open-r1-eval-leaderboard
/
eval_results
/AI-MO
/mistral-7b-sft
/aimo_v03.00
/math
/results_2024-04-21T22-00-32.902902.json
{ | |
"config_general": { | |
"lighteval_sha": "?", | |
"num_fewshot_seeds": 1, | |
"override_batch_size": 4, | |
"max_samples": null, | |
"job_id": "", | |
"start_time": 882532.794319516, | |
"end_time": 891148.601742387, | |
"total_evaluation_time_secondes": "8615.807422871003", | |
"model_name": "AI-MO/mistral-7b-sft", | |
"model_sha": "159047b1ab76bbb7c9369ee71bfef1d441fc029e", | |
"model_dtype": "torch.bfloat16", | |
"model_size": "13.99 GB", | |
"config": null | |
}, | |
"results": { | |
"lighteval|math:algebra|0": { | |
"qem": 0.46166807076663857, | |
"qem_stderr": 0.01447596901495004 | |
}, | |
"lighteval|math:counting_and_probability|0": { | |
"qem": 0.18143459915611815, | |
"qem_stderr": 0.017719692309092615 | |
}, | |
"lighteval|math:geometry|0": { | |
"qem": 0.23382045929018788, | |
"qem_stderr": 0.019359430691791527 | |
}, | |
"lighteval|math:intermediate_algebra|0": { | |
"qem": 0.12624584717607973, | |
"qem_stderr": 0.011058593855296428 | |
}, | |
"lighteval|math:number_theory|0": { | |
"qem": 0.16111111111111112, | |
"qem_stderr": 0.015835091780678594 | |
}, | |
"lighteval|math:prealgebra|0": { | |
"qem": 0.46842709529276694, | |
"qem_stderr": 0.01691775751043896 | |
}, | |
"lighteval|math:precalculus|0": { | |
"qem": 0.1575091575091575, | |
"qem_stderr": 0.015604046923319667 | |
}, | |
"lighteval|math:_average|0": { | |
"qem": 0.25574519147172287, | |
"qem_stderr": 0.01585294029793826 | |
}, | |
"all": { | |
"qem": 0.25574519147172287, | |
"qem_stderr": 0.01585294029793826 | |
} | |
}, | |
"versions": { | |
"lighteval|math:algebra|0": 0, | |
"lighteval|math:counting_and_probability|0": 0, | |
"lighteval|math:geometry|0": 0, | |
"lighteval|math:intermediate_algebra|0": 0, | |
"lighteval|math:number_theory|0": 0, | |
"lighteval|math:prealgebra|0": 0, | |
"lighteval|math:precalculus|0": 0 | |
}, | |
"config_tasks": { | |
"lighteval|math:algebra": { | |
"name": "math:algebra", | |
"prompt_function": "math", | |
"hf_repo": "lighteval/MATH", | |
"hf_subset": "algebra", | |
"metric": [ | |
"quasi_exact_match_math" | |
], | |
"hf_avail_splits": [ | |
"train", | |
"test", | |
"validation" | |
], | |
"evaluation_splits": [ | |
"test" | |
], | |
"few_shots_split": null, | |
"few_shots_select": null, | |
"generation_size": 2048, | |
"stop_sequence": [ | |
"\n" | |
], | |
"output_regex": null, | |
"frozen": false, | |
"suite": [ | |
"lighteval", | |
"math" | |
], | |
"original_num_docs": 1187, | |
"effective_num_docs": 1187, | |
"trust_dataset": true, | |
"must_remove_duplicate_docs": null | |
}, | |
"lighteval|math:counting_and_probability": { | |
"name": "math:counting_and_probability", | |
"prompt_function": "math", | |
"hf_repo": "lighteval/MATH", | |
"hf_subset": "counting_and_probability", | |
"metric": [ | |
"quasi_exact_match_math" | |
], | |
"hf_avail_splits": [ | |
"train", | |
"test", | |
"validation" | |
], | |
"evaluation_splits": [ | |
"test" | |
], | |
"few_shots_split": null, | |
"few_shots_select": null, | |
"generation_size": 2048, | |
"stop_sequence": [ | |
"\n" | |
], | |
"output_regex": null, | |
"frozen": false, | |
"suite": [ | |
"lighteval", | |
"math" | |
], | |
"original_num_docs": 474, | |
"effective_num_docs": 474, | |
"trust_dataset": true, | |
