open-r1-eval-leaderboard
/
eval_results
/meta-llama
/Llama-2-70b-chat-hf
/main
/bbh
/results_2024-03-28T16-59-33.292407.json
| { | |
| "config_general": { | |
| "lighteval_sha": "?", | |
| "num_fewshot_seeds": 1, | |
| "override_batch_size": 1, | |
| "max_samples": null, | |
| "job_id": "", | |
| "start_time": 2079825.461451948, | |
| "end_time": 2080892.035621395, | |
| "total_evaluation_time_secondes": "1066.5741694471799", | |
| "model_name": "meta-llama/Llama-2-70b-chat-hf", | |
| "model_sha": "8b17e6f4e86be78cf54afd49ddb517d4e274c13f", | |
| "model_dtype": "torch.bfloat16", | |
| "model_size": "128.64 GB", | |
| "config": null | |
| }, | |
| "results": { | |
| "lighteval|bigbench:causal_judgment|0": { | |
| "acc": 0.5421052631578948, | |
| "acc_stderr": 0.03624046284425958 | |
| }, | |
| "lighteval|bigbench:date_understanding|0": { | |
| "acc": 0.024390243902439025, | |
| "acc_stderr": 0.008041225285314473 | |
| }, | |
| "lighteval|bigbench:disambiguation_qa|0": { | |
| "acc": 0.47674418604651164, | |
| "acc_stderr": 0.031155388583850818 | |
| }, | |
| "lighteval|bigbench:geometric_shapes|0": { | |
| "acc": 0.1, | |
| "acc_stderr": 0.015833394419077815 | |
| }, | |
| "lighteval|bigbench:logical_deduction_five_objects|0": { | |
| "acc": 0.254, | |
| "acc_stderr": 0.01948659680164338 | |
| }, | |
| "lighteval|bigbench:logical_deduction_seven_objects|0": { | |
| "acc": 0.15142857142857144, | |
| "acc_stderr": 0.013558433870404417 | |
| }, | |
| "lighteval|bigbench:logical_deduction_three_objects|0": { | |
| "acc": 0.4033333333333333, | |
| "acc_stderr": 0.028370197016959944 | |
| }, | |
| "lighteval|bigbench:movie_recommendation|0": { | |
| "acc": 0.358, | |
| "acc_stderr": 0.021461434862859122 | |
| }, | |
| "lighteval|bigbench:navigate|0": { | |
| "acc": 0.5, | |
| "acc_stderr": 0.015819299929208316 | |
| }, | |
| "lighteval|bigbench:reasoning_about_colored_objects|0": { | |
| "acc": 0.19, | |
| "acc_stderr": 0.008774308761784201 | |
| }, | |
| "lighteval|bigbench:ruin_names|0": { | |
| "acc": 0.2700892857142857, | |
| "acc_stderr": 0.021000749078822385 | |
| }, | |
| "lighteval|bigbench:salient_translation_error_detection|0": { | |
| "acc": 0.18537074148296592, | |
| "acc_stderr": 0.012307023122568653 | |
| }, | |
| "lighteval|bigbench:snarks|0": { | |
| "acc": 0.5359116022099447, | |
| "acc_stderr": 0.037171551011396645 | |
| }, | |
| "lighteval|bigbench:sports_understanding|0": { | |
| "acc": 0.68, | |
| "acc_stderr": 0.014758652303574886 | |
| }, | |
| "lighteval|bigbench:temporal_sequences|0": { | |
| "acc": 0.049, | |
| "acc_stderr": 0.006829761756140931 | |
| }, | |
| "lighteval|bigbench:tracking_shuffled_objects_five_objects|0": { | |
| "acc": 0.2008, | |
| "acc_stderr": 0.011335178737965248 | |
| }, | |
| "lighteval|bigbench:tracking_shuffled_objects_seven_objects|0": { | |
| "acc": 0.14114285714285715, | |
| "acc_stderr": 0.00832520647820533 | |
| }, | |
| "lighteval|bigbench:tracking_shuffled_objects_three_objects|0": { | |
| "acc": 0.4033333333333333, | |
| "acc_stderr": 0.028370197016959944 | |
| }, | |
| "lighteval|bigbench:_average|0": { | |
| "acc": 0.30364718987511874, | |
| "acc_stderr": 0.018824392326722 | |
| }, | |
| "all": { | |
| "acc": 0.30364718987511874, | |
| "acc_stderr": 0.018824392326722 | |
| } | |
| }, | |
| "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": "4c7f2f76409f70f4", | |
| "hash_input_tokens": "dfde6ce631994bcf", | |
| "hash_cont_tokens": "b568f0838688d45f" | |
| }, | |
| "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": "e83dd502c036bded", | |
| "hash_input_tokens": "9464bd17bbfbf52f", | |
| "hash_cont_tokens": "148a2d93f0a3fed7" | |
| }, | |
| "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": "9cf3c0f75fc49921", | |
| "hash_input_tokens": "bffd456de920d0b5", | |
| "hash_cont_tokens": "0bb9ee243095a267" | |
| }, | |
| "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": "862e0cf9b72a95a4", | |
| "hash_input_tokens": "9f7b4f4296011f1e", | |
| "hash_cont_tokens": "f687789618cc5b4e" | |
| }, | |
| "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": "238b886ee1e7b36b", | |
| "hash_input_tokens": "582759660fad6bb0", | |
| "hash_cont_tokens": "cc346802490e368d" | |
| }, | |
| "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": "ec2aefb5b9986a54", | |
