open-r1-eval-leaderboard
/
eval_results
/kaist-ai
/mistral-orpo-capybara-7k
/main
/bbh
/results_2024-03-28T19-14-39.416637.json
| { | |
| "config_general": { | |
| "lighteval_sha": "?", | |
| "num_fewshot_seeds": 1, | |
| "override_batch_size": 1, | |
| "max_samples": null, | |
| "job_id": "", | |
| "start_time": 1871307.779225898, | |
| "end_time": 1871541.417488566, | |
| "total_evaluation_time_secondes": "233.63826266792603", | |
| "model_name": "kaist-ai/mistral-orpo-capybara-7k", | |
| "model_sha": "24c1172060658a1923c9b454796857e2cc59fbeb", | |
| "model_dtype": "torch.bfloat16", | |
| "model_size": "13.99 GB", | |
| "config": null | |
| }, | |
| "results": { | |
| "lighteval|bigbench:causal_judgment|0": { | |
| "acc": 0.5157894736842106, | |
| "acc_stderr": 0.036351509398643456 | |
| }, | |
| "lighteval|bigbench:date_understanding|0": { | |
| "acc": 0.15989159891598917, | |
| "acc_stderr": 0.019105419467312108 | |
| }, | |
| "lighteval|bigbench:disambiguation_qa|0": { | |
| "acc": 0.5310077519379846, | |
| "acc_stderr": 0.031129109784128743 | |
| }, | |
| "lighteval|bigbench:geometric_shapes|0": { | |
| "acc": 0.12777777777777777, | |
| "acc_stderr": 0.017619517641103456 | |
| }, | |
| "lighteval|bigbench:logical_deduction_five_objects|0": { | |
| "acc": 0.244, | |
| "acc_stderr": 0.01922673489361458 | |
| }, | |
| "lighteval|bigbench:logical_deduction_seven_objects|0": { | |
| "acc": 0.17142857142857143, | |
| "acc_stderr": 0.014255025282953354 | |
| }, | |
| "lighteval|bigbench:logical_deduction_three_objects|0": { | |
| "acc": 0.43333333333333335, | |
| "acc_stderr": 0.028657565120703117 | |
| }, | |
| "lighteval|bigbench:movie_recommendation|0": { | |
| "acc": 0.616, | |
| "acc_stderr": 0.021772369465547198 | |
| }, | |
| "lighteval|bigbench:navigate|0": { | |
| "acc": 0.522, | |
| "acc_stderr": 0.01580397942816196 | |
| }, | |
| "lighteval|bigbench:reasoning_about_colored_objects|0": { | |
| "acc": 0.263, | |
| "acc_stderr": 0.009847029094655511 | |
| }, | |
| "lighteval|bigbench:ruin_names|0": { | |
| "acc": 0.35714285714285715, | |
| "acc_stderr": 0.022663368463226877 | |
| }, | |
| "lighteval|bigbench:salient_translation_error_detection|0": { | |
| "acc": 0.2725450901803607, | |
| "acc_stderr": 0.01410180295347219 | |
| }, | |
| "lighteval|bigbench:snarks|0": { | |
| "acc": 0.5082872928176796, | |
| "acc_stderr": 0.037262680226389885 | |
| }, | |
| "lighteval|bigbench:sports_understanding|0": { | |
| "acc": 0.679, | |
| "acc_stderr": 0.014770821817934644 | |
| }, | |
| "lighteval|bigbench:temporal_sequences|0": { | |
| "acc": 0.069, | |
| "acc_stderr": 0.008018934050315165 | |
| }, | |
| "lighteval|bigbench:tracking_shuffled_objects_five_objects|0": { | |
| "acc": 0.1632, | |
| "acc_stderr": 0.010456580706141529 | |
| }, | |
| "lighteval|bigbench:tracking_shuffled_objects_seven_objects|0": { | |
| "acc": 0.112, | |
| "acc_stderr": 0.0075408549910848345 | |
| }, | |
| "lighteval|bigbench:tracking_shuffled_objects_three_objects|0": { | |
| "acc": 0.43333333333333335, | |
| "acc_stderr": 0.028657565120703117 | |
| }, | |
| "lighteval|bigbench:_average|0": { | |
| "acc": 0.3432631711417832, | |
| "acc_stderr": 0.01984671488367176 | |
| }, | |
| "all": { | |
| "acc": 0.3432631711417832, | |
| "acc_stderr": 0.01984671488367176 | |
| } | |
| }, | |
| "versions": { | |
| "lighteval|bigbench:causal_judgment|0": 0, | |
| "lighteval|bigbench:date_understanding|0": 0, | |
| "lighteval|bigbench:disambiguation_qa|0": 0, | |
| "lighteval|bigbench:geometric_shapes|0": 0, | |
| "lighteval|bigbench:logical_deduction_five_objects|0": 0, | |
