open-r1-eval-leaderboard
/
eval_results
/HuggingFaceH4
/zephyr-7b-beta-ift
/v1.2
/bbh
/results_2024-03-19T09-26-04.565494.json
| { | |
| "config_general": { | |
| "lighteval_sha": "?", | |
| "num_fewshot_seeds": 1, | |
| "override_batch_size": 1, | |
| "max_samples": null, | |
| "job_id": "", | |
| "start_time": 213044.735679574, | |
| "end_time": 213308.843495247, | |
| "total_evaluation_time_secondes": "264.10781567302183", | |
| "model_name": "HuggingFaceH4/zephyr-7b-beta-ift", | |
| "model_sha": "c69f744d31d928e63787be75c0c5ea9283c4a826", | |
| "model_dtype": "torch.bfloat16", | |
| "model_size": "13.99 GB", | |
| "config": null | |
| }, | |
| "results": { | |
| "harness|bbh:causal_judgment|3": { | |
| "em": 0.45454545454545453, | |
| "em_stderr": 0.03650996949556813, | |
| "qem": 0.45454545454545453, | |
| "qem_stderr": 0.03650996949556813, | |
| "pem": 0.5240641711229946, | |
| "pem_stderr": 0.03661929361528703, | |
| "pqem": 0.5240641711229946, | |
| "pqem_stderr": 0.03661929361528703, | |
| "perfect_em": 0.45454545454545453, | |
| "perfect_em_stderr": 0.03650996949556813 | |
| }, | |
| "harness|bbh:date_understanding|3": { | |
| "em": 0.472, | |
| "em_stderr": 0.031636489531544396, | |
| "qem": 0.472, | |
| "qem_stderr": 0.031636489531544396, | |
| "pem": 0.472, | |
| "pem_stderr": 0.031636489531544396, | |
| "pqem": 0.584, | |
| "pqem_stderr": 0.031235856237014553, | |
| "perfect_em": 0.472, | |
| "perfect_em_stderr": 0.031636489531544396 | |
| }, | |
| "harness|bbh:disambiguation_qa|3": { | |
| "em": 0.608, | |
| "em_stderr": 0.030938207620401195, | |
| "qem": 0.608, | |
| "qem_stderr": 0.030938207620401195, | |
| "pem": 0.608, | |
| "pem_stderr": 0.030938207620401195, | |
| "pqem": 0.712, | |
| "pqem_stderr": 0.028697004587398215, | |
| "perfect_em": 0.608, | |
| "perfect_em_stderr": 0.030938207620401195 | |
| }, | |
| "harness|bbh:geometric_shapes|3": { | |
| "em": 0.264, | |
| "em_stderr": 0.027934518957690908, | |
| "qem": 0.264, | |
| "qem_stderr": 0.027934518957690908, | |
| "pem": 0.264, | |
| "pem_stderr": 0.027934518957690908, | |
| "pqem": 0.264, | |
| "pqem_stderr": 0.027934518957690908, | |
| "perfect_em": 0.264, | |
| "perfect_em_stderr": 0.027934518957690908 | |
| }, | |
| "harness|bbh:logical_deduction_five_objects|3": { | |
| "em": 0.348, | |
| "em_stderr": 0.03018656846451169, | |
| "qem": 0.348, | |
| "qem_stderr": 0.03018656846451169, | |
| "pem": 0.348, | |
| "pem_stderr": 0.03018656846451169, | |
| "pqem": 0.444, | |
| "pqem_stderr": 0.03148684942554575, | |
| "perfect_em": 0.348, | |
| "perfect_em_stderr": 0.03018656846451169 | |
| }, | |
| "harness|bbh:logical_deduction_seven_objects|3": { | |
| "em": 0.292, | |
| "em_stderr": 0.028814320402205648, | |
| "qem": 0.292, | |
| "qem_stderr": 0.028814320402205648, | |
| "pem": 0.292, | |
| "pem_stderr": 0.028814320402205648, | |
| "pqem": 0.388, | |
| "pqem_stderr": 0.030881038748993908, | |
| "perfect_em": 0.292, | |
| "perfect_em_stderr": 0.028814320402205648 | |
| }, | |
| "harness|bbh:logical_deduction_three_objects|3": { | |
| "em": 0.492, | |
| "em_stderr": 0.031682156431413803, | |
| "qem": 0.492, | |
| "qem_stderr": 0.031682156431413803, | |
| "pem": 0.492, | |
| "pem_stderr": 0.031682156431413803, | |
| "pqem": 0.728, | |
| "pqem_stderr": 0.02820008829631, | |
| "perfect_em": 0.492, | |
| "perfect_em_stderr": 0.031682156431413803 | |
| }, | |
| "harness|bbh:movie_recommendation|3": { | |
| "em": 0.6666666666666666, | |
| "em_stderr": 0.02993421700446251, | |
| "qem": 0.6666666666666666, | |
| "qem_stderr": 0.02993421700446251, | |
| "pem": 0.6827309236947792, | |
| "pem_stderr": 0.02955376521544915, | |
| "pqem": 0.8032128514056225, | |
| "pqem_stderr": 0.02524572518448645, | |
| "perfect_em": 0.6666666666666666, | |
| "perfect_em_stderr": 0.02993421700446251 | |
| }, | |
| "harness|bbh:navigate|3": { | |
| "em": 0.544, | |
| "em_stderr": 0.0315632850612134, | |
| "qem": 0.544, | |
