{ "results": { "blimp": { "acc,none": 0.6829850746268656, "acc_stderr,none": 0.23934230120935937, "alias": "blimp" }, "blimp_adjunct_island": { "acc,none": 0.645, "acc_stderr,none": 0.015139491543780529, "alias": " - blimp_adjunct_island" }, "blimp_anaphor_gender_agreement": { "acc,none": 0.778, "acc_stderr,none": 0.013148721948877364, "alias": " - blimp_anaphor_gender_agreement" }, "blimp_anaphor_number_agreement": { "acc,none": 0.93, "acc_stderr,none": 0.008072494358323502, "alias": " - blimp_anaphor_number_agreement" }, "blimp_animate_subject_passive": { "acc,none": 0.75, "acc_stderr,none": 0.013699915608779773, "alias": " - blimp_animate_subject_passive" }, "blimp_animate_subject_trans": { "acc,none": 0.823, "acc_stderr,none": 0.012075463420375061, "alias": " - blimp_animate_subject_trans" }, "blimp_causative": { "acc,none": 0.67, "acc_stderr,none": 0.014876872027456732, "alias": " - blimp_causative" }, "blimp_complex_NP_island": { "acc,none": 0.46, "acc_stderr,none": 0.01576859691439438, "alias": " - blimp_complex_NP_island" }, "blimp_coordinate_structure_constraint_complex_left_branch": { "acc,none": 0.34, "acc_stderr,none": 0.014987482264363935, "alias": " - blimp_coordinate_structure_constraint_complex_left_branch" }, "blimp_coordinate_structure_constraint_object_extraction": { "acc,none": 0.661, "acc_stderr,none": 0.01497675877162034, "alias": " - blimp_coordinate_structure_constraint_object_extraction" }, "blimp_determiner_noun_agreement_1": { "acc,none": 0.961, "acc_stderr,none": 0.0061250727764261, "alias": " - blimp_determiner_noun_agreement_1" }, "blimp_determiner_noun_agreement_2": { "acc,none": 0.974, "acc_stderr,none": 0.005034813735318201, "alias": " - blimp_determiner_noun_agreement_2" }, "blimp_determiner_noun_agreement_irregular_1": { "acc,none": 0.821, "acc_stderr,none": 0.0121287306057191, "alias": " - blimp_determiner_noun_agreement_irregular_1" }, "blimp_determiner_noun_agreement_irregular_2": { "acc,none": 0.886, "acc_stderr,none": 0.01005510343582333, "alias": " - blimp_determiner_noun_agreement_irregular_2" }, "blimp_determiner_noun_agreement_with_adj_2": { "acc,none": 0.93, "acc_stderr,none": 0.008072494358323497, "alias": " - blimp_determiner_noun_agreement_with_adj_2" }, "blimp_determiner_noun_agreement_with_adj_irregular_1": { "acc,none": 0.781, "acc_stderr,none": 0.013084731950262014, "alias": " - blimp_determiner_noun_agreement_with_adj_irregular_1" }, "blimp_determiner_noun_agreement_with_adj_irregular_2": { "acc,none": 0.853, "acc_stderr,none": 0.011203415395160331, "alias": " - blimp_determiner_noun_agreement_with_adj_irregular_2" }, "blimp_determiner_noun_agreement_with_adjective_1": { "acc,none": 0.943, "acc_stderr,none": 0.007335175853706819, "alias": " - blimp_determiner_noun_agreement_with_adjective_1" }, "blimp_distractor_agreement_relational_noun": { "acc,none": 0.453, "acc_stderr,none": 0.015749255189977593, "alias": " - blimp_distractor_agreement_relational_noun" }, "blimp_distractor_agreement_relative_clause": { "acc,none": 0.405, "acc_stderr,none": 0.015531136990453035, "alias": " - blimp_distractor_agreement_relative_clause" }, "blimp_drop_argument": { "acc,none": 0.721, "acc_stderr,none": 0.014190150117612035, "alias": " - blimp_drop_argument" }, "blimp_ellipsis_n_bar_1": { "acc,none": 0.688, "acc_stderr,none": 0.014658474370509012, "alias": " - blimp_ellipsis_n_bar_1" }, "blimp_ellipsis_n_bar_2": { "acc,none": 0.555, "acc_stderr,none": 0.015723301886760934, "alias": " - blimp_ellipsis_n_bar_2" }, "blimp_existential_there_object_raising": { "acc,none": 0.807, "acc_stderr,none": 0.012486268734370145, "alias": " - blimp_existential_there_object_raising" }, "blimp_existential_there_quantifiers_1": { "acc,none": 0.971, "acc_stderr,none": 0.0053091606857569775, "alias": " - blimp_existential_there_quantifiers_1" }, "blimp_existential_there_quantifiers_2": { "acc,none": 