open-r1-eval-leaderboard
/
eval_results
/HuggingFaceH4
/zephyr-7b-beta
/main
/bbh
/results_2024-03-18T20-36-18.052319.json
| { | |
| "config_general": { | |
| "lighteval_sha": "?", | |
| "num_fewshot_seeds": 1, | |
| "override_batch_size": 1, | |
| "max_samples": null, | |
| "job_id": "", | |
| "start_time": 5122577.949751535, | |
| "end_time": 5122918.221141459, | |
| "total_evaluation_time_secondes": "340.2713899239898", | |
| "model_name": "HuggingFaceH4/zephyr-7b-beta", | |
| "model_sha": "b70e0c9a2d9e14bd1e812d3c398e5f313e93b473", | |
| "model_dtype": "torch.bfloat16", | |
| "model_size": "13.99 GB", | |
| "config": null | |
| }, | |
| "results": { | |
| "harness|bbh:causal_judgment|3": { | |
| "em": 0.15508021390374332, | |
| "em_stderr": 0.026541744475095306, | |
| "qem": 0.15508021390374332, | |
| "qem_stderr": 0.026541744475095306, | |
| "pem": 0.5614973262032086, | |
| "pem_stderr": 0.03638341809400995, | |
| "pqem": 0.5614973262032086, | |
| "pqem_stderr": 0.03638341809400995, | |
| "perfect_em": 0.15508021390374332, | |
| "perfect_em_stderr": 0.026541744475095306 | |
| }, | |
| "harness|bbh:date_understanding|3": { | |
| "em": 0.468, | |
| "em_stderr": 0.03162125257572551, | |
| "qem": 0.468, | |
| "qem_stderr": 0.03162125257572551, | |
| "pem": 0.476, | |
| "pem_stderr": 0.03164968895968782, | |
| "pqem": 0.596, | |
| "pqem_stderr": 0.03109668818482529, | |
| "perfect_em": 0.468, | |
| "perfect_em_stderr": 0.03162125257572551 | |
| }, | |
| "harness|bbh:disambiguation_qa|3": { | |
| "em": 0.62, | |
| "em_stderr": 0.030760116042626046, | |
| "qem": 0.62, | |
| "qem_stderr": 0.030760116042626046, | |
| "pem": 0.62, | |
| "pem_stderr": 0.030760116042626046, | |
| "pqem": 0.716, | |
| "pqem_stderr": 0.02857695873043741, | |
| "perfect_em": 0.62, | |
| "perfect_em_stderr": 0.030760116042626046 | |
| }, | |
| "harness|bbh:geometric_shapes|3": { | |
| "em": 0.14, | |
| "em_stderr": 0.021989409645240265, | |
| "qem": 0.14, | |
| "qem_stderr": 0.021989409645240265, | |
| "pem": 0.212, | |
| "pem_stderr": 0.025901884690541162, | |
| "pqem": 0.212, | |
| "pqem_stderr": 0.025901884690541162, | |
| "perfect_em": 0.14, | |
| "perfect_em_stderr": 0.021989409645240265 | |
| }, | |
| "harness|bbh:logical_deduction_five_objects|3": { | |
| "em": 0.36, | |
| "em_stderr": 0.03041876402517499, | |
| "qem": 0.36, | |
| "qem_stderr": 0.03041876402517499, | |
| "pem": 0.364, | |
| "pem_stderr": 0.030491555220405562, | |
| "pqem": 0.5, | |
| "pqem_stderr": 0.031686212526223896, | |
| "perfect_em": 0.36, | |
| "perfect_em_stderr": 0.03041876402517499 | |
| }, | |
| "harness|bbh:logical_deduction_seven_objects|3": { | |
| "em": 0.28, | |
| "em_stderr": 0.02845414827783232, | |
| "qem": 0.28, | |
| "qem_stderr": 0.02845414827783232, | |
| "pem": 0.284, | |
| "pem_stderr": 0.028576958730437408, | |
| "pqem": 0.42, | |
| "pqem_stderr": 0.03127799950463662, | |
| "perfect_em": 0.28, | |
| "perfect_em_stderr": 0.02845414827783232 | |
| }, | |
| "harness|bbh:logical_deduction_three_objects|3": { | |
| "em": 0.544, | |
| "em_stderr": 0.03156328506121339, | |
| "qem": 0.544, | |
| "qem_stderr": 0.03156328506121339, | |
| "pem": 0.548, | |
| "pem_stderr": 0.03153986449255662, | |
| "pqem": 0.784, | |
| "pqem_stderr": 0.026078657663732716, | |
| "perfect_em": 0.544, | |
| "perfect_em_stderr": 0.03156328506121339 | |
| }, | |
| "harness|bbh:movie_recommendation|3": { | |
| "em": 0.5662650602409639, | |
| "em_stderr": 0.03146996403326648, | |
| "qem": 0.5662650602409639, | |
| "qem_stderr": 0.03146996403326648, | |
| "pem": 0.6345381526104418, | |
| "pem_stderr": 0.030579054441446426, | |
| "pqem": 0.7670682730923695, | |
| "pqem_stderr": 0.02684142031987394, | |
| "perfect_em": 0.5662650602409639, | |
| "perfect_em_stderr": 0.03146996403326648 | |
| }, | |
| "harness|bbh:navigate|3": { | |
| "em": 0.596, | |
| "em_stderr": 0.031096688184825298, | |
| "qem": 0.596, | |
