open-r1-eval-leaderboard
/
eval_results
/openchat
/openchat-3.5-0106
/main
/mmlu
/results_2024-03-02T19-10-28.713106.json
| { | |
| "config_general": { | |
| "lighteval_sha": "?", | |
| "num_fewshot_seeds": 1, | |
| "override_batch_size": 1, | |
| "max_samples": null, | |
| "job_id": "", | |
| "start_time": 6408201.692824575, | |
| "end_time": 6409078.687741753, | |
| "total_evaluation_time_secondes": "876.9949171775952", | |
| "model_name": "openchat/openchat-3.5-0106", | |
| "model_sha": "9619fb7d2a8e25fa6b0633c0f57f7f4aa79b45c4", | |
| "model_dtype": "torch.bfloat16", | |
| "model_size": "13.61 GB", | |
| "config": null | |
| }, | |
| "results": { | |
| "lighteval|mmlu:abstract_algebra|5": { | |
| "acc": 0.31, | |
| "acc_stderr": 0.04648231987117316 | |
| }, | |
| "lighteval|mmlu:anatomy|5": { | |
| "acc": 0.5925925925925926, | |
| "acc_stderr": 0.04244633238353228 | |
| }, | |
| "lighteval|mmlu:astronomy|5": { | |
| "acc": 0.6907894736842105, | |
| "acc_stderr": 0.037610708698674805 | |
| }, | |
| "lighteval|mmlu:business_ethics|5": { | |
| "acc": 0.63, | |
| "acc_stderr": 0.04852365870939099 | |
| }, | |
| "lighteval|mmlu:clinical_knowledge|5": { | |
| "acc": 0.6867924528301886, | |
| "acc_stderr": 0.028544793319055326 | |
| }, | |
| "lighteval|mmlu:college_biology|5": { | |
| "acc": 0.7222222222222222, | |
| "acc_stderr": 0.037455547914624555 | |
| }, | |
| "lighteval|mmlu:college_chemistry|5": { | |
| "acc": 0.48, | |
| "acc_stderr": 0.050211673156867795 | |
| }, | |
| "lighteval|mmlu:college_computer_science|5": { | |
| "acc": 0.53, | |
| "acc_stderr": 0.05016135580465919 | |
| }, | |
| "lighteval|mmlu:college_mathematics|5": { | |
| "acc": 0.36, | |
| "acc_stderr": 0.048241815132442176 | |
| }, | |
| "lighteval|mmlu:college_medicine|5": { | |
| "acc": 0.6647398843930635, | |
| "acc_stderr": 0.03599586301247078 | |
| }, | |
| "lighteval|mmlu:college_physics|5": { | |
| "acc": 0.4019607843137255, | |
| "acc_stderr": 0.04878608714466996 | |
| }, | |
| "lighteval|mmlu:computer_security|5": { | |
| "acc": 0.77, | |
| "acc_stderr": 0.042295258468165065 | |
| }, | |
| "lighteval|mmlu:conceptual_physics|5": { | |
| "acc": 0.5446808510638298, | |
| "acc_stderr": 0.032555253593403555 | |
| }, | |
| "lighteval|mmlu:econometrics|5": { | |
| "acc": 0.41228070175438597, | |
| "acc_stderr": 0.046306532033665956 | |
| }, | |
| "lighteval|mmlu:electrical_engineering|5": { | |
| "acc": 0.5379310344827586, | |
| "acc_stderr": 0.04154659671707548 | |
| }, | |
| "lighteval|mmlu:elementary_mathematics|5": { | |
| "acc": 0.3968253968253968, | |
| "acc_stderr": 0.025197101074246483 | |
| }, | |
| "lighteval|mmlu:formal_logic|5": { | |
| "acc": 0.4444444444444444, | |
| "acc_stderr": 0.04444444444444449 | |
| }, | |
| "lighteval|mmlu:global_facts|5": { | |
| "acc": 0.34, | |
| "acc_stderr": 0.04760952285695235 | |
| }, | |
| "lighteval|mmlu:high_school_biology|5": { | |
| "acc": 0.7580645161290323, | |
| "acc_stderr": 0.024362599693031096 | |
| }, | |
| "lighteval|mmlu:high_school_chemistry|5": { | |
| "acc": 0.4975369458128079, | |
| "acc_stderr": 0.03517945038691063 | |
| }, | |
| "lighteval|mmlu:high_school_computer_science|5": { | |
| "acc": 0.71, | |
| "acc_stderr": 0.045604802157206845 | |
| }, | |
| "lighteval|mmlu:high_school_european_history|5": { | |
| "acc": 0.806060606060606, | |
| "acc_stderr": 0.03087414513656208 | |
| }, | |
| "lighteval|mmlu:high_school_geography|5": { | |