0.95, "acc_stderr,none": 0.006895472974897896, "alias": " - blimp_existential_there_quantifiers_2" }, "blimp_existential_there_subject_raising": { "acc,none": 0.836, "acc_stderr,none": 0.011715000693181309, "alias": " - blimp_existential_there_subject_raising" }, "blimp_expletive_it_object_raising": { "acc,none": 0.734, "acc_stderr,none": 0.013979965645145162, "alias": " - blimp_expletive_it_object_raising" }, "blimp_inchoative": { "acc,none": 0.499, "acc_stderr,none": 0.015819268290576814, "alias": " - blimp_inchoative" }, "blimp_intransitive": { "acc,none": 0.688, "acc_stderr,none": 0.014658474370509008, "alias": " - blimp_intransitive" }, "blimp_irregular_past_participle_adjectives": { "acc,none": 0.882, "acc_stderr,none": 0.010206869264381798, "alias": " - blimp_irregular_past_participle_adjectives" }, "blimp_irregular_past_participle_verbs": { "acc,none": 0.884, "acc_stderr,none": 0.010131468138756986, "alias": " - blimp_irregular_past_participle_verbs" }, "blimp_irregular_plural_subject_verb_agreement_1": { "acc,none": 0.78, "acc_stderr,none": 0.013106173040661778, "alias": " - blimp_irregular_plural_subject_verb_agreement_1" }, "blimp_irregular_plural_subject_verb_agreement_2": { "acc,none": 0.795, "acc_stderr,none": 0.012772554096113126, "alias": " - blimp_irregular_plural_subject_verb_agreement_2" }, "blimp_left_branch_island_echo_question": { "acc,none": 0.384, "acc_stderr,none": 0.015387682761897068, "alias": " - blimp_left_branch_island_echo_question" }, "blimp_left_branch_island_simple_question": { "acc,none": 0.48, "acc_stderr,none": 0.015806639423035167, "alias": " - blimp_left_branch_island_simple_question" }, "blimp_matrix_question_npi_licensor_present": { "acc,none": 0.255, "acc_stderr,none": 0.013790038620872839, "alias": " - blimp_matrix_question_npi_licensor_present" }, "blimp_npi_present_1": { "acc,none": 0.399, "acc_stderr,none": 0.01549319331316291, "alias": " - blimp_npi_present_1" }, "blimp_npi_present_2": { "acc,none": 0.454, "acc_stderr,none": 0.01575221038877184, "alias": " - blimp_npi_present_2" }, "blimp_only_npi_licensor_present": { "acc,none": 0.0, "acc_stderr,none": 0.0, "alias": " - blimp_only_npi_licensor_present" }, "blimp_only_npi_scope": { "acc,none": 0.091, "acc_stderr,none": 0.009099549538400222, "alias": " - blimp_only_npi_scope" }, "blimp_passive_1": { "acc,none": 0.889, "acc_stderr,none": 0.009938701010583726, "alias": " - blimp_passive_1" }, "blimp_passive_2": { "acc,none": 0.883, "acc_stderr,none": 0.010169287802713329, "alias": " - blimp_passive_2" }, "blimp_principle_A_c_command": { "acc,none": 0.499, "acc_stderr,none": 0.015819268290576817, "alias": " - blimp_principle_A_c_command" }, "blimp_principle_A_case_1": { "acc,none": 1.0, "acc_stderr,none": 0.0, "alias": " - blimp_principle_A_case_1" }, "blimp_principle_A_case_2": { "acc,none": 0.914, "acc_stderr,none": 0.008870325962594766, "alias": " - blimp_principle_A_case_2" }, "blimp_principle_A_domain_1": { "acc,none": 0.972, "acc_stderr,none": 0.005219506034410043, "alias": " - blimp_principle_A_domain_1" }, "blimp_principle_A_domain_2": { "acc,none": 0.666, "acc_stderr,none": 0.014922019523732965, "alias": " - blimp_principle_A_domain_2" }, "blimp_principle_A_domain_3": { "acc,none": 0.478, "acc_stderr,none": 0.015803979428161957, "alias": " - blimp_principle_A_domain_3" }, "blimp_principle_A_reconstruction": { "acc,none": 0.208, "acc_stderr,none": 0.012841374572096914, "alias": " - blimp_principle_A_reconstruction" }, "blimp_regular_plural_subject_verb_agreement_1": { "acc,none": 0.878, "acc_stderr,none": 0.010354864712936717, "alias": " - blimp_regular_plural_subject_verb_agreement_1" }, "blimp_regular_plural_subject_verb_agreement_2": { "acc,none": 0.788, "acc_stderr,none": 0.012931481864938038, "alias": " - blimp_regular_plural_subject_verb_agreement_2" }, "blimp_sentential_negation_npi_licensor_present": { "acc,none": 