| "lighteval|bigbench:logical_deduction_seven_objects|0": 0, | |
| "lighteval|bigbench:logical_deduction_three_objects|0": 0, | |
| "lighteval|bigbench:movie_recommendation|0": 0, | |
| "lighteval|bigbench:navigate|0": 0, | |
| "lighteval|bigbench:reasoning_about_colored_objects|0": 0, | |
| "lighteval|bigbench:ruin_names|0": 0, | |
| "lighteval|bigbench:salient_translation_error_detection|0": 0, | |
| "lighteval|bigbench:snarks|0": 0, | |
| "lighteval|bigbench:sports_understanding|0": 0, | |
| "lighteval|bigbench:temporal_sequences|0": 0, | |
| "lighteval|bigbench:tracking_shuffled_objects_five_objects|0": 0, | |
| "lighteval|bigbench:tracking_shuffled_objects_seven_objects|0": 0, | |
| "lighteval|bigbench:tracking_shuffled_objects_three_objects|0": 0 | |
| }, | |
| "config_tasks": { | |
| "lighteval|bigbench:causal_judgment": { | |
| "name": "bigbench:causal_judgment", | |
| "prompt_function": "bbh_lighteval", | |
| "hf_repo": "lighteval/bbh", | |
| "hf_subset": "causal_judgement", | |
| "metric": [ | |
| "loglikelihood_acc_single_token" | |
| ], | |
| "hf_avail_splits": [ | |
| "train" | |
| ], | |
| "evaluation_splits": [ | |
| "train" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": null, | |
| "generation_size": -1, | |
| "stop_sequence": [ | |
| "</s>", | |
| "Q:", | |
| "\n\n" | |
| ], | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "lighteval" | |
| ], | |
| "original_num_docs": 190, | |
| "effective_num_docs": 190, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "lighteval|bigbench:date_understanding": { | |
| "name": "bigbench:date_understanding", | |
| "prompt_function": "bbh_lighteval", | |
| "hf_repo": "lighteval/bbh", | |
| "hf_subset": "date_understanding", | |
| "metric": [ | |
| "loglikelihood_acc_single_token" | |
| ], | |
| "hf_avail_splits": [ | |
| "train" | |
| ], | |
| "evaluation_splits": [ | |
| "train" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": null, | |
| "generation_size": -1, | |
| "stop_sequence": [ | |
| "</s>", | |
| "Q:", | |
| "\n\n" | |
| ], | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "lighteval" | |
| ], | |
| "original_num_docs": 369, | |
| "effective_num_docs": 369, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "lighteval|bigbench:disambiguation_qa": { | |
| "name": "bigbench:disambiguation_qa", | |
| "prompt_function": "bbh_lighteval", | |
| "hf_repo": "lighteval/bbh", | |
| "hf_subset": "disambiguation_qa", | |
| "metric": [ | |
| "loglikelihood_acc_single_token" | |
| ], | |
| "hf_avail_splits": [ | |
| "train" | |
| ], | |
| "evaluation_splits": [ | |
| "train" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": null, | |
| "generation_size": -1, | |
| "stop_sequence": [ | |
| "</s>", | |
| "Q:", | |
| "\n\n" | |
| ], | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "lighteval" | |
| ], | |
| "original_num_docs": 258, | |
| "effective_num_docs": 258, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "lighteval|bigbench:geometric_shapes": { | |
| "name": "bigbench:geometric_shapes", | |
| "prompt_function": "bbh_lighteval", | |
| "hf_repo": "lighteval/bbh", | |
| "hf_subset": "geometric_shapes", | |
| "metric": [ | |
| "loglikelihood_acc_single_token" | |
| ], | |
| "hf_avail_splits": [ | |
| "train" | |
| ], | |
| "evaluation_splits": [ | |
| "train" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": null, | |
| "generation_size": -1, | |
| "stop_sequence": [ | |
| "</s>", | |
| "Q:", | |
| "\n\n" | |
| ], | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "lighteval" | |
| ], | |
| "original_num_docs": 360, | |
