{ "config_general": { "lighteval_sha": "?", "num_fewshot_seeds": 1, "override_batch_size": 1, "max_samples": null, "job_id": "", "start_time": 1256111.789890816, "end_time": 1256925.970734101, "total_evaluation_time_secondes": "814.1808432850521", "model_name": "abacaj/phi-2-super", "model_sha": "f1e578c868e6cc20fb1ea8eeee427ddf6e0e2ee4", "model_dtype": "torch.bfloat16", "model_size": "5.19 GB", "config": null }, "results": { "lighteval|mmlu:abstract_algebra|5": { "acc": 0.31, "acc_stderr": 0.04648231987117316 }, "lighteval|mmlu:anatomy|5": { "acc": 0.4888888888888889, "acc_stderr": 0.04318275491977976 }, "lighteval|mmlu:astronomy|5": { "acc": 0.6052631578947368, "acc_stderr": 0.039777499346220734 }, "lighteval|mmlu:business_ethics|5": { "acc": 0.6, "acc_stderr": 0.049236596391733084 }, "lighteval|mmlu:clinical_knowledge|5": { "acc": 0.6113207547169811, "acc_stderr": 0.03000048544867599 }, "lighteval|mmlu:college_biology|5": { "acc": 0.6458333333333334, "acc_stderr": 0.039994111357535424 }, "lighteval|mmlu:college_chemistry|5": { "acc": 0.39, "acc_stderr": 0.04902071300001975 }, "lighteval|mmlu:college_computer_science|5": { "acc": 0.4, "acc_stderr": 0.049236596391733084 }, "lighteval|mmlu:college_mathematics|5": { "acc": 0.42, "acc_stderr": 0.04960449637488584 }, "lighteval|mmlu:college_medicine|5": { "acc": 0.6069364161849711, "acc_stderr": 0.0372424959581773 }, "lighteval|mmlu:college_physics|5": { "acc": 0.30392156862745096, "acc_stderr": 0.045766654032077636 }, "lighteval|mmlu:computer_security|5": { "acc": 0.68, "acc_stderr": 0.046882617226215034 }, "lighteval|mmlu:conceptual_physics|5": { "acc": 0.5063829787234042, "acc_stderr": 0.03268335899936336 }, "lighteval|mmlu:econometrics|5": { "acc": 0.35964912280701755, "acc_stderr": 0.045144961328736334 }, "lighteval|mmlu:electrical_engineering|5": { "acc": 0.503448275862069, "acc_stderr": 0.04166567577101579 }, "lighteval|mmlu:elementary_mathematics|5": { "acc": 0.4312169312169312, "acc_stderr": 0.025506481698138204 }, "lighteval|mmlu:formal_logic|5": { "acc": 0.40476190476190477, "acc_stderr": 0.04390259265377562 }, "lighteval|mmlu:global_facts|5": { "acc": 0.34, "acc_stderr": 0.04760952285695235 }, "lighteval|mmlu:high_school_biology|5": { "acc": 0.6838709677419355, "acc_stderr": 0.02645087448904277 }, "lighteval|mmlu:high_school_chemistry|5": { "acc": 0.5024630541871922, "acc_stderr": 0.03517945038691063 }, "lighteval|mmlu:high_school_computer_science|5": { "acc": 0.66, "acc_stderr": 0.04760952285695237 }, "lighteval|mmlu:high_school_european_history|5": { "acc": 0.703030303030303, "acc_stderr": 0.0356796977226805 }, "lighteval|mmlu:high_school_geography|5": { "acc": 0.7575757575757576, "acc_stderr": 0.030532892233932026 }, "lighteval|mmlu:high_school_government_and_politics|5": { "acc": 0.8082901554404145, "acc_stderr": 0.028408953626245285 }, "lighteval|mmlu:high_school_macroeconomics|5": { "acc": 0.6230769230769231, "acc_stderr": 0.024570975364225995 }, "lighteval|mmlu:high_school_mathematics|5": { "acc": 0.35185185185185186, "acc_stderr": 0.029116617606083025 }, "lighteval|mmlu:high_school_microeconomics|5": { "acc": 0.6092436974789915, "acc_stderr": 0.031693802357129965 }, "lighteval|mmlu:high_school_physics|5": { "acc": 0.3443708609271523, "acc_stderr": 0.03879687024073327 }, "lighteval|mmlu:high_school_psychology|5": { "acc": 0.7963302752293578, "acc_stderr": 0.017266742087630804 }, "lighteval|mmlu:high_school_statistics|5": { "acc": 0.44907407407407407, "acc_stderr": 0.03392238405321616 }, "lighteval|mmlu:high_school_us_history|5": { "acc": 0.7107843137254902, "acc_stderr": 0.031822318676475544 }, "lighteval|mmlu:high_school_world_history|5": { "acc": 0.729957805907173, "acc_stderr": 0.028900721906293426 }, "lighteval|mmlu:human_aging|5": { "acc": 0.6322869955156951, "acc_stderr": 0.03236198350928275 }, "lighteval|mmlu:human_sexuality|5": { "acc": 0.6793893129770993, "acc_stderr": 0.04093329229834278 }, "lighteval|mmlu:international_law|5": { "acc": 0.7355371900826446, "acc_stderr": 0.040261875275912046 }, "lighteval|mmlu:jurisprudence|5": { "acc": 0.7407407407407407, "acc_stderr": 0.04236511258094633 }, "lighteval|mmlu:logical_fallacies|5": { "acc": 0.7668711656441718, "acc_stderr": 0.033220157957767414 }, "lighteval|mmlu:machine_learning|5": { "acc": 0.4642857142857143, "acc_stderr": 0.04733667890053756 }, "lighteval|mmlu:management|5": { "acc": 0.7669902912621359, "acc_stderr": 0.041858325989283136 }, "lighteval|mmlu:marketing|5": { "acc": 0.8247863247863247, "acc_stderr": 0.02490443909891824 }, "lighteval|mmlu:medical_genetics|5": { "acc": 0.61, "acc_stderr": 0.04902071300001975 }, "lighteval|mmlu:miscellaneous|5": { "acc": 0.7037037037037037, "acc_stderr": 0.016328814422102052 }, "lighteval|mmlu:moral_disputes|5": { "acc": 0.630057803468208, "acc_stderr": 0.025992472029306397 }, "lighteval|mmlu:moral_scenarios|5": { "acc": 0.27262569832402234, "acc_stderr": 0.014893391735249619 }, "lighteval|mmlu:nutrition|5": { "acc": 0.6339869281045751, "acc_stderr": 0.02758281141515961 }, "lighteval|mmlu:philosophy|5": { "acc": 0.6270096463022508, "acc_stderr": 0.027466610213140112 }, "lighteval|mmlu:prehistory|5": { "acc": 0.6296296296296297, "acc_stderr": 0.02686949074481526 }, "lighteval|mmlu:professional_accounting|5": { "acc": 0.4326241134751773, "acc_stderr": 0.02955545423677886 }, "lighteval|mmlu:professional_law|5": { "acc": 0.4172099087353325, "acc_stderr": 0.012593959992906419 }, "lighteval|mmlu:professional_medicine|5": { "acc": 0.4889705882352941, "acc_stderr": 0.030365446477275675 }, "lighteval|mmlu:professional_psychology|5": { "acc": 0.5424836601307189, "acc_stderr": 0.02015468571259089 }, "lighteval|mmlu:public_relations|5": { "acc": 0.7090909090909091, "acc_stderr": 0.04350271442923243 }, "lighteval|mmlu:security_studies|5": { "acc": 0.6938775510204082, "acc_stderr": 0.029504896454595957 }, "lighteval|mmlu:sociology|5": { "acc": 0.7860696517412935, "acc_stderr": 0.028996909693328927 }, "lighteval|mmlu:us_foreign_policy|5": { "acc": 0.76, "acc_stderr": 0.04292346959909281 }, "lighteval|mmlu:virology|5": { "acc": 0.463855421686747, "acc_stderr": 0.03882310850890594 }, "lighteval|mmlu:world_religions|5": { "acc": 0.7017543859649122, "acc_stderr": 0.03508771929824563 }, "lighteval|mmlu:_average|5": { "acc": 0.5798487843538951, "acc_stderr": 0.03518940861065782 } }, "versions": { "lighteval|mmlu:abstract_algebra|5": 0, "lighteval|mmlu:anatomy|5": 0, "lighteval|mmlu:astronomy|5": 0, "lighteval|mmlu:business_ethics|5": 0, "lighteval|mmlu:clinical_knowledge|5": 0, "lighteval|mmlu:college_biology|5": 0, "lighteval|mmlu:college_chemistry|5": 0, "lighteval|mmlu:college_computer_science|5": 0, "lighteval|mmlu:college_mathematics|5": 0, "lighteval|mmlu:college_medicine|5": 0, "lighteval|mmlu:college_physics|5": 0, "lighteval|mmlu:computer_security|5": 0, "lighteval|mmlu:conceptual_physics|5": 0, "lighteval|mmlu:econometrics|5": 0, "lighteval|mmlu:electrical_engineering|5": 0, "lighteval|mmlu:elementary_mathematics|5": 0, "lighteval|mmlu:formal_logic|5": 0, "lighteval|mmlu:global_facts|5": 0, "lighteval|mmlu:high_school_biology|5": 0, "lighteval|mmlu:high_school_chemistry|5": 0, "lighteval|mmlu:high_school_computer_science|5": 0, "lighteval|mmlu:high_school_european_history|5": 0, "lighteval|mmlu:high_school_geography|5": 0, "lighteval|mmlu:high_school_government_and_politics|5": 0, "lighteval|mmlu:high_school_macroeconomics|5": 0, "lighteval|mmlu:high_school_mathematics|5": 0, "lighteval|mmlu:high_school_microeconomics|5": 0, "lighteval|mmlu:high_school_physics|5": 0, "lighteval|mmlu:high_school_psychology|5": 0, "lighteval|mmlu:high_school_statistics|5": 0, "lighteval|mmlu:high_school_us_history|5": 0, "lighteval|mmlu:high_school_world_history|5": 0, "lighteval|mmlu:human_aging|5": 0, "lighteval|mmlu:human_sexuality|5": 0, "lighteval|mmlu:international_law|5": 0, "lighteval|mmlu:jurisprudence|5": 0, "lighteval|mmlu:logical_fallacies|5": 0, "lighteval|mmlu:machine_learning|5": 0, "lighteval|mmlu:management|5": 0, "lighteval|mmlu:marketing|5": 0, "lighteval|mmlu:medical_genetics|5": 0, "lighteval|mmlu:miscellaneous|5": 0, "lighteval|mmlu:moral_disputes|5": 0, "lighteval|mmlu:moral_scenarios|5": 0, "lighteval|mmlu:nutrition|5": 0, "lighteval|mmlu:philosophy|5": 0, "lighteval|mmlu:prehistory|5": 0, "lighteval|mmlu:professional_accounting|5": 0, "lighteval|mmlu:professional_law|5": 0, "lighteval|mmlu:professional_medicine|5": 0, "lighteval|mmlu:professional_psychology|5": 0, "lighteval|mmlu:public_relations|5": 0, "lighteval|mmlu:security_studies|5": 0, "lighteval|mmlu:sociology|5": 0, "lighteval|mmlu:us_foreign_policy|5": 0, "lighteval|mmlu:virology|5": 0, "lighteval|mmlu:world_religions|5": 0 }, "config_tasks": { "lighteval|mmlu:abstract_algebra": { "name": "mmlu:abstract_algebra", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "abstract_algebra", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 100, "effective_num_docs": 100 }, "lighteval|mmlu:anatomy": { "name": "mmlu:anatomy", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "anatomy", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 135, "effective_num_docs": 135 }, "lighteval|mmlu:astronomy": { "name": "mmlu:astronomy", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "astronomy", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 152, "effective_num_docs": 152 }, "lighteval|mmlu:business_ethics": { "name": "mmlu:business_ethics", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "business_ethics", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 100, "effective_num_docs": 100 }, "lighteval|mmlu:clinical_knowledge": { "name": "mmlu:clinical_knowledge", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "clinical_knowledge", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 265, "effective_num_docs": 265 }, "lighteval|mmlu:college_biology": { "name": "mmlu:college_biology", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "college_biology", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 144, "effective_num_docs": 144 }, "lighteval|mmlu:college_chemistry": { "name": "mmlu:college_chemistry", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "college_chemistry", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 100, "effective_num_docs": 100 }, "lighteval|mmlu:college_computer_science": { "name": "mmlu:college_computer_science", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "college_computer_science", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 100, "effective_num_docs": 100 }, "lighteval|mmlu:college_mathematics": { "name": "mmlu:college_mathematics", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "college_mathematics", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 100, "effective_num_docs": 100 }, "lighteval|mmlu:college_medicine": { "name": "mmlu:college_medicine", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "college_medicine", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 173, "effective_num_docs": 173 }, "lighteval|mmlu:college_physics": { "name": "mmlu:college_physics", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "college_physics", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 102, "effective_num_docs": 102 }, "lighteval|mmlu:computer_security": { "name": "mmlu:computer_security", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "computer_security", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 100, "effective_num_docs": 100 }, "lighteval|mmlu:conceptual_physics": { "name": "mmlu:conceptual_physics", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "conceptual_physics", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 235, "effective_num_docs": 235 }, "lighteval|mmlu:econometrics": { "name": "mmlu:econometrics", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "econometrics", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 114, "effective_num_docs": 114 }, "lighteval|mmlu:electrical_engineering": { "name": "mmlu:electrical_engineering", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "electrical_engineering", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 145, "effective_num_docs": 145 }, "lighteval|mmlu:elementary_mathematics": { "name": "mmlu:elementary_mathematics", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "elementary_mathematics", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 378, "effective_num_docs": 378 }, "lighteval|mmlu:formal_logic": { "name": "mmlu:formal_logic", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "formal_logic", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 126, "effective_num_docs": 126 }, "lighteval|mmlu:global_facts": { "name": "mmlu:global_facts", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "global_facts", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 100, "effective_num_docs": 100 }, "lighteval|mmlu:high_school_biology": { "name": "mmlu:high_school_biology", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "high_school_biology", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 310, "effective_num_docs": 310 }, "lighteval|mmlu:high_school_chemistry": { "name": "mmlu:high_school_chemistry", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "high_school_chemistry", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 203, "effective_num_docs": 203 }, "lighteval|mmlu:high_school_computer_science": { "name": "mmlu:high_school_computer_science", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "high_school_computer_science", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 100, "effective_num_docs": 100 }, "lighteval|mmlu:high_school_european_history": { "name": "mmlu:high_school_european_history", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "high_school_european_history", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 165, "effective_num_docs": 165 }, "lighteval|mmlu:high_school_geography": { "name": "mmlu:high_school_geography", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "high_school_geography", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 198, "effective_num_docs": 198 }, "lighteval|mmlu:high_school_government_and_politics": { "name": "mmlu:high_school_government_and_politics", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "high_school_government_and_politics", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 193, "effective_num_docs": 193 }, "lighteval|mmlu:high_school_macroeconomics": { "name": "mmlu:high_school_macroeconomics", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "high_school_macroeconomics", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 390, "effective_num_docs": 390 }, "lighteval|mmlu:high_school_mathematics": { "name": "mmlu:high_school_mathematics", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "high_school_mathematics", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 270, "effective_num_docs": 270 }, "lighteval|mmlu:high_school_microeconomics": { "name": "mmlu:high_school_microeconomics", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "high_school_microeconomics", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 