| { |
| "model_id": "m14", |
| "checkpoint": "results/m14/final.pt", |
| "trained_step": 15258, |
| "num_fewshot": 0, |
| "limit": null, |
| "results": { |
| "lambada_openai": { |
| "name": "lambada_openai", |
| "alias": "lambada_openai", |
| "sample_len": 5153, |
| "perplexity,none": 404.5905237231641, |
| "perplexity_stderr,none": 21.19147790393626, |
| "acc,none": 0.18105957694546865, |
| "acc_stderr,none": 0.005364743691197477 |
| }, |
| "hellaswag": { |
| "name": "hellaswag", |
| "alias": "hellaswag", |
| "sample_len": 10042, |
| "acc,none": 0.2801234813782115, |
| "acc_stderr,none": 0.004481416180499928, |
| "acc_norm,none": 0.293168691495718, |
| "acc_norm_stderr,none": 0.0045428479356060215 |
| }, |
| "piqa": { |
| "name": "piqa", |
| "alias": "piqa", |
| "sample_len": 1838, |
| "acc,none": 0.6093579978237215, |
| "acc_stderr,none": 0.011383377760053567, |
| "acc_norm,none": 0.6066376496191512, |
| "acc_norm_stderr,none": 0.011397418546689315 |
| }, |
| "sciq": { |
| "name": "sciq", |
| "alias": "sciq", |
| "sample_len": 1000, |
| "acc,none": 0.685, |
| "acc_stderr,none": 0.014696631960792617, |
| "acc_norm,none": 0.594, |
| "acc_norm_stderr,none": 0.01553722643863458 |
| }, |
| "arc_easy": { |
| "name": "arc_easy", |
| "alias": "arc_easy", |
| "sample_len": 2376, |
| "acc,none": 0.4654882154882155, |
| "acc_stderr,none": 0.010235314238969277, |
| "acc_norm,none": 0.4313973063973064, |
| "acc_norm_stderr,none": 0.010162752847747605 |
| }, |
| "winogrande": { |
| "name": "winogrande", |
| "alias": "winogrande", |
| "sample_len": 1267, |
| "acc,none": 0.5122336227308603, |
| "acc_stderr,none": 0.014048278820405729 |
| }, |
| "blimp_adjunct_island": { |
| "name": "blimp_adjunct_island", |
| "alias": "blimp_adjunct_island", |
| "sample_len": 1000, |
| "acc,none": 0.804, |
| "acc_stderr,none": 0.012559527926707347 |
| }, |
| "blimp_anaphor_gender_agreement": { |
| "name": "blimp_anaphor_gender_agreement", |
| "alias": "blimp_anaphor_gender_agreement", |
| "sample_len": 1000, |
| "acc,none": 0.962, |
| "acc_stderr,none": 0.0060491811505849775 |
| }, |
| "blimp_anaphor_number_agreement": { |
| "name": "blimp_anaphor_number_agreement", |
| "alias": "blimp_anaphor_number_agreement", |
| "sample_len": 1000, |
| "acc,none": 0.984, |
| "acc_stderr,none": 0.003969856390319398 |
| }, |
| "blimp_animate_subject_passive": { |
| "name": "blimp_animate_subject_passive", |
| "alias": "blimp_animate_subject_passive", |
| "sample_len": 1000, |
| "acc,none": 0.719, |
| "acc_stderr,none": 0.01422115470843494 |
| }, |
| "blimp_animate_subject_trans": { |
| "name": "blimp_animate_subject_trans", |
| "alias": "blimp_animate_subject_trans", |
| "sample_len": 1000, |
| "acc,none": 0.907, |
| "acc_stderr,none": 0.009188875634996669 |
| }, |
| "blimp_causative": { |
| "name": "blimp_causative", |
| "alias": "blimp_causative", |
| "sample_len": 1000, |
| "acc,none": 0.749, |
| "acc_stderr,none": 0.013718133516888775 |
| }, |
