open-r1-eval-leaderboard
/
eval_results
/abhishek
/autotrain-mixtral-8x7b-orpo-v1
/main
/agieval
/results_2024-05-01T17-53-21.084520.json
| { | |
| "config_general": { | |
| "lighteval_sha": "?", | |
| "num_fewshot_seeds": 1, | |
| "override_batch_size": 4, | |
| "max_samples": null, | |
| "job_id": "", | |
| "start_time": 155101.916908373, | |
| "end_time": 158135.917280056, | |
| "total_evaluation_time_secondes": "3034.0003716829815", | |
| "model_name": "abhishek/autotrain-mixtral-8x7b-orpo-v1", | |
| "model_sha": "a8be37cf01ad767a0c71e0ba3af29c0b3ebcb559", | |
| "model_dtype": "torch.bfloat16", | |
| "model_size": "87.49 GB", | |
| "config": null | |
| }, | |
| "results": { | |
| "lighteval|agieval:aqua-rat|0": { | |
| "acc": 0.28346456692913385, | |
| "acc_stderr": 0.02833400492130763, | |
| "acc_norm": 0.2637795275590551, | |
| "acc_norm_stderr": 0.027705387412897232 | |
| }, | |
| "lighteval|agieval:gaokao-biology|0": { | |
| "acc": 0.2619047619047619, | |
| "acc_stderr": 0.030412684459928764, | |
| "acc_norm": 0.2904761904761905, | |
| "acc_norm_stderr": 0.03140260048069879 | |
| }, | |
| "lighteval|agieval:gaokao-chemistry|0": { | |
| "acc": 0.23671497584541062, | |
| "acc_stderr": 0.029615742669460064, | |
| "acc_norm": 0.28502415458937197, | |
| "acc_norm_stderr": 0.031452336098736614 | |
| }, | |
| "lighteval|agieval:gaokao-chinese|0": { | |
| "acc": 0.2967479674796748, | |
| "acc_stderr": 0.029185445861037915, | |
| "acc_norm": 0.2967479674796748, | |
| "acc_norm_stderr": 0.029185445861037915 | |
| }, | |
| "lighteval|agieval:gaokao-english|0": { | |
| "acc": 0.7581699346405228, | |
| "acc_stderr": 0.024518195641879334, | |
| "acc_norm": 0.7058823529411765, | |
| "acc_norm_stderr": 0.026090162504279053 | |
| }, | |
| "lighteval|agieval:gaokao-geography|0": { | |
| "acc": 0.4321608040201005, | |
| "acc_stderr": 0.035204872502584535, | |
| "acc_norm": 0.3869346733668342, | |
| "acc_norm_stderr": 0.034613029011343124 | |
| }, | |
| "lighteval|agieval:gaokao-history|0": { | |
| "acc": 0.4340425531914894, | |
| "acc_stderr": 0.03240038086792747, | |
| "acc_norm": 0.39574468085106385, | |
| "acc_norm_stderr": 0.03196758697835363 | |
| }, | |
| "lighteval|agieval:gaokao-mathqa|0": { | |
| "acc": 0.2905982905982906, | |
| "acc_stderr": 0.024269376594480002, | |
| "acc_norm": 0.2962962962962963, | |
| "acc_norm_stderr": 0.024407539882901102 | |
| }, | |
| "lighteval|agieval:gaokao-physics|0": { | |
| "acc": 0.28, | |
| "acc_stderr": 0.031828687164775826, | |
| "acc_norm": 0.315, | |
| "acc_norm_stderr": 0.032928657464464904 | |
| }, | |
| "lighteval|agieval:logiqa-en|0": { | |
| "acc": 0.42857142857142855, | |
| "acc_stderr": 0.019410463442478737, | |
| "acc_norm": 0.4039938556067588, | |
| "acc_norm_stderr": 0.019246690834000657 | |
| }, | |
| "lighteval|agieval:logiqa-zh|0": { | |
| "acc": 0.2857142857142857, | |
| "acc_stderr": 0.01771924779845829, | |
| "acc_norm": 0.3225806451612903, | |
| "acc_norm_stderr": 0.018335437421251717 | |
| }, | |
| "lighteval|agieval:lsat-ar|0": { | |
| "acc": 0.21304347826086956, | |
| "acc_stderr": 0.02705775438993618, | |
| "acc_norm": 0.1782608695652174, | |
| "acc_norm_stderr": 0.025291655246273914 | |
| }, | |
| "lighteval|agieval:lsat-lr|0": { | |
| "acc": 0.5117647058823529, | |
| "acc_stderr": 0.02215597466931114, | |
| "acc_norm": 0.43529411764705883, | |
| "acc_norm_stderr": 0.02197574787518572 | |
| }, | |
| "lighteval|agieval:lsat-rc|0": { | |