238, "effective_num_docs": 238 }, "lighteval|mmlu:high_school_physics": { "name": "mmlu:high_school_physics", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "high_school_physics", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 151, "effective_num_docs": 151 }, "lighteval|mmlu:high_school_psychology": { "name": "mmlu:high_school_psychology", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "high_school_psychology", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 545, "effective_num_docs": 545 }, "lighteval|mmlu:high_school_statistics": { "name": "mmlu:high_school_statistics", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "high_school_statistics", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 216, "effective_num_docs": 216 }, "lighteval|mmlu:high_school_us_history": { "name": "mmlu:high_school_us_history", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "high_school_us_history", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 204, "effective_num_docs": 204 }, "lighteval|mmlu:high_school_world_history": { "name": "mmlu:high_school_world_history", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "high_school_world_history", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 237, "effective_num_docs": 237 }, "lighteval|mmlu:human_aging": { "name": "mmlu:human_aging", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "human_aging", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 223, "effective_num_docs": 223 }, "lighteval|mmlu:human_sexuality": { "name": "mmlu:human_sexuality", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "human_sexuality", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 131, "effective_num_docs": 131 }, "lighteval|mmlu:international_law": { "name": "mmlu:international_law", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "international_law", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 121, "effective_num_docs": 121 }, "lighteval|mmlu:jurisprudence": { "name": "mmlu:jurisprudence", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "jurisprudence", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 108, "effective_num_docs": 108 }, "lighteval|mmlu:logical_fallacies": { "name": "mmlu:logical_fallacies", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "logical_fallacies", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 163, "effective_num_docs": 163 }, "lighteval|mmlu:machine_learning": { "name": "mmlu:machine_learning", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "machine_learning", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 112, "effective_num_docs": 112 }, "lighteval|mmlu:management": { "name": "mmlu:management", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "management", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 103, "effective_num_docs": 103 }, "lighteval|mmlu:marketing": { "name": "mmlu:marketing", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "marketing", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 234, "effective_num_docs": 234 }, "lighteval|mmlu:medical_genetics": { "name": "mmlu:medical_genetics", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "medical_genetics", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 100, "effective_num_docs": 100 }, "lighteval|mmlu:miscellaneous": { "name": "mmlu:miscellaneous", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "miscellaneous", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 783, "effective_num_docs": 783 }, "lighteval|mmlu:moral_disputes": { "name": "mmlu:moral_disputes", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "moral_disputes", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 346, "effective_num_docs": 346 }, "lighteval|mmlu:moral_scenarios": { "name": "mmlu:moral_scenarios", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "moral_scenarios", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 895, "effective_num_docs": 895 }, "lighteval|mmlu:nutrition": { "name": "mmlu:nutrition", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "nutrition", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 306, "effective_num_docs": 306 }, "lighteval|mmlu:philosophy": { "name": "mmlu:philosophy", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "philosophy", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 311, "effective_num_docs": 311 }, "lighteval|mmlu:prehistory": { "name": "mmlu:prehistory", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "prehistory", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 324, "effective_num_docs": 324 }, "lighteval|mmlu:professional_accounting": { "name": "mmlu:professional_accounting", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "professional_accounting", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 282, "effective_num_docs": 282 }, "lighteval|mmlu:professional_law": { "name": "mmlu:professional_law", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "professional_law", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 