| "effective_num_docs": 360, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "lighteval|bigbench:logical_deduction_five_objects": { | |
| "name": "bigbench:logical_deduction_five_objects", | |
| "prompt_function": "bbh_lighteval", | |
| "hf_repo": "lighteval/bbh", | |
| "hf_subset": "logical_deduction_five_objects", | |
| "metric": [ | |
| "loglikelihood_acc_single_token" | |
| ], | |
| "hf_avail_splits": [ | |
| "train" | |
| ], | |
| "evaluation_splits": [ | |
| "train" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": null, | |
| "generation_size": -1, | |
| "stop_sequence": [ | |
| "</s>", | |
| "Q:", | |
| "\n\n" | |
| ], | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "lighteval" | |
| ], | |
| "original_num_docs": 500, | |
| "effective_num_docs": 500, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "lighteval|bigbench:logical_deduction_seven_objects": { | |
| "name": "bigbench:logical_deduction_seven_objects", | |
| "prompt_function": "bbh_lighteval", | |
| "hf_repo": "lighteval/bbh", | |
| "hf_subset": "logical_deduction_seven_objects", | |
| "metric": [ | |
| "loglikelihood_acc_single_token" | |
| ], | |
| "hf_avail_splits": [ | |
| "train" | |
| ], | |
| "evaluation_splits": [ | |
| "train" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": null, | |
| "generation_size": -1, | |
| "stop_sequence": [ | |
| "</s>", | |
| "Q:", | |
| "\n\n" | |
| ], | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "lighteval" | |
| ], | |
| "original_num_docs": 700, | |
| "effective_num_docs": 700, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "lighteval|bigbench:logical_deduction_three_objects": { | |
| "name": "bigbench:logical_deduction_three_objects", | |
| "prompt_function": "bbh_lighteval", | |
| "hf_repo": "lighteval/bbh", | |
| "hf_subset": "logical_deduction_three_objects", | |
| "metric": [ | |
| "loglikelihood_acc_single_token" | |
| ], | |
| "hf_avail_splits": [ | |
| "train" | |
| ], | |
| "evaluation_splits": [ | |
| "train" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": null, | |
| "generation_size": -1, | |
| "stop_sequence": [ | |
| "</s>", | |
| "Q:", | |
| "\n\n" | |
| ], | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "lighteval" | |
| ], | |
| "original_num_docs": 300, | |
| "effective_num_docs": 300, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "lighteval|bigbench:movie_recommendation": { | |
| "name": "bigbench:movie_recommendation", | |
| "prompt_function": "bbh_lighteval", | |
| "hf_repo": "lighteval/bbh", | |
| "hf_subset": "movie_recommendation", | |
| "metric": [ | |
| "loglikelihood_acc_single_token" | |
| ], | |
| "hf_avail_splits": [ | |
| "train" | |
| ], | |
| "evaluation_splits": [ | |
| "train" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": null, | |
| "generation_size": -1, | |
| "stop_sequence": [ | |
| "</s>", | |
| "Q:", | |
| "\n\n" | |
| ], | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "lighteval" | |
| ], | |
| "original_num_docs": 500, | |
| "effective_num_docs": 500, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "lighteval|bigbench:navigate": { | |
| "name": "bigbench:navigate", | |
| "prompt_function": "bbh_lighteval", | |
| "hf_repo": "lighteval/bbh", | |
| "hf_subset": "navigate", | |
| "metric": [ | |
| "loglikelihood_acc_single_token" | |
| ], | |
| "hf_avail_splits": [ | |
| "train" | |
| ], | |
| "evaluation_splits": [ | |
| "train" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": null, | |
| "generation_size": -1, | |
| "stop_sequence": [ | |
| "</s>", | |
| "Q:", | |
| "\n\n" | |
| ], | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "lighteval" | |
