PyPI - crfm-helm - Versions diffs - 0.5.6__py3-none-any.whl → 0.5.8__py3-none-any.whl - Mend

crfm-helm 0.5.6py3-none-any.whl → 0.5.8py3-none-any.whl

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Potentially problematic release.

This version of crfm-helm might be problematic. Click here for more details.

Files changed (311) hide show

{crfm_helm-0.5.6.dist-info → crfm_helm-0.5.8.dist-info}/METADATA +60 -125
{crfm_helm-0.5.6.dist-info → crfm_helm-0.5.8.dist-info}/RECORD +293 -229
helm/benchmark/adaptation/adapter_spec.py +5 -0
helm/benchmark/adaptation/adapters/multimodal/multiple_choice_joint_multimodal_adapter.py +11 -3
helm/benchmark/adaptation/adapters/multiple_choice_joint_adapter.py +11 -8
helm/benchmark/annotation/aci_bench_annotator.py +11 -22
helm/benchmark/annotation/air_bench_annotator.py +1 -1
helm/benchmark/annotation/alrage_annotator.py +90 -0
helm/benchmark/annotation/chw_care_plan_annotator.py +10 -21
helm/benchmark/annotation/dischargeme_annotator.py +11 -22
helm/benchmark/annotation/live_qa_annotator.py +1 -1
helm/benchmark/annotation/med_dialog_annotator.py +11 -22
helm/benchmark/annotation/medalign_annotator.py +11 -22
helm/benchmark/annotation/medi_qa_annotator.py +11 -22
helm/benchmark/annotation/medication_qa_annotator.py +11 -22
helm/benchmark/annotation/mental_health_annotator.py +11 -22
helm/benchmark/annotation/mimic_bhc_annotator.py +11 -22
helm/benchmark/annotation/mimic_rrs_annotator.py +11 -22
helm/benchmark/annotation/model_as_judge.py +23 -18
helm/benchmark/annotation/mtsamples_procedures_annotator.py +11 -22
helm/benchmark/annotation/mtsamples_replicate_annotator.py +11 -22
helm/benchmark/annotation/starr_patient_instructions_annotator.py +11 -22
helm/benchmark/metrics/air_bench_metrics.py +3157 -1
helm/benchmark/metrics/alrage_metric.py +35 -0
helm/benchmark/metrics/basic_metrics.py +267 -2
helm/benchmark/metrics/classification_metrics.py +19 -1
helm/benchmark/metrics/codeinsights_code_efficiency_metrics.py +186 -0
helm/benchmark/metrics/codeinsights_code_evaluation_metrics.py +477 -0
helm/benchmark/metrics/codeinsights_correct_code_metrics.py +366 -0
helm/benchmark/metrics/codeinsights_edge_case_metrics.py +92 -0
helm/benchmark/metrics/codeinsights_metric_specs.py +51 -0
helm/benchmark/metrics/comet_metric.py +1 -1
helm/benchmark/metrics/conv_fin_qa_calc_metrics.py +12 -1
helm/benchmark/metrics/copyright_metrics.py +1 -1
helm/benchmark/metrics/decodingtrust_stereotype_bias_metrics.py +1 -1
helm/benchmark/metrics/dry_run_metrics.py +30 -1
helm/benchmark/metrics/efficiency_metrics.py +74 -0
helm/benchmark/metrics/ehr_sql_metrics.py +57 -1
helm/benchmark/metrics/evaluate_reference_metrics.py +300 -1
helm/benchmark/metrics/gpqa_chain_of_thought_metric.py +13 -1
helm/benchmark/metrics/helpdesk_call_summarization_metrics.py +13 -1
helm/benchmark/metrics/ifeval_metrics.py +13 -1
helm/benchmark/metrics/image_generation/clip_score_metrics.py +13 -2
helm/benchmark/metrics/image_generation/fractal_dimension/fractal_dimension_util.py +1 -1
helm/benchmark/metrics/instruction_following_critique_metrics.py +41 -1
helm/benchmark/metrics/kpi_edgar_metrics.py +21 -0
helm/benchmark/metrics/language_modeling_metrics.py +13 -1
helm/benchmark/metrics/live_qa_metrics.py +13 -1
helm/benchmark/metrics/llm_jury_metrics.py +13 -1
helm/benchmark/metrics/lmkt_metric_specs.py +12 -0
helm/benchmark/metrics/lmkt_metrics.py +47 -0
helm/benchmark/metrics/medcalc_bench_metrics.py +14 -1
helm/benchmark/metrics/medec_metrics.py +25 -2
helm/benchmark/metrics/melt_toxicity_metric.py +1 -1
helm/benchmark/metrics/metric.py +25 -0
