open-r1-eval-leaderboard
/
eval_results
/mistralai
/Mistral-7B-Instruct-v0.2
/main
/agieval
/results_2024-03-28T16-44-41.848289.json
| { | |
| "config_general": { | |
| "lighteval_sha": "?", | |
| "num_fewshot_seeds": 1, | |
| "override_batch_size": 1, | |
| "max_samples": null, | |
| "job_id": "", | |
| "start_time": 8646450.977722593, | |
| "end_time": 8646730.167740777, | |
| "total_evaluation_time_secondes": "279.19001818448305", | |
| "model_name": "mistralai/Mistral-7B-Instruct-v0.2", | |
| "model_sha": "41b61a33a2483885c981aa79e0df6b32407ed873", | |
| "model_dtype": "torch.bfloat16", | |
| "model_size": "13.99 GB", | |
| "config": null | |
| }, | |
| "results": { | |
| "lighteval|agieval:aqua-rat|0": { | |
| "acc": 0.20866141732283464, | |
| "acc_stderr": 0.02554712225493389, | |
| "acc_norm": 0.2125984251968504, | |
| "acc_norm_stderr": 0.025722779833723057 | |
| }, | |
| "lighteval|agieval:gaokao-biology|0": { | |
| "acc": 0.3, | |
| "acc_stderr": 0.031698338899620884, | |
| "acc_norm": 0.319047619047619, | |
| "acc_norm_stderr": 0.03224133248962465 | |
| }, | |
| "lighteval|agieval:gaokao-chemistry|0": { | |
| "acc": 0.28019323671497587, | |
| "acc_stderr": 0.031289827964521094, | |
| "acc_norm": 0.30917874396135264, | |
| "acc_norm_stderr": 0.03219986494000449 | |
| }, | |
| "lighteval|agieval:gaokao-chinese|0": { | |
| "acc": 0.3170731707317073, | |
| "acc_stderr": 0.029729239035124134, | |
| "acc_norm": 0.3170731707317073, | |
| "acc_norm_stderr": 0.029729239035124134 | |
| }, | |
| "lighteval|agieval:gaokao-english|0": { | |
| "acc": 0.7581699346405228, | |
| "acc_stderr": 0.024518195641879334, | |
| "acc_norm": 0.7647058823529411, | |
| "acc_norm_stderr": 0.0242886194660461 | |
| }, | |
| "lighteval|agieval:gaokao-geography|0": { | |
| "acc": 0.39195979899497485, | |
| "acc_stderr": 0.034693995271705115, | |
| "acc_norm": 0.39195979899497485, | |
| "acc_norm_stderr": 0.034693995271705115 | |
| }, | |
| "lighteval|agieval:gaokao-history|0": { | |
| "acc": 0.4, | |
| "acc_stderr": 0.03202563076101737, | |
| "acc_norm": 0.40425531914893614, | |
| "acc_norm_stderr": 0.03208115750788684 | |
| }, | |
| "lighteval|agieval:gaokao-mathqa|0": { | |
| "acc": 0.28205128205128205, | |
| "acc_stderr": 0.024053414152940704, | |
| "acc_norm": 0.28205128205128205, | |
| "acc_norm_stderr": 0.0240534141529407 | |
| }, | |
| "lighteval|agieval:gaokao-physics|0": { | |
| "acc": 0.245, | |
| "acc_stderr": 0.0304880732921142, | |
| "acc_norm": 0.25, | |
| "acc_norm_stderr": 0.030695456590127176 | |
| }, | |
| "lighteval|agieval:logiqa-en|0": { | |
| "acc": 0.3655913978494624, | |
| "acc_stderr": 0.01888973378759108, | |
| "acc_norm": 0.37019969278033793, | |
| "acc_norm_stderr": 0.01893925025913045 | |
| }, | |
| "lighteval|agieval:logiqa-zh|0": { | |
| "acc": 0.3579109062980031, | |
| "acc_stderr": 0.01880305578483482, | |
| "acc_norm": 0.35944700460829493, | |
| "acc_norm_stderr": 0.018820809084481263 | |
| }, | |
| "lighteval|agieval:lsat-ar|0": { | |
| "acc": 0.21304347826086956, | |
| "acc_stderr": 0.027057754389936198, | |
| "acc_norm": 0.19130434782608696, | |
| "acc_norm_stderr": 0.025991852462828483 | |
| }, | |
| "lighteval|agieval:lsat-lr|0": { | |
| "acc": 0.3980392156862745, | |
| "acc_stderr": 0.02169642300116407, | |
| "acc_norm": 0.38823529411764707, | |
| "acc_norm_stderr": 0.021601346576260526 | |
| }, | |
| "lighteval|agieval:lsat-rc|0": { | |
| "acc": 0.5650557620817844, | |
| "acc_stderr": 0.030282731632881116, | |