| "acc": 0.6133828996282528, | |
| "acc_stderr": 0.029746711725453, | |
| "acc_norm": 0.5315985130111525, | |
| "acc_norm_stderr": 0.030481309049990975 | |
| }, | |
| "lighteval|agieval:sat-en|0": { | |
| "acc": 0.7864077669902912, | |
| "acc_stderr": 0.02862461766771065, | |
| "acc_norm": 0.7281553398058253, | |
| "acc_norm_stderr": 0.031073880563247468 | |
| }, | |
| "lighteval|agieval:sat-en-without-passage|0": { | |
| "acc": 0.5145631067961165, | |
| "acc_stderr": 0.03490669905098906, | |
| "acc_norm": 0.44660194174757284, | |
| "acc_norm_stderr": 0.034721796582639484 | |
| }, | |
| "lighteval|agieval:sat-math|0": { | |
| "acc": 0.44545454545454544, | |
| "acc_stderr": 0.03358522134954388, | |
| "acc_norm": 0.35909090909090907, | |
| "acc_norm_stderr": 0.0324174137275263 | |
| }, | |
| "lighteval|agieval:_average|0": { | |
| "acc": 0.4160415336416192, | |
| "acc_stderr": 0.028175063575133087, | |
| "acc_norm": 0.3906742373644382, | |
| "acc_norm_stderr": 0.028429216293813443 | |
| }, | |
| "all": { | |
| "acc": 0.4160415336416192, | |
| "acc_stderr": 0.028175063575133087, | |
| "acc_norm": 0.3906742373644382, | |
| "acc_norm_stderr": 0.028429216293813443 | |
| } | |
| }, | |
| "versions": { | |
| "lighteval|agieval:aqua-rat|0": 0, | |
| "lighteval|agieval:gaokao-biology|0": 0, | |
| "lighteval|agieval:gaokao-chemistry|0": 0, | |
| "lighteval|agieval:gaokao-chinese|0": 0, | |
| "lighteval|agieval:gaokao-english|0": 0, | |
| "lighteval|agieval:gaokao-geography|0": 0, | |
| "lighteval|agieval:gaokao-history|0": 0, | |
| "lighteval|agieval:gaokao-mathqa|0": 0, | |
| "lighteval|agieval:gaokao-physics|0": 0, | |
| "lighteval|agieval:logiqa-en|0": 0, | |
| "lighteval|agieval:logiqa-zh|0": 0, | |
| "lighteval|agieval:lsat-ar|0": 0, | |
| "lighteval|agieval:lsat-lr|0": 0, | |
| "lighteval|agieval:lsat-rc|0": 0, | |
| "lighteval|agieval:sat-en|0": 0, | |
| "lighteval|agieval:sat-en-without-passage|0": 0, | |
| "lighteval|agieval:sat-math|0": 0 | |
| }, | |
| "config_tasks": { | |
| "lighteval|agieval:aqua-rat": { | |
| "name": "agieval:aqua-rat", | |
| "prompt_function": "agieval", | |
| "hf_repo": "dmayhem93/agieval-aqua-rat", | |
| "hf_subset": "default", | |
| "metric": [ | |
| "loglikelihood_acc", | |
| "loglikelihood_acc_norm_nospace" | |
| ], | |
| "hf_avail_splits": [ | |
| "test" | |
| ], | |
| "evaluation_splits": [ | |
| "test" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": "random_sampling", | |
| "generation_size": 1, | |
| "stop_sequence": null, | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "lighteval" | |
| ], | |
| "original_num_docs": 254, | |
| "effective_num_docs": 254, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "lighteval|agieval:gaokao-biology": { | |
| "name": "agieval:gaokao-biology", | |
| "prompt_function": "agieval", | |
| "hf_repo": "dmayhem93/agieval-gaokao-biology", | |
| "hf_subset": "default", | |
| "metric": [ | |
| "loglikelihood_acc", | |
| "loglikelihood_acc_norm_nospace" | |
| ], | |
| "hf_avail_splits": [ | |
| "test" | |
| ], | |
| "evaluation_splits": [ | |
| "test" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": "random_sampling", | |
| "generation_size": 1, | |
| "stop_sequence": null, | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "lighteval" | |
| ], | |
| "original_num_docs": 210, | |
| "effective_num_docs": 210, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "lighteval|agieval:gaokao-chemistry": { | |
| "name": "agieval:gaokao-chemistry", | |
