open-r1-eval-leaderboard
/
eval_results
/HuggingFaceH4
/starcoder2-15b-dpo
/v4.1
/bbh
/results_2024-03-28T16-46-39.712895.json
| { | |
| "config_general": { | |
| "lighteval_sha": "?", | |
| "num_fewshot_seeds": 1, | |
| "override_batch_size": 1, | |
| "max_samples": null, | |
| "job_id": "", | |
| "start_time": 549736.407698777, | |
| "end_time": 550052.573380215, | |
| "total_evaluation_time_secondes": "316.1656814379385", | |
| "model_name": "HuggingFaceH4/starcoder2-15b-dpo", | |
| "model_sha": "3fea85086795d1c7043849eb550e631fbf8426f4", | |
| "model_dtype": "torch.bfloat16", | |
| "model_size": "30.04 GB", | |
| "config": null | |
| }, | |
| "results": { | |
| "lighteval|bigbench:causal_judgment|0": { | |
| "acc": 0.5473684210526316, | |
| "acc_stderr": 0.03620607045823047 | |
| }, | |
| "lighteval|bigbench:date_understanding|0": { | |
| "acc": 0.4173441734417344, | |
| "acc_stderr": 0.025705692903559226 | |
| }, | |
| "lighteval|bigbench:disambiguation_qa|0": { | |
| "acc": 0.4263565891472868, | |
| "acc_stderr": 0.030848988785650445 | |
| }, | |
| "lighteval|bigbench:geometric_shapes|0": { | |
| "acc": 0.21944444444444444, | |
| "acc_stderr": 0.021843236126455383 | |
| }, | |
| "lighteval|bigbench:logical_deduction_five_objects|0": { | |
| "acc": 0.322, | |
| "acc_stderr": 0.020916668330019882 | |
| }, | |
| "lighteval|bigbench:logical_deduction_seven_objects|0": { | |
| "acc": 0.28285714285714286, | |
| "acc_stderr": 0.017035223107187112 | |
| }, | |
| "lighteval|bigbench:logical_deduction_three_objects|0": { | |
| "acc": 0.36666666666666664, | |
| "acc_stderr": 0.027868673283383914 | |
| }, | |
| "lighteval|bigbench:movie_recommendation|0": { | |
| "acc": 0.346, | |
| "acc_stderr": 0.02129495127723464 | |
| }, | |
| "lighteval|bigbench:navigate|0": { | |
| "acc": 0.597, | |
| "acc_stderr": 0.015518757419066533 | |
| }, | |
| "lighteval|bigbench:reasoning_about_colored_objects|0": { | |
| "acc": 0.515, | |
| "acc_stderr": 0.011178102477052811 | |
| }, | |
| "lighteval|bigbench:ruin_names|0": { | |
| "acc": 0.20982142857142858, | |
| "acc_stderr": 0.01925900217665581 | |
| }, | |
| "lighteval|bigbench:salient_translation_error_detection|0": { | |
| "acc": 0.48897795591182364, | |
| "acc_stderr": 0.01583131092997039 | |
| }, | |
| "lighteval|bigbench:snarks|0": { | |
| "acc": 0.4972375690607735, | |
| "acc_stderr": 0.03726723083765757 | |
| }, | |
| "lighteval|bigbench:sports_understanding|0": { | |
| "acc": 0.546, | |
| "acc_stderr": 0.01575221038877184 | |
| }, | |
| "lighteval|bigbench:temporal_sequences|0": { | |
| "acc": 0.417, | |
| "acc_stderr": 0.015599819048769618 | |
| }, | |
| "lighteval|bigbench:tracking_shuffled_objects_five_objects|0": { | |
| "acc": 0.1664, | |
| "acc_stderr": 0.010538390680863754 | |
| }, | |
| "lighteval|bigbench:tracking_shuffled_objects_seven_objects|0": { | |
| "acc": 0.084, | |
| "acc_stderr": 0.006632732388430605 | |
| }, | |
| "lighteval|bigbench:tracking_shuffled_objects_three_objects|0": { | |
| "acc": 0.36666666666666664, | |
| "acc_stderr": 0.027868673283383914 | |
| }, | |
| "lighteval|bigbench:_average|0": { | |
| "acc": 0.37867450321225554, | |
| "acc_stderr": 0.02095365188346355 | |
| }, | |
| "all": { | |
| "acc": 0.37867450321225554, | |
| "acc_stderr": 0.02095365188346355 | |
| } | |
| }, | |
| "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": "7292c47f5bf2ba48", | |
| "hash_input_tokens": "25626d06e33da51e", | |
| "hash_cont_tokens": "9048fca902c56e3b" | |
| }, | |
| "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": "4db646afa4176c07", | |
| "hash_input_tokens": "feb040823960f4e5", | |
| "hash_cont_tokens": "7a23294cc54d4601" | |
| }, | |
| "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": "12d668cf5edc9542", | |
| "hash_input_tokens": "666cc2ec82589591", | |
| "hash_cont_tokens": "5045873fe712fa7b" | |
| }, | |
| "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": "51dfb12a121e7a69", | |
| "hash_input_tokens": "e1fceec900e245a0", | |
| "hash_cont_tokens": "06c6f05eeee15a68" | |
| }, | |
| "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": "b6e4a71663bc3e1c", | |