| "blimp_complex_NP_island": { |
| "name": "blimp_complex_NP_island", |
| "alias": "blimp_complex_NP_island", |
| "sample_len": 1000, |
| "acc,none": 0.522, |
| "acc_stderr,none": 0.01580397942816194 |
| }, |
| "blimp_coordinate_structure_constraint_complex_left_branch": { |
| "name": "blimp_coordinate_structure_constraint_complex_left_branch", |
| "alias": "blimp_coordinate_structure_constraint_complex_left_branch", |
| "sample_len": 1000, |
| "acc,none": 0.511, |
| "acc_stderr,none": 0.01581547119529257 |
| }, |
| "blimp_coordinate_structure_constraint_object_extraction": { |
| "name": "blimp_coordinate_structure_constraint_object_extraction", |
| "alias": "blimp_coordinate_structure_constraint_object_extraction", |
| "sample_len": 1000, |
| "acc,none": 0.816, |
| "acc_stderr,none": 0.01225945734093864 |
| }, |
| "blimp_determiner_noun_agreement_1": { |
| "name": "blimp_determiner_noun_agreement_1", |
| "alias": "blimp_determiner_noun_agreement_1", |
| "sample_len": 1000, |
| "acc,none": 0.976, |
| "acc_stderr,none": 0.004842256441727084 |
| }, |
| "blimp_determiner_noun_agreement_2": { |
| "name": "blimp_determiner_noun_agreement_2", |
| "alias": "blimp_determiner_noun_agreement_2", |
| "sample_len": 1000, |
| "acc,none": 0.954, |
| "acc_stderr,none": 0.006627814717380679 |
| }, |
| "blimp_determiner_noun_agreement_irregular_1": { |
| "name": "blimp_determiner_noun_agreement_irregular_1", |
| "alias": "blimp_determiner_noun_agreement_irregular_1", |
| "sample_len": 1000, |
| "acc,none": 0.903, |
| "acc_stderr,none": 0.009363689373248062 |
| }, |
| "blimp_determiner_noun_agreement_irregular_2": { |
| "name": "blimp_determiner_noun_agreement_irregular_2", |
| "alias": "blimp_determiner_noun_agreement_irregular_2", |
| "sample_len": 1000, |
| "acc,none": 0.931, |
| "acc_stderr,none": 0.008018934050315124 |
| }, |
| "blimp_determiner_noun_agreement_with_adj_2": { |
| "name": "blimp_determiner_noun_agreement_with_adj_2", |
| "alias": "blimp_determiner_noun_agreement_with_adj_2", |
| "sample_len": 1000, |
| "acc,none": 0.927, |
| "acc_stderr,none": 0.008230354715243976 |
| }, |
| "blimp_determiner_noun_agreement_with_adj_irregular_1": { |
| "name": "blimp_determiner_noun_agreement_with_adj_irregular_1", |
| "alias": "blimp_determiner_noun_agreement_with_adj_irregular_1", |
| "sample_len": 1000, |
| "acc,none": 0.878, |
| "acc_stderr,none": 0.010354864712936776 |
| }, |
| "blimp_determiner_noun_agreement_with_adj_irregular_2": { |
| "name": "blimp_determiner_noun_agreement_with_adj_irregular_2", |
| "alias": "blimp_determiner_noun_agreement_with_adj_irregular_2", |
| "sample_len": 1000, |
| "acc,none": 0.891, |
| "acc_stderr,none": 0.009859828407037124 |
| }, |
| "blimp_determiner_noun_agreement_with_adjective_1": { |
| "name": "blimp_determiner_noun_agreement_with_adjective_1", |
| "alias": "blimp_determiner_noun_agreement_with_adjective_1", |
| "sample_len": 1000, |
| "acc,none": 0.934, |
| "acc_stderr,none": 0.007855297938697532 |
