open-r1-eval-leaderboard
/
eval_results
/Qwen
/Qwen1.5-14B-Chat
/main
/bbh
/results_2024-03-18T20-23-46.729702.json
| { | |
| "config_general": { | |
| "lighteval_sha": "?", | |
| "num_fewshot_seeds": 1, | |
| "override_batch_size": 1, | |
| "max_samples": null, | |
| "job_id": "", | |
| "start_time": 1563676.841818657, | |
| "end_time": 1564004.190045505, | |
| "total_evaluation_time_secondes": "327.3482268480584", | |
| "model_name": "Qwen/Qwen1.5-14B-Chat", | |
| "model_sha": "17e11c306ed235e970c9bb8e5f7233527140cdcf", | |
| "model_dtype": "torch.bfloat16", | |
| "model_size": "27.01 GB", | |
| "config": null | |
| }, | |
| "results": { | |
| "harness|bbh:causal_judgment|3": { | |
| "em": 0.5882352941176471, | |
| "em_stderr": 0.03608640563085621, | |
| "qem": 0.5882352941176471, | |
| "qem_stderr": 0.03608640563085621, | |
| "pem": 0.5882352941176471, | |
| "pem_stderr": 0.03608640563085621, | |
| "pqem": 0.5882352941176471, | |
| "pqem_stderr": 0.03608640563085621, | |
| "perfect_em": 0.5882352941176471, | |
| "perfect_em_stderr": 0.03608640563085621 | |
| }, | |
| "harness|bbh:date_understanding|3": { | |
| "em": 0.46, | |
| "em_stderr": 0.031584653891499, | |
| "qem": 0.46, | |
| "qem_stderr": 0.031584653891499, | |
| "pem": 0.46, | |
| "pem_stderr": 0.031584653891499, | |
| "pqem": 0.58, | |
| "pqem_stderr": 0.03127799950463661, | |
| "perfect_em": 0.46, | |
| "perfect_em_stderr": 0.031584653891499 | |
| }, | |
| "harness|bbh:disambiguation_qa|3": { | |
| "em": 0.528, | |
| "em_stderr": 0.031636489531544396, | |
| "qem": 0.528, | |
| "qem_stderr": 0.031636489531544396, | |
| "pem": 0.62, | |
| "pem_stderr": 0.03076011604262605, | |
| "pqem": 0.788, | |
| "pqem_stderr": 0.02590188469054116, | |
| "perfect_em": 0.528, | |
| "perfect_em_stderr": 0.031636489531544396 | |
| }, | |
| "harness|bbh:geometric_shapes|3": { | |
| "em": 0.1, | |
| "em_stderr": 0.019011727515734378, | |
| "qem": 0.1, | |
| "qem_stderr": 0.019011727515734378, | |
| "pem": 0.116, | |
| "pem_stderr": 0.020293429803083865, | |
| "pqem": 0.116, | |
| "pqem_stderr": 0.020293429803083865, | |
| "perfect_em": 0.1, | |
| "perfect_em_stderr": 0.019011727515734378 | |
| }, | |
| "harness|bbh:logical_deduction_five_objects|3": { | |
| "em": 0.432, | |
| "em_stderr": 0.031391810765429407, | |
| "qem": 0.432, | |
| "qem_stderr": 0.031391810765429407, | |
| "pem": 0.448, | |
| "pem_stderr": 0.03151438761115355, | |
| "pqem": 0.572, | |
| "pqem_stderr": 0.031355968923772605, | |
| "perfect_em": 0.432, | |
| "perfect_em_stderr": 0.031391810765429407 | |
| }, | |
| "harness|bbh:logical_deduction_seven_objects|3": { | |
| "em": 0.464, | |
| "em_stderr": 0.03160397514522375, | |
| "qem": 0.464, | |
| "qem_stderr": 0.03160397514522375, | |
| "pem": 0.464, | |
| "pem_stderr": 0.03160397514522375, | |
| "pqem": 0.56, | |
| "pqem_stderr": 0.031457244522235646, | |
| "perfect_em": 0.464, | |
| "perfect_em_stderr": 0.03160397514522375 | |
| }, | |
| "harness|bbh:logical_deduction_three_objects|3": { | |
| "em": 0.588, | |
| "em_stderr": 0.031191596026022898, | |
| "qem": 0.588, | |
| "qem_stderr": 0.031191596026022898, | |
| "pem": 0.592, | |
| "pem_stderr": 0.0311452098465485, | |
| "pqem": 0.78, | |
| "pqem_stderr": 0.026251792824605824, | |
| "perfect_em": 0.588, | |
| "perfect_em_stderr": 0.031191596026022898 | |
| }, | |
| "harness|bbh:movie_recommendation|3": { | |
| "em": 0.7309236947791165, | |
| "em_stderr": 0.028160989317009937, | |
| "qem": 0.7309236947791165, | |
| "qem_stderr": 0.028160989317009937, | |
| "pem": 0.7349397590361446, | |
| "pem_stderr": 0.02802672325167473, | |
| "pqem": 0.8152610441767069, | |
| "pqem_stderr": 0.0246434636057179, | |
| "perfect_em": 0.7309236947791165, | |
| "perfect_em_stderr": 0.028160989317009937 | |
| }, | |
| "harness|bbh:navigate|3": { | |
| "em": 0.644, | |
| "em_stderr": 0.030343680657153208, | |
| "qem": 0.644, | |
| "qem_stderr": 0.030343680657153208, | |
| "pem": 0.644, | |
| "pem_stderr": 0.030343680657153208, | |
| "pqem": 0.644, | |