| "qem_stderr": 0.0315632850612134, | |
| "pem": 0.544, | |
| "pem_stderr": 0.0315632850612134, | |
| "pqem": 0.544, | |
| "pqem_stderr": 0.0315632850612134, | |
| "perfect_em": 0.544, | |
| "perfect_em_stderr": 0.0315632850612134 | |
| }, | |
| "harness|bbh:reasoning_about_colored_objects|3": { | |
| "em": 0.376, | |
| "em_stderr": 0.030696336267394587, | |
| "qem": 0.376, | |
| "qem_stderr": 0.030696336267394587, | |
| "pem": 0.384, | |
| "pem_stderr": 0.030821679117375374, | |
| "pqem": 0.544, | |
| "pqem_stderr": 0.03156328506121339, | |
| "perfect_em": 0.376, | |
| "perfect_em_stderr": 0.030696336267394587 | |
| }, | |
| "harness|bbh:ruin_names|3": { | |
| "em": 0.3790322580645161, | |
| "em_stderr": 0.030869111522661374, | |
| "qem": 0.3790322580645161, | |
| "qem_stderr": 0.030869111522661374, | |
| "pem": 0.3790322580645161, | |
| "pem_stderr": 0.030869111522661374, | |
| "pqem": 0.4596774193548387, | |
| "pqem_stderr": 0.03171061518395055, | |
| "perfect_em": 0.3790322580645161, | |
| "perfect_em_stderr": 0.030869111522661374 | |
| }, | |
| "harness|bbh:salient_translation_error_detection|3": { | |
| "em": 0.28, | |
| "em_stderr": 0.02845414827783233, | |
| "qem": 0.28, | |
| "qem_stderr": 0.02845414827783233, | |
| "pem": 0.28, | |
| "pem_stderr": 0.02845414827783233, | |
| "pqem": 0.42, | |
| "pqem_stderr": 0.03127799950463661, | |
| "perfect_em": 0.28, | |
| "perfect_em_stderr": 0.02845414827783233 | |
| }, | |
| "harness|bbh:snarks|3": { | |
| "em": 0.4550561797752809, | |
| "em_stderr": 0.03743016495716993, | |
| "qem": 0.4550561797752809, | |
| "qem_stderr": 0.03743016495716993, | |
| "pem": 0.4550561797752809, | |
| "pem_stderr": 0.03743016495716993, | |
| "pqem": 0.550561797752809, | |
| "pqem_stderr": 0.03738964966056965, | |
| "perfect_em": 0.4550561797752809, | |
| "perfect_em_stderr": 0.03743016495716993 | |
| }, | |
| "harness|bbh:sports_understanding|3": { | |
| "em": 0.356, | |
| "em_stderr": 0.030343680657153208, | |
| "qem": 0.356, | |
| "qem_stderr": 0.030343680657153208, | |
| "pem": 0.74, | |
| "pem_stderr": 0.0277973157526443, | |
| "pqem": 0.74, | |
| "pqem_stderr": 0.0277973157526443, | |
| "perfect_em": 0.356, | |
| "perfect_em_stderr": 0.030343680657153208 | |
| }, | |
| "harness|bbh:temporal_sequences|3": { | |
| "em": 0.076, | |
| "em_stderr": 0.016793573067859648, | |
| "qem": 0.076, | |
| "qem_stderr": 0.016793573067859648, | |
| "pem": 0.076, | |
| "pem_stderr": 0.016793573067859648, | |
| "pqem": 0.348, | |
| "pqem_stderr": 0.03018656846451168, | |
| "perfect_em": 0.076, | |
| "perfect_em_stderr": 0.016793573067859648 | |
| }, | |
| "harness|bbh:tracking_shuffled_objects_five_objects|3": { | |
| "em": 0.188, | |
| "em_stderr": 0.02476037772775049, | |
| "qem": 0.188, | |
| "qem_stderr": 0.02476037772775049, | |
| "pem": 0.188, | |
| "pem_stderr": 0.02476037772775049, | |
| "pqem": 0.372, | |
| "pqem_stderr": 0.03063032594455832, | |
| "perfect_em": 0.188, | |
| "perfect_em_stderr": 0.02476037772775049 | |
| }, | |
| "harness|bbh:tracking_shuffled_objects_seven_objects|3": { | |
| "em": 0.112, | |
| "em_stderr": 0.019985536939171444, | |
| "qem": 0.112, | |
| "qem_stderr": 0.019985536939171444, | |
| "pem": 0.112, | |
| "pem_stderr": 0.019985536939171444, | |
| "pqem": 0.256, | |
| "pqem_stderr": 0.027657108718204912, | |
| "perfect_em": 0.112, | |
| "perfect_em_stderr": 0.019985536939171444 | |
| }, | |
| "harness|bbh:tracking_shuffled_objects_three_objects|3": { | |
| "em": 0.38, | |
| "em_stderr": 0.030760116042626042, | |
| "qem": 0.38, | |
| "qem_stderr": 0.030760116042626042, | |
| "pem": 0.38, | |
| "pem_stderr": 0.030760116042626042, | |
| "pqem": 0.684, | |
| "pqem_stderr": 0.029462657598578666, | |
| "perfect_em": 0.38, | |
| "perfect_em_stderr": 0.030760116042626042 | |
| }, | |
| "harness|bbh:_average|3": { | |
| "em": 0.3746278088362176, | |
| "em_stderr": 0.029405154357146152, | |
| "qem": 0.3746278088362176, | |