| }, |
| "blimp_distractor_agreement_relational_noun": { |
| "name": "blimp_distractor_agreement_relational_noun", |
| "alias": "blimp_distractor_agreement_relational_noun", |
| "sample_len": 1000, |
| "acc,none": 0.757, |
| "acc_stderr,none": 0.013569640199177543 |
| }, |
| "blimp_distractor_agreement_relative_clause": { |
| "name": "blimp_distractor_agreement_relative_clause", |
| "alias": "blimp_distractor_agreement_relative_clause", |
| "sample_len": 1000, |
| "acc,none": 0.617, |
| "acc_stderr,none": 0.0153801023256527 |
| }, |
| "blimp_drop_argument": { |
| "name": "blimp_drop_argument", |
| "alias": "blimp_drop_argument", |
| "sample_len": 1000, |
| "acc,none": 0.771, |
| "acc_stderr,none": 0.013294199326613684 |
| }, |
| "blimp_ellipsis_n_bar_1": { |
| "name": "blimp_ellipsis_n_bar_1", |
| "alias": "blimp_ellipsis_n_bar_1", |
| "sample_len": 1000, |
| "acc,none": 0.801, |
| "acc_stderr,none": 0.01263164908309911 |
| }, |
| "blimp_ellipsis_n_bar_2": { |
| "name": "blimp_ellipsis_n_bar_2", |
| "alias": "blimp_ellipsis_n_bar_2", |
| "sample_len": 1000, |
| "acc,none": 0.838, |
| "acc_stderr,none": 0.011657267771304384 |
| }, |
| "blimp_existential_there_object_raising": { |
| "name": "blimp_existential_there_object_raising", |
| "alias": "blimp_existential_there_object_raising", |
| "sample_len": 1000, |
| "acc,none": 0.774, |
| "acc_stderr,none": 0.013232501619085452 |
| }, |
| "blimp_existential_there_quantifiers_1": { |
| "name": "blimp_existential_there_quantifiers_1", |
| "alias": "blimp_existential_there_quantifiers_1", |
| "sample_len": 1000, |
| "acc,none": 0.97, |
| "acc_stderr,none": 0.0053971408290991564 |
| }, |
| "blimp_existential_there_quantifiers_2": { |
| "name": "blimp_existential_there_quantifiers_2", |
| "alias": "blimp_existential_there_quantifiers_2", |
| "sample_len": 1000, |
| "acc,none": 0.326, |
| "acc_stderr,none": 0.014830507204541047 |
| }, |
| "blimp_existential_there_subject_raising": { |
| "name": "blimp_existential_there_subject_raising", |
| "alias": "blimp_existential_there_subject_raising", |
| "sample_len": 1000, |
| "acc,none": 0.845, |
| "acc_stderr,none": 0.011450157470799522 |
| }, |
| "blimp_expletive_it_object_raising": { |
| "name": "blimp_expletive_it_object_raising", |
| "alias": "blimp_expletive_it_object_raising", |
| "sample_len": 1000, |
| "acc,none": 0.791, |
| "acc_stderr,none": 0.012864077288499337 |
| }, |
| "blimp_inchoative": { |
| "name": "blimp_inchoative", |
| "alias": "blimp_inchoative", |
| "sample_len": 1000, |
| "acc,none": 0.691, |
| "acc_stderr,none": 0.014619600977206347 |
| }, |
| "blimp_intransitive": { |
| "name": "blimp_intransitive", |
| "alias": "blimp_intransitive", |
| "sample_len": 1000, |
| "acc,none": 0.782, |
| "acc_stderr,none": 0.013063179040595235 |
| }, |
| "blimp_irregular_past_participle_adjectives": { |
| "name": "blimp_irregular_past_participle_adjectives", |
| "alias": "blimp_irregular_past_participle_adjectives", |