| "hash_input_tokens": "c17ae0cbd369459a", | |
| "hash_cont_tokens": "32aea888033ad0dc" | |
| }, | |
| "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": "d0c82c066345c294", | |
| "hash_input_tokens": "baee5f5de503a467", | |
| "hash_cont_tokens": "ba5a8d269b45e12a" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 700, | |
| "padded": 699, | |
| "non_padded": 1, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|bigbench:logical_deduction_three_objects|0": { | |
| "hashes": { | |
| "hash_examples": "0509e5712ab9bcdb", | |
| "hash_full_prompts": "396c1e56901b46ed", | |
| "hash_input_tokens": "d9429939dfdae94b", | |
| "hash_cont_tokens": "e3ce74c69ea4c7e3" | |
| }, | |
| "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": "e821384b2a44e36b", | |
| "hash_input_tokens": "10097f61f3af57b6", | |
| "hash_cont_tokens": "3edc9acda2c08d09" | |
| }, | |
| "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": "43248e6945903d81", | |
| "hash_input_tokens": "6f2b8316b98d62e2", | |
| "hash_cont_tokens": "0553671f47471f4a" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 1000, | |
| "padded": 1000, | |
| "non_padded": 0, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|bigbench:reasoning_about_colored_objects|0": { | |
| "hashes": { | |
| "hash_examples": "39be1ab1677a651d", | |
| "hash_full_prompts": "7f7a503aaa70068f", | |
| "hash_input_tokens": "b055191640d37f0f", | |
| "hash_cont_tokens": "776322177f1453e1" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 2000, | |
| "padded": 1990, | |
| "non_padded": 10, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|bigbench:ruin_names|0": { | |
| "hashes": { | |
| "hash_examples": "e9b96b31d2154941", | |
| "hash_full_prompts": "ae8931c806192844", | |
| "hash_input_tokens": "1185c8959f28c736", | |
| "hash_cont_tokens": "f296e5a757f45e57" | |
| }, | |
| "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": "643d82c4ce3fab01", | |
| "hash_input_tokens": "0b89fee17433e59a", | |
| "hash_cont_tokens": "58bc9bf03f6a0a43" | |
| }, | |
| "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": "b12bcea4b9bc9027", | |
| "hash_input_tokens": "3e9eda61f6cb3f9e", | |
| "hash_cont_tokens": "33ab16c8600245cd" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 181, | |
| "padded": 181, | |
| "non_padded": 0, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|bigbench:sports_understanding|0": { | |
| "hashes": { | |
| "hash_examples": "bd65741f00770373", | |
| "hash_full_prompts": "39d7688aa2d209e1", | |
| "hash_input_tokens": "8618146aeca9d12b", | |
| "hash_cont_tokens": "f893ba0bdfd798ba" | |
| }, | |
| "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": "1a874610f00343dc", | |
| "hash_input_tokens": "7260a3f4a9a8ccc4", | |
| "hash_cont_tokens": "f134565be0a3da6d" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 1000, | |
| "padded": 986, | |
| "non_padded": 14, | |
| "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": "392b486c4039dca8", | |
| "hash_input_tokens": "85e35dab42b0c7eb", | |
| "hash_cont_tokens": "28a2a00edb2e8f71" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 1250, | |
| "padded": 1250, | |
| "non_padded": 0, | |
| "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": "1bad8a693cc74da1", | |
| "hash_input_tokens": "96aaed79d152ec7a", | |
| "hash_cont_tokens": "407da553a98fc49f" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 1750, | |
| "padded": 1742, | |
| "non_padded": 8, | |
| "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": "396c1e56901b46ed", | |
| "hash_input_tokens": "118a36e811393884", | |
| "hash_cont_tokens": "78a91d437956ff62" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 300, | |
| "padded": 298, | |
| "non_padded": 2, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| } | |
| }, | |
| "summary_general": { | |
| "hashes": { | |
| "hash_examples": "51a30c4501ba4586", | |
| "hash_full_prompts": "96a511cab844bc38", | |
| "hash_input_tokens": "6f0fdd8e5937e7ad", | |
| "hash_cont_tokens": "508f88fb7741ce7b" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 13104, | |
| "padded": 13068, | |
| "non_padded": 36, | |
| "num_truncated_few_shots": 0 | |
| } | |
| } |