{"batch_policy":{"choice_candidates":16,"code_test_workers":4,"generation_examples":8,"padded_token_budget":8192},"batch_timings":[{"candidates":16,"kind":"choice","padded_tokens":448,"seconds":0.12951591901946813},{"candidates":16,"kind":"choice","padded_tokens":496,"seconds":0.03552447399124503},{"candidates":16,"kind":"choice","padded_tokens":528,"seconds":0.040724562015384436},{"candidates":16,"kind":"choice","padded_tokens":592,"seconds":0.04952972300816327},{"candidates":16,"kind":"choice","padded_tokens":608,"seconds":0.04472782800439745},{"candidates":16,"kind":"choice","padded_tokens":768,"seconds":0.05515940301120281}],"dataset_revision":"cdb5c3d5eed6ead6e5a341c8e56e669bb666725b","device":"cuda:0","evaluator_source":true,"evidence":"CodeSOTA execution on White","execution_profile":"batched-v2","finished":1792198122,"gpu":0,"id":"bddd3154bd9cf23f3dece3fb","measured_cost_usd":null,"metric":"accuracy","model":"Qwen/Qwen2.5-1.5B","model_revision":"8faed761d45a263340a0528343f099c05c9a4323","n":32,"notes":"Pilot subset, CodeSOTA batched-v2 execution. Separate scope from serial v1; unchanged prompt/tokenization/metric contracts.","peak_gpu_bytes":3151138816,"profile":"pilot-32","protocol":{"adapter":"entailment","area":"Language","benchmark":"SNLI","contract":"Premise + hypothesis \u2192 entailment / neutral / contradiction","dataset":"stanfordnlp/snli","dataset_config":"plain_text","dataset_revision":"cdb5c3d5eed6ead6e5a341c8e56e669bb666725b","dataset_url":"https://huggingface.co/datasets/stanfordnlp/snli","direction":"max","id":"entailment","license":["cc-by-sa-4.0"],"limit":32,"metric":"accuracy","models":[{"id":"Qwen/Qwen2.5-0.5B","revision":"060db6499f32faf8b98477b0a26969ef7d8b9987"},{"id":"Qwen/Qwen2.5-1.5B","revision":"8faed761d45a263340a0528343f099c05c9a4323"}],"name":"Natural language inference","protocol_version":"codesota-pilot-v2-batched","runner_file":"evaluate_fast.py","source":"https://nlp.stanford.edu/projects/snli/","split":"test"},"resource_metrics":{"basis":"Evaluator subprocess elapsed \u00d7 one reserved GPU; includes data/model loading and CPU scoring; not kernel-active time","gpu_count":1,"gpu_slot_seconds":7.377454819041304,"measurement_version":"gpu-slot-v1"},"runner_sha256":"bc899fb160f6881408863cd8bcdf96102dd6d8b10559a5f781a8cd816103da77","scope":"3ef5086324e4a200ffa81c1058aab801250aedfad5e4f11c7c4b8a63c1bf5b3e","score":0.375,"seconds":1.6738499279599637,"seed":42,"selection":"first 32 valid examples in pinned split order","status":"completed","task":"entailment","worker":"white-gpu0"}