| "sample_len": 1000, |
| "acc,none": 0.934, |
| "acc_stderr,none": 0.007855297938697532 |
| }, |
| "blimp_irregular_past_participle_verbs": { |
| "name": "blimp_irregular_past_participle_verbs", |
| "alias": "blimp_irregular_past_participle_verbs", |
| "sample_len": 1000, |
| "acc,none": 0.9, |
| "acc_stderr,none": 0.00949157995752499 |
| }, |
| "blimp_irregular_plural_subject_verb_agreement_1": { |
| "name": "blimp_irregular_plural_subject_verb_agreement_1", |
| "alias": "blimp_irregular_plural_subject_verb_agreement_1", |
| "sample_len": 1000, |
| "acc,none": 0.911, |
| "acc_stderr,none": 0.009008893392651516 |
| }, |
| "blimp_irregular_plural_subject_verb_agreement_2": { |
| "name": "blimp_irregular_plural_subject_verb_agreement_2", |
| "alias": "blimp_irregular_plural_subject_verb_agreement_2", |
| "sample_len": 1000, |
| "acc,none": 0.896, |
| "acc_stderr,none": 0.009658016218524242 |
| }, |
| "blimp_left_branch_island_echo_question": { |
| "name": "blimp_left_branch_island_echo_question", |
| "alias": "blimp_left_branch_island_echo_question", |
| "sample_len": 1000, |
| "acc,none": 0.576, |
| "acc_stderr,none": 0.015635487471405186 |
| }, |
| "blimp_left_branch_island_simple_question": { |
| "name": "blimp_left_branch_island_simple_question", |
| "alias": "blimp_left_branch_island_simple_question", |
| "sample_len": 1000, |
| "acc,none": 0.615, |
| "acc_stderr,none": 0.015395194445410945 |
| }, |
| "blimp_matrix_question_npi_licensor_present": { |
| "name": "blimp_matrix_question_npi_licensor_present", |
| "alias": "blimp_matrix_question_npi_licensor_present", |
| "sample_len": 1000, |
| "acc,none": 0.521, |
| "acc_stderr,none": 0.015805341148131185 |
| }, |
| "blimp_npi_present_1": { |
| "name": "blimp_npi_present_1", |
| "alias": "blimp_npi_present_1", |
| "sample_len": 1000, |
| "acc,none": 0.614, |
| "acc_stderr,none": 0.015402637476784335 |
| }, |
| "blimp_npi_present_2": { |
| "name": "blimp_npi_present_2", |
| "alias": "blimp_npi_present_2", |
| "sample_len": 1000, |
| "acc,none": 0.691, |
| "acc_stderr,none": 0.014619600977206347 |
| }, |
| "blimp_only_npi_licensor_present": { |
| "name": "blimp_only_npi_licensor_present", |
| "alias": "blimp_only_npi_licensor_present", |
| "sample_len": 1000, |
| "acc,none": 0.976, |
| "acc_stderr,none": 0.004842256441727084 |
| }, |
| "blimp_only_npi_scope": { |
| "name": "blimp_only_npi_scope", |
| "alias": "blimp_only_npi_scope", |
| "sample_len": 1000, |
| "acc,none": 0.582, |
| "acc_stderr,none": 0.015605111967541904 |
| }, |
| "blimp_passive_1": { |
| "name": "blimp_passive_1", |
| "alias": "blimp_passive_1", |
| "sample_len": 1000, |
| "acc,none": 0.87, |
| "acc_stderr,none": 0.010640169792499236 |
| }, |
| "blimp_passive_2": { |
| "name": "blimp_passive_2", |
| "alias": "blimp_passive_2", |
| "sample_len": 1000, |
| "acc,none": 0.825, |
| "acc_stderr,none": 0.012021627157731998 |
| }, |
| "blimp_principle_A_c_command": { |
| "name": "blimp_principle_A_c_command", |