1534, "effective_num_docs": 1534 }, "lighteval|mmlu:professional_medicine": { "name": "mmlu:professional_medicine", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "professional_medicine", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 272, "effective_num_docs": 272 }, "lighteval|mmlu:professional_psychology": { "name": "mmlu:professional_psychology", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "professional_psychology", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 612, "effective_num_docs": 612 }, "lighteval|mmlu:public_relations": { "name": "mmlu:public_relations", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "public_relations", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 110, "effective_num_docs": 110 }, "lighteval|mmlu:security_studies": { "name": "mmlu:security_studies", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "security_studies", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 245, "effective_num_docs": 245 }, "lighteval|mmlu:sociology": { "name": "mmlu:sociology", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "sociology", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 201, "effective_num_docs": 201 }, "lighteval|mmlu:us_foreign_policy": { "name": "mmlu:us_foreign_policy", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "us_foreign_policy", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 100, "effective_num_docs": 100 }, "lighteval|mmlu:virology": { "name": "mmlu:virology", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "virology", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 166, "effective_num_docs": 166 }, "lighteval|mmlu:world_religions": { "name": "mmlu:world_religions", "prompt_function": "mmlu_harness", "hf_repo": "lighteval/mmlu", "hf_subset": "world_religions", "metric": [ "loglikelihood_acc" ], "hf_avail_splits": [ "auxiliary_train", "test", "validation", "dev" ], "evaluation_splits": [ "test" ], "few_shots_split": "dev", "few_shots_select": "sequential", "generation_size": 1, "stop_sequence": [ "\n" ], "output_regex": null, "frozen": false, "suite": [ "lighteval", "mmlu" ], "original_num_docs": 171, "effective_num_docs": 171 } }, "summary_tasks": { "lighteval|mmlu:abstract_algebra|5": { "hashes": { "hash_examples": "4c76229e00c9c0e9", "hash_full_prompts": "8560328c37a286da", "hash_input_tokens": "bf08b6654c5ce175", "hash_cont_tokens": "844bd0bf669e8136" }, "truncated": 0, "non_truncated": 100, "padded": 400, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:anatomy|5": { "hashes": { "hash_examples": "6a1f8104dccbd33b", "hash_full_prompts": "07a5b58569dc0c0e", "hash_input_tokens": "ea648227a04a3cc3", "hash_cont_tokens": "aa3ffb1a6e4356f5" }, "truncated": 0, "non_truncated": 135, "padded": 540, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:astronomy|5": { "hashes": { "hash_examples": "1302effa3a76ce4c", "hash_full_prompts": "951d0fa240300281", "hash_input_tokens": "4e52aeeb91a30146", "hash_cont_tokens": "18cfffb76bc8f0d1" }, "truncated": 0, "non_truncated": 152, "padded": 608, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:business_ethics|5": { "hashes": { "hash_examples": "03cb8bce5336419a", "hash_full_prompts": "2878153ae2f0aad4", "hash_input_tokens": "1b745cb0dad581a2", "hash_cont_tokens": "844bd0bf669e8136" }, "truncated": 0, "non_truncated": 100, "padded": 400, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:clinical_knowledge|5": { "hashes": { "hash_examples": "ffbb9c7b2be257f9", "hash_full_prompts": "9d9a3111c002b207", "hash_input_tokens": "846b0c7f392f2be6", "hash_cont_tokens": "cd61f7de0830a75a" }, "truncated": 0, "non_truncated": 265, "padded": 1060, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:college_biology|5": { "hashes": { "hash_examples": "3ee77f176f38eb8e", "hash_full_prompts": "b8bee76c9dba6d68", "hash_input_tokens": "426f7b456868cadc", "hash_cont_tokens": "16b3626c8a5e3797" }, "truncated": 0, "non_truncated": 144, "padded": 576, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:college_chemistry|5": { "hashes": { "hash_examples": "ce61a69c46d47aeb", "hash_full_prompts": "3459dc743c60cea4", "hash_input_tokens": "5db3ef6f010d8bcd", "hash_cont_tokens": "844bd0bf669e8136" }, "truncated": 0, "non_truncated": 100, "padded": 394, "non_padded": 6, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:college_computer_science|5": { "hashes": { "hash_examples": "32805b52d7d5daab", "hash_full_prompts": "df78f0d04538c284", "hash_input_tokens": "b711aaa4c36ca53e", "hash_cont_tokens": "844bd0bf669e8136" }, "truncated": 0, "non_truncated": 100, "padded": 400, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:college_mathematics|5": { "hashes": { "hash_examples": "55da1a0a0bd33722", "hash_full_prompts": "aab9e2bebe16495b", "hash_input_tokens": "605eb966540f6ed2", "hash_cont_tokens": "844bd0bf669e8136" }, "truncated": 0, "non_truncated": 100, "padded": 400, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:college_medicine|5": { "hashes": { "hash_examples": "c33e143163049176", "hash_full_prompts": "725140cc2920f343", "hash_input_tokens": "f98a1bf8729099a3", "hash_cont_tokens": "62bb469d2a319d91" }, "truncated": 