| "pqem_stderr": 0.030343680657153208, | |
| "perfect_em": 0.644, | |
| "perfect_em_stderr": 0.030343680657153208 | |
| }, | |
| "harness|bbh:reasoning_about_colored_objects|3": { | |
| "em": 0.164, | |
| "em_stderr": 0.02346526100207676, | |
| "qem": 0.164, | |
| "qem_stderr": 0.02346526100207676, | |
| "pem": 0.548, | |
| "pem_stderr": 0.03153986449255662, | |
| "pqem": 0.608, | |
| "pqem_stderr": 0.030938207620401195, | |
| "perfect_em": 0.164, | |
| "perfect_em_stderr": 0.02346526100207676 | |
| }, | |
| "harness|bbh:ruin_names|3": { | |
| "em": 0.16532258064516128, | |
| "em_stderr": 0.023636161221813987, | |
| "qem": 0.16532258064516128, | |
| "qem_stderr": 0.023636161221813987, | |
| "pem": 0.6088709677419355, | |
| "pem_stderr": 0.031050899338481038, | |
| "pqem": 0.717741935483871, | |
| "pqem_stderr": 0.02863906688337371, | |
| "perfect_em": 0.16532258064516128, | |
| "perfect_em_stderr": 0.023636161221813987 | |
| }, | |
| "harness|bbh:salient_translation_error_detection|3": { | |
| "em": 0.32, | |
| "em_stderr": 0.029561724955241033, | |
| "qem": 0.32, | |
| "qem_stderr": 0.029561724955241033, | |
| "pem": 0.48, | |
| "pem_stderr": 0.031660853408495185, | |
| "pqem": 0.548, | |
| "pqem_stderr": 0.03153986449255663, | |
| "perfect_em": 0.32, | |
| "perfect_em_stderr": 0.029561724955241033 | |
| }, | |
| "harness|bbh:snarks|3": { | |
| "em": 0.7696629213483146, | |
| "em_stderr": 0.03164794946543344, | |
| "qem": 0.7696629213483146, | |
| "qem_stderr": 0.03164794946543344, | |
| "pem": 0.7696629213483146, | |
| "pem_stderr": 0.03164794946543344, | |
| "pqem": 0.8539325842696629, | |
| "pqem_stderr": 0.026546167628776974, | |
| "perfect_em": 0.7696629213483146, | |
| "perfect_em_stderr": 0.03164794946543344 | |
| }, | |
| "harness|bbh:sports_understanding|3": { | |
| "em": 0.628, | |
| "em_stderr": 0.03063032594455831, | |
| "qem": 0.628, | |
| "qem_stderr": 0.03063032594455831, | |
| "pem": 0.696, | |
| "pem_stderr": 0.029150213374159673, | |
| "pqem": 0.696, | |
| "pqem_stderr": 0.029150213374159673, | |
| "perfect_em": 0.628, | |
| "perfect_em_stderr": 0.03063032594455831 | |
| }, | |
| "harness|bbh:temporal_sequences|3": { | |
| "em": 0.58, | |
| "em_stderr": 0.031277999504636625, | |
| "qem": 0.58, | |
| "qem_stderr": 0.031277999504636625, | |
| "pem": 0.58, | |
| "pem_stderr": 0.031277999504636625, | |
| "pqem": 0.732, | |
| "pqem_stderr": 0.028068762382526688, | |
| "perfect_em": 0.58, | |
| "perfect_em_stderr": 0.031277999504636625 | |
| }, | |
| "harness|bbh:tracking_shuffled_objects_five_objects|3": { | |
| "em": 0.08, | |
| "em_stderr": 0.017192507941462976, | |
| "qem": 0.08, | |
| "qem_stderr": 0.017192507941462976, | |
| "pem": 0.16, | |
| "pem_stderr": 0.023232714782060654, | |
| "pqem": 0.32, | |
| "pqem_stderr": 0.02956172495524105, | |
| "perfect_em": 0.08, | |
| "perfect_em_stderr": 0.017192507941462976 | |
| }, | |
| "harness|bbh:tracking_shuffled_objects_seven_objects|3": { | |
| "em": 0.08, | |
| "em_stderr": 0.017192507941462966, | |
| "qem": 0.08, | |
| "qem_stderr": 0.017192507941462966, | |
| "pem": 0.152, | |
| "pem_stderr": 0.02275202449176547, | |
| "pqem": 0.296, | |
| "pqem_stderr": 0.02892893938837962, | |
| "perfect_em": 0.08, | |
| "perfect_em_stderr": 0.017192507941462966 | |
| }, | |
| "harness|bbh:tracking_shuffled_objects_three_objects|3": { | |
| "em": 0.032, | |
| "em_stderr": 0.011153546809230828, | |
| "qem": 0.032, | |
| "qem_stderr": 0.011153546809230828, | |
| "pem": 0.288, | |
| "pem_stderr": 0.028697004587398205, | |
| "pqem": 0.604, | |
| "pqem_stderr": 0.03099319785457785, | |
| "perfect_em": 0.032, | |
| "perfect_em_stderr": 0.011153546809230828 | |
| }, | |
| "harness|bbh:_average|3": { | |
| "em": 0.40856358282723554, | |
| "em_stderr": 0.027042739625910563, | |
| "qem": 0.40856358282723554, | |
| "qem_stderr": 0.027042739625910563, | |
| "pem": 0.49720605234689114, | |
| "pem_stderr": 0.029576005851378102, | |
| "pqem": 0.601065047669327, | |
| "pqem_stderr": 0.02899877859681092, | |