| "acc_norm": 0.5539033457249071, | |
| "acc_norm_stderr": 0.030364356394504122 | |
| }, | |
| "lighteval|agieval:sat-en|0": { | |
| "acc": 0.7087378640776699, | |
| "acc_stderr": 0.031732764025283834, | |
| "acc_norm": 0.6990291262135923, | |
| "acc_norm_stderr": 0.032035605718474115 | |
| }, | |
| "lighteval|agieval:sat-en-without-passage|0": { | |
| "acc": 0.39805825242718446, | |
| "acc_stderr": 0.03418799390613399, | |
| "acc_norm": 0.38349514563106796, | |
| "acc_norm_stderr": 0.033960279445866395 | |
| }, | |
| "lighteval|agieval:sat-math|0": { | |
| "acc": 0.38636363636363635, | |
| "acc_stderr": 0.032902705393166656, | |
| "acc_norm": 0.35909090909090907, | |
| "acc_norm_stderr": 0.032417413727526295 | |
| }, | |
| "lighteval|agieval:_average|0": { | |
| "acc": 0.3868181972647755, | |
| "acc_stderr": 0.028211588187932263, | |
| "acc_norm": 0.3856220651457946, | |
| "acc_norm_stderr": 0.028225692526838465 | |
| }, | |
| "all": { | |
| "acc": 0.3868181972647755, | |
| "acc_stderr": 0.028211588187932263, | |
| "acc_norm": 0.3856220651457946, | |
| "acc_norm_stderr": 0.028225692526838465 | |
| } | |
| }, | |
| "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": "3861b27efd5a6170", | |
| "hash_input_tokens": "24b3053366cab5d6", | |
| "hash_cont_tokens": "7c2bcf33ed1e5ae3" | |
| }, | |
| "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": "1ef14dcf5933ff33", | |
| "hash_input_tokens": "3553b0507d1c6868", | |
| "hash_cont_tokens": "b3febf85776696e1" | |
| }, | |
| "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": "324a4d07d3b7de40", | |
| "hash_input_tokens": "62a14b0cfb722cf4", | |
| "hash_cont_tokens": "711fb41221b6515e" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 207, | |
| "padded": 830, | |
| "non_padded": 1, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|agieval:gaokao-chinese|0": { | |
| "hashes": { | |
| "hash_examples": "1010b21fde4726ab", | |
| "hash_full_prompts": "938611fd6ecdbf4e", | |
| "hash_input_tokens": "19eb2367f1cb72de", | |
| "hash_cont_tokens": "393b1820b8f4534f" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 246, | |
| "padded": 981, | |
| "non_padded": 3, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|agieval:gaokao-english|0": { | |
| "hashes": { | |
| "hash_examples": "4864e492a350ae93", | |
| "hash_full_prompts": "17369402cc020876", | |
| "hash_input_tokens": "066e65f806b474dd", | |
| "hash_cont_tokens": "dd72ceb1d7224598" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 306, | |
| "padded": 1222, | |
| "non_padded": 2, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|agieval:gaokao-geography|0": { | |
| "hashes": { | |
| "hash_examples": "ec3a021e37650e7d", | |
| "hash_full_prompts": "79c20e1e52638229", | |
| "hash_input_tokens": "2e676a0441f390c7", | |
| "hash_cont_tokens": "e06462bcee629ea8" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 199, | |
| "padded": 794, | |
| "non_padded": 2, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|agieval:gaokao-history|0": { | |
| "hashes": { | |
| "hash_examples": "b3fad1596f1ae1f9", | |
| "hash_full_prompts": "541c270a0bbab3fd", | |
| "hash_input_tokens": "cf7e8a42c7bdf5cd", | |
| "hash_cont_tokens": "87cce58a4ec6cfd8" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 235, | |
| "padded": 935, | |
| "non_padded": 5, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|agieval:gaokao-mathqa|0": { | |
| "hashes": { | |
| "hash_examples": "1d1088556861b0b0", | |
| "hash_full_prompts": "6002200193304ad6", | |