| ], | |
| "original_num_docs": 1000, | |
| "effective_num_docs": 1000, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "lighteval|bigbench:reasoning_about_colored_objects": { | |
| "name": "bigbench:reasoning_about_colored_objects", | |
| "prompt_function": "bbh_lighteval", | |
| "hf_repo": "lighteval/bbh", | |
| "hf_subset": "reasoning_about_colored_objects", | |
| "metric": [ | |
| "loglikelihood_acc_single_token" | |
| ], | |
| "hf_avail_splits": [ | |
| "train" | |
| ], | |
| "evaluation_splits": [ | |
| "train" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": null, | |
| "generation_size": -1, | |
| "stop_sequence": [ | |
| "</s>", | |
| "Q:", | |
| "\n\n" | |
| ], | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "lighteval" | |
| ], | |
| "original_num_docs": 2000, | |
| "effective_num_docs": 2000, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "lighteval|bigbench:ruin_names": { | |
| "name": "bigbench:ruin_names", | |
| "prompt_function": "bbh_lighteval", | |
| "hf_repo": "lighteval/bbh", | |
| "hf_subset": "ruin_names", | |
| "metric": [ | |
| "loglikelihood_acc_single_token" | |
| ], | |
| "hf_avail_splits": [ | |
| "train" | |
| ], | |
| "evaluation_splits": [ | |
| "train" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": null, | |
| "generation_size": -1, | |
| "stop_sequence": [ | |
| "</s>", | |
| "Q:", | |
| "\n\n" | |
| ], | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "lighteval" | |
| ], | |
| "original_num_docs": 448, | |
| "effective_num_docs": 448, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "lighteval|bigbench:salient_translation_error_detection": { | |
| "name": "bigbench:salient_translation_error_detection", | |
| "prompt_function": "bbh_lighteval", | |
| "hf_repo": "lighteval/bbh", | |
| "hf_subset": "salient_translation_error_detection", | |
| "metric": [ | |
| "loglikelihood_acc_single_token" | |
| ], | |
| "hf_avail_splits": [ | |
| "train" | |
| ], | |
| "evaluation_splits": [ | |
| "train" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": null, | |
| "generation_size": -1, | |
| "stop_sequence": [ | |
| "</s>", | |
| "Q:", | |
| "\n\n" | |
| ], | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "lighteval" | |
| ], | |
| "original_num_docs": 998, | |
| "effective_num_docs": 998, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "lighteval|bigbench:snarks": { | |
| "name": "bigbench:snarks", | |
| "prompt_function": "bbh_lighteval", | |
| "hf_repo": "lighteval/bbh", | |
| "hf_subset": "snarks", | |
| "metric": [ | |
| "loglikelihood_acc_single_token" | |
| ], | |
| "hf_avail_splits": [ | |
| "train" | |
| ], | |
| "evaluation_splits": [ | |
| "train" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": null, | |
| "generation_size": -1, | |
| "stop_sequence": [ | |
| "</s>", | |
| "Q:", | |
| "\n\n" | |
| ], | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "lighteval" | |
| ], | |
| "original_num_docs": 181, | |
| "effective_num_docs": 181, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "lighteval|bigbench:sports_understanding": { | |
| "name": "bigbench:sports_understanding", | |
| "prompt_function": "bbh_lighteval", | |
| "hf_repo": "lighteval/bbh", | |
| "hf_subset": "sports_understanding", | |
| "metric": [ | |
| "loglikelihood_acc_single_token" | |
| ], | |
| "hf_avail_splits": [ | |
| "train" | |
| ], | |
| "evaluation_splits": [ | |
| "train" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": null, | |
| "generation_size": -1, | |
| "stop_sequence": [ | |
| "</s>", | |
| "Q:", | |
| "\n\n" | |
| ], | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "lighteval" | |
| ], | |
| "original_num_docs": 1000, | |
| "effective_num_docs": 1000, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "lighteval|bigbench:temporal_sequences": { | |
| "name": "bigbench:temporal_sequences", | |
| "prompt_function": "bbh_lighteval", | |
| "hf_repo": "lighteval/bbh", | |
| "hf_subset": "temporal_sequences", | |
| "metric": [ | |
| "loglikelihood_acc_single_token" | |
| ], | |
| "hf_avail_splits": [ | |
| "train" | |
| ], | |
| "evaluation_splits": [ | |
| "train" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": null, | |
| "generation_size": -1, | |
| "stop_sequence": [ | |
| "</s>", | |
| "Q:", | |
| "\n\n" | |
| ], | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "lighteval" | |
| ], | |
| "original_num_docs": 1000, | |
| "effective_num_docs": 1000, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "lighteval|bigbench:tracking_shuffled_objects_five_objects": { | |
| "name": "bigbench:tracking_shuffled_objects_five_objects", | |
| "prompt_function": "bbh_lighteval", | |
| "hf_repo": "lighteval/bbh", | |
| "hf_subset": "tracking_shuffled_objects_five_objects", | |
| "metric": [ | |
| "loglikelihood_acc_single_token" | |
| ], | |
| "hf_avail_splits": [ | |
| "train" | |
| ], | |
| "evaluation_splits": [ | |
| "train" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": null, | |
| "generation_size": -1, | |
| "stop_sequence": [ | |
| "</s>", | |
| "Q:", | |
| "\n\n" | |
| ], | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "lighteval" | |
| ], | |
| "original_num_docs": 1250, | |
| "effective_num_docs": 1250, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "lighteval|bigbench:tracking_shuffled_objects_seven_objects": { | |
| "name": "bigbench:tracking_shuffled_objects_seven_objects", | |
| "prompt_function": "bbh_lighteval", | |
| "hf_repo": "lighteval/bbh", | |
| "hf_subset": "tracking_shuffled_objects_seven_objects", | |
| "metric": [ | |
| "loglikelihood_acc_single_token" | |
| ], | |
| "hf_avail_splits": [ | |
| "train" | |
| ], | |
| "evaluation_splits": [ | |
| "train" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": null, | |
| "generation_size": -1, | |
| "stop_sequence": [ | |
| "</s>", | |
| "Q:", | |
| "\n\n" | |
| ], | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "lighteval" | |
| ], | |
| "original_num_docs": 1750, | |
| "effective_num_docs": 1750, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "lighteval|bigbench:tracking_shuffled_objects_three_objects": { | |
| "name": "bigbench:tracking_shuffled_objects_three_objects", | |
| "prompt_function": "bbh_lighteval", | |
| "hf_repo": "lighteval/bbh", | |
| "hf_subset": "tracking_shuffled_objects_three_objects", | |
| "metric": [ | |
| "loglikelihood_acc_single_token" | |
| ], | |
| "hf_avail_splits": [ | |
| "train" | |
| ], | |
| "evaluation_splits": [ | |
| "train" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": null, | |
| "generation_size": -1, | |
| "stop_sequence": [ | |
| "</s>", | |
| "Q:", | |
| "\n\n" | |
| ], | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "lighteval" | |
| ], | |
| "original_num_docs": 300, | |
| "effective_num_docs": 300, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| } | |
| }, | |
| "summary_tasks": { | |
| "lighteval|bigbench:causal_judgment|0": { | |
| "hashes": { | |
| "hash_examples": "dfb1ae47218f2850", | |
| "hash_full_prompts": "92b1cf75ca896127", | |
| "hash_input_tokens": "3fbebfde3354c6ac", | |
| "hash_cont_tokens": "ac670c3ea513a639" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 190, | |
| "padded": 189, | |