helm/benchmark/metrics/mimiciv_billing_code_metrics.py +32 -1
helm/benchmark/metrics/omni_math_metrics.py +13 -1
helm/benchmark/metrics/seahelm_metrics.py +14 -1
helm/benchmark/metrics/summac/model_summac.py +3 -3
helm/benchmark/metrics/summarization_metrics.py +129 -1
helm/benchmark/metrics/toxicity_metrics.py +31 -1
helm/benchmark/metrics/wildbench_metrics.py +21 -1
helm/benchmark/model_deployment_registry.py +11 -19
helm/benchmark/presentation/create_plots.py +11 -2
helm/benchmark/presentation/schema.py +10 -22
helm/benchmark/presentation/summarize.py +189 -14
helm/benchmark/presentation/taxonomy_info.py +20 -0
helm/benchmark/presentation/test_create_plots.py +4 -1
helm/benchmark/run.py +7 -1
helm/benchmark/run_expander.py +4 -0
helm/benchmark/run_specs/arabic_run_specs.py +191 -0
helm/benchmark/run_specs/bluex_run_specs.py +40 -0
helm/benchmark/run_specs/classic_run_specs.py +2 -55
helm/benchmark/run_specs/codeinsights_run_specs.py +192 -0
helm/benchmark/run_specs/healthqa_br_run_specs.py +40 -0
helm/benchmark/run_specs/heim_run_specs.py +3 -1
helm/benchmark/run_specs/lmkt_run_specs.py +144 -0
helm/benchmark/run_specs/long_context_run_specs.py +48 -1
helm/benchmark/run_specs/medhelm/__init__.py +0 -0
helm/benchmark/run_specs/medhelm/benchmark_config.py +219 -0
helm/benchmark/run_specs/medhelm_run_specs.py +360 -50
helm/benchmark/run_specs/multilingual_run_specs.py +50 -0
helm/benchmark/run_specs/speech_disorder_audio_run_specs.py +5 -11
helm/benchmark/scenarios/aci_bench_scenario.py +23 -0
helm/benchmark/scenarios/air_bench_scenario.py +21 -0
helm/benchmark/scenarios/alghafa_scenario.py +126 -0
helm/benchmark/scenarios/alrage_scenario.py +54 -0
helm/benchmark/scenarios/anthropic_hh_rlhf_scenario.py +23 -1
helm/benchmark/scenarios/arabic_exams_scenario.py +114 -0
helm/benchmark/scenarios/arabic_mmlu_scenario.py +82 -0
helm/benchmark/scenarios/aratrust_scenario.py +95 -0
helm/benchmark/scenarios/audio_language/casual_conversations2_scenario.py +1 -1
helm/benchmark/scenarios/audio_language/mustard_scenario.py +1 -1
helm/benchmark/scenarios/audio_language/{ultra_suite_asr_classification.py → ultra_suite_asr_classification_scenario.py} +9 -8
helm/benchmark/scenarios/audio_language/ultra_suite_asr_transcription_scenario.py +99 -0
helm/benchmark/scenarios/audio_language/ultra_suite_classification_scenario.py +13 -5
helm/benchmark/scenarios/audio_language/ultra_suite_disorder_breakdown_scenario.py +13 -5
helm/benchmark/scenarios/audio_language/ultra_suite_disorder_symptoms_scenario.py +13 -5
helm/benchmark/scenarios/babi_qa_scenario.py +15 -0
helm/benchmark/scenarios/bbq_scenario.py +15 -0
helm/benchmark/scenarios/best_chatgpt_prompts.yaml +473 -0
helm/benchmark/scenarios/bluex_scenario.py +70 -0
helm/benchmark/scenarios/bold_scenario.py +15 -0
helm/benchmark/scenarios/boolq_scenario.py +20 -0
helm/benchmark/scenarios/chw_care_plan_scenario.py +23 -0
helm/benchmark/scenarios/civil_comments_scenario.py +13 -0
helm/benchmark/scenarios/clear_scenario.py +23 -0
helm/benchmark/scenarios/cleva_scenario.py +480 -1
helm/benchmark/scenarios/code_scenario.py +28 -0
helm/benchmark/scenarios/codeinsights_code_efficiency_scenario.py +197 -0
helm/benchmark/scenarios/codeinsights_correct_code_scenario.py +78 -0
helm/benchmark/scenarios/codeinsights_edge_case_scenario.py +192 -0
helm/benchmark/scenarios/codeinsights_student_coding_scenario.py +162 -0
helm/benchmark/scenarios/codeinsights_student_mistake_scenario.py +188 -0
helm/benchmark/scenarios/commonsense_scenario.py +26 -0
helm/benchmark/scenarios/compositional_instructions.yaml +70 -0
helm/benchmark/scenarios/conv_fin_qa_calc_scenario.py +21 -0
