open-r1-eval-leaderboard
/
eval_results
/HuggingFaceH4
/zephyr-7b-beta-ift
/v1.4
/bbh
/results_2024-03-19T09-26-27.067508.json
| { | |
| "config_general": { | |
| "lighteval_sha": "?", | |
| "num_fewshot_seeds": 1, | |
| "override_batch_size": 1, | |
| "max_samples": null, | |
| "job_id": "", | |
| "start_time": 421804.303392443, | |
| "end_time": 422069.48693289, | |
| "total_evaluation_time_secondes": "265.183540447033", | |
| "model_name": "HuggingFaceH4/zephyr-7b-beta-ift", | |
| "model_sha": "0a50b8076f15d8735364b82bcc234a142998519a", | |
| "model_dtype": "torch.bfloat16", | |
| "model_size": "13.99 GB", | |
| "config": null | |
| }, | |
| "results": { | |
| "harness|bbh:causal_judgment|3": { | |
| "em": 0.36363636363636365, | |
| "em_stderr": 0.03527198153014412, | |
| "qem": 0.36363636363636365, | |
| "qem_stderr": 0.03527198153014412, | |
| "pem": 0.5561497326203209, | |
| "pem_stderr": 0.03642987131924727, | |
| "pqem": 0.5561497326203209, | |
| "pqem_stderr": 0.03642987131924727, | |
| "perfect_em": 0.36363636363636365, | |
| "perfect_em_stderr": 0.03527198153014412 | |
| }, | |
| "harness|bbh:date_understanding|3": { | |
| "em": 0.48, | |
| "em_stderr": 0.031660853408495185, | |
| "qem": 0.48, | |
| "qem_stderr": 0.031660853408495185, | |
| "pem": 0.48, | |
| "pem_stderr": 0.031660853408495185, | |
| "pqem": 0.572, | |
| "pqem_stderr": 0.031355968923772605, | |
| "perfect_em": 0.48, | |
| "perfect_em_stderr": 0.031660853408495185 | |
| }, | |
| "harness|bbh:disambiguation_qa|3": { | |
| "em": 0.636, | |
| "em_stderr": 0.030491555220405555, | |
| "qem": 0.636, | |
| "qem_stderr": 0.030491555220405555, | |
| "pem": 0.636, | |
| "pem_stderr": 0.030491555220405555, | |
| "pqem": 0.712, | |
| "pqem_stderr": 0.02869700458739822, | |
| "perfect_em": 0.636, | |
| "perfect_em_stderr": 0.030491555220405555 | |
| }, | |
| "harness|bbh:geometric_shapes|3": { | |
| "em": 0.228, | |
| "em_stderr": 0.026587432487268452, | |
| "qem": 0.228, | |
| "qem_stderr": 0.026587432487268452, | |
| "pem": 0.228, | |
| "pem_stderr": 0.026587432487268452, | |
| "pqem": 0.228, | |
| "pqem_stderr": 0.026587432487268452, | |
| "perfect_em": 0.228, | |
| "perfect_em_stderr": 0.026587432487268452 | |
| }, | |
| "harness|bbh:logical_deduction_five_objects|3": { | |
| "em": 0.368, | |
| "em_stderr": 0.030562070620993163, | |
| "qem": 0.368, | |
| "qem_stderr": 0.030562070620993163, | |
| "pem": 0.368, | |
| "pem_stderr": 0.030562070620993163, | |
| "pqem": 0.456, | |
| "pqem_stderr": 0.03156328506121339, | |
| "perfect_em": 0.368, | |
| "perfect_em_stderr": 0.030562070620993163 | |
| }, | |
| "harness|bbh:logical_deduction_seven_objects|3": { | |
| "em": 0.272, | |
| "em_stderr": 0.028200088296310006, | |
| "qem": 0.272, | |
| "qem_stderr": 0.028200088296310006, | |
| "pem": 0.272, | |
| "pem_stderr": 0.028200088296310006, | |
| "pqem": 0.388, | |
| "pqem_stderr": 0.03088103874899391, | |
| "perfect_em": 0.272, | |
| "perfect_em_stderr": 0.028200088296310006 | |
| }, | |
| "harness|bbh:logical_deduction_three_objects|3": { | |
| "em": 0.524, | |
| "em_stderr": 0.03164968895968782, | |
| "qem": 0.524, | |
| "qem_stderr": 0.03164968895968782, | |
| "pem": 0.524, | |
| "pem_stderr": 0.03164968895968782, | |
| "pqem": 0.712, | |
| "pqem_stderr": 0.02869700458739821, | |
| "perfect_em": 0.524, | |
| "perfect_em_stderr": 0.03164968895968782 | |