| "hash_input_tokens": "2548836d56fce744", | |
| "hash_cont_tokens": "2a94547c7122ac8a" | |
| }, | |
| "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": "f369763a9ab136d8", | |
| "hash_input_tokens": "97e768427d717e33", | |
| "hash_cont_tokens": "0810414105730f5a" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 300, | |
| "padded": 283, | |
| "non_padded": 17, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|bigbench:movie_recommendation|0": { | |
| "hashes": { | |
| "hash_examples": "530cc6f737830f45", | |
| "hash_full_prompts": "7aee02c6af822316", | |
| "hash_input_tokens": "ff6f5e54dccb748a", | |
| "hash_cont_tokens": "39b78764b1f62817" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 500, | |
| "padded": 496, | |
| "non_padded": 4, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|bigbench:navigate|0": { | |
| "hashes": { | |
| "hash_examples": "7962ef85d0058b9a", | |
| "hash_full_prompts": "b6eff183ee6f8e88", | |
| "hash_input_tokens": "6e6625533763e732", | |
| "hash_cont_tokens": "929d37605ce0219d" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 1000, | |
| "padded": 994, | |
| "non_padded": 6, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|bigbench:reasoning_about_colored_objects|0": { | |
| "hashes": { | |
| "hash_examples": "39be1ab1677a651d", | |
| "hash_full_prompts": "897f7f976f40ec00", | |
| "hash_input_tokens": "8ccf9a502879e6a4", | |
| "hash_cont_tokens": "60f23832d085ccc8" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 2000, | |
| "padded": 2000, | |
| "non_padded": 0, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|bigbench:ruin_names|0": { | |
| "hashes": { | |
| "hash_examples": "e9b96b31d2154941", | |
| "hash_full_prompts": "cad13f7801d6f123", | |
| "hash_input_tokens": "18ee7a859c4683fe", | |
| "hash_cont_tokens": "3a14d53f73c7b38b" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 448, | |
| "padded": 448, | |
| "non_padded": 0, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|bigbench:salient_translation_error_detection|0": { | |
| "hashes": { | |
| "hash_examples": "951ac59f7ad0427d", | |
| "hash_full_prompts": "7c45b95866240537", | |
| "hash_input_tokens": "c77c5cccd00f938c", | |
| "hash_cont_tokens": "1195a841bd9abbcc" | |
| }, | |
| "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": "713ab9a15534316c", | |
| "hash_input_tokens": "dd3e22f508de2312", | |
| "hash_cont_tokens": "c1ec66334133c92a" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 181, | |
| "padded": 176, | |
| "non_padded": 5, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|bigbench:sports_understanding|0": { | |
| "hashes": { | |
| "hash_examples": "bd65741f00770373", | |
| "hash_full_prompts": "241b2cf353df5ecf", | |
| "hash_input_tokens": "4191638b06174ffa", | |
| "hash_cont_tokens": "96ca07f613d55c35" | |
| }, | |
| "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": "729db25fed2f940e", | |
| "hash_input_tokens": "58c2ebacc1a03ca0", | |
| "hash_cont_tokens": "fd792f95bc07e152" | |
| }, | |
| "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": "74fbca2d47f8529e", | |
| "hash_input_tokens": "c3596088a6d4bb85", | |
| "hash_cont_tokens": "823f94ba98681e0e" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 1250, | |
| "padded": 1233, | |
| "non_padded": 17, | |
| "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": "bb13b4b0abdc1636", | |
| "hash_input_tokens": "b586533e6a04f73d", | |
| "hash_cont_tokens": "76f29108032b2ebb" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 1750, | |
| "padded": 1729, | |
| "non_padded": 21, | |
| "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": "f369763a9ab136d8", | |
| "hash_input_tokens": "19f2be77920da3b7", | |
| "hash_cont_tokens": "faf55b66443be9d9" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 300, | |
| "padded": 282, | |
| "non_padded": 18, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| } | |
| }, | |
| "summary_general": { | |
| "hashes": { | |
| "hash_examples": "51a30c4501ba4586", | |
| "hash_full_prompts": "a5ca2bdc97de22da", | |
| "hash_input_tokens": "52cc4a335d912c7a", | |
| "hash_cont_tokens": "b7c4466916e24dad" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 13104, | |
| "padded": 13015, | |
| "non_padded": 89, | |
| "num_truncated_few_shots": 0 | |
| } | |
| } |