| "perfect_em": 0.40856358282723554, | |
| "perfect_em_stderr": 0.027042739625910563 | |
| } | |
| }, | |
| "versions": { | |
| "harness|bbh:causal_judgment|3": 0, | |
| "harness|bbh:date_understanding|3": 0, | |
| "harness|bbh:disambiguation_qa|3": 0, | |
| "harness|bbh:geometric_shapes|3": 0, | |
| "harness|bbh:logical_deduction_five_objects|3": 0, | |
| "harness|bbh:logical_deduction_seven_objects|3": 0, | |
| "harness|bbh:logical_deduction_three_objects|3": 0, | |
| "harness|bbh:movie_recommendation|3": 0, | |
| "harness|bbh:navigate|3": 0, | |
| "harness|bbh:reasoning_about_colored_objects|3": 0, | |
| "harness|bbh:ruin_names|3": 0, | |
| "harness|bbh:salient_translation_error_detection|3": 0, | |
| "harness|bbh:snarks|3": 0, | |
| "harness|bbh:sports_understanding|3": 0, | |
| "harness|bbh:temporal_sequences|3": 0, | |
| "harness|bbh:tracking_shuffled_objects_five_objects|3": 0, | |
| "harness|bbh:tracking_shuffled_objects_seven_objects|3": 0, | |
| "harness|bbh:tracking_shuffled_objects_three_objects|3": 0 | |
| }, | |
| "config_tasks": { | |
| "harness|bbh:causal_judgment": { | |
| "name": "bbh:causal_judgment", | |
| "prompt_function": "bbh_causal_judgment", | |
| "hf_repo": "lukaemon/bbh", | |
| "hf_subset": "causal_judgement", | |
| "metric": [ | |
| "exact_match", | |
| "quasi_exact_match", | |
| "prefix_exact_match", | |
| "prefix_quasi_exact_match", | |
| "perfect_exact_match" | |
| ], | |
| "hf_avail_splits": [ | |
| "test" | |
| ], | |
| "evaluation_splits": [ | |
| "test" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": null, | |
| "generation_size": 20, | |
| "stop_sequence": [ | |
| "</s>", | |
| "Q:", | |
| "\n\n" | |
| ], | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "harness" | |
| ], | |
| "original_num_docs": 187, | |
| "effective_num_docs": 187, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "harness|bbh:date_understanding": { | |
| "name": "bbh:date_understanding", | |
| "prompt_function": "bbh_date_understanding", | |
| "hf_repo": "lukaemon/bbh", | |
| "hf_subset": "date_understanding", | |
| "metric": [ | |
| "exact_match", | |
| "quasi_exact_match", | |
| "prefix_exact_match", | |
| "prefix_quasi_exact_match", | |
| "perfect_exact_match" | |
| ], | |
| "hf_avail_splits": [ | |
| "test" | |
| ], | |
| "evaluation_splits": [ | |
| "test" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": null, | |
| "generation_size": 20, | |
| "stop_sequence": [ | |
| "</s>", | |
| "Q:", | |
| "\n\n" | |
| ], | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "harness" | |
| ], | |
| "original_num_docs": 250, | |
| "effective_num_docs": 250, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "harness|bbh:disambiguation_qa": { | |
| "name": "bbh:disambiguation_qa", | |
| "prompt_function": "bbh_disambiguation_qa", | |
| "hf_repo": "lukaemon/bbh", | |
| "hf_subset": "disambiguation_qa", | |
| "metric": [ | |
| "exact_match", | |
| "quasi_exact_match", | |
| "prefix_exact_match", | |
| "prefix_quasi_exact_match", | |
| "perfect_exact_match" | |
| ], | |
| "hf_avail_splits": [ | |
| "test" | |
| ], | |
| "evaluation_splits": [ | |
| "test" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": null, | |
| "generation_size": 20, | |
| "stop_sequence": [ | |
| "</s>", | |
| "Q:", | |
| "\n\n" | |
| ], | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "harness" | |
| ], | |
| "original_num_docs": 250, | |
| "effective_num_docs": 250, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "harness|bbh:geometric_shapes": { | |
| "name": "bbh:geometric_shapes", | |
| "prompt_function": "bbh_geometric_shapes", | |
| "hf_repo": "lukaemon/bbh", | |
| "hf_subset": "geometric_shapes", | |
| "metric": [ | |
| "exact_match", | |
| "quasi_exact_match", | |
| "prefix_exact_match", | |
| "prefix_quasi_exact_match", | |
| "perfect_exact_match" | |
| ], | |
| "hf_avail_splits": [ | |
| "test" | |
| ], | |
| "evaluation_splits": [ | |
| "test" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": null, | |
| "generation_size": 20, | |