| "prompt_function": "agieval", | |
| "hf_repo": "dmayhem93/agieval-gaokao-chemistry", | |
| "hf_subset": "default", | |
| "metric": [ | |
| "loglikelihood_acc", | |
| "loglikelihood_acc_norm_nospace" | |
| ], | |
| "hf_avail_splits": [ | |
| "test" | |
| ], | |
| "evaluation_splits": [ | |
| "test" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": "random_sampling", | |
| "generation_size": 1, | |
| "stop_sequence": null, | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "lighteval" | |
| ], | |
| "original_num_docs": 207, | |
| "effective_num_docs": 207, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "lighteval|agieval:gaokao-chinese": { | |
| "name": "agieval:gaokao-chinese", | |
| "prompt_function": "agieval", | |
| "hf_repo": "dmayhem93/agieval-gaokao-chinese", | |
| "hf_subset": "default", | |
| "metric": [ | |
| "loglikelihood_acc", | |
| "loglikelihood_acc_norm_nospace" | |
| ], | |
| "hf_avail_splits": [ | |
| "test" | |
| ], | |
| "evaluation_splits": [ | |
| "test" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": "random_sampling", | |
| "generation_size": 1, | |
| "stop_sequence": null, | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "lighteval" | |
| ], | |
| "original_num_docs": 246, | |
| "effective_num_docs": 246, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "lighteval|agieval:gaokao-english": { | |
| "name": "agieval:gaokao-english", | |
| "prompt_function": "agieval", | |
| "hf_repo": "dmayhem93/agieval-gaokao-english", | |
| "hf_subset": "default", | |
| "metric": [ | |
| "loglikelihood_acc", | |
| "loglikelihood_acc_norm_nospace" | |
| ], | |
| "hf_avail_splits": [ | |
| "test" | |
| ], | |
| "evaluation_splits": [ | |
| "test" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": "random_sampling", | |
| "generation_size": 1, | |
| "stop_sequence": null, | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "lighteval" | |
| ], | |
| "original_num_docs": 306, | |
| "effective_num_docs": 306, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "lighteval|agieval:gaokao-geography": { | |
| "name": "agieval:gaokao-geography", | |
| "prompt_function": "agieval", | |
| "hf_repo": "dmayhem93/agieval-gaokao-geography", | |
| "hf_subset": "default", | |
| "metric": [ | |
| "loglikelihood_acc", | |
| "loglikelihood_acc_norm_nospace" | |
| ], | |
| "hf_avail_splits": [ | |
| "test" | |
| ], | |
| "evaluation_splits": [ | |
| "test" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": "random_sampling", | |
| "generation_size": 1, | |
| "stop_sequence": null, | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "lighteval" | |
| ], | |
| "original_num_docs": 199, | |
| "effective_num_docs": 199, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "lighteval|agieval:gaokao-history": { | |
| "name": "agieval:gaokao-history", | |
| "prompt_function": "agieval", | |
| "hf_repo": "dmayhem93/agieval-gaokao-history", | |
| "hf_subset": "default", | |
| "metric": [ | |
| "loglikelihood_acc", | |
| "loglikelihood_acc_norm_nospace" | |
| ], | |
| "hf_avail_splits": [ | |
| "test" | |
| ], | |
| "evaluation_splits": [ | |
| "test" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": "random_sampling", | |
| "generation_size": 1, | |
| "stop_sequence": null, | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "lighteval" | |
| ], | |
| "original_num_docs": 235, | |
| "effective_num_docs": 235, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "lighteval|agieval:gaokao-mathqa": { | |