"must_remove_duplicate_docs": null | |
}, | |
"lighteval|math:geometry": { | |
"name": "math:geometry", | |
"prompt_function": "math", | |
"hf_repo": "lighteval/MATH", | |
"hf_subset": "geometry", | |
"metric": [ | |
"quasi_exact_match_math" | |
], | |
"hf_avail_splits": [ | |
"train", | |
"test", | |
"validation" | |
], | |
"evaluation_splits": [ | |
"test" | |
], | |
"few_shots_split": null, | |
"few_shots_select": null, | |
"generation_size": 2048, | |
"stop_sequence": [ | |
"\n" | |
], | |
"output_regex": null, | |
"frozen": false, | |
"suite": [ | |
"lighteval", | |
"math" | |
], | |
"original_num_docs": 479, | |
"effective_num_docs": 479, | |
"trust_dataset": true, | |
"must_remove_duplicate_docs": null | |
}, | |
"lighteval|math:intermediate_algebra": { | |
"name": "math:intermediate_algebra", | |
"prompt_function": "math", | |
"hf_repo": "lighteval/MATH", | |
"hf_subset": "intermediate_algebra", | |
"metric": [ | |
"quasi_exact_match_math" | |
], | |
"hf_avail_splits": [ | |
"train", | |
"test", | |
"validation" | |
], | |
"evaluation_splits": [ | |
"test" | |
], | |
"few_shots_split": null, | |
"few_shots_select": null, | |
"generation_size": 2048, | |
"stop_sequence": [ | |
"\n" | |
], | |
"output_regex": null, | |
"frozen": false, | |
"suite": [ | |
"lighteval", | |
"math" | |
], | |
"original_num_docs": 903, | |
"effective_num_docs": 903, | |
"trust_dataset": true, | |
"must_remove_duplicate_docs": null | |
}, | |
"lighteval|math:number_theory": { | |
"name": "math:number_theory", | |
"prompt_function": "math", | |
"hf_repo": "lighteval/MATH", | |
"hf_subset": "number_theory", | |
"metric": [ | |
"quasi_exact_match_math" | |
], | |
"hf_avail_splits": [ | |
"train", | |
"test", | |
"validation" | |
], | |
"evaluation_splits": [ | |
"test" | |
], | |
"few_shots_split": null, | |
"few_shots_select": null, | |
"generation_size": 2048, | |
"stop_sequence": [ | |
"\n" | |
], | |
"output_regex": null, | |
"frozen": false, | |
"suite": [ | |
"lighteval", | |
"math" | |
], | |
"original_num_docs": 540, | |
"effective_num_docs": 540, | |
"trust_dataset": true, | |
"must_remove_duplicate_docs": null | |
}, | |
"lighteval|math:prealgebra": { | |
"name": "math:prealgebra", | |
"prompt_function": "math", | |
"hf_repo": "lighteval/MATH", | |
"hf_subset": "prealgebra", | |
"metric": [ | |
"quasi_exact_match_math" | |
], | |
"hf_avail_splits": [ | |
"train", | |
"test", | |
"validation" | |
], | |
"evaluation_splits": [ | |
"test" | |
], | |
"few_shots_split": null, | |
"few_shots_select": null, | |
"generation_size": 2048, | |
"stop_sequence": [ | |
"\n" | |
], | |
"output_regex": null, | |
"frozen": false, | |
"suite": [ | |
"lighteval", | |
"math" | |
], | |
"original_num_docs": 871, | |
"effective_num_docs": 871, | |
"trust_dataset": true, | |
"must_remove_duplicate_docs": null | |
}, | |
"lighteval|math:precalculus": { | |
"name": "math:precalculus", | |
"prompt_function": "math", | |
"hf_repo": "lighteval/MATH", | |
"hf_subset": "precalculus", | |
"metric": [ | |
"quasi_exact_match_math" | |
], | |
"hf_avail_splits": [ | |
"train", | |
"test", | |
"validation" | |
], | |
"evaluation_splits": [ | |
"test" | |
], | |
"few_shots_split": null, | |
"few_shots_select": null, | |
"generation_size": 2048, | |
"stop_sequence": [ | |
"\n" | |
], | |
"output_regex": null, | |
"frozen": false, | |