| "stop_sequence": [ | |
| "</s>", | |
| "Q:", | |
| "\n\n" | |
| ], | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "harness" | |
| ], | |
| "original_num_docs": 250, | |
| "effective_num_docs": 250, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "harness|bbh:logical_deduction_five_objects": { | |
| "name": "bbh:logical_deduction_five_objects", | |
| "prompt_function": "bbh_logical_deduction_five_objects", | |
| "hf_repo": "lukaemon/bbh", | |
| "hf_subset": "logical_deduction_five_objects", | |
| "metric": [ | |
| "exact_match", | |
| "quasi_exact_match", | |
| "prefix_exact_match", | |
| "prefix_quasi_exact_match", | |
| "perfect_exact_match" | |
| ], | |
| "hf_avail_splits": [ | |
| "test" | |
| ], | |
| "evaluation_splits": [ | |
| "test" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": null, | |
| "generation_size": 20, | |
| "stop_sequence": [ | |
| "</s>", | |
| "Q:", | |
| "\n\n" | |
| ], | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "harness" | |
| ], | |
| "original_num_docs": 250, | |
| "effective_num_docs": 250, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "harness|bbh:logical_deduction_seven_objects": { | |
| "name": "bbh:logical_deduction_seven_objects", | |
| "prompt_function": "bbh_logical_deduction_seven_objects", | |
| "hf_repo": "lukaemon/bbh", | |
| "hf_subset": "logical_deduction_seven_objects", | |
| "metric": [ | |
| "exact_match", | |
| "quasi_exact_match", | |
| "prefix_exact_match", | |
| "prefix_quasi_exact_match", | |
| "perfect_exact_match" | |
| ], | |
| "hf_avail_splits": [ | |
| "test" | |
| ], | |
| "evaluation_splits": [ | |
| "test" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": null, | |
| "generation_size": 20, | |
| "stop_sequence": [ | |
| "</s>", | |
| "Q:", | |
| "\n\n" | |
| ], | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "harness" | |
| ], | |
| "original_num_docs": 250, | |
| "effective_num_docs": 250, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "harness|bbh:logical_deduction_three_objects": { | |
| "name": "bbh:logical_deduction_three_objects", | |
| "prompt_function": "bbh_logical_deduction_three_objects", | |
| "hf_repo": "lukaemon/bbh", | |
| "hf_subset": "logical_deduction_three_objects", | |
| "metric": [ | |
| "exact_match", | |
| "quasi_exact_match", | |
| "prefix_exact_match", | |
| "prefix_quasi_exact_match", | |
| "perfect_exact_match" | |
| ], | |
| "hf_avail_splits": [ | |
| "test" | |
| ], | |
| "evaluation_splits": [ | |
| "test" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": null, | |
| "generation_size": 20, | |
| "stop_sequence": [ | |
| "</s>", | |
| "Q:", | |
| "\n\n" | |
| ], | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "harness" | |
| ], | |
| "original_num_docs": 250, | |
| "effective_num_docs": 250, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "harness|bbh:movie_recommendation": { | |
| "name": "bbh:movie_recommendation", | |
| "prompt_function": "bbh_movie_recommendation", | |
| "hf_repo": "lukaemon/bbh", | |
| "hf_subset": "movie_recommendation", | |
| "metric": [ | |
| "exact_match", | |
| "quasi_exact_match", | |
| "prefix_exact_match", | |
| "prefix_quasi_exact_match", | |
| "perfect_exact_match" | |
| ], | |
| "hf_avail_splits": [ | |
| "test" | |
| ], | |
| "evaluation_splits": [ | |
| "test" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": null, | |
| "generation_size": 20, | |
| "stop_sequence": [ | |
| "</s>", | |
| "Q:", | |
| "\n\n" | |
| ], | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "harness" | |
| ], | |
| "original_num_docs": 249, | |
| "effective_num_docs": 249, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "harness|bbh:navigate": { | |
| "name": "bbh:navigate", | |
| "prompt_function": "bbh_navigate", | |
| "hf_repo": "lukaemon/bbh", | |
| "hf_subset": "navigate", | |
| "metric": [ | |
| "exact_match", | |
| "quasi_exact_match", | |
| "prefix_exact_match", | |
| "prefix_quasi_exact_match", | |
| "perfect_exact_match" | |
| ], | |
| "hf_avail_splits": [ | |
| "test" | |
