| """Validate the published dataset hashes, counts and response structure.""" |
| import hashlib,json |
| from pathlib import Path |
| root=Path(__file__).resolve().parent |
| spec={ |
| "train":(1138,"33a9949ef2e43f97581ba6e371007298027c35c4266924c40a5dc8c3dc25ada0"), |
| "validation":(126,"dc1d73be7d960fa952303951aac2ed45b1068e3aeda17c2a28b8a0407fd6a47f"), |
| } |
| groups={} |
| for split,(count,digest) in spec.items(): |
| blob=(root/"data"/(split+".jsonl")).read_bytes() |
| assert hashlib.sha256(blob).hexdigest()==digest,split |
| rows=[json.loads(x) for x in blob.decode("utf-8").splitlines() if x.strip()] |
| assert len(rows)==count |
| groups[split]={r["group_id"] for r in rows} |
| for r in rows: |
| assert [m["role"] for m in r["messages"]]==["system","user","assistant"] |
| assert json.loads(r["messages"][-1]["content"])==r["response"] |
| assert set(r["response"])=={"answer","practice_question","practice_answer","new_words","encouragement"} |
| print(split,count,digest) |
| assert not groups["train"] & groups["validation"],"Concept-group leakage" |
| print("Hashes, counts, messages and concept separation verified.") |
|
|