| "name": "agieval:gaokao-mathqa", | |
| "prompt_function": "agieval", | |
| "hf_repo": "dmayhem93/agieval-gaokao-mathqa", | |
| "hf_subset": "default", | |
| "metric": [ | |
| "loglikelihood_acc", | |
| "loglikelihood_acc_norm_nospace" | |
| ], | |
| "hf_avail_splits": [ | |
| "test" | |
| ], | |
| "evaluation_splits": [ | |
| "test" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": "random_sampling", | |
| "generation_size": 1, | |
| "stop_sequence": null, | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "lighteval" | |
| ], | |
| "original_num_docs": 351, | |
| "effective_num_docs": 351, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "lighteval|agieval:gaokao-physics": { | |
| "name": "agieval:gaokao-physics", | |
| "prompt_function": "agieval", | |
| "hf_repo": "dmayhem93/agieval-gaokao-physics", | |
| "hf_subset": "default", | |
| "metric": [ | |
| "loglikelihood_acc", | |
| "loglikelihood_acc_norm_nospace" | |
| ], | |
| "hf_avail_splits": [ | |
| "test" | |
| ], | |
| "evaluation_splits": [ | |
| "test" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": "random_sampling", | |
| "generation_size": 1, | |
| "stop_sequence": null, | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "lighteval" | |
| ], | |
| "original_num_docs": 200, | |
| "effective_num_docs": 200, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "lighteval|agieval:logiqa-en": { | |
| "name": "agieval:logiqa-en", | |
| "prompt_function": "agieval", | |
| "hf_repo": "dmayhem93/agieval-logiqa-en", | |
| "hf_subset": "default", | |
| "metric": [ | |
| "loglikelihood_acc", | |
| "loglikelihood_acc_norm_nospace" | |
| ], | |
| "hf_avail_splits": [ | |
| "test" | |
| ], | |
| "evaluation_splits": [ | |
| "test" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": "random_sampling", | |
| "generation_size": 1, | |
| "stop_sequence": null, | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "lighteval" | |
| ], | |
| "original_num_docs": 651, | |
| "effective_num_docs": 651, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "lighteval|agieval:logiqa-zh": { | |
| "name": "agieval:logiqa-zh", | |
| "prompt_function": "agieval", | |
| "hf_repo": "dmayhem93/agieval-logiqa-zh", | |
| "hf_subset": "default", | |
| "metric": [ | |
| "loglikelihood_acc", | |
| "loglikelihood_acc_norm_nospace" | |
| ], | |
| "hf_avail_splits": [ | |
| "test" | |
| ], | |
| "evaluation_splits": [ | |
| "test" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": "random_sampling", | |
| "generation_size": 1, | |
| "stop_sequence": null, | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "lighteval" | |
| ], | |
| "original_num_docs": 651, | |
| "effective_num_docs": 651, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "lighteval|agieval:lsat-ar": { | |
| "name": "agieval:lsat-ar", | |
| "prompt_function": "agieval", | |
| "hf_repo": "dmayhem93/agieval-lsat-ar", | |
| "hf_subset": "default", | |
| "metric": [ | |
| "loglikelihood_acc", | |
| "loglikelihood_acc_norm_nospace" | |
| ], | |
| "hf_avail_splits": [ | |
| "test" | |
| ], | |
| "evaluation_splits": [ | |
| "test" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": "random_sampling", | |
| "generation_size": 1, | |
| "stop_sequence": null, | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "lighteval" | |
| ], | |
| "original_num_docs": 230, | |
| "effective_num_docs": 230, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "lighteval|agieval:lsat-lr": { | |