| ], | |
| "evaluation_splits": [ | |
| "test" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": null, | |
| "generation_size": 20, | |
| "stop_sequence": [ | |
| "</s>", | |
| "Q:", | |
| "\n\n" | |
| ], | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "harness" | |
| ], | |
| "original_num_docs": 250, | |
| "effective_num_docs": 250, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "harness|bbh:reasoning_about_colored_objects": { | |
| "name": "bbh:reasoning_about_colored_objects", | |
| "prompt_function": "bbh_reasoning_about_colored_objects", | |
| "hf_repo": "lukaemon/bbh", | |
| "hf_subset": "reasoning_about_colored_objects", | |
| "metric": [ | |
| "exact_match", | |
| "quasi_exact_match", | |
| "prefix_exact_match", | |
| "prefix_quasi_exact_match", | |
| "perfect_exact_match" | |
| ], | |
| "hf_avail_splits": [ | |
| "test" | |
| ], | |
| "evaluation_splits": [ | |
| "test" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": null, | |
| "generation_size": 20, | |
| "stop_sequence": [ | |
| "</s>", | |
| "Q:", | |
| "\n\n" | |
| ], | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "harness" | |
| ], | |
| "original_num_docs": 250, | |
| "effective_num_docs": 250, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "harness|bbh:ruin_names": { | |
| "name": "bbh:ruin_names", | |
| "prompt_function": "bbh_ruin_names", | |
| "hf_repo": "lukaemon/bbh", | |
| "hf_subset": "ruin_names", | |
| "metric": [ | |
| "exact_match", | |
| "quasi_exact_match", | |
| "prefix_exact_match", | |
| "prefix_quasi_exact_match", | |
| "perfect_exact_match" | |
| ], | |
| "hf_avail_splits": [ | |
| "test" | |
| ], | |
| "evaluation_splits": [ | |
| "test" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": null, | |
| "generation_size": 20, | |
| "stop_sequence": [ | |
| "</s>", | |
| "Q:", | |
| "\n\n" | |
| ], | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "harness" | |
| ], | |
| "original_num_docs": 248, | |
| "effective_num_docs": 248, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "harness|bbh:salient_translation_error_detection": { | |
| "name": "bbh:salient_translation_error_detection", | |
| "prompt_function": "bbh_salient_translation_error_detection", | |
| "hf_repo": "lukaemon/bbh", | |
| "hf_subset": "salient_translation_error_detection", | |
| "metric": [ | |
| "exact_match", | |
| "quasi_exact_match", | |
| "prefix_exact_match", | |
| "prefix_quasi_exact_match", | |
| "perfect_exact_match" | |
| ], | |
| "hf_avail_splits": [ | |
| "test" | |
| ], | |
| "evaluation_splits": [ | |
| "test" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": null, | |
| "generation_size": 20, | |
| "stop_sequence": [ | |
| "</s>", | |
| "Q:", | |
| "\n\n" | |
| ], | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "harness" | |
| ], | |
| "original_num_docs": 250, | |
| "effective_num_docs": 250, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "harness|bbh:snarks": { | |
| "name": "bbh:snarks", | |
| "prompt_function": "bbh_snarks", | |
| "hf_repo": "lukaemon/bbh", | |
| "hf_subset": "snarks", | |
| "metric": [ | |
| "exact_match", | |
| "quasi_exact_match", | |
| "prefix_exact_match", | |
| "prefix_quasi_exact_match", | |
| "perfect_exact_match" | |
| ], | |
| "hf_avail_splits": [ | |
| "test" | |
| ], | |
| "evaluation_splits": [ | |
| "test" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": null, | |
| "generation_size": 20, | |
| "stop_sequence": [ | |
| "</s>", | |
| "Q:", | |
| "\n\n" | |
| ], | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "harness" | |
| ], | |
| "original_num_docs": 178, | |
| "effective_num_docs": 178, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "harness|bbh:sports_understanding": { | |
| "name": "bbh:sports_understanding", | |
| "prompt_function": "bbh_sports_understanding", | |
| "hf_repo": "lukaemon/bbh", | |
| "hf_subset": "sports_understanding", | |
| "metric": [ | |
| "exact_match", | |
| "quasi_exact_match", | |
| "prefix_exact_match", | |
| "prefix_quasi_exact_match", | |
