{"dataset_revision":"388097ea7776314e93a529163e0fea805b8a6454","device":"cuda:0","evaluator_source":true,"evidence":"CodeSOTA execution on White","finished":1792193843,"gpu":1,"id":"40d1459ca17c38921b30a6ea","measured_cost_usd":null,"metric":"accuracy","model":"Qwen/Qwen2.5-0.5B","model_revision":"060db6499f32faf8b98477b0a26969ef7d8b9987","n":32,"notes":"32-example CodeSOTA declarative multiple-choice pilot. Not an official full-benchmark score.","peak_gpu_bytes":1026005504,"profile":"pilot-32","protocol":{"adapter":"choice","area":"Language","benchmark":"OpenBookQA main (test)","contract":"Question + candidate answers \u2192 selected answer","dataset":"allenai/openbookqa","dataset_config":"main","dataset_revision":"388097ea7776314e93a529163e0fea805b8a6454","dataset_url":"https://huggingface.co/datasets/allenai/openbookqa","direction":"max","id":"openbookqa-main-choice","license":["unknown"],"limit":32,"metric":"accuracy","models":[{"id":"Qwen/Qwen2.5-0.5B","revision":"060db6499f32faf8b98477b0a26969ef7d8b9987"},{"id":"Qwen/Qwen2.5-1.5B","revision":"8faed761d45a263340a0528343f099c05c9a4323"}],"name":"OpenBookQA main (test)","parent_task":"reading","protocol_version":"codesota-choice-pilot-v1","recipe":{"answer_field":"answerKey","answer_type":"label","choice_labels_field":"choices.label","choices_field":"choices.text","data_file":"main/test-00000-of-00001.parquet","question_field":"question_stem"},"runner_file":"evaluate_choice.py","source":"https://huggingface.co/datasets/allenai/openbookqa","split":"test"},"runner_sha256":"4f6810ca652d8b80ae0c5e19f2a7e313e68357bb645a88a4d866e4c1141a0ed9","scope":"4be725c6398a1c6edfd89751b16ad115044ec5c7273cb39dd1c5c0eb2dfd2c79","score":0.40625,"seconds":2.220320450956933,"seed":42,"selection":"first 32 examples in pinned parquet order","status":"completed","task":"openbookqa-main-choice","worker":"white-gpu1"}