| "acc": 0.7828282828282829, | |
| "acc_stderr": 0.029376616484945633 | |
| }, | |
| "lighteval|mmlu:high_school_government_and_politics|5": { | |
| "acc": 0.8756476683937824, | |
| "acc_stderr": 0.02381447708659356 | |
| }, | |
| "lighteval|mmlu:high_school_macroeconomics|5": { | |
| "acc": 0.6358974358974359, | |
| "acc_stderr": 0.024396672985094757 | |
| }, | |
| "lighteval|mmlu:high_school_mathematics|5": { | |
| "acc": 0.3592592592592593, | |
| "acc_stderr": 0.02925290592725198 | |
| }, | |
| "lighteval|mmlu:high_school_microeconomics|5": { | |
| "acc": 0.6386554621848739, | |
| "acc_stderr": 0.03120469122515002 | |
| }, | |
| "lighteval|mmlu:high_school_physics|5": { | |
| "acc": 0.3443708609271523, | |
| "acc_stderr": 0.038796870240733264 | |
| }, | |
| "lighteval|mmlu:high_school_psychology|5": { | |
| "acc": 0.8348623853211009, | |
| "acc_stderr": 0.015919557829976044 | |
| }, | |
| "lighteval|mmlu:high_school_statistics|5": { | |
| "acc": 0.4675925925925926, | |
| "acc_stderr": 0.03402801581358966 | |
| }, | |
| "lighteval|mmlu:high_school_us_history|5": { | |
| "acc": 0.8186274509803921, | |
| "acc_stderr": 0.027044621719474082 | |
| }, | |
| "lighteval|mmlu:high_school_world_history|5": { | |
| "acc": 0.810126582278481, | |
| "acc_stderr": 0.025530100460233494 | |
| }, | |
| "lighteval|mmlu:human_aging|5": { | |
| "acc": 0.7040358744394619, | |
| "acc_stderr": 0.030636591348699813 | |
| }, | |
| "lighteval|mmlu:human_sexuality|5": { | |
| "acc": 0.7709923664122137, | |
| "acc_stderr": 0.036853466317118506 | |
| }, | |
| "lighteval|mmlu:international_law|5": { | |
| "acc": 0.8347107438016529, | |
| "acc_stderr": 0.03390780612972776 | |
| }, | |
| "lighteval|mmlu:jurisprudence|5": { | |
| "acc": 0.7685185185185185, | |
| "acc_stderr": 0.04077494709252627 | |
| }, | |
| "lighteval|mmlu:logical_fallacies|5": { | |
| "acc": 0.7300613496932515, | |
| "acc_stderr": 0.03487825168497892 | |
| }, | |
| "lighteval|mmlu:machine_learning|5": { | |
| "acc": 0.5089285714285714, | |
| "acc_stderr": 0.04745033255489123 | |
| }, | |
| "lighteval|mmlu:management|5": { | |
| "acc": 0.7961165048543689, | |
| "acc_stderr": 0.0398913985953177 | |
| }, | |
| "lighteval|mmlu:marketing|5": { | |
| "acc": 0.8675213675213675, | |
| "acc_stderr": 0.022209309073165616 | |
| }, | |
| "lighteval|mmlu:medical_genetics|5": { | |
| "acc": 0.71, | |
| "acc_stderr": 0.04560480215720684 | |
| }, | |
| "lighteval|mmlu:miscellaneous|5": { | |
| "acc": 0.8160919540229885, | |
| "acc_stderr": 0.013853724170922531 | |
| }, | |
| "lighteval|mmlu:moral_disputes|5": { | |
| "acc": 0.7283236994219653, | |
| "acc_stderr": 0.02394851290546836 | |
| }, | |
| "lighteval|mmlu:moral_scenarios|5": { | |
| "acc": 0.31508379888268156, | |
| "acc_stderr": 0.015536850852473636 | |
| }, | |
| "lighteval|mmlu:nutrition|5": { | |
| "acc": 0.7418300653594772, | |
| "acc_stderr": 0.025058503316958143 | |
| }, | |
| "lighteval|mmlu:philosophy|5": { | |
| "acc": 0.7202572347266881, | |
| "acc_stderr": 0.02549425935069491 | |
| }, | |
| "lighteval|mmlu:prehistory|5": { | |
| "acc": 0.7314814814814815, | |
| "acc_stderr": 0.024659685185967294 | |
| }, | |
| "lighteval|mmlu:professional_accounting|5": { | |
| "acc": 0.4787234042553192, | |
| "acc_stderr": 0.029800481645628693 | |
| }, | |
| "lighteval|mmlu:professional_law|5": { | |
| "acc": 0.47783572359843546, | |
| "acc_stderr": 0.012757683047716175 | |
