File size: 4,948 Bytes
896cc39 | 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 | """Offline integrity and metric audit for the public six-arm release (stdlib only)."""
from collections import Counter
import hashlib
from html.parser import HTMLParser
import json
from pathlib import Path
import re
import tarfile
def main():
root = Path(__file__).resolve().parent
release = root / 'e1-luna-six-way-20261003'
def sha(data):
return hashlib.sha256(data).hexdigest()
checksums = {}
for line in (root / 'SHA256SUMS').read_text().splitlines():
expected, name = line.split(' ', 1)
assert not Path(name).is_absolute() and '..' not in Path(name).parts
assert sha((root / name).read_bytes()) == expected, name
checksums[name] = expected
manifest = json.loads((release / 'manifest.json').read_text())
expected_members = {(m['archive'], m['path']): m for m in manifest['members'] if 'archive' in m}
seen, archive_totals = set(), {}
credential = re.compile(rb'\b(?:hf_[A-Za-z0-9]{10,}|sk-[A-Za-z0-9_-]{10,})\b')
for name in checksums:
path = root / name
if path.suffix != '.gz':
assert not credential.search(path.read_bytes()), 'Credential-like string in ' + name
for archive_path in sorted((release / 'raw').glob('*.tar.gz')):
arm = archive_path.name.removesuffix('.tar.gz')
totals = Counter()
with tarfile.open(archive_path, 'r:gz') as archive:
for member in archive:
assert member.isfile() and not Path(member.name).is_absolute() and '..' not in Path(member.name).parts
key = (str(archive_path.relative_to(release)), member.name)
assert key not in seen and key in expected_members
seen.add(key)
raw = archive.extractfile(member).read()
assert sha(raw) == expected_members[key]['published_sha256'], key
assert not credential.search(raw), 'Credential-like string in ' + member.name
value = json.loads(raw)
if member.name.endswith('/episode.json') or (arm.startswith('harbor') and member.name.endswith('/summary.json')):
totals['n'] += 1
if arm.startswith('paper'):
passed, tokens, calls = value['clean_pass'], value['tokens_in'] + value['tokens_out'], value['llm_calls']
elif arm.startswith('native'):
passed, tokens, calls = value['status'] == 'pass', value['usage']['total_tokens'], value['usage']['model_responses']
else:
passed, tokens, calls = value['grade']['status'] == 'pass', value['tokens']['total_tokens'], value['api_calls']
totals['passed'] += int(passed)
totals['total_tokens'] += tokens
totals['model_calls'] += calls
archive_totals[arm] = totals
assert seen == set(expected_members)
rows = [json.loads(line) for line in (release / 'results.jsonl').read_text().splitlines()]
assert len(rows) == len({(r['arm'], r['task']) for r in rows}) == 588
summary = json.loads((release / 'summary.json').read_text())
tasks = set(json.loads((release / 'tasks/selected-task-ids.json').read_text()))
assert len(tasks) == 98
assert {json.loads(line)['tid'] for line in (release / 'tasks/tasks.jsonl').read_text().splitlines()} == tasks
for arm, expected in summary['groups'].items():
group = [r for r in rows if r['arm'] == arm]
assert len(group) == 98 and {r['task'] for r in group} == tasks
assert all(r['total_tokens'] == r['input_tokens'] + r['output_tokens'] for r in group)
totals = {'n': len(group), 'passed': sum(r['status'] == 'pass' for r in group),
'total_tokens': sum(r['total_tokens'] for r in group), 'model_calls': sum(r['api_calls'] for r in group)}
assert totals == dict(archive_totals[arm]), arm
assert all(expected[k] == v for k, v in totals.items()), arm
class Links(HTMLParser):
def __init__(self):
super().__init__()
self.hrefs, self.ids = [], set()
def handle_starttag(self, tag, attrs):
a = dict(attrs)
if 'id' in a:
self.ids.add(a['id'])
if tag == 'a' and 'href' in a:
self.hrefs.append(a['href'])
for path in release.glob('*.html'):
parsed = Links()
parsed.feed(path.read_text())
for href in parsed.hrefs:
if href.startswith('#'):
assert href[1:] in parsed.ids, href
elif not href.startswith(('http:', 'https:')):
assert (path.parent / href.split('#')[0]).exists(), href
print(json.dumps({'verified_files': len(checksums), 'verified_archive_members': len(seen),
'episodes': len(rows), 'groups': archive_totals, 'status': 'PASS'}, indent=2))
if __name__ == '__main__':
main()
|