| "name": "agieval:lsat-lr", | |
| "prompt_function": "agieval", | |
| "hf_repo": "dmayhem93/agieval-lsat-lr", | |
| "hf_subset": "default", | |
| "metric": [ | |
| "loglikelihood_acc", | |
| "loglikelihood_acc_norm_nospace" | |
| ], | |
| "hf_avail_splits": [ | |
| "test" | |
| ], | |
| "evaluation_splits": [ | |
| "test" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": "random_sampling", | |
| "generation_size": 1, | |
| "stop_sequence": null, | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "lighteval" | |
| ], | |
| "original_num_docs": 510, | |
| "effective_num_docs": 510, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "lighteval|agieval:lsat-rc": { | |
| "name": "agieval:lsat-rc", | |
| "prompt_function": "agieval", | |
| "hf_repo": "dmayhem93/agieval-lsat-rc", | |
| "hf_subset": "default", | |
| "metric": [ | |
| "loglikelihood_acc", | |
| "loglikelihood_acc_norm_nospace" | |
| ], | |
| "hf_avail_splits": [ | |
| "test" | |
| ], | |
| "evaluation_splits": [ | |
| "test" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": "random_sampling", | |
| "generation_size": 1, | |
| "stop_sequence": null, | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "lighteval" | |
| ], | |
| "original_num_docs": 269, | |
| "effective_num_docs": 269, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "lighteval|agieval:sat-en": { | |
| "name": "agieval:sat-en", | |
| "prompt_function": "agieval", | |
| "hf_repo": "dmayhem93/agieval-sat-en", | |
| "hf_subset": "default", | |
| "metric": [ | |
| "loglikelihood_acc", | |
| "loglikelihood_acc_norm_nospace" | |
| ], | |
| "hf_avail_splits": [ | |
| "test" | |
| ], | |
| "evaluation_splits": [ | |
| "test" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": "random_sampling", | |
| "generation_size": 1, | |
| "stop_sequence": null, | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "lighteval" | |
| ], | |
| "original_num_docs": 206, | |
| "effective_num_docs": 206, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "lighteval|agieval:sat-en-without-passage": { | |
| "name": "agieval:sat-en-without-passage", | |
| "prompt_function": "agieval", | |
| "hf_repo": "dmayhem93/agieval-sat-en-without-passage", | |
| "hf_subset": "default", | |
| "metric": [ | |
| "loglikelihood_acc", | |
| "loglikelihood_acc_norm_nospace" | |
| ], | |
| "hf_avail_splits": [ | |
| "test" | |
| ], | |
| "evaluation_splits": [ | |
| "test" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": "random_sampling", | |
| "generation_size": 1, | |
| "stop_sequence": null, | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "lighteval" | |
| ], | |
| "original_num_docs": 206, | |
| "effective_num_docs": 206, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "lighteval|agieval:sat-math": { | |
| "name": "agieval:sat-math", | |
| "prompt_function": "agieval", | |
| "hf_repo": "dmayhem93/agieval-sat-math", | |
| "hf_subset": "default", | |
| "metric": [ | |
| "loglikelihood_acc", | |
| "loglikelihood_acc_norm_nospace" | |
| ], | |
| "hf_avail_splits": [ | |
| "test" | |
| ], | |
| "evaluation_splits": [ | |
| "test" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": "random_sampling", | |
| "generation_size": 1, | |
| "stop_sequence": null, | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "lighteval" | |
| ], | |
| "original_num_docs": 220, | |
| "effective_num_docs": 220, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| } | |
| }, | |