"suite": [ | |
"lighteval", | |
"math" | |
], | |
"original_num_docs": 546, | |
"effective_num_docs": 546, | |
"trust_dataset": true, | |
"must_remove_duplicate_docs": null | |
} | |
}, | |
"summary_tasks": { | |
"lighteval|math:algebra|0": { | |
"hashes": { | |
"hash_examples": "37a2fd2f076d2e49", | |
"hash_full_prompts": "b594f95fc76837ae", | |
"hash_input_tokens": "268f292e08b20496", | |
"hash_cont_tokens": "5ff3b282e103a786" | |
}, | |
"truncated": 1187, | |
"non_truncated": 0, | |
"padded": 186, | |
"non_padded": 1001, | |
"effective_few_shots": 0.0, | |
"num_truncated_few_shots": 0 | |
}, | |
"lighteval|math:counting_and_probability|0": { | |
"hashes": { | |
"hash_examples": "97b4892e28bc078b", | |
"hash_full_prompts": "7248a30b48b9a71a", | |
"hash_input_tokens": "a734e9fc478accef", | |
"hash_cont_tokens": "34e1403ab557ae3f" | |
}, | |
"truncated": 474, | |
"non_truncated": 0, | |
"padded": 42, | |
"non_padded": 432, | |
"effective_few_shots": 0.0, | |
"num_truncated_few_shots": 0 | |
}, | |
"lighteval|math:geometry|0": { | |
"hashes": { | |
"hash_examples": "9e9f0228b8b3d093", | |
"hash_full_prompts": "b0ba9fa4265b2a7f", | |
"hash_input_tokens": "fd28b84db8b56f2b", | |
"hash_cont_tokens": "79a2cb59ecbb222f" | |
}, | |
"truncated": 479, | |
"non_truncated": 0, | |
"padded": 129, | |
"non_padded": 350, | |
"effective_few_shots": 0.0, | |
"num_truncated_few_shots": 0 | |
}, | |
"lighteval|math:intermediate_algebra|0": { | |
"hashes": { | |
"hash_examples": "cfe73a8e28ae94de", | |
"hash_full_prompts": "0519b385dc4d18c5", | |
"hash_input_tokens": "ac07a8b5e8432e70", | |
"hash_cont_tokens": "b36ca367a0db8943" | |
}, | |
"truncated": 900, | |
"non_truncated": 3, | |
"padded": 54, | |
"non_padded": 849, | |
"effective_few_shots": 0.0, | |
"num_truncated_few_shots": 0 | |
}, | |
"lighteval|math:number_theory|0": { | |
"hashes": { | |
"hash_examples": "4ee5237cf144afac", | |
"hash_full_prompts": "e34e2d9e3719c6b3", | |
"hash_input_tokens": "05c6e6efc4dd2f8e", | |
"hash_cont_tokens": "d07397c221f94650" | |
}, | |
"truncated": 540, | |
"non_truncated": 0, | |
"padded": 7, | |
"non_padded": 533, | |
"effective_few_shots": 0.0, | |
"num_truncated_few_shots": 0 | |
}, | |
"lighteval|math:prealgebra|0": { | |
"hashes": { | |
"hash_examples": "3fb3afeb885f73d8", | |
"hash_full_prompts": "dd429103e5accb7a", | |
"hash_input_tokens": "a674b0bd864f5275", | |
"hash_cont_tokens": "fa2fff6b43f88d13" | |
}, | |
"truncated": 871, | |
"non_truncated": 0, | |
"padded": 55, | |
"non_padded": 816, | |
"effective_few_shots": 0.0, | |
"num_truncated_few_shots": 0 | |
}, | |
"lighteval|math:precalculus|0": { | |
"hashes": { | |
"hash_examples": "753e25ab9ec4b46c", | |
"hash_full_prompts": "5437a265a758ad19", | |
"hash_input_tokens": "b75d4bd55f3f5325", | |
"hash_cont_tokens": "8e1e6abecde2e5f0" | |
}, | |
"truncated": 546, | |
"non_truncated": 0, | |
"padded": 24, | |
"non_padded": 522, | |
"effective_few_shots": 0.0, | |
"num_truncated_few_shots": 0 | |
} | |
}, | |
"summary_general": { | |
"hashes": { | |
"hash_examples": "b76099aa9092a203", | |
"hash_full_prompts": "fc120bd09022ad3f", | |
"hash_input_tokens": "c7fcbcc2ee4328b7", | |
"hash_cont_tokens": "152aa86ad3998642" | |
}, | |
"truncated": 4997, | |
"non_truncated": 3, | |
"padded": 497, | |
"non_padded": 4503, | |
"num_truncated_few_shots": 0 | |
} | |
} |