| "alias": "blimp_principle_A_c_command", |
| "sample_len": 1000, |
| "acc,none": 0.58, |
| "acc_stderr,none": 0.015615500115072966 |
| }, |
| "blimp_principle_A_case_1": { |
| "name": "blimp_principle_A_case_1", |
| "alias": "blimp_principle_A_case_1", |
| "sample_len": 1000, |
| "acc,none": 1.0, |
| "acc_stderr,none": 0.0 |
| }, |
| "blimp_principle_A_case_2": { |
| "name": "blimp_principle_A_case_2", |
| "alias": "blimp_principle_A_case_2", |
| "sample_len": 1000, |
| "acc,none": 0.915, |
| "acc_stderr,none": 0.008823426366942316 |
| }, |
| "blimp_principle_A_domain_1": { |
| "name": "blimp_principle_A_domain_1", |
| "alias": "blimp_principle_A_domain_1", |
| "sample_len": 1000, |
| "acc,none": 0.877, |
| "acc_stderr,none": 0.010391293421849803 |
| }, |
| "blimp_principle_A_domain_2": { |
| "name": "blimp_principle_A_domain_2", |
| "alias": "blimp_principle_A_domain_2", |
| "sample_len": 1000, |
| "acc,none": 0.764, |
| "acc_stderr,none": 0.013434451402438602 |
| }, |
| "blimp_principle_A_domain_3": { |
| "name": "blimp_principle_A_domain_3", |
| "alias": "blimp_principle_A_domain_3", |
| "sample_len": 1000, |
| "acc,none": 0.637, |
| "acc_stderr,none": 0.01521389044467136 |
| }, |
| "blimp_principle_A_reconstruction": { |
| "name": "blimp_principle_A_reconstruction", |
| "alias": "blimp_principle_A_reconstruction", |
| "sample_len": 1000, |
| "acc,none": 0.51, |
| "acc_stderr,none": 0.015816135752773183 |
| }, |
| "blimp_regular_plural_subject_verb_agreement_1": { |
| "name": "blimp_regular_plural_subject_verb_agreement_1", |
| "alias": "blimp_regular_plural_subject_verb_agreement_1", |
| "sample_len": 1000, |
| "acc,none": 0.92, |
| "acc_stderr,none": 0.00858333697775368 |
| }, |
| "blimp_regular_plural_subject_verb_agreement_2": { |
| "name": "blimp_regular_plural_subject_verb_agreement_2", |
| "alias": "blimp_regular_plural_subject_verb_agreement_2", |
| "sample_len": 1000, |
| "acc,none": 0.878, |
| "acc_stderr,none": 0.010354864712936776 |
| }, |
| "blimp_sentential_negation_npi_licensor_present": { |
| "name": "blimp_sentential_negation_npi_licensor_present", |
| "alias": "blimp_sentential_negation_npi_licensor_present", |
| "sample_len": 1000, |
| "acc,none": 0.987, |
| "acc_stderr,none": 0.003583830889403662 |
| }, |
| "blimp_sentential_negation_npi_scope": { |
| "name": "blimp_sentential_negation_npi_scope", |
| "alias": "blimp_sentential_negation_npi_scope", |
| "sample_len": 1000, |
| "acc,none": 0.652, |
| "acc_stderr,none": 0.015070604603768328 |
| }, |
| "blimp_sentential_subject_island": { |
| "name": "blimp_sentential_subject_island", |
| "alias": "blimp_sentential_subject_island", |
| "sample_len": 1000, |
| "acc,none": 0.506, |
| "acc_stderr,none": 0.015818160898606836 |
| }, |
| "blimp_superlative_quantifiers_1": { |
| "name": "blimp_superlative_quantifiers_1", |
| "alias": "blimp_superlative_quantifiers_1", |
| "sample_len": 1000, |
| "acc,none": 0.934, |