| }, | |
| "lighteval|mmlu:professional_medicine|5": { | |
| "acc": 0.7058823529411765, | |
| "acc_stderr": 0.027678468642144714 | |
| }, | |
| "lighteval|mmlu:professional_psychology|5": { | |
| "acc": 0.6519607843137255, | |
| "acc_stderr": 0.019270998708223977 | |
| }, | |
| "lighteval|mmlu:public_relations|5": { | |
| "acc": 0.6545454545454545, | |
| "acc_stderr": 0.04554619617541054 | |
| }, | |
| "lighteval|mmlu:security_studies|5": { | |
| "acc": 0.7387755102040816, | |
| "acc_stderr": 0.028123429335142783 | |
| }, | |
| "lighteval|mmlu:sociology|5": { | |
| "acc": 0.8159203980099502, | |
| "acc_stderr": 0.027403859410786848 | |
| }, | |
| "lighteval|mmlu:us_foreign_policy|5": { | |
| "acc": 0.84, | |
| "acc_stderr": 0.03684529491774708 | |
| }, | |
| "lighteval|mmlu:virology|5": { | |
| "acc": 0.5301204819277109, | |
| "acc_stderr": 0.03885425420866767 | |
| }, | |
| "lighteval|mmlu:world_religions|5": { | |
| "acc": 0.8304093567251462, | |
| "acc_stderr": 0.02878210810540171 | |
| }, | |
| "lighteval|mmlu:_average|5": { | |
| "acc": 0.6372441558659353, | |
| "acc_stderr": 0.03378283520026763 | |
| } | |
| }, | |
| "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": "90e44aeaf456614e", | |
| "hash_input_tokens": "4fafe067b499a101", | |
| "hash_cont_tokens": "17b868b63507f9a3" | |
| }, | |
| "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": "f2fc6fd9210201a1", | |
| "hash_input_tokens": "2ad4dc1a8a3541eb", | |
| "hash_cont_tokens": "a52a4f60d98cbe5c" | |
| }, | |
| "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": "a7461a5194918a14", | |
| "hash_input_tokens": "6faf9e29ba3953e0", | |
| "hash_cont_tokens": "10f7d8eeba97841d" | |
| }, | |
| "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": "5668d68cd4c12414", | |
| "hash_input_tokens": "086d1597af775ff0", | |
| "hash_cont_tokens": "17b868b63507f9a3" | |
| }, | |
| "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": "7a65ffa4791dcfd8", | |
| "hash_input_tokens": "c4ed21a37c0d08a8", | |
| "hash_cont_tokens": "edef9975ba9165b5" | |
| }, | |
| "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": "e947aca04f9c2faf", | |
| "hash_input_tokens": "8d30a11a8aeca6ef", | |
| "hash_cont_tokens": "0aa103ec6602280b" | |
| }, | |
| "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": "0184616710ebc367", | |
| "hash_input_tokens": "edda1ea2a9e92600", | |
| "hash_cont_tokens": "17b868b63507f9a3" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 100, | |
| "padded": 400, | |
| "non_padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|mmlu:college_computer_science|5": { | |
| "hashes": { | |
| "hash_examples": "32805b52d7d5daab", | |
| "hash_full_prompts": "c066320f7773b4e2", | |
| "hash_input_tokens": "abc446443b07d0c4", | |
| "hash_cont_tokens": "17b868b63507f9a3" | |
| }, | |
| "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": "8da594fdac89351c", | |
| "hash_input_tokens": "ca86463992448e9f", | |
| "hash_cont_tokens": "17b868b63507f9a3" | |
| }, | |
| "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": "52f47d914e6e90c8", | |
| "hash_input_tokens": "93015213490133e4", | |
| "hash_cont_tokens": "1979021dbc698754" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 173, | |
| "padded": 692, | |
| "non_padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|mmlu:college_physics|5": { | |
| "hashes": { | |
| "hash_examples": "ebdab1cdb7e555df", | |