| }, | |
| "harness|bbh:movie_recommendation|3": { | |
| "em": 0.6465863453815262, | |
| "em_stderr": 0.030354919577080145, | |
| "qem": 0.6465863453815262, | |
| "qem_stderr": 0.030354919577080145, | |
| "pem": 0.6506024096385542, | |
| "pem_stderr": 0.030275543348462008, | |
| "pqem": 0.7028112449799196, | |
| "pqem_stderr": 0.029020821148267553, | |
| "perfect_em": 0.6465863453815262, | |
| "perfect_em_stderr": 0.030354919577080145 | |
| }, | |
| "harness|bbh:navigate|3": { | |
| "em": 0.592, | |
| "em_stderr": 0.031145209846548505, | |
| "qem": 0.592, | |
| "qem_stderr": 0.031145209846548505, | |
| "pem": 0.592, | |
| "pem_stderr": 0.031145209846548505, | |
| "pqem": 0.592, | |
| "pqem_stderr": 0.031145209846548505, | |
| "perfect_em": 0.592, | |
| "perfect_em_stderr": 0.031145209846548505 | |
| }, | |
| "harness|bbh:reasoning_about_colored_objects|3": { | |
| "em": 0.296, | |
| "em_stderr": 0.028928939388379617, | |
| "qem": 0.296, | |
| "qem_stderr": 0.028928939388379617, | |
| "pem": 0.364, | |
| "pem_stderr": 0.030491555220405562, | |
| "pqem": 0.5, | |
| "pqem_stderr": 0.031686212526223896, | |
| "perfect_em": 0.296, | |
| "perfect_em_stderr": 0.028928939388379617 | |
| }, | |
| "harness|bbh:ruin_names|3": { | |
| "em": 0.36693548387096775, | |
| "em_stderr": 0.03066693445085009, | |
| "qem": 0.36693548387096775, | |
| "qem_stderr": 0.03066693445085009, | |
| "pem": 0.36693548387096775, | |
| "pem_stderr": 0.03066693445085009, | |
| "pqem": 0.4717741935483871, | |
| "pqem_stderr": 0.03176350513565306, | |
| "perfect_em": 0.36693548387096775, | |
| "perfect_em_stderr": 0.03066693445085009 | |
| }, | |
| "harness|bbh:salient_translation_error_detection|3": { | |
| "em": 0.308, | |
| "em_stderr": 0.029256928606501864, | |
| "qem": 0.308, | |
| "qem_stderr": 0.029256928606501864, | |
| "pem": 0.308, | |
| "pem_stderr": 0.029256928606501864, | |
| "pqem": 0.444, | |
| "pqem_stderr": 0.03148684942554574, | |
| "perfect_em": 0.308, | |
| "perfect_em_stderr": 0.029256928606501864 | |
| }, | |
| "harness|bbh:snarks|3": { | |
| "em": 0.5056179775280899, | |
| "em_stderr": 0.037579929004759816, | |
| "qem": 0.5056179775280899, | |
| "qem_stderr": 0.037579929004759816, | |
| "pem": 0.5056179775280899, | |
| "pem_stderr": 0.037579929004759816, | |
| "pqem": 0.5280898876404494, | |
| "pqem_stderr": 0.03752294651708461, | |
| "perfect_em": 0.5056179775280899, | |
| "perfect_em_stderr": 0.037579929004759816 | |
| }, | |
| "harness|bbh:sports_understanding|3": { | |
| "em": 0.02, | |
| "em_stderr": 0.008872139507342685, | |
| "qem": 0.02, | |
| "qem_stderr": 0.008872139507342685, | |
| "pem": 0.78, | |
| "pem_stderr": 0.026251792824605824, | |
| "pqem": 0.78, | |
| "pqem_stderr": 0.026251792824605824, | |
| "perfect_em": 0.02, | |
| "perfect_em_stderr": 0.008872139507342685 | |
| }, | |
| "harness|bbh:temporal_sequences|3": { | |
| "em": 0.088, | |
| "em_stderr": 0.017953084777052885, | |
| "qem": 0.088, | |
| "qem_stderr": 0.017953084777052885, | |
| "pem": 0.088, | |
| "pem_stderr": 0.017953084777052885, | |
| "pqem": 0.348, | |
| "pqem_stderr": 0.030186568464511683, | |
| "perfect_em": 0.088, | |
| "perfect_em_stderr": 0.017953084777052885 | |
| }, | |
| "harness|bbh:tracking_shuffled_objects_five_objects|3": { | |
| "em": 0.176, | |
| "em_stderr": 0.024133497525457116, | |
