Download code/src/prepare_pilot_data.py from PratikBuilds/modelclinic-experiment-001: direct link, hf CLI and curl.
- Browser
- Download file 1.34 kB
-
https://huggingface.co/datasets/PratikBuilds/modelclinic-experiment-001/resolve/main/code/src/prepare_pilot_data.py
- Command line
-
hf download hf://datasets/PratikBuilds/modelclinic-experiment-001/code/src/prepare_pilot_data.py
-
curl -L -o prepare_pilot_data.py https://huggingface.co/datasets/PratikBuilds/modelclinic-experiment-001/resolve/main/code/src/prepare_pilot_data.py
1.34 kB
| """Freeze 20 MATH-500 test items; they are not novel to published benchmark.""" | |
| import hashlib,json,random,requests | |
| from pathlib import Path | |
| rev='6e4ed1a2a79af7d8630a6b768ec859cb5af4d3be' | |
| url=f'https://huggingface.co/datasets/HuggingFaceH4/MATH-500/resolve/{rev}/test.jsonl' | |
| r=requests.get(url,timeout=30);r.raise_for_status(); raw=r.content | |
| items=[json.loads(line) for line in raw.splitlines() if line.strip()] | |
| if len(items)!=500:raise ValueError(f'Unexpected dataset size: {len(items)}') | |
| indices=sorted(random.Random(18012026).sample(range(len(items)),20)) | |
| out=Path('experiments/math500_pilot_20.jsonl');out.write_text(''.join(json.dumps({'source_row':i,**items[i]},ensure_ascii=False)+'\n' for i in indices),encoding='utf-8') | |
| Path('experiments/math500_pilot_20_manifest.json').write_text(json.dumps({'dataset':'HuggingFaceH4/MATH-500','revision':rev,'source_url':url,'source_sha256':hashlib.sha256(raw).hexdigest(),'selection':'Python random.Random(18012026).sample(range(500),20), then sorted','row_indices':indices,'selection_sha256':hashlib.sha256(out.read_bytes()).hexdigest(),'caveat':'Held out from this session direction discovery; these questions are part of the published benchmark and cannot be claimed unseen by the original authors or model.'},indent=2)) | |
| print('Prepared 20 selected rows; no evaluation run.') | |