| "qem_stderr": 0.031096688184825298, | |
| "pem": 0.6, | |
| "pem_stderr": 0.031046021028253257, | |
| "pqem": 0.6, | |
| "pqem_stderr": 0.031046021028253257, | |
| "perfect_em": 0.596, | |
| "perfect_em_stderr": 0.031096688184825298 | |
| }, | |
| "harness|bbh:reasoning_about_colored_objects|3": { | |
| "em": 0.252, | |
| "em_stderr": 0.027513851933031352, | |
| "qem": 0.252, | |
| "qem_stderr": 0.027513851933031352, | |
| "pem": 0.308, | |
| "pem_stderr": 0.029256928606501864, | |
| "pqem": 0.452, | |
| "pqem_stderr": 0.03153986449255662, | |
| "perfect_em": 0.252, | |
| "perfect_em_stderr": 0.027513851933031352 | |
| }, | |
| "harness|bbh:ruin_names|3": { | |
| "em": 0.3064516129032258, | |
| "em_stderr": 0.02933397125939671, | |
| "qem": 0.3064516129032258, | |
| "qem_stderr": 0.02933397125939671, | |
| "pem": 0.3548387096774194, | |
| "pem_stderr": 0.030443961913180523, | |
| "pqem": 0.5564516129032258, | |
| "pqem_stderr": 0.03161081756372432, | |
| "perfect_em": 0.3064516129032258, | |
| "perfect_em_stderr": 0.02933397125939671 | |
| }, | |
| "harness|bbh:salient_translation_error_detection|3": { | |
| "em": 0.384, | |
| "em_stderr": 0.030821679117375385, | |
| "qem": 0.384, | |
| "qem_stderr": 0.030821679117375385, | |
| "pem": 0.384, | |
| "pem_stderr": 0.030821679117375385, | |
| "pqem": 0.524, | |
| "pqem_stderr": 0.03164968895968782, | |
| "perfect_em": 0.384, | |
| "perfect_em_stderr": 0.030821679117375385 | |
| }, | |
| "harness|bbh:snarks|3": { | |
| "em": 0.5112359550561798, | |
| "em_stderr": 0.03757281091983853, | |
| "qem": 0.5112359550561798, | |
| "qem_stderr": 0.03757281091983853, | |
| "pem": 0.550561797752809, | |
| "pem_stderr": 0.037389649660569645, | |
| "pqem": 0.6853932584269663, | |
| "pqem_stderr": 0.03490336455459108, | |
| "perfect_em": 0.5112359550561798, | |
| "perfect_em_stderr": 0.03757281091983853 | |
| }, | |
| "harness|bbh:sports_understanding|3": { | |
| "em": 0.36, | |
| "em_stderr": 0.030418764025174985, | |
| "qem": 0.36, | |
| "qem_stderr": 0.030418764025174985, | |
| "pem": 0.784, | |
| "pem_stderr": 0.02607865766373272, | |
| "pqem": 0.784, | |
| "pqem_stderr": 0.02607865766373272, | |
| "perfect_em": 0.36, | |
| "perfect_em_stderr": 0.030418764025174985 | |
| }, | |
| "harness|bbh:temporal_sequences|3": { | |
| "em": 0.152, | |
| "em_stderr": 0.022752024491765464, | |
| "qem": 0.152, | |
| "qem_stderr": 0.022752024491765464, | |
| "pem": 0.156, | |
| "pem_stderr": 0.022995023034068748, | |
| "pqem": 0.428, | |
| "pqem_stderr": 0.03135596892377261, | |
| "perfect_em": 0.152, | |
| "perfect_em_stderr": 0.022752024491765464 | |
| }, | |
| "harness|bbh:tracking_shuffled_objects_five_objects|3": { | |
| "em": 0.148, | |
| "em_stderr": 0.022503547243806144, | |
| "qem": 0.148, | |
| "qem_stderr": 0.022503547243806144, | |
| "pem": 0.148, | |
| "pem_stderr": 0.022503547243806144, | |
| "pqem": 0.348, | |
| "pqem_stderr": 0.030186568464511707, | |
| "perfect_em": 0.148, | |
| "perfect_em_stderr": 0.022503547243806144 | |
| }, | |
| "harness|bbh:tracking_shuffled_objects_seven_objects|3": { | |
| "em": 0.108, | |
| "em_stderr": 0.019669559381568755, | |
| "qem": 0.108, | |
| "qem_stderr": 0.019669559381568755, | |
| "pem": 0.116, | |
| "pem_stderr": 0.020293429803083844, | |
| "pqem": 0.26, | |
| "pqem_stderr": 0.027797315752644308, | |
| "perfect_em": 0.108, | |
| "perfect_em_stderr": 0.019669559381568755 | |
| }, | |
| "harness|bbh:tracking_shuffled_objects_three_objects|3": { | |
| "em": 0.34, | |
| "em_stderr": 0.030020073605457904, | |
| "qem": 0.34, | |
| "qem_stderr": 0.030020073605457904, | |
| "pem": 0.34, | |
| "pem_stderr": 0.030020073605457904, | |
| "pqem": 0.652, | |
| "pqem_stderr": 0.030186568464511683, | |
| "perfect_em": 0.34, | |
| "perfect_em_stderr": 0.030020073605457904 | |
| }, | |
| "harness|bbh:_average|3": { | |
| "em": 0.3495018245613396, | |
| "em_stderr": 0.028584536349911933, | |
| "qem": 0.3495018245613396, | |