0.999, "acc_stderr,none": 0.0010000000000000115, "alias": " - blimp_sentential_negation_npi_licensor_present" }, "blimp_sentential_negation_npi_scope": { "acc,none": 0.446, "acc_stderr,none": 0.015726771166750354, "alias": " - blimp_sentential_negation_npi_scope" }, "blimp_sentential_subject_island": { "acc,none": 0.368, "acc_stderr,none": 0.0152580735615218, "alias": " - blimp_sentential_subject_island" }, "blimp_superlative_quantifiers_1": { "acc,none": 0.656, "acc_stderr,none": 0.015029633724408947, "alias": " - blimp_superlative_quantifiers_1" }, "blimp_superlative_quantifiers_2": { "acc,none": 0.859, "acc_stderr,none": 0.011010914595992445, "alias": " - blimp_superlative_quantifiers_2" }, "blimp_tough_vs_raising_1": { "acc,none": 0.38, "acc_stderr,none": 0.015356947477797585, "alias": " - blimp_tough_vs_raising_1" }, "blimp_tough_vs_raising_2": { "acc,none": 0.839, "acc_stderr,none": 0.011628164696727204, "alias": " - blimp_tough_vs_raising_2" }, "blimp_transitive": { "acc,none": 0.787, "acc_stderr,none": 0.01295371756673722, "alias": " - blimp_transitive" }, "blimp_wh_island": { "acc,none": 0.617, "acc_stderr,none": 0.015380102325652702, "alias": " - blimp_wh_island" }, "blimp_wh_questions_object_gap": { "acc,none": 0.596, "acc_stderr,none": 0.015524980677122581, "alias": " - blimp_wh_questions_object_gap" }, "blimp_wh_questions_subject_gap": { "acc,none": 0.895, "acc_stderr,none": 0.009698921026024977, "alias": " - blimp_wh_questions_subject_gap" }, "blimp_wh_questions_subject_gap_long_distance": { "acc,none": 0.92, "acc_stderr,none": 0.008583336977753653, "alias": " - blimp_wh_questions_subject_gap_long_distance" }, "blimp_wh_vs_that_no_gap": { "acc,none": 0.961, "acc_stderr,none": 0.006125072776426117, "alias": " - blimp_wh_vs_that_no_gap" }, "blimp_wh_vs_that_no_gap_long_distance": { "acc,none": 0.974, "acc_stderr,none": 0.005034813735318215, "alias": " - blimp_wh_vs_that_no_gap_long_distance" }, "blimp_wh_vs_that_with_gap": { "acc,none": 0.289, "acc_stderr,none": 0.014341711358296184, "alias": " - blimp_wh_vs_that_with_gap" }, "blimp_wh_vs_that_with_gap_long_distance": { "acc,none": 0.082, "acc_stderr,none": 0.00868051561552374, "alias": " - blimp_wh_vs_that_with_gap_long_distance" } }, "groups": { "blimp": { "acc,none": 0.6829850746268656, "acc_stderr,none": 0.23934230120935937, "alias": "blimp" } }, "configs": { "blimp_adjunct_island": { "task": "blimp_adjunct_island", "group": "blimp", "dataset_path": "blimp", "dataset_name": "adjunct_island", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_anaphor_gender_agreement": { "task": "blimp_anaphor_gender_agreement", "group": "blimp", "dataset_path": "blimp", "dataset_name": "anaphor_gender_agreement", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_anaphor_number_agreement": { "task": "blimp_anaphor_number_agreement", "group": "blimp", "dataset_path": "blimp", "dataset_name": "anaphor_number_agreement", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_animate_subject_passive": { "task": "blimp_animate_subject_passive", "group": "blimp", "dataset_path": "blimp", "dataset_name": "animate_subject_passive", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_animate_subject_trans": { "task": "blimp_animate_subject_trans", "group": "blimp", "dataset_path": "blimp", "dataset_name": "animate_subject_trans", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_causative": { "task": "blimp_causative", "group": "blimp", "dataset_path": "blimp", "dataset_name": "causative", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_complex_NP_island": { "task": "blimp_complex_NP_island", "group": "blimp", "dataset_path": "blimp", "dataset_name": "complex_NP_island", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_coordinate_structure_constraint_complex_left_branch": { "task": "blimp_coordinate_structure_constraint_complex_left_branch", "group": "blimp", "dataset_path": "blimp", "dataset_name": "coordinate_structure_constraint_complex_left_branch", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_coordinate_structure_constraint_object_extraction": { "task": "blimp_coordinate_structure_constraint_object_extraction", "group": "blimp", "dataset_path": "blimp", "dataset_name": "coordinate_structure_constraint_object_extraction", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_determiner_noun_agreement_1": { "task": "blimp_determiner_noun_agreement_1", "group": "blimp", "dataset_path": "blimp", "dataset_name": "determiner_noun_agreement_1", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_determiner_noun_agreement_2": { "task": "blimp_determiner_noun_agreement_2", "group": "blimp", "dataset_path": "blimp", "dataset_name": "determiner_noun_agreement_2", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_determiner_noun_agreement_irregular_1": { "task": "blimp_determiner_noun_agreement_irregular_1", "group": "blimp", "dataset_path": "blimp", "dataset_name": "determiner_noun_agreement_irregular_1", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_determiner_noun_agreement_irregular_2": { "task": "blimp_determiner_noun_agreement_irregular_2", "group": "blimp", "dataset_path": "blimp", "dataset_name": "determiner_noun_agreement_irregular_2", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_determiner_noun_agreement_with_adj_2": { "task": "blimp_determiner_noun_agreement_with_adj_2", "group": "blimp", "dataset_path": "blimp", "dataset_name": "determiner_noun_agreement_with_adj_2", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_determiner_noun_agreement_with_adj_irregular_1": { "task": "blimp_determiner_noun_agreement_with_adj_irregular_1", "group": "blimp", "dataset_path": "blimp", "dataset_name": "determiner_noun_agreement_with_adj_irregular_1", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_determiner_noun_agreement_with_adj_irregular_2": { "task": "blimp_determiner_noun_agreement_with_adj_irregular_2", "group": "blimp", "dataset_path": "blimp", "dataset_name": "determiner_noun_agreement_with_adj_irregular_2", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_determiner_noun_agreement_with_adjective_1": { "task": "blimp_determiner_noun_agreement_with_adjective_1", "group": "blimp", "dataset_path": "blimp", "dataset_name": "determiner_noun_agreement_with_adjective_1", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_distractor_agreement_relational_noun": { "task": "blimp_distractor_agreement_relational_noun", "group": "blimp", "dataset_path": "blimp", "dataset_name": "distractor_agreement_relational_noun", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_distractor_agreement_relative_clause": { "task": "blimp_distractor_agreement_relative_clause", "group": "blimp", "dataset_path": "blimp", "dataset_name": "distractor_agreement_relative_clause", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_drop_argument": { "task": "blimp_drop_argument", "group": "blimp", "dataset_path": "blimp", "dataset_name": "drop_argument", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_ellipsis_n_bar_1": { "task": "blimp_ellipsis_n_bar_1", "group": "blimp", "dataset_path": "blimp", "dataset_name": "ellipsis_n_bar_1", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_ellipsis_n_bar_2": { "task": "blimp_ellipsis_n_bar_2", "group": "blimp", "dataset_path": "blimp", "dataset_name": "ellipsis_n_bar_2", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_existential_there_object_raising": { "task": "blimp_existential_there_object_raising", "group": "blimp", "dataset_path": "blimp", "dataset_name": "existential_there_object_raising", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_existential_there_quantifiers_1": { "task": "blimp_existential_there_quantifiers_1", "group": "blimp", "dataset_path": "blimp", "dataset_name": "existential_there_quantifiers_1", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_existential_there_quantifiers_2": { "task": "blimp_existential_there_quantifiers_2", "group": "blimp", "dataset_path": "blimp", "dataset_name": "existential_there_quantifiers_2", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_existential_there_subject_raising": { "task": "blimp_existential_there_subject_raising", "group": "blimp", "dataset_path": "blimp", "dataset_name": "existential_there_subject_raising", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_expletive_it_object_raising": { "task": "blimp_expletive_it_object_raising", "group": "blimp", "dataset_path": "blimp", "dataset_name": "expletive_it_object_raising", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_inchoative": { "task": "blimp_inchoative", "group": "blimp", "dataset_path": "blimp", "dataset_name": "inchoative", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_intransitive": { "task": "blimp_intransitive", "group": "blimp", "dataset_path": "blimp", "dataset_name": "intransitive", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_irregular_past_participle_adjectives": { "task": "blimp_irregular_past_participle_adjectives", "group": "blimp", "dataset_path": "blimp", "dataset_name": "irregular_past_participle_adjectives", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_irregular_past_participle_verbs": { "task": "blimp_irregular_past_participle_verbs", "group": "blimp", "dataset_path": "blimp", "dataset_name": "irregular_past_participle_verbs", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_irregular_plural_subject_verb_agreement_1": { "task": "blimp_irregular_plural_subject_verb_agreement_1", "group": "blimp", "dataset_path": "blimp", "dataset_name": "irregular_plural_subject_verb_agreement_1", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_irregular_plural_subject_verb_agreement_2": { "task": "blimp_irregular_plural_subject_verb_agreement_2", "group": "blimp", "dataset_path": "blimp", "dataset_name": "irregular_plural_subject_verb_agreement_2", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_left_branch_island_echo_question": { "task": "blimp_left_branch_island_echo_question", "group": "blimp", "dataset_path": "blimp", "dataset_name": "left_branch_island_echo_question", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_left_branch_island_simple_question": { "task": "blimp_left_branch_island_simple_question", "group": "blimp", "dataset_path": "blimp", "dataset_name": "left_branch_island_simple_question", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_matrix_question_npi_licensor_present": { "task": "blimp_matrix_question_npi_licensor_present", "group": "blimp", "dataset_path": "blimp", "dataset_name": "matrix_question_npi_licensor_present", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_npi_present_1": { "task": "blimp_npi_present_1", "group": "blimp", "dataset_path": "blimp", "dataset_name": "npi_present_1", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_npi_present_2": { "task": "blimp_npi_present_2", "group": "blimp", "dataset_path": "blimp", "dataset_name": "npi_present_2", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_only_npi_licensor_present": { "task": "blimp_only_npi_licensor_present", "group": "blimp", "dataset_path": "blimp", "dataset_name": "only_npi_licensor_present", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_only_npi_scope": { "task": "blimp_only_npi_scope", "group": "blimp", "dataset_path": "blimp", "dataset_name": "only_npi_scope", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_passive_1": { "task": "blimp_passive_1", "group": "blimp", "dataset_path": "blimp", "dataset_name": "passive_1", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