Datasets:
File size: 6,288 Bytes
8271cc3 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 | """Validate the Test-only archive and records. This is not an answer scorer."""
from __future__ import annotations
import argparse
import hashlib
import json
from collections import Counter
from pathlib import Path, PurePosixPath
import zipfile
TASKS = {"assembly_decision": 238, "atomic_action": 662, "minimal_rework": 100}
FAMILIES = {"gearbox_structure", "luban_12_pillars", "parametric_box_lock_01",
"parametric_tower_lock_10", "spherical_interlocking_structure",
"triangle_dovetail_structure"}
META = {"schema_version", "task_family", "split", "assembly_id", "variant_id",
"episode_id", "problem_id", "experiment_group_id", "source_sample_id",
"label_contract", "task_contract_version"}
def file_hash(path):
with Path(path).open("rb") as f:
return hashlib.file_digest(f, "sha256").hexdigest()
def safe_image_path(value):
p = PurePosixPath(value)
return (not p.is_absolute() and len(p.parts) > 1 and p.parts[0] == "images"
and ".." not in p.parts and "\\" not in value and ":" not in value
and p.suffix == ".png" and p.as_posix() == value)
def verify(root):
root = Path(root).resolve()
rows = [json.loads(l) for l in (root / "test.jsonl").read_text(encoding="utf-8").splitlines() if l.strip()]
errors = []
if len(rows) != 1000:
errors.append("Expected 1000 Test records")
if len({r["id"] for r in rows}) != len(rows):
errors.append("Duplicate sample IDs")
counts = Counter(r["metadata"]["task_family"] for r in rows)
if counts != Counter(TASKS):
errors.append("Task counts do not match the Test split")
if {r["metadata"]["assembly_id"] for r in rows} != FAMILIES:
errors.append("Unexpected assembly families")
for row in rows:
if set(row) != {"id", "messages", "images", "metadata"}:
errors.append(f"Unexpected top-level fields: {row['id']}")
if set(row["metadata"]) != META or row["metadata"]["split"] != "test":
errors.append(f"Non-public metadata or non-Test row: {row['id']}")
if not row["messages"] or any(set(m) != {"role", "content"} or m["role"] != "user" or not isinstance(m["content"], str) for m in row["messages"]):
errors.append(f"Non-user message or invalid prompt: {row['id']}")
if not row["images"] or any(not safe_image_path(p) for p in row["images"]):
errors.append(f"Invalid image references: {row['id']}")
refs = {p for r in rows for p in r["images"]}
if len(refs) != 1140:
errors.append("Expected 1140 unique Test images")
index = json.loads((root / "image_manifest.json").read_text(encoding="utf-8"))
entries = index["images"]
if len(entries) != len({e['path'] for e in entries}) or {e['path'] for e in entries} != refs:
errors.append("Image index does not exactly match Test references")
expected = {e["path"]: e for e in entries}
with zipfile.ZipFile(root / "images.zip") as archive:
names = archive.namelist()
if len(names) != len(set(names)) or set(names) != refs:
errors.append("Archive contains duplicate, missing or extra files")
for info in archive.infolist():
if not safe_image_path(info.filename) or info.filename not in expected:
errors.append(f"Unexpected archive entry: {info.filename}")
continue
content = archive.read(info)
entry = expected[info.filename]
if not content.startswith(b"\x89PNG\r\n\x1a\n"):
errors.append(f"Not a PNG: {info.filename}")
if len(content) != entry['bytes'] or hashlib.sha256(content).hexdigest() != entry['sha256']:
errors.append(f"Image hash mismatch: {info.filename}")
declared = {}
for line in (root / "MANIFEST.sha256").read_text(encoding="utf-8").splitlines():
digest, name = line.split(" ", 1)
if Path(name).name != name or "/" in name or "\\" in name or name in declared:
errors.append(f"Invalid/duplicate manifest entry: {name}")
continue
declared[name] = digest
if not (root / name).is_file() or file_hash(root / name) != digest:
errors.append(f"Release hash mismatch: {name}")
actual = {p.name for p in root.iterdir() if p.is_file() and p.name != "MANIFEST.sha256"}
if actual != set(declared):
errors.append("Release files differ from the checksum manifest")
for path in root.iterdir():
if path.is_dir() and path.name not in {".git", ".cache", "__pycache__", "images"}:
errors.append(f"Unexpected directory in upload root: {path.name}")
# If users manually extracted images, they must still be Test-only and intact.
if (root / 'images').exists():
extracted = {p.relative_to(root).as_posix(): p for p in (root/'images').rglob('*') if p.is_file()}
if set(extracted) != refs:
errors.append("Extracted images contain missing or extra files")
else:
for name, path in extracted.items():
if file_hash(path) != expected[name]['sha256']:
errors.append(f"Extracted image hash mismatch: {name}")
release = json.loads((root/'release_manifest.json').read_text(encoding='utf-8'))
if release['counts']['test_rows'] != len(rows) or release['counts']['unique_images'] != len(refs):
errors.append("Release manifest counts mismatch")
if release['test_jsonl_sha256'] != file_hash(root/'test.jsonl'):
errors.append("Release manifest JSONL hash mismatch")
return {"test_rows": len(rows), "task_counts": dict(counts), "families": len(FAMILIES),
"unique_images": len(refs), "assistant_messages": sum(m['role']=='assistant' for r in rows for m in r['messages']),
"metadata_allowlist_checked": True, "errors": errors}
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument("--root", type=Path, default=Path(__file__).resolve().parent)
try:
result = verify(parser.parse_args().root)
except Exception as exc:
result = {"errors": [f"{type(exc).__name__}: {exc}"]}
print(json.dumps(result, ensure_ascii=False, indent=2))
raise SystemExit(bool(result["errors"]))
|