Datasets:
Languages:
English
Size:
n<1K
ArXiv:
Tags:
physical-reasoning
benchmark
vision-language-models
physics-simulation
minimal-pairs
evaluation
License:
DominoBench release: DominoBench-1K, -140 subset, -50, generator, evaluation harness (part 9)
572a861 verified Download eval/parse.py from emilioferrara/DominoBench: direct link, hf CLI and curl.
- Browser
- Download file 2.02 kB
-
https://huggingface.co/datasets/emilioferrara/DominoBench/resolve/main/eval/parse.py
- Command line
-
hf download hf://datasets/emilioferrara/DominoBench/eval/parse.py
-
curl -L -o parse.py https://huggingface.co/datasets/emilioferrara/DominoBench/resolve/main/eval/parse.py
2.02 kB
| """Extract the final JSON answer from a model response.""" | |
| from __future__ import annotations | |
| import json | |
| import re | |
| FENCE = re.compile(r"```(?:json)?\s*(\{.*?\})\s*```", re.S) | |
| def _candidates(text: str): | |
| for m in reversed(FENCE.findall(text)): | |
| yield m | |
| # fallback: the last balanced {...} in the text | |
| depth, end = 0, None | |
| for i in range(len(text) - 1, -1, -1): | |
| c = text[i] | |
| if c == "}": | |
| if depth == 0: | |
| end = i | |
| depth += 1 | |
| elif c == "{" and depth: | |
| depth -= 1 | |
| if depth == 0 and end is not None: | |
| yield text[i:end + 1] | |
| end = None | |
| def parse(text: str, people: list[str]) -> dict: | |
| """Return {"answer": {...}, "p_dies": {...}, "deciding_factors": str, "ok": bool, "error": str|None}.""" | |
| for cand in _candidates(text or ""): | |
| try: | |
| obj = json.loads(cand) | |
| except json.JSONDecodeError: | |
| continue | |
| ans = obj.get("answer") | |
| if not isinstance(ans, dict): | |
| continue | |
| norm = {} | |
| for k, v in ans.items(): | |
| k2 = str(k).strip().upper()[:1] | |
| v2 = str(v).strip().lower() | |
| norm[k2] = "dies" if v2.startswith(("die", "dead", "kill")) else "survives" if v2.startswith(("surv", "live", "alive", "safe")) else v2 | |
| missing = [p for p in people if p not in norm] | |
| p = obj.get("p_dies") or {} | |
| pd = {} | |
| for k, v in p.items(): | |
| try: | |
| pd[str(k).strip().upper()[:1]] = min(1.0, max(0.0, float(v))) | |
| except (TypeError, ValueError): | |
| pass | |
| return {"answer": norm, "p_dies": pd, "deciding_factors": str(obj.get("deciding_factors", "")), | |
| "ok": not missing and all(norm[q] in ("dies", "survives") for q in people), | |
| "error": f"missing {missing}" if missing else None} | |
| return {"answer": {}, "p_dies": {}, "deciding_factors": "", "ok": False, "error": "no JSON answer found"} | |