| "hash_full_prompts": "667e05c14196e18e", | |
| "hash_input_tokens": "1db843c9f2536b63", | |
| "hash_cont_tokens": "7cf7fe2bab00acbd" | |
| }, | |
| "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": "2ae5f8390b24d740", | |
| "hash_input_tokens": "32c91fbdbf5589e1", | |
| "hash_cont_tokens": "17b868b63507f9a3" | |
| }, | |
| "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": "9a387a05d8fca622", | |
| "hash_input_tokens": "993c77da999f4489", | |
| "hash_cont_tokens": "903f64eed2b0d217" | |
| }, | |
| "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": "ab99797199fcfa2e", | |
| "hash_input_tokens": "a9e85fbbfea4a113", | |
| "hash_cont_tokens": "721ae6c5302c4bf2" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 114, | |
| "padded": 455, | |
| "non_padded": 1, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|mmlu:electrical_engineering|5": { | |
| "hashes": { | |
| "hash_examples": "acbc5def98c19b3f", | |
| "hash_full_prompts": "c331f4e23c83df9d", | |
| "hash_input_tokens": "1493b7dbdf013d64", | |
| "hash_cont_tokens": "15a738960ed3e587" | |
| }, | |
| "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": "2b9e1987d86fd16b", | |
| "hash_input_tokens": "98a28064ab220a14", | |
| "hash_cont_tokens": "c96470462fc71683" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 378, | |
| "padded": 1512, | |
| "non_padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|mmlu:formal_logic|5": { | |
| "hashes": { | |
| "hash_examples": "8635216e1909a03f", | |
| "hash_full_prompts": "40d847cbdbb4cf3c", | |
| "hash_input_tokens": "bbf11dc49bbaf8ee", | |
| "hash_cont_tokens": "0e1ce025c9d6ee7e" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 126, | |
| "padded": 500, | |
| "non_padded": 4, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|mmlu:global_facts|5": { | |
| "hashes": { | |
| "hash_examples": "30b315aa6353ee47", | |
| "hash_full_prompts": "d131d968f8e10d7a", | |
| "hash_input_tokens": "2c2e38e5efd29c07", | |
| "hash_cont_tokens": "17b868b63507f9a3" | |
| }, | |
| "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": "8fbfbe5891f0058f", | |
| "hash_input_tokens": "ada0e17375806202", | |
| "hash_cont_tokens": "e34d57f7d3c4ca16" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 310, | |
| "padded": 1240, | |
| "non_padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|mmlu:high_school_chemistry|5": { | |
| "hashes": { | |
| "hash_examples": "b0661bfa1add6404", | |
| "hash_full_prompts": "08c8fd1313cdca6c", | |
| "hash_input_tokens": "c612ebec6c0b43b7", | |
| "hash_cont_tokens": "e8482d44df4b3740" | |
| }, | |
| "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": "6f8cc7e6dec3b5cc", | |
| "hash_input_tokens": "61834c21a63f2c9e", | |
| "hash_cont_tokens": "17b868b63507f9a3" | |
| }, | |
| "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": "72b650fd34178f24", | |
| "hash_input_tokens": "af4f29164bd9819a", | |
| "hash_cont_tokens": "d63e679a49418339" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 165, | |
| "padded": 656, | |
| "non_padded": 4, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|mmlu:high_school_geography|5": { | |
| "hashes": { | |
| "hash_examples": "7dc963c7acd19ad8", | |
| "hash_full_prompts": "d9feb1aab8140a29", | |
| "hash_input_tokens": "9403d0bc29d5694e", | |
| "hash_cont_tokens": "d78483e286d06f1a" | |
| }, | |
| "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": "41f65e7e4c037c7e", | |