| "qem_stderr": 0.028584536349911933, | |
| "pem": 0.41341311034688205, | |
| "pem_stderr": 0.029262861797096727, | |
| "pqem": 0.5470228039236539, | |
| "pqem_stderr": 0.030233226421237065, | |
| "perfect_em": 0.3495018245613396, | |
| "perfect_em_stderr": 0.028584536349911933 | |
| } | |
| }, | |
| "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": "fa8168f39a475fb0", | |
| "hash_input_tokens": "787f75e06fd43c0d", | |
| "hash_cont_tokens": "02b5f6876479ee6f" | |
| }, | |
| "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": "2cceeea606638d49", | |
| "hash_input_tokens": "10c13d6fb8af7c22", | |
| "hash_cont_tokens": "da80e7905b3d4564" | |
| }, | |
| "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": "d8f1ba70c22ae578", | |
| "hash_input_tokens": "c21a88707f480cab", | |
| "hash_cont_tokens": "6e64a77989631706" | |
| }, | |
| "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": "52a60ed1d0113b8b", | |
| "hash_input_tokens": "10e113b2cf3fa584", | |
| "hash_cont_tokens": "70da2b36605721bc" | |
| }, | |
| "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": "253aa9791c941909", | |
| "hash_input_tokens": "0bc166cab0aed76a", | |
| "hash_cont_tokens": "e0f40cb91e3668e4" | |
| }, | |
| "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": "aa6117f601cd268e", | |
| "hash_input_tokens": "ab99c78b48e3a0bb", | |
| "hash_cont_tokens": "e15546cc120f758f" | |
| }, | |
| "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": "1892b050bc7848a4", | |
| "hash_input_tokens": "a720b56aa7c52551", | |
| "hash_cont_tokens": "19ceb44a47d6a83e" | |
| }, | |
| "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": "8e00606ed3407167", | |
| "hash_input_tokens": "c825ab1c99245a17", | |
| "hash_cont_tokens": "3db5d490f746e528" | |
| }, | |
| "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": "8d50c5baf1df7aef", | |
| "hash_input_tokens": "f234e6b28ea1fa49", | |
| "hash_cont_tokens": "b75cef2daa7b5aa7" | |
| }, | |
| "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": "3d2441a21c12a960", | |
| "hash_input_tokens": "f3b577892955aa84", | |
| "hash_cont_tokens": "3a39b267c7438e86" | |
| }, | |
| "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": "ba95caa786f313b1", | |
| "hash_input_tokens": "9954b30d4205604a", | |
| "hash_cont_tokens": "f0ed57d273446ff6" | |
| }, | |
| "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": "a8512d174e1cab8f", | |
| "hash_input_tokens": "3e738df24b7eddf8", | |
| "hash_cont_tokens": "303a352b96e7ada1" | |
| }, | |
| "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": "ff91d81466b9041f", | |
| "hash_input_tokens": "21388b09e13d0208", | |
| "hash_cont_tokens": "f1b8acd779f9be16" | |
| }, | |
| "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": "a59324d9eb37e0f5", | |
| "hash_input_tokens": "0ad41bb8d2290a5b", | |
| "hash_cont_tokens": "4349ad0a7c0ef88c" | |
| }, | |
| "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": "1b3971192bf481e7", | |
| "hash_input_tokens": "3051b60940ccceab", | |
| "hash_cont_tokens": "672a26a849c36b02" | |
| }, | |
| "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": "7ef4567d2fcf5094", | |
| "hash_input_tokens": "b841310ee5531238", | |
| "hash_cont_tokens": "77d91b7062568db8" | |
| }, | |
| "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": "196a0f8712857624", | |
| "hash_input_tokens": "3e738df24b7eddf8", | |
| "hash_cont_tokens": "7451006fd3eba0cf" | |
| }, | |
| "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": "592a03f0518f17b6", | |
| "hash_input_tokens": "19e0ef1dd5ae9d33", | |
| "hash_cont_tokens": "3c3725d0040edbd9" | |
| }, | |
| "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": "0d80ce968d89d4ef", | |
| "hash_input_tokens": "72bda1e7aeb34786", | |
| "hash_cont_tokens": "72154389368339d0" | |
| }, | |
| "truncated": 4362, | |
| "non_truncated": 0, | |
| "padded": 0, | |
| "non_padded": 4362, | |
| "num_truncated_few_shots": 0 | |
| } | |
| } |