Datasets:
File size: 3,512 Bytes
cb288cf | 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 | """Score frozen generation/source tasks. CPU only; no model or network access."""
import argparse, collections, json, sys
from pathlib import Path
ROOT = Path(__file__).resolve().parents[1]
sys.path.insert(0, str(ROOT / "scoring"))
from common_scoring import score_generation, score_source
def rows(path):
return [json.loads(line) for line in Path(path).read_text().splitlines() if line.strip()]
def main():
p = argparse.ArgumentParser()
p.add_argument("--suite", required=True, choices=["generation", "source"])
p.add_argument("--predictions", required=True)
p.add_argument("--output", required=True)
a = p.parse_args()
cases = rows(ROOT / "inputs" / (a.suite + ".jsonl"))
refs = {x["id"]: x["reference"] for x in rows(ROOT / "references" / (a.suite + ".jsonl"))}
key = "unit_id" if a.suite == "generation" else "id"
by_id = {x[key]: x for x in cases}
predictions = rows(a.predictions)
ids = [x["id"] for x in predictions]
if len(ids) != len(set(ids)):
raise ValueError("duplicate prediction IDs")
if set(ids) - by_id.keys():
raise ValueError("unknown prediction IDs")
values = []
for pred in predictions:
case = by_id[pred["id"]]
raw = pred["raw_text"]
if not isinstance(raw, str):
raise TypeError("raw_text must be a string")
eos = pred.get("terminal_eos_text", "")
if not isinstance(eos, str):
raise TypeError("terminal_eos_text must be a string")
if eos and pred.get("terminal_eos_verified") is not True:
raise ValueError("verify terminal token ID before specifying terminal_eos_text")
score = (score_generation(raw, refs[pred["id"]], case.get("tools", []), eos)
if a.suite == "generation" else score_source(case, refs[pred["id"]], raw, eos))
values.append({"id": pred["id"], "kind": case.get("diagnostic_kind", case.get("kind")), "scores": score})
groups = {}
metric_names = ({"nonempty","json_valid","reference_top_level_keys_present","exact_json_reference_match",
"tool_schema_valid","tool_names_match_reference","tool_arguments_match_reference"}
if a.suite == "generation" else
{"json_valid","identifiers_match","schema_fields_present","all_nonempty_spans_supported",
"numeric_literals_exact","mechanical_grounding_pass","quote_match",
"interval_matches_source","exact_reference_position"})
for kind in sorted({x["kind"] for x in values}):
group = [x["scores"] for x in values if x["kind"] == kind]
groups[kind] = {m: {"passed": sum(s[m] for s in group if type(s.get(m)) is bool),
"scored": sum(type(s.get(m)) is bool for s in group)}
for m in sorted(metric_names) if any(type(s.get(m)) is bool for s in group)}
out = {"suite": a.suite, "expected": len(cases), "submitted": len(values),
"complete": len(values) == len(cases), "missing_ids": sorted(by_id.keys() - set(ids)),
"scientific_content_review_completed": False, "metrics_by_kind": groups, "items": values}
dest = Path(a.output)
if dest.exists():
raise FileExistsError("refusing to overwrite scores")
dest.write_text(json.dumps(out, ensure_ascii=False, indent=2) + "\n")
print(json.dumps({k:v for k,v in out.items() if k not in ["items","missing_ids"]}, ensure_ascii=False))
if __name__ == "__main__":
main()
|