| "hash_input_tokens": "deed809433f5a0d7", | |
| "hash_cont_tokens": "691cdff71ff5fe57" | |
| }, | |
| "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": "68dd67998fc6c7cc", | |
| "hash_input_tokens": "9730035d00aa266f", | |
| "hash_cont_tokens": "d5ad4c5bdca967ad" | |
| }, | |
| "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": "00360113e4d448ac", | |
| "hash_input_tokens": "fe851ceba8e0a2f4", | |
| "hash_cont_tokens": "8f631ca5687dd0d4" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 270, | |
| "padded": 1080, | |
| "non_padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|mmlu:high_school_microeconomics|5": { | |
| "hashes": { | |
| "hash_examples": "2118f21f71d87d84", | |
| "hash_full_prompts": "3789cd87399c8406", | |
| "hash_input_tokens": "100886cc179ce460", | |
| "hash_cont_tokens": "7321048a28451473" | |
| }, | |
| "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": "ae212c00989651c4", | |
| "hash_input_tokens": "67c76ed19fc0dee1", | |
| "hash_cont_tokens": "bb137581f269861c" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 151, | |
| "padded": 604, | |
| "non_padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|mmlu:high_school_psychology|5": { | |
| "hashes": { | |
| "hash_examples": "c8d1d98a40e11f2f", | |
| "hash_full_prompts": "e5b37bd3c1d978ce", | |
| "hash_input_tokens": "284239071e87204e", | |
| "hash_cont_tokens": "b455cab2675bd863" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 545, | |
| "padded": 2180, | |
| "non_padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|mmlu:high_school_statistics|5": { | |
| "hashes": { | |
| "hash_examples": "666c8759b98ee4ff", | |
| "hash_full_prompts": "e1528ff7f1afe6b8", | |
| "hash_input_tokens": "d3385e9bc84abbf3", | |
| "hash_cont_tokens": "1b3196fec7e58037" | |
| }, | |
| "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": "0efba1956c24e8d8", | |
| "hash_input_tokens": "89db03114b6c007b", | |
| "hash_cont_tokens": "a331dedc2aa01b3e" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 204, | |
| "padded": 816, | |
| "non_padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|mmlu:high_school_world_history|5": { | |
| "hashes": { | |
| "hash_examples": "7e5085b6184b0322", | |
| "hash_full_prompts": "c42e068217fa2f37", | |
| "hash_input_tokens": "73647e70917167cf", | |
| "hash_cont_tokens": "d0fbe030b8c8c2bf" | |
| }, | |
| "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": "4cd428818f949370", | |
| "hash_input_tokens": "508abd59b7696876", | |
| "hash_cont_tokens": "1dd29c3755494850" | |
| }, | |
| "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": "5772acea8cdc617b", | |
| "hash_input_tokens": "2e18c7505bf1630c", | |
| "hash_cont_tokens": "c85573f663c10691" | |
| }, | |
| "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": "b8f7c52e80a6bcbf", | |
| "hash_input_tokens": "eb746d99406379a7", | |
| "hash_cont_tokens": "d263804ba918154f" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 121, | |
| "padded": 476, | |
| "non_padded": 8, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|mmlu:jurisprudence|5": { | |
| "hashes": { | |
| "hash_examples": "e956f86b124076fe", | |
| "hash_full_prompts": "7acdaaa21e55d4c9", | |
| "hash_input_tokens": "48439d0f3b679614", | |
| "hash_cont_tokens": "581986691a84ece8" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 108, | |
| "padded": 425, | |