| "hash_input_tokens": "a2cfd284204fffcc", | |
| "hash_cont_tokens": "771402d59229cbae" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 351, | |
| "padded": 1401, | |
| "non_padded": 3, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|agieval:gaokao-physics|0": { | |
| "hashes": { | |
| "hash_examples": "eb05f035c7bfca2f", | |
| "hash_full_prompts": "7b606fe0e48d1136", | |
| "hash_input_tokens": "798a188685f3c7ba", | |
| "hash_cont_tokens": "f1574dddbe4231e1" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 200, | |
| "padded": 797, | |
| "non_padded": 3, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|agieval:logiqa-en|0": { | |
| "hashes": { | |
| "hash_examples": "0a688a45f69c21e0", | |
| "hash_full_prompts": "70df665a01ea64f7", | |
| "hash_input_tokens": "4ef64d022f86d868", | |
| "hash_cont_tokens": "106c1564fdd5ff8e" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 651, | |
| "padded": 2597, | |
| "non_padded": 7, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|agieval:logiqa-zh|0": { | |
| "hashes": { | |
| "hash_examples": "620d6888b6012ea5", | |
| "hash_full_prompts": "11b751123ca8b711", | |
| "hash_input_tokens": "d30860e1a31f27b3", | |
| "hash_cont_tokens": "22d3640f03c44bb0" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 651, | |
| "padded": 2581, | |
| "non_padded": 23, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|agieval:lsat-ar|0": { | |
| "hashes": { | |
| "hash_examples": "627c8f5ccd5da209", | |
| "hash_full_prompts": "254286f37d4a76ce", | |
| "hash_input_tokens": "dc24a391ad28e575", | |
| "hash_cont_tokens": "9e13af9ad8f5f78e" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 230, | |
| "padded": 1140, | |
| "non_padded": 10, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|agieval:lsat-lr|0": { | |
| "hashes": { | |
| "hash_examples": "794641c86de172f5", | |
| "hash_full_prompts": "cf1badbb8c42f6a8", | |
| "hash_input_tokens": "2252f2846ec8cecf", | |
| "hash_cont_tokens": "86333847359ceee5" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 510, | |
| "padded": 2525, | |
| "non_padded": 25, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|agieval:lsat-rc|0": { | |
| "hashes": { | |
| "hash_examples": "35981ed917ea01cf", | |
| "hash_full_prompts": "5fbc2eae9474469b", | |
| "hash_input_tokens": "768fcfff60a6a668", | |
| "hash_cont_tokens": "cb47c6b984067525" | |
| }, | |
| "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": "171c266251542e36", | |
| "hash_input_tokens": "c5342c28aac7b5e9", | |
| "hash_cont_tokens": "cb01422bf828aefe" | |
| }, | |
| "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": "66bc76f2704c6ca7", | |
| "hash_input_tokens": "582f986b74383001", | |
| "hash_cont_tokens": "cb01422bf828aefe" | |
| }, | |
| "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": "6016bbecd8642dbe", | |
| "hash_input_tokens": "0f1ec1ab676fb3e4", | |
| "hash_cont_tokens": "0c4980b69a75cb83" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 220, | |
| "padded": 876, | |
| "non_padded": 4, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| } | |
| }, | |
| "summary_general": { | |
| "hashes": { | |
| "hash_examples": "da3af66181f18ddf", | |
| "hash_full_prompts": "e341944cee05878c", | |
| "hash_input_tokens": "5f2600d57bfbb6aa", | |
| "hash_cont_tokens": "7f70e41f6e93ac40" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 5151, | |
| "padded": 21772, | |
| "non_padded": 92, | |
| "num_truncated_few_shots": 0 | |
| } | |
| } |