_passive_2": { "task": "blimp_passive_2", "group": "blimp", "dataset_path": "blimp", "dataset_name": "passive_2", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_principle_A_c_command": { "task": "blimp_principle_A_c_command", "group": "blimp", "dataset_path": "blimp", "dataset_name": "principle_A_c_command", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_principle_A_case_1": { "task": "blimp_principle_A_case_1", "group": "blimp", "dataset_path": "blimp", "dataset_name": "principle_A_case_1", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_principle_A_case_2": { "task": "blimp_principle_A_case_2", "group": "blimp", "dataset_path": "blimp", "dataset_name": "principle_A_case_2", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_principle_A_domain_1": { "task": "blimp_principle_A_domain_1", "group": "blimp", "dataset_path": "blimp", "dataset_name": "principle_A_domain_1", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_principle_A_domain_2": { "task": "blimp_principle_A_domain_2", "group": "blimp", "dataset_path": "blimp", "dataset_name": "principle_A_domain_2", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_principle_A_domain_3": { "task": "blimp_principle_A_domain_3", "group": "blimp", "dataset_path": "blimp", "dataset_name": "principle_A_domain_3", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_principle_A_reconstruction": { "task": "blimp_principle_A_reconstruction", "group": "blimp", "dataset_path": "blimp", "dataset_name": "principle_A_reconstruction", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_regular_plural_subject_verb_agreement_1": { "task": "blimp_regular_plural_subject_verb_agreement_1", "group": "blimp", "dataset_path": "blimp", "dataset_name": "regular_plural_subject_verb_agreement_1", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_regular_plural_subject_verb_agreement_2": { "task": "blimp_regular_plural_subject_verb_agreement_2", "group": "blimp", "dataset_path": "blimp", "dataset_name": "regular_plural_subject_verb_agreement_2", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_sentential_negation_npi_licensor_present": { "task": "blimp_sentential_negation_npi_licensor_present", "group": "blimp", "dataset_path": "blimp", "dataset_name": "sentential_negation_npi_licensor_present", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_sentential_negation_npi_scope": { "task": "blimp_sentential_negation_npi_scope", "group": "blimp", "dataset_path": "blimp", "dataset_name": "sentential_negation_npi_scope", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_sentential_subject_island": { "task": "blimp_sentential_subject_island", "group": "blimp", "dataset_path": "blimp", "dataset_name": "sentential_subject_island", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_superlative_quantifiers_1": { "task": "blimp_superlative_quantifiers_1", "group": "blimp", "dataset_path": "blimp", "dataset_name": "superlative_quantifiers_1", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_superlative_quantifiers_2": { "task": "blimp_superlative_quantifiers_2", "group": "blimp", "dataset_path": "blimp", "dataset_name": "superlative_quantifiers_2", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_tough_vs_raising_1": { "task": "blimp_tough_vs_raising_1", "group": "blimp", "dataset_path": "blimp", "dataset_name": "tough_vs_raising_1", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_tough_vs_raising_2": { "task": "blimp_tough_vs_raising_2", "group": "blimp", "dataset_path": "blimp", "dataset_name": "tough_vs_raising_2", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_transitive": { "task": "blimp_transitive", "group": "blimp", "dataset_path": "blimp", "dataset_name": "transitive", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_wh_island": { "task": "blimp_wh_island", "group": "blimp", "dataset_path": "blimp", "dataset_name