| "non_padded": 7, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|mmlu:logical_fallacies|5": { | |
| "hashes": { | |
| "hash_examples": "956e0e6365ab79f1", | |
| "hash_full_prompts": "c45d3e37f815195d", | |
| "hash_input_tokens": "6a6da454effee292", | |
| "hash_cont_tokens": "55a858b28bbda458" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 163, | |
| "padded": 632, | |
| "non_padded": 20, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|mmlu:machine_learning|5": { | |
| "hashes": { | |
| "hash_examples": "397997cc6f4d581e", | |
| "hash_full_prompts": "466217f1bc6aebd7", | |
| "hash_input_tokens": "cd2373efe91fca82", | |
| "hash_cont_tokens": "e99d3d3efd4ac7a3" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 112, | |
| "padded": 444, | |
| "non_padded": 4, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|mmlu:management|5": { | |
| "hashes": { | |
| "hash_examples": "2bcbe6f6ca63d740", | |
| "hash_full_prompts": "c8ca0e9acf6e1ed0", | |
| "hash_input_tokens": "b67ec3cadfeb972d", | |
| "hash_cont_tokens": "13d9dc56bca34726" | |
| }, | |
| "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": "93bae49cc96e2754", | |
| "hash_input_tokens": "16edef7045b3845b", | |
| "hash_cont_tokens": "2700ea26933916a2" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 234, | |
| "padded": 912, | |
| "non_padded": 24, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|mmlu:medical_genetics|5": { | |
| "hashes": { | |
| "hash_examples": "182a71f4763d2cea", | |
| "hash_full_prompts": "bc058984481865da", | |
| "hash_input_tokens": "4efa81487c3b7f4b", | |
| "hash_cont_tokens": "17b868b63507f9a3" | |
| }, | |
| "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": "c843e21d2085ad27", | |
| "hash_input_tokens": "6f182621a95157d9", | |
| "hash_cont_tokens": "7bf4341c79587250" | |
| }, | |
| "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": "b0f5f04906b51f2e", | |
| "hash_input_tokens": "36259959b88501df", | |
| "hash_cont_tokens": "38a48e9de6976f00" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 346, | |
| "padded": 1384, | |
| "non_padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|mmlu:moral_scenarios|5": { | |
| "hashes": { | |
| "hash_examples": "fd8b0431fbdd75ef", | |
| "hash_full_prompts": "2977055b16d1a565", | |
| "hash_input_tokens": "07c73c15596dca9f", | |
| "hash_cont_tokens": "761c4dc187689d89" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 895, | |
| "padded": 3528, | |
| "non_padded": 52, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|mmlu:nutrition|5": { | |
| "hashes": { | |
| "hash_examples": "71e55e2b829b6528", | |
| "hash_full_prompts": "2af2dd632e1139e9", | |
| "hash_input_tokens": "293f93556834b15a", | |
| "hash_cont_tokens": "65005bd7d6f6012a" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 306, | |
| "padded": 1220, | |
| "non_padded": 4, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|mmlu:philosophy|5": { | |
| "hashes": { | |
| "hash_examples": "a6d489a8d208fa4b", | |
| "hash_full_prompts": "c27d3328965a9913", | |
| "hash_input_tokens": "e0866f1a0348746f", | |
| "hash_cont_tokens": "0b47934fb6314dec" | |
| }, | |
| "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": "238f18f1179be0fa", | |
| "hash_input_tokens": "411a46e95ce0915e", | |
| "hash_cont_tokens": "3f20acd855ee0a29" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 324, | |
| "padded": 1294, | |
| "non_padded": 2, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|mmlu:professional_accounting|5": { | |