| "perfect_exact_match" | |
| ], | |
| "hf_avail_splits": [ | |
| "test" | |
| ], | |
| "evaluation_splits": [ | |
| "test" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": null, | |
| "generation_size": 20, | |
| "stop_sequence": [ | |
| "</s>", | |
| "Q:", | |
| "\n\n" | |
| ], | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "harness" | |
| ], | |
| "original_num_docs": 250, | |
| "effective_num_docs": 250, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "harness|bbh:temporal_sequences": { | |
| "name": "bbh:temporal_sequences", | |
| "prompt_function": "bbh_temporal_sequences", | |
| "hf_repo": "lukaemon/bbh", | |
| "hf_subset": "temporal_sequences", | |
| "metric": [ | |
| "exact_match", | |
| "quasi_exact_match", | |
| "prefix_exact_match", | |
| "prefix_quasi_exact_match", | |
| "perfect_exact_match" | |
| ], | |
| "hf_avail_splits": [ | |
| "test" | |
| ], | |
| "evaluation_splits": [ | |
| "test" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": null, | |
| "generation_size": 20, | |
| "stop_sequence": [ | |
| "</s>", | |
| "Q:", | |
| "\n\n" | |
| ], | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "harness" | |
| ], | |
| "original_num_docs": 250, | |
| "effective_num_docs": 250, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "harness|bbh:tracking_shuffled_objects_five_objects": { | |
| "name": "bbh:tracking_shuffled_objects_five_objects", | |
| "prompt_function": "bbh_tracking_shuffled_objects_five_objects", | |
| "hf_repo": "lukaemon/bbh", | |
| "hf_subset": "tracking_shuffled_objects_five_objects", | |
| "metric": [ | |
| "exact_match", | |
| "quasi_exact_match", | |
| "prefix_exact_match", | |
| "prefix_quasi_exact_match", | |
| "perfect_exact_match" | |
| ], | |
| "hf_avail_splits": [ | |
| "test" | |
| ], | |
| "evaluation_splits": [ | |
| "test" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": null, | |
| "generation_size": 20, | |
| "stop_sequence": [ | |
| "</s>", | |
| "Q:", | |
| "\n\n" | |
| ], | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "harness" | |
| ], | |
| "original_num_docs": 250, | |
| "effective_num_docs": 250, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "harness|bbh:tracking_shuffled_objects_seven_objects": { | |
| "name": "bbh:tracking_shuffled_objects_seven_objects", | |
| "prompt_function": "bbh_tracking_shuffled_objects_seven_objects", | |
| "hf_repo": "lukaemon/bbh", | |
| "hf_subset": "tracking_shuffled_objects_seven_objects", | |
| "metric": [ | |
| "exact_match", | |
| "quasi_exact_match", | |
| "prefix_exact_match", | |
| "prefix_quasi_exact_match", | |
| "perfect_exact_match" | |
| ], | |
| "hf_avail_splits": [ | |
| "test" | |
| ], | |
| "evaluation_splits": [ | |
| "test" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": null, | |
| "generation_size": 20, | |
| "stop_sequence": [ | |
| "</s>", | |
| "Q:", | |
| "\n\n" | |
| ], | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "harness" | |
| ], | |
| "original_num_docs": 250, | |
| "effective_num_docs": 250, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| }, | |
| "harness|bbh:tracking_shuffled_objects_three_objects": { | |
| "name": "bbh:tracking_shuffled_objects_three_objects", | |
| "prompt_function": "bbh_tracking_shuffled_objects_three_objects", | |
| "hf_repo": "lukaemon/bbh", | |
| "hf_subset": "tracking_shuffled_objects_three_objects", | |
| "metric": [ | |
| "exact_match", | |
| "quasi_exact_match", | |
| "prefix_exact_match", | |
| "prefix_quasi_exact_match", | |
| "perfect_exact_match" | |
| ], | |
| "hf_avail_splits": [ | |
| "test" | |
| ], | |
| "evaluation_splits": [ | |
| "test" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": null, | |
| "generation_size": 20, | |
| "stop_sequence": [ | |
| "</s>", | |
| "Q:", | |
| "\n\n" | |
| ], | |
| "output_regex": null, | |
| "frozen": false, | |
| "suite": [ | |
| "harness" | |
| ], | |
| "original_num_docs": 250, | |
| "effective_num_docs": 250, | |
| "trust_dataset": true, | |
| "must_remove_duplicate_docs": null | |
