train_files={len(files)} workers={args.workers}")
for worker, chunk in enumerate(chunks):
if not chunk:
continue
out_dir = args.out_dir / f"worker_{worker}"
out_dir.mkdir(parents=True, exist_ok=True)
log_path = args.log_dir / f"filter_european_hplt_{args.screen_prefix}_w{worker}.log"
data_files = ",".join(chunk)
cmd = (
"python3 src/filter_european_hplt.py "
f"--repo-id {args.repo_id} "
f"--data-files '{data_files}' "
f"--out-dir '{out_dir}' "
f"--source-name european_hplt_pl_w{worker} "
f"--max-records {args.max_records_per_worker} "
f"--max-tokens {target_tokens} "
f"--shard-records {args.shard_records} "
f"--shard-tokens {args.shard_tokens} "
f"--progress-every {args.progress_every} "
f"> '{log_path}' 2>&1"
)
print(
f"worker_{worker}: files={len(chunk)} target_tokens={target_tokens:,} "
f"first={chunk[0]} last={chunk[-1]} log={log_path}"
)
if not args.dry_run:
subprocess.run(
["screen", "-dmS", f"{args.screen_prefix}_w{worker}", "zsh", "-lc", cmd],
check=True,
)
if __name__ == "__main__":
main()
|