0, "non_truncated": 173, "padded": 680, "non_padded": 12, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:college_physics|5": { "hashes": { "hash_examples": "ebdab1cdb7e555df", "hash_full_prompts": "08850f27e43ce208", "hash_input_tokens": "360d63078e885151", "hash_cont_tokens": "bf103c9a1f61ec12" }, "truncated": 0, "non_truncated": 102, "padded": 408, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:computer_security|5": { "hashes": { "hash_examples": "a24fd7d08a560921", "hash_full_prompts": "70ba7f73960b536c", "hash_input_tokens": "8241df1daf843287", "hash_cont_tokens": "844bd0bf669e8136" }, "truncated": 0, "non_truncated": 100, "padded": 400, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:conceptual_physics|5": { "hashes": { "hash_examples": "8300977a79386993", "hash_full_prompts": "7e879ba2829dbeb7", "hash_input_tokens": "018d09540fad98b1", "hash_cont_tokens": "ff5ca3d84bb47a0b" }, "truncated": 0, "non_truncated": 235, "padded": 940, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:econometrics|5": { "hashes": { "hash_examples": "ddde36788a04a46f", "hash_full_prompts": "dd8d0b14bc4b73de", "hash_input_tokens": "3b5f360ce4c4ef56", "hash_cont_tokens": "21f0989f5760198a" }, "truncated": 0, "non_truncated": 114, "padded": 456, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:electrical_engineering|5": { "hashes": { "hash_examples": "acbc5def98c19b3f", "hash_full_prompts": "4ad87171757447a2", "hash_input_tokens": "b881c8ddaa5861c6", "hash_cont_tokens": "35bf6c0c1a7ee403" }, "truncated": 0, "non_truncated": 145, "padded": 580, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:elementary_mathematics|5": { "hashes": { "hash_examples": "146e61d07497a9bd", "hash_full_prompts": "e6ad5f3437016940", "hash_input_tokens": "fb3706917cdcfa28", "hash_cont_tokens": "f7d801bfd913884d" }, "truncated": 0, "non_truncated": 378, "padded": 1496, "non_padded": 16, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:formal_logic|5": { "hashes": { "hash_examples": "8635216e1909a03f", "hash_full_prompts": "1cd9b9c21440d627", "hash_input_tokens": "5f56f6bcfd3d5473", "hash_cont_tokens": "23f9089575432d5a" }, "truncated": 0, "non_truncated": 126, "padded": 504, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:global_facts|5": { "hashes": { "hash_examples": "30b315aa6353ee47", "hash_full_prompts": "435179047fe97427", "hash_input_tokens": "321ed2a4e80ad788", "hash_cont_tokens": "844bd0bf669e8136" }, "truncated": 0, "non_truncated": 100, "padded": 400, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:high_school_biology|5": { "hashes": { "hash_examples": "c9136373af2180de", "hash_full_prompts": "2277f3e0bcc9291d", "hash_input_tokens": "ded71925ac98bdd4", "hash_cont_tokens": "04b8293f2ab7fbbf" }, "truncated": 0, "non_truncated": 310, "padded": 1220, "non_padded": 20, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:high_school_chemistry|5": { "hashes": { "hash_examples": "b0661bfa1add6404", "hash_full_prompts": "dd245889f5b5aebb", "hash_input_tokens": "499b44d887506c20", "hash_cont_tokens": "c3deabee1deab3a3" }, "truncated": 0, "non_truncated": 203, "padded": 812, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:high_school_computer_science|5": { "hashes": { "hash_examples": "80fc1d623a3d665f", "hash_full_prompts": "48878c696523f767", "hash_input_tokens": "25339dd3cc3243bb", "hash_cont_tokens": "844bd0bf669e8136" }, "truncated": 0, "non_truncated": 100, "padded": 400, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:high_school_european_history|5": { "hashes": { "hash_examples": "854da6e5af0fe1a1", "hash_full_prompts": "93884d98c336e0ee", "hash_input_tokens": "0cec093b878a8308", "hash_cont_tokens": "c4f2565ca36881d5" }, "truncated": 660, "non_truncated": -495, "padded": 0, "non_padded": 660, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:high_school_geography|5": { "hashes": { "hash_examples": "7dc963c7acd19ad8", "hash_full_prompts": "24ca3ac3c033d252", "hash_input_tokens": "c69c3f0e8c89ccfc", "hash_cont_tokens": "780e569058de22be" }, "truncated": 0, "non_truncated": 198, "padded": 792, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:high_school_government_and_politics|5": { "hashes": { "hash_examples": "1f675dcdebc9758f", "hash_full_prompts": "898a46330dd6f826", "hash_input_tokens": "b50b930c30f84216", "hash_cont_tokens": "7994d94bfa36d003" }, "truncated": 0, "non_truncated": 193, "padded": 772, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:high_school_macroeconomics|5": { "hashes": { "hash_examples": "2fb32cf2d80f0b35", "hash_full_prompts": "c6ff197ef73e969a", "hash_input_tokens": "60d727ae1bae88a2", "hash_cont_tokens": "8f5c8baf02161f10" }, "truncated": 0, "non_truncated": 390, "padded": 1560, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:high_school_mathematics|5": { "hashes": { "hash_examples": "fd6646fdb5d58a1f", "hash_full_prompts": "1ff366c269c5671c", "hash_input_tokens": "9f5665aa6ac080d7", "hash_cont_tokens": "a2c91752be5b1798" }, "truncated": 