| "acc_stderr,none": 0.007855297938697532 |
| }, |
| "blimp_superlative_quantifiers_2": { |
| "name": "blimp_superlative_quantifiers_2", |
| "alias": "blimp_superlative_quantifiers_2", |
| "sample_len": 1000, |
| "acc,none": 0.683, |
| "acc_stderr,none": 0.014721675438880174 |
| }, |
| "blimp_tough_vs_raising_1": { |
| "name": "blimp_tough_vs_raising_1", |
| "alias": "blimp_tough_vs_raising_1", |
| "sample_len": 1000, |
| "acc,none": 0.673, |
| "acc_stderr,none": 0.014842213153411162 |
| }, |
| "blimp_tough_vs_raising_2": { |
| "name": "blimp_tough_vs_raising_2", |
| "alias": "blimp_tough_vs_raising_2", |
| "sample_len": 1000, |
| "acc,none": 0.876, |
| "acc_stderr,none": 0.010427498872343873 |
| }, |
| "blimp_transitive": { |
| "name": "blimp_transitive", |
| "alias": "blimp_transitive", |
| "sample_len": 1000, |
| "acc,none": 0.809, |
| "acc_stderr,none": 0.012436787112179482 |
| }, |
| "blimp_wh_island": { |
| "name": "blimp_wh_island", |
| "alias": "blimp_wh_island", |
| "sample_len": 1000, |
| "acc,none": 0.855, |
| "acc_stderr,none": 0.0111399775178902 |
| }, |
| "blimp_wh_questions_object_gap": { |
| "name": "blimp_wh_questions_object_gap", |
| "alias": "blimp_wh_questions_object_gap", |
| "sample_len": 1000, |
| "acc,none": 0.767, |
| "acc_stderr,none": 0.01337497251921999 |
| }, |
| "blimp_wh_questions_subject_gap": { |
| "name": "blimp_wh_questions_subject_gap", |
| "alias": "blimp_wh_questions_subject_gap", |
| "sample_len": 1000, |
| "acc,none": 0.917, |
| "acc_stderr,none": 0.008728527206074756 |
| }, |
| "blimp_wh_questions_subject_gap_long_distance": { |
| "name": "blimp_wh_questions_subject_gap_long_distance", |
| "alias": "blimp_wh_questions_subject_gap_long_distance", |
| "sample_len": 1000, |
| "acc,none": 0.847, |
| "acc_stderr,none": 0.011389500459665504 |
| }, |
| "blimp_wh_vs_that_no_gap": { |
| "name": "blimp_wh_vs_that_no_gap", |
| "alias": "blimp_wh_vs_that_no_gap", |
| "sample_len": 1000, |
| "acc,none": 0.969, |
| "acc_stderr,none": 0.005483527064679181 |
| }, |
| "blimp_wh_vs_that_no_gap_long_distance": { |
| "name": "blimp_wh_vs_that_no_gap_long_distance", |
| "alias": "blimp_wh_vs_that_no_gap_long_distance", |
| "sample_len": 1000, |
| "acc,none": 0.95, |
| "acc_stderr,none": 0.006895472974897918 |
| }, |
| "blimp_wh_vs_that_with_gap": { |
| "name": "blimp_wh_vs_that_with_gap", |
| "alias": "blimp_wh_vs_that_with_gap", |
| "sample_len": 1000, |
| "acc,none": 0.528, |
| "acc_stderr,none": 0.015794475789511517 |
| }, |
| "blimp_wh_vs_that_with_gap_long_distance": { |
| "name": "blimp_wh_vs_that_with_gap_long_distance", |
| "alias": "blimp_wh_vs_that_with_gap_long_distance", |
| "sample_len": 1000, |
| "acc,none": 0.284, |
| "acc_stderr,none": 0.014267009061031241 |
| }, |
| "blimp": { |
| "alias": "blimp", |
| "name": "blimp", |
| "sample_len": 67000, |
| "acc,none": 0.7844776119402985, |
| "acc_stderr,none": 0.0014535278370615976, |
| "sample_count": { |
| "acc,none": 67000 |
| } |
| } |
| } |
| } |