| "qem_stderr": 0.029405154357146152, | |
| "pem": 0.40116019625875393, | |
| "pem_stderr": 0.029255590483600453, | |
| "pqem": 0.5203064577575703, | |
| "pqem_stderr": 0.0305299547779338, | |
| "perfect_em": 0.3746278088362176, | |
| "perfect_em_stderr": 0.029405154357146152 | |
| } | |
| }, | |
| "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": "7303fa1d0fe0b29a", | |
| "hash_input_tokens": "79663e73bb5ce6ac", | |
| "hash_cont_tokens": "ef9327cee35d8029" | |
| }, | |
| "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": "69e60d10afa5a6f1", | |
| "hash_input_tokens": "e9bd5760c58a1104", | |
| "hash_cont_tokens": "71849ecf8584dab2" | |
| }, | |
| "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": "ae0a8fd428f9aee3", | |
| "hash_input_tokens": "b3625dcc25d708b2", | |
| "hash_cont_tokens": "f56e2ae6e51ffd35" | |
| }, | |
| "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": "76633257f67207f9", | |
| "hash_input_tokens": "c16e8768d8c9056f", | |
| "hash_cont_tokens": "6f82cfc57fad51ed" | |
| }, | |
| "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": "3c96645848786efd", | |
| "hash_input_tokens": "915443ee37f164dc", | |
| "hash_cont_tokens": "3ba76cd31c2b11fc" | |
| }, | |
| "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": "185c5851c101ee66", | |
| "hash_input_tokens": "66d532c31ef57236", | |
| "hash_cont_tokens": "ef5df9f78cac0d98" | |
| }, | |
| "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": "8ba2d94357e589d0", | |
| "hash_input_tokens": "d51c6ad06efbf88b", | |
| "hash_cont_tokens": "ec35259b14a29994" | |
| }, | |
| "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": "a411e216d0f5f626", | |
| "hash_input_tokens": "e17a3080d43ae54f", | |
| "hash_cont_tokens": "fa36c8a51d19d1b6" | |
| }, | |
| "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": "ebb3084ecc78a46a", | |
| "hash_input_tokens": "90854b0ca565c8f5", | |
| "hash_cont_tokens": "f0b81383ef175dc7" | |
| }, | |
| "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": "38328d016a4ebef3", | |
| "hash_input_tokens": "b45b5a8a531e8bf5", | |
| "hash_cont_tokens": "1a6066f33559a166" | |
| }, | |
| "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": "9c7d0493c37182d6", | |
| "hash_input_tokens": "627b6058879c9350", | |
| "hash_cont_tokens": "4954998436a16a37" | |
| }, | |
| "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": "edccd4061b168b78", | |
| "hash_input_tokens": "7d4d7e481ad8766b", | |
| "hash_cont_tokens": "dba092317d566767" | |
| }, | |
| "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": "31cafd95ab850a44", | |
| "hash_input_tokens": "616900bacd0ba7ca", | |
| "hash_cont_tokens": "9009295fa2b6acd2" | |
| }, | |
| "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": "3d46581e9bbec2d0", | |
| "hash_input_tokens": "8e9e99c22dd3a8d2", | |
| "hash_cont_tokens": "135c4017ca5b989e" | |
| }, | |
| "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": "4a54db144a5dd222", | |
| "hash_input_tokens": "24789970b2290dd3", | |
| "hash_cont_tokens": "d2ca778f9b56259d" | |
| }, | |
| "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": "e3079106787cc311", | |
| "hash_input_tokens": "9036045cff895b08", | |
| "hash_cont_tokens": "8fda2075208098ac" | |
| }, | |
| "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": "6364e5b860590ec8", | |
| "hash_input_tokens": "7100c488aa0764ff", | |
| "hash_cont_tokens": "79b78fa6382fde9b" | |
| }, | |
| "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": "01aef56c4d1fe9fe", | |
| "hash_input_tokens": "b9690a5d32a586fc", | |
| "hash_cont_tokens": "e20b3d52b7a6ebc5" | |
| }, | |
| "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": "1cbeab0a00117cb8", | |
| "hash_input_tokens": "3608679dab4ce40e", | |
| "hash_cont_tokens": "686cee659724e963" | |
| }, | |
| "truncated": 4362, | |
| "non_truncated": 0, | |
| "padded": 0, | |
| "non_padded": 4362, | |
| "num_truncated_few_shots": 0 | |
| } | |
| } |