| "qem": 0.176, | |
| "qem_stderr": 0.024133497525457116, | |
| "pem": 0.176, | |
| "pem_stderr": 0.024133497525457116, | |
| "pqem": 0.376, | |
| "pqem_stderr": 0.030696336267394583, | |
| "perfect_em": 0.176, | |
| "perfect_em_stderr": 0.024133497525457116 | |
| }, | |
| "harness|bbh:tracking_shuffled_objects_seven_objects|3": { | |
| "em": 0.112, | |
| "em_stderr": 0.01998553693917145, | |
| "qem": 0.112, | |
| "qem_stderr": 0.01998553693917145, | |
| "pem": 0.112, | |
| "pem_stderr": 0.01998553693917145, | |
| "pqem": 0.244, | |
| "pqem_stderr": 0.02721799546455318, | |
| "perfect_em": 0.112, | |
| "perfect_em_stderr": 0.01998553693917145 | |
| }, | |
| "harness|bbh:tracking_shuffled_objects_three_objects|3": { | |
| "em": 0.376, | |
| "em_stderr": 0.030696336267394583, | |
| "qem": 0.376, | |
| "qem_stderr": 0.030696336267394583, | |
| "pem": 0.376, | |
| "pem_stderr": 0.030696336267394583, | |
| "pqem": 0.688, | |
| "pqem_stderr": 0.029361067575219817, | |
| "perfect_em": 0.376, | |
| "perfect_em_stderr": 0.030696336267394583 | |
| }, | |
| "harness|bbh:_average|3": { | |
| "em": 0.3532653428009415, | |
| "em_stderr": 0.02799984035632461, | |
| "qem": 0.3532653428009415, | |
| "qem_stderr": 0.02799984035632461, | |
| "pem": 0.410183644647663, | |
| "pem_stderr": 0.02911210606242318, | |
| "pqem": 0.5166013921549487, | |
| "pqem_stderr": 0.030586161717272256, | |
| "perfect_em": 0.3532653428009415, | |
| "perfect_em_stderr": 0.02799984035632461 | |
| } | |
| }, | |
| "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": "372cb5cf1df956ef" | |
| }, | |
| "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": "8c0865bfd5d6ecf9" | |
| }, | |
| "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": "5de2352d7fcfa99e" | |
| }, | |
| "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": "09918400c9ea6b83" | |
| }, | |
| "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": "1b95ed6253158cb7" | |
| }, | |
| "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": "ee899db484d19856" | |
| }, | |
| "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": "c78fc8827c2409a9" | |
| }, | |
| "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": "abe57201c291f601" | |
| }, | |
| "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": "106bfc7a0d9f53bd" | |
| }, | |
| "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": "19f3d4375e80f9c8" | |
| }, | |
| "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": "b13730f22563b515" | |
| }, | |
| "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": "50734314f7cae042" | |
| }, | |
| "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": "b774b909eab451ee" | |
| }, | |
| "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": "25f31a0722d36082" | |
| }, | |
| "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": "799becc48ad28c99" | |
| }, | |
| "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": "fb54b12a82bb6d2d" | |
| }, | |
| "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": "4657ed62d033183f" | |
| }, | |
| "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": "8ead035cb04fb1cf" | |
| }, | |
| "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": "e566def04c5d82fa" | |
| }, | |
| "truncated": 4362, | |
| "non_truncated": 0, | |
| "padded": 0, | |
| "non_padded": 4362, | |
| "num_truncated_few_shots": 0 | |
| } | |
| } |