| "non_padded": 1, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|bigbench:date_understanding|0": { | |
| "hashes": { | |
| "hash_examples": "2b823c41500a6ec2", | |
| "hash_full_prompts": "a086589baadb24a5", | |
| "hash_input_tokens": "262880ca53810c51", | |
| "hash_cont_tokens": "e7711b87d7f90d38" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 369, | |
| "padded": 369, | |
| "non_padded": 0, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|bigbench:disambiguation_qa|0": { | |
| "hashes": { | |
| "hash_examples": "2a4c3d41db198cea", | |
| "hash_full_prompts": "407f0b9a565699a7", | |
| "hash_input_tokens": "7159e757a625c662", | |
| "hash_cont_tokens": "de89f8a6e5dac00c" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 258, | |
| "padded": 258, | |
| "non_padded": 0, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|bigbench:geometric_shapes|0": { | |
| "hashes": { | |
| "hash_examples": "24aa261103911b72", | |
| "hash_full_prompts": "c7a3189ee1642ab5", | |
| "hash_input_tokens": "2bbd71a22a0967ff", | |
| "hash_cont_tokens": "e51eec73c3eb26c9" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 360, | |
| "padded": 360, | |
| "non_padded": 0, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|bigbench:logical_deduction_five_objects|0": { | |
| "hashes": { | |
| "hash_examples": "cb5bdc92afc41f83", | |
| "hash_full_prompts": "199211ceee2c8d60", | |
| "hash_input_tokens": "0c67894c26b553e1", | |
| "hash_cont_tokens": "4c9e9d2d14981c58" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 500, | |
| "padded": 500, | |
| "non_padded": 0, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|bigbench:logical_deduction_seven_objects|0": { | |
| "hashes": { | |
| "hash_examples": "b6805ea696739f9f", | |
| "hash_full_prompts": "38f66304971bdfb0", | |
| "hash_input_tokens": "fe8815876ceefc9f", | |
| "hash_cont_tokens": "1745fa6fd92f0e0d" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 700, | |
| "padded": 700, | |
| "non_padded": 0, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|bigbench:logical_deduction_three_objects|0": { | |
| "hashes": { | |
| "hash_examples": "0509e5712ab9bcdb", | |
| "hash_full_prompts": "0a9439e283842405", | |
| "hash_input_tokens": "bdaf9a4ad6d5308d", | |
| "hash_cont_tokens": "2b5b679169d7bcf1" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 300, | |
| "padded": 300, | |
| "non_padded": 0, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|bigbench:movie_recommendation|0": { | |
| "hashes": { | |
| "hash_examples": "530cc6f737830f45", | |
| "hash_full_prompts": "e7d59f843d80e6ba", | |
| "hash_input_tokens": "d107040c803cc28a", | |
| "hash_cont_tokens": "be520838bf2427bc" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 500, | |
| "padded": 500, | |
| "non_padded": 0, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|bigbench:navigate|0": { | |
| "hashes": { | |
| "hash_examples": "7962ef85d0058b9a", | |
| "hash_full_prompts": "d58b607419984968", | |
| "hash_input_tokens": "f163e125b820f371", | |
| "hash_cont_tokens": "04e3a57b821a3dd8" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 1000, | |
| "padded": 988, | |
| "non_padded": 12, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|bigbench:reasoning_about_colored_objects|0": { | |
| "hashes": { | |
| "hash_examples": "39be1ab1677a651d", | |
| "hash_full_prompts": "157347bb0e7fe3f1", | |
| "hash_input_tokens": "562dab78f82dce11", | |
| "hash_cont_tokens": "3fe982d2154a001a" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 2000, | |
| "padded": 1969, | |
| "non_padded": 31, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|bigbench:ruin_names|0": { | |
