File size: 4,576 Bytes
f764cdf 02ebb32 f764cdf | 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 | """Audit dependency samples against real local The Stack v2 repositories."""
from __future__ import annotations
import argparse
from collections import Counter
import json
import os
from pathlib import Path
import time
import tiktoken
from audit_repoinfo_stack_v2 import load_bodies, make_repo, sample_repositories
from stackbuild import BuildConfig, build_samples
ROOT = Path(__file__).resolve().parents[1]
DATA_ROOT = Path(os.environ.get(
"STACKGO_DATA_ROOT", "/home/ldj/tgai/data/pretrain-data"
))
DEFAULT_DATA = DATA_ROOT / "the-stack-v2-contents-star-gt10"
DEFAULT_REPORT = ROOT / "reports/stackbuild-audit.json"
def audit_repo(repo, samples, max_tokens: int) -> tuple[list[str], dict]:
errors: list[str] = []
source_paths = {file.path.lstrip("/") for file in repo.files}
covered = {file.path for sample in samples for file in sample.files}
if covered != source_paths:
errors.append(
f"coverage mismatch missing={len(source_paths - covered)} "
f"extra={len(covered - source_paths)}"
)
if any(sample.kind != "dependency" for sample in samples):
errors.append("non-dependency sample emitted")
ids = [sample.sample_id for sample in samples]
if len(ids) != len(set(ids)):
errors.append("duplicate sample ids")
oversized = [sample.sample_id for sample in samples if sample.token_count > max_tokens]
if oversized:
errors.append(f"{len(oversized)} samples exceed token budget")
malformed = [
sample.sample_id for sample in samples
if not sample.text.startswith(f"<｜begin▁of▁repo▁name｜>{repo.repo_id}<｜end▁of▁repo▁name｜>\n")
or sample.text.count("<｜begin▁of▁file▁name｜>") != len(sample.files)
or "<|endoftext|>" in sample.text
or "<|fim_" in sample.text
]
if malformed:
errors.append(f"{len(malformed)} samples have malformed boundary tokens")
return errors, {
"files": len(source_paths),
"samples": len(samples),
"covered_files": len(covered),
"max_sample_tokens": max((sample.token_count for sample in samples), default=0),
}
def main() -> None:
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument("--data-root", type=Path, default=DEFAULT_DATA)
parser.add_argument("--repositories", type=int, default=10)
parser.add_argument("--seed", type=int, default=20261030)
parser.add_argument("--max-tokens", type=int, default=8192)
parser.add_argument("--report", type=Path, default=DEFAULT_REPORT)
args = parser.parse_args()
sampled = sample_repositories(args.data_root, 1, args.repositories, args.seed)
encoding = tiktoken.get_encoding("cl100k_base")
config = BuildConfig(max_tokens=args.max_tokens)
report = {"seed": args.seed, "max_tokens": args.max_tokens, "repositories": []}
totals = Counter()
maximum_sample_tokens = 0
started = time.perf_counter()
for shard, rows in sampled:
bodies = load_bodies(shard, rows)
for row in rows:
repo, source_errors = make_repo(row, bodies)
before = time.perf_counter()
samples = build_samples(repo, config=config, tokenizer=encoding)
errors, stats = audit_repo(repo, samples, args.max_tokens)
errors[:0] = source_errors
record = {
"repo_id": repo.repo_id,
"revision_id": repo.revision_id,
"elapsed_seconds": round(time.perf_counter() - before, 4),
"stats": stats,
"errors": errors,
}
report["repositories"].append(record)
maximum_sample_tokens = max(maximum_sample_tokens, stats["max_sample_tokens"])
additive = {key: value for key, value in stats.items() if key != "max_sample_tokens"}
totals.update(repositories=1, errors=len(errors), **additive)
print(repo.repo_id, stats, f"errors={len(errors)}", flush=True)
totals["elapsed_seconds"] = round(time.perf_counter() - started, 3)
totals["max_sample_tokens"] = maximum_sample_tokens
report["totals"] = dict(totals)
args.report.parent.mkdir(parents=True, exist_ok=True)
temporary = args.report.with_suffix(args.report.suffix + ".part")
temporary.write_text(json.dumps(report, ensure_ascii=False, indent=2) + "\n")
temporary.replace(args.report)
print(json.dumps(report["totals"], indent=2), flush=True)
if totals["errors"]:
raise SystemExit(1)
if __name__ == "__main__":
main()
|