": "wh_island", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_wh_questions_object_gap": { "task": "blimp_wh_questions_object_gap", "group": "blimp", "dataset_path": "blimp", "dataset_name": "wh_questions_object_gap", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_wh_questions_subject_gap": { "task": "blimp_wh_questions_subject_gap", "group": "blimp", "dataset_path": "blimp", "dataset_name": "wh_questions_subject_gap", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_wh_questions_subject_gap_long_distance": { "task": "blimp_wh_questions_subject_gap_long_distance", "group": "blimp", "dataset_path": "blimp", "dataset_name": "wh_questions_subject_gap_long_distance", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_wh_vs_that_no_gap": { "task": "blimp_wh_vs_that_no_gap", "group": "blimp", "dataset_path": "blimp", "dataset_name": "wh_vs_that_no_gap", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_wh_vs_that_no_gap_long_distance": { "task": "blimp_wh_vs_that_no_gap_long_distance", "group": "blimp", "dataset_path": "blimp", "dataset_name": "wh_vs_that_no_gap_long_distance", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_wh_vs_that_with_gap": { "task": "blimp_wh_vs_that_with_gap", "group": "blimp", "dataset_path": "blimp", "dataset_name": "wh_vs_that_with_gap", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } }, "blimp_wh_vs_that_with_gap_long_distance": { "task": "blimp_wh_vs_that_with_gap_long_distance", "group": "blimp", "dataset_path": "blimp", "dataset_name": "wh_vs_that_with_gap_long_distance", "validation_split": "train", "doc_to_text": "", "doc_to_target": 0, "doc_to_choice": "{{[sentence_good, sentence_bad]}}", "description": "", "target_delimiter": " ", "fewshot_delimiter": "\n\n", "num_fewshot": 0, "metric_list": [ { "metric": "acc" } ], "output_type": "multiple_choice", "repeats": 1, "should_decontaminate": true, "doc_to_decontamination_query": "{{sentence_good}} {{sentence_bad}}", "metadata": { "version": 1.0 } } }, "versions": { "blimp": "N/A", "blimp_adjunct_island": 1.0, "blimp_anaphor_gender_agreement": 1.0, "blimp_anaphor_number_agreement": 1.0, "blimp_animate_subject_passive": 1.0, "blimp_animate_subject_trans": 1.0, "blimp_causative": 1.0, "blimp_complex_NP_island": 1.0, "blimp_coordinate_structure_constraint_complex_left_branch": 1.0, "blimp_coordinate_structure_constraint_object_extraction": 1.0, "blimp_determiner_noun_agreement_1": 1.0, "blimp_determiner_noun_agreement_2": 1.0, "blimp_determiner_noun_agreement_irregular_1": 1.0, "blimp_determiner_noun_agreement_irregular_2": 1.0, "blimp_determiner_noun_agreement_with_adj_2": 1.0, "blimp_determiner_noun_agreement_with_adj_irregular_1": 1.0, "blimp_determiner_noun_agreement_with_adj_irregular_2": 1.0, "blimp_determiner_noun_agreement_with_adjective_1": 1.0, "blimp_distractor_agreement_relational_noun": 1.0, "blimp_distractor_agreement_relative_clause": 1.0, "blimp_drop_argument": 1.0, "blimp_ellipsis_n_bar_1": 1.0, "blimp_ellipsis_n_bar_2": 1.0, "blimp_existential_there_object_raising": 1.0, "blimp_existential_there_quantifiers_1": 1.0, "blimp_existential_there_quantifiers_2": 1.0, "blimp_existential_there_subject_raising": 1.0, "blimp_expletive_it_object_raising": 1.0, "blimp_inchoative": 1.0, "blimp_intransitive": 1.0, "blimp_irregular_past_participle_adjectives": 1.0, "blimp_irregular_past_participle_verbs": 1.0, "blimp_irregular_plural_subject_verb_agreement_1": 1.0, "blimp_irregular_plural_subject_verb_agreement_2": 1.0, "blimp_left_branch_island_echo_question": 1.0, "blimp_left_branch_island_simple_question": 1.0, "blimp_matrix_question_npi_licensor_present": 1.0, "blimp_npi_present_1": 1.0, "blimp_npi_present_2": 1.0, "blimp_only_npi_licensor_present": 1.0, "blimp_only_npi_scope": 1.0, "blimp_passive_1": 1.0, "blimp_passive_2": 1.0, "blimp_principle_A_c_command": 