| "hashes": { | |
| "hash_examples": "e9b96b31d2154941", | |
| "hash_full_prompts": "9cb73d2fcaf5ee1e", | |
| "hash_input_tokens": "07268ad2915fab08", | |
| "hash_cont_tokens": "046bbbbddb05b429" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 448, | |
| "padded": 442, | |
| "non_padded": 6, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|bigbench:salient_translation_error_detection|0": { | |
| "hashes": { | |
| "hash_examples": "951ac59f7ad0427d", | |
| "hash_full_prompts": "3598ce066bb83298", | |
| "hash_input_tokens": "d514272792c1a43b", | |
| "hash_cont_tokens": "e78fb6d09071e0f6" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 998, | |
| "padded": 998, | |
| "non_padded": 0, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|bigbench:snarks|0": { | |
| "hashes": { | |
| "hash_examples": "3a53eb9b9d758534", | |
| "hash_full_prompts": "fb6c17d84dd479d6", | |
| "hash_input_tokens": "a714129b2023dfb4", | |
| "hash_cont_tokens": "f5cb71a436613293" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 181, | |
| "padded": 178, | |
| "non_padded": 3, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|bigbench:sports_understanding|0": { | |
| "hashes": { | |
| "hash_examples": "bd65741f00770373", | |
| "hash_full_prompts": "467a508a87ae3ce4", | |
| "hash_input_tokens": "c1c3655837258597", | |
| "hash_cont_tokens": "02230fac16464d15" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 1000, | |
| "padded": 1000, | |
| "non_padded": 0, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|bigbench:temporal_sequences|0": { | |
| "hashes": { | |
| "hash_examples": "1d13139f47cb2df7", | |
| "hash_full_prompts": "c1b66e439c72477b", | |
| "hash_input_tokens": "d47a810e1658bcc6", | |
| "hash_cont_tokens": "88c86d8bfb960c7d" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 1000, | |
| "padded": 1000, | |
| "non_padded": 0, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|bigbench:tracking_shuffled_objects_five_objects|0": { | |
| "hashes": { | |
| "hash_examples": "8770a702a9646648", | |
| "hash_full_prompts": "8c353eaa84f712ff", | |
| "hash_input_tokens": "e940ffaad712c055", | |
| "hash_cont_tokens": "7cf11d867348e0b1" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 1250, | |
| "padded": 1180, | |
| "non_padded": 70, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|bigbench:tracking_shuffled_objects_seven_objects|0": { | |
| "hashes": { | |
| "hash_examples": "b469b7d073824a59", | |
| "hash_full_prompts": "902f5b74467353eb", | |
| "hash_input_tokens": "9d52e492d2f6cd34", | |
| "hash_cont_tokens": "f76ba63a583d749e" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 1750, | |
| "padded": 1701, | |
| "non_padded": 49, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|bigbench:tracking_shuffled_objects_three_objects|0": { | |
| "hashes": { | |
| "hash_examples": "0509e5712ab9bcdb", | |
| "hash_full_prompts": "0a9439e283842405", | |
| "hash_input_tokens": "5867c63542b2b703", | |
| "hash_cont_tokens": "b2cce0a4a2edc859" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 300, | |
| "padded": 294, | |
| "non_padded": 6, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| } | |
| }, | |
| "summary_general": { | |
| "hashes": { | |
| "hash_examples": "51a30c4501ba4586", | |
| "hash_full_prompts": "2f8425bef1f1b307", | |
| "hash_input_tokens": "b84ccc3699968295", | |
| "hash_cont_tokens": "2f0ff7c19ccc0d8e" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 13104, | |
| "padded": 12926, | |
| "non_padded": 178, | |
| "num_truncated_few_shots": 0 | |
| } | |
| } |