| "summary_tasks": { | |
| "lighteval|agieval:aqua-rat|0": { | |
| "hashes": { | |
| "hash_examples": "f09607f69e5b7525", | |
| "hash_full_prompts": "f0af1499da980246", | |
| "hash_input_tokens": "23ad8501912f9a35", | |
| "hash_cont_tokens": "5cb1512d1fe45a87" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 254, | |
| "padded": 1270, | |
| "non_padded": 0, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|agieval:gaokao-biology|0": { | |
| "hashes": { | |
| "hash_examples": "f262eaf4a72db963", | |
| "hash_full_prompts": "4027de4b1cdd1c67", | |
| "hash_input_tokens": "dec771248a9b4d26", | |
| "hash_cont_tokens": "a469b12b65e9474e" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 210, | |
| "padded": 840, | |
| "non_padded": 0, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|agieval:gaokao-chemistry|0": { | |
| "hashes": { | |
| "hash_examples": "47f2e649f58d9da5", | |
| "hash_full_prompts": "39de31ab927f9675", | |
| "hash_input_tokens": "ac2e1e2f92e86b27", | |
| "hash_cont_tokens": "43337f7821e2165f" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 207, | |
| "padded": 831, | |
| "non_padded": 0, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|agieval:gaokao-chinese|0": { | |
| "hashes": { | |
| "hash_examples": "1010b21fde4726ab", | |
| "hash_full_prompts": "3b4313f1bd85fd2e", | |
| "hash_input_tokens": "4337bd80a77f1b96", | |
| "hash_cont_tokens": "7c9a23bda503006a" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 246, | |
| "padded": 982, | |
| "non_padded": 2, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|agieval:gaokao-english|0": { | |
| "hashes": { | |
| "hash_examples": "4864e492a350ae93", | |
| "hash_full_prompts": "547863254a606496", | |
| "hash_input_tokens": "02dd45b62519e29a", | |
| "hash_cont_tokens": "2939fab73e305a01" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 306, | |
| "padded": 1224, | |
| "non_padded": 0, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|agieval:gaokao-geography|0": { | |
| "hashes": { | |
| "hash_examples": "ec3a021e37650e7d", | |
| "hash_full_prompts": "0a7cffbf555ab29e", | |
| "hash_input_tokens": "4905dc7fbf95a714", | |
| "hash_cont_tokens": "b627d5a2a58e1b4e" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 199, | |
| "padded": 796, | |
| "non_padded": 0, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|agieval:gaokao-history|0": { | |
| "hashes": { | |
| "hash_examples": "b3fad1596f1ae1f9", | |
| "hash_full_prompts": "b8aca4146c3435af", | |
| "hash_input_tokens": "45dd622494c7eba6", | |
| "hash_cont_tokens": "29af98bdef705752" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 235, | |
| "padded": 940, | |
| "non_padded": 0, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|agieval:gaokao-mathqa|0": { | |
| "hashes": { | |
| "hash_examples": "1d1088556861b0b0", | |
| "hash_full_prompts": "1441e196c635c040", | |
| "hash_input_tokens": "a6db369430f70502", | |
| "hash_cont_tokens": "e503c4178e232557" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 351, | |
| "padded": 1404, | |
| "non_padded": 0, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|agieval:gaokao-physics|0": { | |
| "hashes": { | |
| "hash_examples": "eb05f035c7bfca2f", | |
| "hash_full_prompts": "be15722274b1466d", | |
| "hash_input_tokens": "20cf0004ad23a395", | |
| "hash_cont_tokens": "74730137f1150712" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 200, | |
| "padded": 800, | |