| } | |
| }, | |
| "summary_tasks": { | |
| "harness|bbh:causal_judgment|3": { | |
| "hashes": { | |
| "hash_examples": "63218f5ae055ab2b", | |
| "hash_full_prompts": "148a4c45a8d2b858", | |
| "hash_input_tokens": "41f3903cf7efd5be", | |
| "hash_cont_tokens": "208a981a9b3bf6f2" | |
| }, | |
| "truncated": 187, | |
| "non_truncated": 0, | |
| "padded": 0, | |
| "non_padded": 187, | |
| "effective_few_shots": 3.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|bbh:date_understanding|3": { | |
| "hashes": { | |
| "hash_examples": "f145c7a06def3c8e", | |
| "hash_full_prompts": "e79a3237877b106e", | |
| "hash_input_tokens": "ae3f6744fd7add8b", | |
| "hash_cont_tokens": "27776b5e1982fbeb" | |
| }, | |
| "truncated": 250, | |
| "non_truncated": 0, | |
| "padded": 0, | |
| "non_padded": 250, | |
| "effective_few_shots": 3.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|bbh:disambiguation_qa|3": { | |
| "hashes": { | |
| "hash_examples": "19677fd1773f7eb9", | |
| "hash_full_prompts": "9458fa1926b438bb", | |
| "hash_input_tokens": "a9ea8e42e7ca4f0f", | |
| "hash_cont_tokens": "1961c493faca2d51" | |
| }, | |
| "truncated": 250, | |
| "non_truncated": 0, | |
| "padded": 0, | |
| "non_padded": 250, | |
| "effective_few_shots": 3.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|bbh:geometric_shapes|3": { | |
| "hashes": { | |
| "hash_examples": "76c7b11a13cc72a9", | |
| "hash_full_prompts": "78ead1e22de562a8", | |
| "hash_input_tokens": "ded49f2dd58c24f6", | |
| "hash_cont_tokens": "38278a0798c297a3" | |
| }, | |
| "truncated": 250, | |
| "non_truncated": 0, | |
| "padded": 0, | |
| "non_padded": 250, | |
| "effective_few_shots": 3.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|bbh:logical_deduction_five_objects|3": { | |
| "hashes": { | |
| "hash_examples": "0e958c856332a745", | |
| "hash_full_prompts": "0d7cff0e511b49e7", | |
| "hash_input_tokens": "d16bae7ef2034333", | |
| "hash_cont_tokens": "b88f4093e9753dde" | |
| }, | |
| "truncated": 250, | |
| "non_truncated": 0, | |
| "padded": 0, | |
| "non_padded": 250, | |
| "effective_few_shots": 3.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|bbh:logical_deduction_seven_objects|3": { | |
| "hashes": { | |
| "hash_examples": "ab9de25a5eb40d09", | |
| "hash_full_prompts": "db7b7b19919ef4a7", | |
| "hash_input_tokens": "8b1a10d2204d99c3", | |
| "hash_cont_tokens": "8cb6d01f6b0c44de" | |
| }, | |
| "truncated": 250, | |
| "non_truncated": 0, | |
| "padded": 0, | |
| "non_padded": 250, | |
| "effective_few_shots": 3.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|bbh:logical_deduction_three_objects|3": { | |
| "hashes": { | |
| "hash_examples": "3c6bf52517714218", | |
| "hash_full_prompts": "fd6a5580415c1e21", | |
| "hash_input_tokens": "d48e4b55d3c0ab6c", | |
| "hash_cont_tokens": "317427caa93f2464" | |
| }, | |
| "truncated": 250, | |
| "non_truncated": 0, | |
| "padded": 0, | |
| "non_padded": 250, | |
| "effective_few_shots": 3.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|bbh:movie_recommendation|3": { | |
| "hashes": { | |
| "hash_examples": "2d9dc4975935d31a", | |
| "hash_full_prompts": "b7c9fdf1a2ad8106", | |
| "hash_input_tokens": "019f6011d8ebb7b4", | |
| "hash_cont_tokens": "aac911199f7504cc" | |
| }, | |
| "truncated": 249, | |
| "non_truncated": 0, | |
| "padded": 0, | |
| "non_padded": 249, | |
| "effective_few_shots": 3.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|bbh:navigate|3": { | |
| "hashes": { | |
| "hash_examples": "ba91dcdb9a064255", | |
| "hash_full_prompts": "35aa68650803f91c", | |
| "hash_input_tokens": "fbf7bbc927857899", | |
| "hash_cont_tokens": "42ce692268fe33f3" | |
| }, | |
| "truncated": 250, | |
| "non_truncated": 0, | |
| "padded": 0, | |
| "non_padded": 250, | |
| "effective_few_shots": 3.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|bbh:reasoning_about_colored_objects|3": { | |
| "hashes": { | |
| "hash_examples": "a6ba328c4c3385d2", | |
| "hash_full_prompts": "954618143d9d5c6d", | |
| "hash_input_tokens": "eab386685d65e6d7", | |