1.0, "blimp_principle_A_case_1": 1.0, "blimp_principle_A_case_2": 1.0, "blimp_principle_A_domain_1": 1.0, "blimp_principle_A_domain_2": 1.0, "blimp_principle_A_domain_3": 1.0, "blimp_principle_A_reconstruction": 1.0, "blimp_regular_plural_subject_verb_agreement_1": 1.0, "blimp_regular_plural_subject_verb_agreement_2": 1.0, "blimp_sentential_negation_npi_licensor_present": 1.0, "blimp_sentential_negation_npi_scope": 1.0, "blimp_sentential_subject_island": 1.0, "blimp_superlative_quantifiers_1": 1.0, "blimp_superlative_quantifiers_2": 1.0, "blimp_tough_vs_raising_1": 1.0, "blimp_tough_vs_raising_2": 1.0, "blimp_transitive": 1.0, "blimp_wh_island": 1.0, "blimp_wh_questions_object_gap": 1.0, "blimp_wh_questions_subject_gap": 1.0, "blimp_wh_questions_subject_gap_long_distance": 1.0, "blimp_wh_vs_that_no_gap": 1.0, "blimp_wh_vs_that_no_gap_long_distance": 1.0, "blimp_wh_vs_that_with_gap": 1.0, "blimp_wh_vs_that_with_gap_long_distance": 1.0 }, "n-shot": { "blimp": 0, "blimp_adjunct_island": 0, "blimp_anaphor_gender_agreement": 0, "blimp_anaphor_number_agreement": 0, "blimp_animate_subject_passive": 0, "blimp_animate_subject_trans": 0, "blimp_causative": 0, "blimp_complex_NP_island": 0, "blimp_coordinate_structure_constraint_complex_left_branch": 0, "blimp_coordinate_structure_constraint_object_extraction": 0, "blimp_determiner_noun_agreement_1": 0, "blimp_determiner_noun_agreement_2": 0, "blimp_determiner_noun_agreement_irregular_1": 0, "blimp_determiner_noun_agreement_irregular_2": 0, "blimp_determiner_noun_agreement_with_adj_2": 0, "blimp_determiner_noun_agreement_with_adj_irregular_1": 0, "blimp_determiner_noun_agreement_with_adj_irregular_2": 0, "blimp_determiner_noun_agreement_with_adjective_1": 0, "blimp_distractor_agreement_relational_noun": 0, "blimp_distractor_agreement_relative_clause": 0, "blimp_drop_argument": 0, "blimp_ellipsis_n_bar_1": 0, "blimp_ellipsis_n_bar_2": 0, "blimp_existential_there_object_raising": 0, "blimp_existential_there_quantifiers_1": 0, "blimp_existential_there_quantifiers_2": 0, "blimp_existential_there_subject_raising": 0, "blimp_expletive_it_object_raising": 0, "blimp_inchoative": 0, "blimp_intransitive": 0, "blimp_irregular_past_participle_adjectives": 0, "blimp_irregular_past_participle_verbs": 0, "blimp_irregular_plural_subject_verb_agreement_1": 0, "blimp_irregular_plural_subject_verb_agreement_2": 0, "blimp_left_branch_island_echo_question": 0, "blimp_left_branch_island_simple_question": 0, "blimp_matrix_question_npi_licensor_present": 0, "blimp_npi_present_1": 0, "blimp_npi_present_2": 0, "blimp_only_npi_licensor_present": 0, "blimp_only_npi_scope": 0, "blimp_passive_1": 0, "blimp_passive_2": 0, "blimp_principle_A_c_command": 0, "blimp_principle_A_case_1": 0, "blimp_principle_A_case_2": 0, "blimp_principle_A_domain_1": 0, "blimp_principle_A_domain_2": 0, "blimp_principle_A_domain_3": 0, "blimp_principle_A_reconstruction": 0, "blimp_regular_plural_subject_verb_agreement_1": 0, "blimp_regular_plural_subject_verb_agreement_2": 0, "blimp_sentential_negation_npi_licensor_present": 0, "blimp_sentential_negation_npi_scope": 0, "blimp_sentential_subject_island": 0, "blimp_superlative_quantifiers_1": 0, "blimp_superlative_quantifiers_2": 0, "blimp_tough_vs_raising_1": 0, "blimp_tough_vs_raising_2": 0, "blimp_transitive": 0, "blimp_wh_island": 0, "blimp_wh_questions_object_gap": 0, "blimp_wh_questions_subject_gap": 0, "blimp_wh_questions_subject_gap_long_distance": 0, "blimp_wh_vs_that_no_gap": 0, "blimp_wh_vs_that_no_gap_long_distance": 0, "blimp_wh_vs_that_with_gap": 0, "blimp_wh_vs_that_with_gap_long_distance": 0 }, "config": { "model": "hf", "model_args": "pretrained=/home/bastian/Dokumente/teenie_llamas/models/final_20", "batch_size": 1, "batch_sizes": [], "device": "cuda", "use_cache": null, "limit": null, "bootstrap_iters": 100000, "gen_kwargs": null }, "git_hash": null }