| "hashes": { | |
| "hash_examples": "50f57ab32f5f6cea", | |
| "hash_full_prompts": "ec5f30cdbd3e2fb7", | |
| "hash_input_tokens": "1a98d55aa50f3668", | |
| "hash_cont_tokens": "8f122ba881355d4b" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 282, | |
| "padded": 1112, | |
| "non_padded": 16, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|mmlu:professional_law|5": { | |
| "hashes": { | |
| "hash_examples": "a8fdc85c64f4b215", | |
| "hash_full_prompts": "dca20f0ec5c404d5", | |
| "hash_input_tokens": "2a0dffd2a4f4dcc5", | |
| "hash_cont_tokens": "90d5df417c4d3fd3" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 1534, | |
| "padded": 6136, | |
| "non_padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|mmlu:professional_medicine|5": { | |
| "hashes": { | |
| "hash_examples": "c373a28a3050a73a", | |
| "hash_full_prompts": "3f68e07a7f24fb39", | |
| "hash_input_tokens": "f2d1dc6931f5116a", | |
| "hash_cont_tokens": "4a2d2988884f7f70" | |
| }, | |
| "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": "ab0809fbed0b3c38", | |
| "hash_input_tokens": "5df72d301ba7b64c", | |
| "hash_cont_tokens": "e0a952cb8a9c81de" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 612, | |
| "padded": 2448, | |
| "non_padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|mmlu:public_relations|5": { | |
| "hashes": { | |
| "hash_examples": "b66d52e28e7d14e0", | |
| "hash_full_prompts": "34b792bee1c8e997", | |
| "hash_input_tokens": "affc6b06625957a9", | |
| "hash_cont_tokens": "1fa77a8dff3922b8" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 110, | |
| "padded": 432, | |
| "non_padded": 8, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|mmlu:security_studies|5": { | |
| "hashes": { | |
| "hash_examples": "514c14feaf000ad9", | |
| "hash_full_prompts": "6d7279e524890405", | |
| "hash_input_tokens": "451c1d85a0a00131", | |
| "hash_cont_tokens": "81fc9cb3cbdd52db" | |
| }, | |
| "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": "e8d724177e268d1b", | |
| "hash_input_tokens": "88cda8cc70937ae4", | |
| "hash_cont_tokens": "2a0493252ed2cf43" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 201, | |
| "padded": 800, | |
| "non_padded": 4, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|mmlu:us_foreign_policy|5": { | |
| "hashes": { | |
| "hash_examples": "ed7b78629db6678f", | |
| "hash_full_prompts": "84a37258f46c988c", | |
| "hash_input_tokens": "59bca315132f387c", | |
| "hash_cont_tokens": "17b868b63507f9a3" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 100, | |
| "padded": 396, | |
| "non_padded": 4, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "lighteval|mmlu:virology|5": { | |
| "hashes": { | |
| "hash_examples": "bc52ffdc3f9b994a", | |
| "hash_full_prompts": "bd6253ed3796d2f5", | |
| "hash_input_tokens": "c7a327a6e4389a42", | |
| "hash_cont_tokens": "5ab892d003b00c98" | |
| }, | |
| "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": "94940a5122a20641", | |
| "hash_input_tokens": "d0be44284510122f", | |
| "hash_cont_tokens": "15a5e5dbdfbb8568" | |
| }, | |
| "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": "9e47cc8e71473bb6", | |
| "hash_input_tokens": "795951d7228c622e", | |
| "hash_cont_tokens": "c24d2aedc1261d61" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 14042, | |
| "padded": 56006, | |
| "non_padded": 162, | |
| "num_truncated_few_shots": 0 | |
| } | |
| } |