0, "non_truncated": 270, "padded": 1064, "non_padded": 16, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:high_school_microeconomics|5": { "hashes": { "hash_examples": "2118f21f71d87d84", "hash_full_prompts": "5aa922958c3e74ab", "hash_input_tokens": "4c58b49637bfe6ec", "hash_cont_tokens": "985403b262df21a4" }, "truncated": 0, "non_truncated": 238, "padded": 952, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:high_school_physics|5": { "hashes": { "hash_examples": "dc3ce06378548565", "hash_full_prompts": "a434155958b8ba4d", "hash_input_tokens": "6208914815bed6c5", "hash_cont_tokens": "db71da66ed82b921" }, "truncated": 0, "non_truncated": 151, "padded": 600, "non_padded": 4, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:high_school_psychology|5": { "hashes": { "hash_examples": "c8d1d98a40e11f2f", "hash_full_prompts": "88756725e3f5d3a4", "hash_input_tokens": "98c88878636fab62", "hash_cont_tokens": "e81cf9738ad7e157" }, "truncated": 0, "non_truncated": 545, "padded": 2168, "non_padded": 12, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:high_school_statistics|5": { "hashes": { "hash_examples": "666c8759b98ee4ff", "hash_full_prompts": "ce2c9ad6242fb787", "hash_input_tokens": "9ca8524d3048f7ce", "hash_cont_tokens": "4a2d5f00cb00d9b7" }, "truncated": 0, "non_truncated": 216, "padded": 864, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:high_school_us_history|5": { "hashes": { "hash_examples": "95fef1c4b7d3f81e", "hash_full_prompts": "6acc502114e95366", "hash_input_tokens": "4d051e2e8115778b", "hash_cont_tokens": "eab825cf8fbdd085" }, "truncated": 816, "non_truncated": -612, "padded": 0, "non_padded": 816, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:high_school_world_history|5": { "hashes": { "hash_examples": "7e5085b6184b0322", "hash_full_prompts": "6f36378c408a4e64", "hash_input_tokens": "08470e967491cfa0", "hash_cont_tokens": "e9bcfaa6beefb456" }, "truncated": 0, "non_truncated": 237, "padded": 948, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:human_aging|5": { "hashes": { "hash_examples": "c17333e7c7c10797", "hash_full_prompts": "5a0ae03a2e92b270", "hash_input_tokens": "db0c9bcb85cbde72", "hash_cont_tokens": "38eafdb22e9fca11" }, "truncated": 0, "non_truncated": 223, "padded": 892, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:human_sexuality|5": { "hashes": { "hash_examples": "4edd1e9045df5e3d", "hash_full_prompts": "87c8aedcf6122d0c", "hash_input_tokens": "aaff5d3845a458ca", "hash_cont_tokens": "11de075f88fc7cd2" }, "truncated": 0, "non_truncated": 131, "padded": 524, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:international_law|5": { "hashes": { "hash_examples": "db2fa00d771a062a", "hash_full_prompts": "1adaf7918f454370", "hash_input_tokens": "4f87ba983046ce69", "hash_cont_tokens": "6f8215a3de7eebd1" }, "truncated": 0, "non_truncated": 121, "padded": 484, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:jurisprudence|5": { "hashes": { "hash_examples": "e956f86b124076fe", "hash_full_prompts": "69c68a735eb3d872", "hash_input_tokens": "8a60a8de04ecd762", "hash_cont_tokens": "5c77c6f472688075" }, "truncated": 0, "non_truncated": 108, "padded": 432, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:logical_fallacies|5": { "hashes": { "hash_examples": "956e0e6365ab79f1", "hash_full_prompts": "2d0087e356f681d0", "hash_input_tokens": "22c8b4788180aba4", "hash_cont_tokens": "25a46284b3589e0d" }, "truncated": 0, "non_truncated": 163, "padded": 652, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:machine_learning|5": { "hashes": { "hash_examples": "397997cc6f4d581e", "hash_full_prompts": "4da1ac14231fba61", "hash_input_tokens": "b5a77040e1bb0728", "hash_cont_tokens": "aacac708cd4c5a61" }, "truncated": 0, "non_truncated": 112, "padded": 448, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:management|5": { "hashes": { "hash_examples": "2bcbe6f6ca63d740", "hash_full_prompts": "89e468b681baaf79", "hash_input_tokens": "8ce16b84cfaead6a", "hash_cont_tokens": "d37808f586a9e9b5" }, "truncated": 0, "non_truncated": 103, "padded": 412, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:marketing|5": { "hashes": { "hash_examples": "8ddb20d964a1b065", "hash_full_prompts": "40953654aee50578", "hash_input_tokens": "482468b4bbf8aefc", "hash_cont_tokens": "95faf210efa02f90" }, "truncated": 0, "non_truncated": 234, "padded": 936, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:medical_genetics|5": { "hashes": { "hash_examples": "182a71f4763d2cea", "hash_full_prompts": "2c4a13f3ed8f6da7", "hash_input_tokens": "588f8a368fb23e28", "hash_cont_tokens": "844bd0bf669e8136" }, "truncated": 0, "non_truncated": 100, "padded": 400, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:miscellaneous|5": { "hashes": { "hash_examples": "4c404fdbb4ca57fc", "hash_full_prompts": "adc833a2ff0318fc", "hash_input_tokens": "d7abe11bb6ccc02f", "hash_cont_tokens": "ef1ae838a09a7521" }, "truncated": 0, "non_truncated": 783, "padded": 