| "hash_cont_tokens": "a1478d7792f71bc5" | |
| }, | |
| "truncated": 250, | |
| "non_truncated": 0, | |
| "padded": 0, | |
| "non_padded": 250, | |
| "effective_few_shots": 3.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|bbh:ruin_names|3": { | |
| "hashes": { | |
| "hash_examples": "2ef28d5f2d4fdd25", | |
| "hash_full_prompts": "fd807f4380c14312", | |
| "hash_input_tokens": "f17c7a83f10826d5", | |
| "hash_cont_tokens": "51c9036e8fdf045c" | |
| }, | |
| "truncated": 248, | |
| "non_truncated": 0, | |
| "padded": 0, | |
| "non_padded": 248, | |
| "effective_few_shots": 3.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|bbh:salient_translation_error_detection|3": { | |
| "hashes": { | |
| "hash_examples": "c13f25ec8ffed496", | |
| "hash_full_prompts": "49ab3bc1ed62613f", | |
| "hash_input_tokens": "4ecdc0361737b984", | |
| "hash_cont_tokens": "1c29c8820c867140" | |
| }, | |
| "truncated": 250, | |
| "non_truncated": 0, | |
| "padded": 0, | |
| "non_padded": 250, | |
| "effective_few_shots": 3.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|bbh:snarks|3": { | |
| "hashes": { | |
| "hash_examples": "5f6db7bff7f6f22e", | |
| "hash_full_prompts": "fa5c1ca26f4a8d48", | |
| "hash_input_tokens": "e0e9ee6065d02980", | |
| "hash_cont_tokens": "1f5acfe20dc4c20f" | |
| }, | |
| "truncated": 178, | |
| "non_truncated": 0, | |
| "padded": 0, | |
| "non_padded": 178, | |
| "effective_few_shots": 3.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|bbh:sports_understanding|3": { | |
| "hashes": { | |
| "hash_examples": "042afbe5d9c1f02d", | |
| "hash_full_prompts": "607b29401b4907ec", | |
| "hash_input_tokens": "e2868a0e92e68765", | |
| "hash_cont_tokens": "0bdb63a319fafd37" | |
| }, | |
| "truncated": 250, | |
| "non_truncated": 0, | |
| "padded": 0, | |
| "non_padded": 250, | |
| "effective_few_shots": 3.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|bbh:temporal_sequences|3": { | |
| "hashes": { | |
| "hash_examples": "803a05f352eb6afc", | |
| "hash_full_prompts": "faab4b5e14b9304e", | |
| "hash_input_tokens": "26a0ab6e389383c3", | |
| "hash_cont_tokens": "654dd18c6ce9b107" | |
| }, | |
| "truncated": 250, | |
| "non_truncated": 0, | |
| "padded": 0, | |
| "non_padded": 250, | |
| "effective_few_shots": 3.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|bbh:tracking_shuffled_objects_five_objects|3": { | |
| "hashes": { | |
| "hash_examples": "2bbac6db7ab0d527", | |
| "hash_full_prompts": "7e689cfb3916666f", | |
| "hash_input_tokens": "d0e6770bedbf925a", | |
| "hash_cont_tokens": "f9059c9721ee68ca" | |
| }, | |
| "truncated": 250, | |
| "non_truncated": 0, | |
| "padded": 0, | |
| "non_padded": 250, | |
| "effective_few_shots": 3.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|bbh:tracking_shuffled_objects_seven_objects|3": { | |
| "hashes": { | |
| "hash_examples": "845caf093ac2b58c", | |
| "hash_full_prompts": "a80a61e259878fa0", | |
| "hash_input_tokens": "9b1288a55911b79d", | |
| "hash_cont_tokens": "9bc0919fdc6f2321" | |
| }, | |
| "truncated": 250, | |
| "non_truncated": 0, | |
| "padded": 0, | |
| "non_padded": 250, | |
| "effective_few_shots": 3.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|bbh:tracking_shuffled_objects_three_objects|3": { | |
| "hashes": { | |
| "hash_examples": "9004f14d5a32b9a8", | |
| "hash_full_prompts": "fc66cf32f54cd46f", | |
| "hash_input_tokens": "1328545243e4b107", | |
| "hash_cont_tokens": "665bd289f598bc8a" | |
| }, | |
| "truncated": 250, | |
| "non_truncated": 0, | |
| "padded": 0, | |
| "non_padded": 250, | |
| "effective_few_shots": 3.0, | |
| "num_truncated_few_shots": 0 | |
| } | |
| }, | |
| "summary_general": { | |
| "hashes": { | |
| "hash_examples": "4ff1e3dc5703575d", | |
| "hash_full_prompts": "3758756e616fd780", | |
| "hash_input_tokens": "3ee0dfd31c556ab0", | |
| "hash_cont_tokens": "e0f90f1a3ddfc4a9" | |
| }, | |
| "truncated": 4362, | |
| "non_truncated": 0, | |
| "padded": 0, | |
| "non_padded": 4362, | |
| "num_truncated_few_shots": 0 | |
| } | |
| } |