helm/benchmark/scenarios/copyright_scenario.py +35 -1
helm/benchmark/scenarios/cti_to_mitre_scenario.py +21 -0
helm/benchmark/scenarios/decodingtrust_adv_demonstration_scenario.py +22 -1
helm/benchmark/scenarios/decodingtrust_adv_robustness_scenario.py +23 -1
helm/benchmark/scenarios/decodingtrust_fairness_scenario.py +22 -1
helm/benchmark/scenarios/decodingtrust_machine_ethics_scenario.py +21 -1
helm/benchmark/scenarios/decodingtrust_ood_robustness_scenario.py +13 -0
helm/benchmark/scenarios/decodingtrust_privacy_scenario.py +13 -1
helm/benchmark/scenarios/decodingtrust_stereotype_bias_scenario.py +13 -1
helm/benchmark/scenarios/decodingtrust_toxicity_prompts_scenario.py +13 -1
helm/benchmark/scenarios/dischargeme_scenario.py +24 -0
helm/benchmark/scenarios/disinformation_scenario.py +22 -0
helm/benchmark/scenarios/dyck_language_scenario.py +15 -0
helm/benchmark/scenarios/ehrshot_scenario.py +22 -0
helm/benchmark/scenarios/enem_challenge_scenario.py +19 -0
helm/benchmark/scenarios/entity_data_imputation_scenario.py +14 -0
helm/benchmark/scenarios/entity_matching_scenario.py +14 -0
helm/benchmark/scenarios/exams_multilingual_scenario.py +115 -0
helm/benchmark/scenarios/financial_phrasebank_scenario.py +21 -0
helm/benchmark/scenarios/gold_commodity_news_scenario.py +21 -0
helm/benchmark/scenarios/gpqa_scenario.py +18 -0
helm/benchmark/scenarios/grammar_scenario.py +20 -1
helm/benchmark/scenarios/gsm_scenario.py +15 -0
helm/benchmark/scenarios/headqa_scenario.py +22 -0
helm/benchmark/scenarios/healthqa_br_scenario.py +80 -0
helm/benchmark/scenarios/helpdesk_call_summarization_scenario.py +13 -0
helm/benchmark/scenarios/ice_scenario.py +21 -1
helm/benchmark/scenarios/ifeval_scenario.py +18 -0
helm/benchmark/scenarios/imdb_scenario.py +15 -0
helm/benchmark/scenarios/infinite_bench_en_mc_scenario.py +90 -0
helm/benchmark/scenarios/infinite_bench_en_qa_scenario.py +1 -1
helm/benchmark/scenarios/koala_scenario.py +21 -1
helm/benchmark/scenarios/kpi_edgar_scenario.py +21 -0
helm/benchmark/scenarios/legal_contract_summarization_scenario.py +20 -0
helm/benchmark/scenarios/legal_summarization_scenario.py +50 -0
helm/benchmark/scenarios/legal_support_scenario.py +13 -0
helm/benchmark/scenarios/legalbench_scenario.py +20 -0
helm/benchmark/scenarios/lex_glue_scenario.py +11 -0
helm/benchmark/scenarios/lextreme_scenario.py +11 -0
helm/benchmark/scenarios/lmkt_scenarios.py +288 -0
helm/benchmark/scenarios/lsat_qa_scenario.py +14 -0
helm/benchmark/scenarios/madinah_qa_scenario.py +73 -0
helm/benchmark/scenarios/math_scenario.py +47 -20
helm/benchmark/scenarios/mbzuai_human_translated_arabic_mmlu.py +68 -0
helm/benchmark/scenarios/med_dialog_scenario.py +32 -1
helm/benchmark/scenarios/med_mcqa_scenario.py +14 -0
helm/benchmark/scenarios/med_qa_scenario.py +14 -0
helm/benchmark/scenarios/medalign_scenario.py +23 -0
helm/benchmark/scenarios/medalign_scenario_helper.py +19 -125
helm/benchmark/scenarios/medbullets_scenario.py +22 -0
helm/benchmark/scenarios/medcalc_bench_scenario.py +22 -0
helm/benchmark/scenarios/medec_scenario.py +23 -0
helm/benchmark/scenarios/medhallu_scenario.py +23 -0
helm/benchmark/scenarios/medhelm/__init__.py +0 -0
helm/benchmark/scenarios/medhelm/judges.yaml +14 -0
helm/benchmark/scenarios/medhelm_configurable_scenario.py +101 -0
helm/benchmark/scenarios/medi_qa_scenario.py +23 -0
helm/benchmark/scenarios/medication_qa_scenario.py +31 -1
helm/benchmark/scenarios/melt_scenarios.py +2 -2
helm/benchmark/scenarios/mental_health_scenario.py +23 -0
helm/benchmark/scenarios/mimic_bhc_scenario.py +25 -1