3132, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:moral_disputes|5": { "hashes": { "hash_examples": "60cbd2baa3fea5c9", "hash_full_prompts": "2da22b1b7b7ac579", "hash_input_tokens": "381a348f09c4fca5", "hash_cont_tokens": "16b6c6e390eb7cea" }, "truncated": 0, "non_truncated": 346, "padded": 1380, "non_padded": 4, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:moral_scenarios|5": { "hashes": { "hash_examples": "fd8b0431fbdd75ef", "hash_full_prompts": "e6fcd7b4353007ad", "hash_input_tokens": "73a92f9c6f36efce", "hash_cont_tokens": "4130880a19c4edb0" }, "truncated": 0, "non_truncated": 895, "padded": 3443, "non_padded": 137, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:nutrition|5": { "hashes": { "hash_examples": "71e55e2b829b6528", "hash_full_prompts": "6390f239537dda22", "hash_input_tokens": "eb6f76dd0a4ef5fd", "hash_cont_tokens": "96b81f570a84328b" }, "truncated": 0, "non_truncated": 306, "padded": 1224, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:philosophy|5": { "hashes": { "hash_examples": "a6d489a8d208fa4b", "hash_full_prompts": "e2991ca3cc1f79b0", "hash_input_tokens": "ee88e581f1e41a6b", "hash_cont_tokens": "dddff9925c9b675a" }, "truncated": 0, "non_truncated": 311, "padded": 1244, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:prehistory|5": { "hashes": { "hash_examples": "6cc50f032a19acaa", "hash_full_prompts": "fc8c775b75957dc6", "hash_input_tokens": "205f6a163b975b1f", "hash_cont_tokens": "e3a7592f84b44888" }, "truncated": 0, "non_truncated": 324, "padded": 1264, "non_padded": 32, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:professional_accounting|5": { "hashes": { "hash_examples": "50f57ab32f5f6cea", "hash_full_prompts": "fb233c016c038494", "hash_input_tokens": "8a7318d07a544234", "hash_cont_tokens": "f9edf462e8201551" }, "truncated": 0, "non_truncated": 282, "padded": 1104, "non_padded": 24, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:professional_law|5": { "hashes": { "hash_examples": "a8fdc85c64f4b215", "hash_full_prompts": "07e47b0f7c6c5ac7", "hash_input_tokens": "60d0178f5ac799f6", "hash_cont_tokens": "a2de48df0afbaff7" }, "truncated": 16, "non_truncated": 1518, "padded": 6120, "non_padded": 16, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:professional_medicine|5": { "hashes": { "hash_examples": "c373a28a3050a73a", "hash_full_prompts": "73ac5eb25066167b", "hash_input_tokens": "d6a7b8999ef67944", "hash_cont_tokens": "ecf7754754c2bb76" }, "truncated": 0, "non_truncated": 272, "padded": 1088, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:professional_psychology|5": { "hashes": { "hash_examples": "bf5254fe818356af", "hash_full_prompts": "baf87cdb55b25904", "hash_input_tokens": "c244e27cc8a644ed", "hash_cont_tokens": "30b07e31cf9b5c6f" }, "truncated": 0, "non_truncated": 612, "padded": 2444, "non_padded": 4, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:public_relations|5": { "hashes": { "hash_examples": "b66d52e28e7d14e0", "hash_full_prompts": "539df7a43876b017", "hash_input_tokens": "9095272590a00974", "hash_cont_tokens": "cf3600a50782c6c5" }, "truncated": 0, "non_truncated": 110, "padded": 440, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:security_studies|5": { "hashes": { "hash_examples": "514c14feaf000ad9", "hash_full_prompts": "2d8da730f4d25a2b", "hash_input_tokens": "7bdfb4c6942e69a6", "hash_cont_tokens": "4d1dc7c4ad251829" }, "truncated": 0, "non_truncated": 245, "padded": 980, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:sociology|5": { "hashes": { "hash_examples": "f6c9bc9d18c80870", "hash_full_prompts": "920257d06a12039d", "hash_input_tokens": "3fc5010d73fccc7a", "hash_cont_tokens": "d36b9d9f0f4424fe" }, "truncated": 0, "non_truncated": 201, "padded": 804, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:us_foreign_policy|5": { "hashes": { "hash_examples": "ed7b78629db6678f", "hash_full_prompts": "f5147e339d0eb52e", "hash_input_tokens": "9b9a81cabbcf26b4", "hash_cont_tokens": "844bd0bf669e8136" }, "truncated": 0, "non_truncated": 100, "padded": 393, "non_padded": 7, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:virology|5": { "hashes": { "hash_examples": "bc52ffdc3f9b994a", "hash_full_prompts": "1031328c1811e714", "hash_input_tokens": "fc26ddbcd55fdc30", "hash_cont_tokens": "30d4fa4828c5468f" }, "truncated": 0, "non_truncated": 166, "padded": 664, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 }, "lighteval|mmlu:world_religions|5": { "hashes": { "hash_examples": "ecdb4a4f94f62930", "hash_full_prompts": "0c021ca991f9409f", "hash_input_tokens": "984b63357fd0546f", "hash_cont_tokens": "a0a7af55ac7ae037" }, "truncated": 0, "non_truncated": 171, "padded": 684, "non_padded": 0, "effective_few_shots": 5.0, "num_truncated_few_shots": 0 } }, "summary_general": { "hashes": { "hash_examples": "341a076d0beb7048", "hash_full_prompts": "4ffe8426e9ad1cf8", "hash_input_tokens": "1ad997d0d144125c", "hash_cont_tokens": "5adc802ef64ae43f" }, "truncated": 1492, "non_truncated": 12550, "padded": 54382, "non_padded": 1786, "num_truncated_few_shots": 0 } }