| "non_padded": 0, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|agieval:logiqa-en|0": { | |
| "hashes": { | |
| "hash_examples": "0a688a45f69c21e0", | |
| "hash_full_prompts": "ca179e67bdc726a6", | |
| "hash_input_tokens": "8020ea34074302e8", | |
| "hash_cont_tokens": "406bb91c565edad6" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 651, | |
| "padded": 2604, | |
| "non_padded": 0, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|agieval:logiqa-zh|0": { | |
| "hashes": { | |
| "hash_examples": "620d6888b6012ea5", | |
| "hash_full_prompts": "55e305ed89c6e580", | |
| "hash_input_tokens": "b6b03fce18b14a92", | |
| "hash_cont_tokens": "9a57711e484917c2" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 651, | |
| "padded": 2603, | |
| "non_padded": 1, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|agieval:lsat-ar|0": { | |
| "hashes": { | |
| "hash_examples": "627c8f5ccd5da209", | |
| "hash_full_prompts": "59e010e22954d5b7", | |
| "hash_input_tokens": "f4504d090f5c0b3f", | |
| "hash_cont_tokens": "35b4b88913597c0b" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 230, | |
| "padded": 1137, | |
| "non_padded": 13, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|agieval:lsat-lr|0": { | |
| "hashes": { | |
| "hash_examples": "794641c86de172f5", | |
| "hash_full_prompts": "efc3c1a3a1586d3e", | |
| "hash_input_tokens": "df779a462088fb9d", | |
| "hash_cont_tokens": "036ded82b481c7ce" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 510, | |
| "padded": 2532, | |
| "non_padded": 18, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|agieval:lsat-rc|0": { | |
| "hashes": { | |
| "hash_examples": "35981ed917ea01cf", | |
| "hash_full_prompts": "b80e2b86e1eb0cea", | |
| "hash_input_tokens": "fa91b05e6e7ce596", | |
| "hash_cont_tokens": "800a77d64122b1b9" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 269, | |
| "padded": 1345, | |
| "non_padded": 0, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|agieval:sat-en|0": { | |
| "hashes": { | |
| "hash_examples": "041c39c646536a1e", | |
| "hash_full_prompts": "4eb610121b313521", | |
| "hash_input_tokens": "99e0785666d42b0c", | |
| "hash_cont_tokens": "70c3acb1407a2ef8" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 206, | |
| "padded": 821, | |
| "non_padded": 0, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|agieval:sat-en-without-passage|0": { | |
| "hashes": { | |
| "hash_examples": "e4d9284367dff68f", | |
| "hash_full_prompts": "532ea18906ff2f4e", | |
| "hash_input_tokens": "0070904e82d507eb", | |
| "hash_cont_tokens": "7f81dee0bb601ee9" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 206, | |
| "padded": 817, | |
| "non_padded": 4, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|agieval:sat-math|0": { | |
| "hashes": { | |
| "hash_examples": "01db7291603fc1a0", | |
| "hash_full_prompts": "1422fad2e0cca51f", | |
| "hash_input_tokens": "6b607d94aa2ca7d6", | |
| "hash_cont_tokens": "c555a90f08036a0f" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 220, | |
| "padded": 877, | |
| "non_padded": 3, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| } | |
| }, | |
| "summary_general": { | |
| "hashes": { | |
| "hash_examples": "da3af66181f18ddf", | |
| "hash_full_prompts": "e89209d4ce68d63a", | |
| "hash_input_tokens": "ca61573f55e495e6", | |
| "hash_cont_tokens": "832044d7eea021d0" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 5151, | |
| "padded": 21823, | |
| "non_padded": 41, | |
| "num_truncated_few_shots": 0 | |
| } | |
| } |