helm/benchmark/scenarios/mimic_rrs_scenario.py +23 -0
helm/benchmark/scenarios/mimiciv_billing_code_scenario.py +22 -0
helm/benchmark/scenarios/mmlu_pro_scenario.py +18 -0
helm/benchmark/scenarios/mmlu_scenario.py +15 -0
helm/benchmark/scenarios/mmmlu_scenario.py +85 -0
helm/benchmark/scenarios/msmarco_scenario.py +30 -0
helm/benchmark/scenarios/mtsamples_procedures_scenario.py +22 -0
helm/benchmark/scenarios/mtsamples_replicate_scenario.py +22 -0
helm/benchmark/scenarios/n2c2_ct_matching_scenario.py +20 -0
helm/benchmark/scenarios/narrativeqa_scenario.py +20 -0
helm/benchmark/scenarios/natural_qa_scenario.py +32 -0
helm/benchmark/scenarios/omni_math_scenario.py +18 -0
helm/benchmark/scenarios/open_assistant_scenario.py +22 -0
helm/benchmark/scenarios/pubmed_qa_scenario.py +22 -0
helm/benchmark/scenarios/quac_scenario.py +14 -0
helm/benchmark/scenarios/race_based_med_scenario.py +23 -0
helm/benchmark/scenarios/raft_scenario.py +15 -0
helm/benchmark/scenarios/real_toxicity_prompts_scenario.py +14 -1
helm/benchmark/scenarios/scenario.py +31 -0
helm/benchmark/scenarios/seahelm_scenario.py +350 -2
helm/benchmark/scenarios/self_instruct_scenario.py +29 -1
helm/benchmark/scenarios/shc_bmt_scenario.py +22 -0
helm/benchmark/scenarios/shc_cdi_scenario.py +20 -0
helm/benchmark/scenarios/shc_conf_scenario.py +23 -0
helm/benchmark/scenarios/shc_ent_scenario.py +21 -0
helm/benchmark/scenarios/shc_gip_scenario.py +20 -0
helm/benchmark/scenarios/shc_privacy_scenario.py +22 -0
helm/benchmark/scenarios/shc_proxy_scenario.py +22 -0
helm/benchmark/scenarios/shc_ptbm_scenario.py +23 -0
helm/benchmark/scenarios/shc_sequoia_scenario.py +21 -0
helm/benchmark/scenarios/situation_prompts.yaml +49 -0
helm/benchmark/scenarios/starr_patient_instructions_scenario.py +22 -0
helm/benchmark/scenarios/summarization_scenario.py +37 -0
helm/benchmark/scenarios/synthetic_efficiency_scenario.py +22 -1
helm/benchmark/scenarios/synthetic_reasoning_natural_scenario.py +13 -0
helm/benchmark/scenarios/test_alghafa_scenario.py +29 -0
helm/benchmark/scenarios/test_alrage_scenario.py +23 -0
helm/benchmark/scenarios/test_arabic_exams_scenario.py +21 -0
helm/benchmark/scenarios/test_aratrust_scenario.py +21 -0
helm/benchmark/scenarios/test_bluex_scenario.py +59 -0
helm/benchmark/scenarios/test_exams_multilingual_scenario.py +29 -0
helm/benchmark/scenarios/test_healtha_br_scenario.py +57 -0
helm/benchmark/scenarios/the_pile_scenario.py +13 -1
helm/benchmark/scenarios/truthful_qa_scenario.py +14 -0
helm/benchmark/scenarios/twitter_aae_scenario.py +20 -1
helm/benchmark/scenarios/vicuna_scenario.py +21 -1
helm/benchmark/scenarios/wikifact_scenario.py +20 -0
helm/benchmark/scenarios/wildbench_scenario.py +18 -0
helm/benchmark/scenarios/wmt_14_scenario.py +12 -0
helm/benchmark/slurm_jobs.py +1 -2
helm/benchmark/slurm_runner.py +8 -1
helm/benchmark/static/schema_arabic.yaml +271 -0
helm/benchmark/static/schema_classic.yaml +0 -17
helm/benchmark/static/schema_long_context.yaml +24 -6
helm/benchmark/static/schema_medhelm.yaml +36 -0
helm/benchmark/static/schema_slp.yaml +219 -0
helm/benchmark/static_build/assets/index-671a5e06.js +10 -0
helm/benchmark/static_build/assets/index-9352595e.css +1 -0
helm/benchmark/static_build/index.html +2 -2
helm/benchmark/window_services/image_generation/clip_window_service.py +1 -3
helm/clients/audio_language/llama_omni/arguments.py +61 -0
helm/clients/audio_language/llama_omni/constants.py +9 -0
helm/clients/audio_language/llama_omni/conversation.py +213 -0
helm/clients/audio_language/llama_omni/model/__init__.py +0 -0
helm/clients/audio_language/llama_omni/model/builder.py +88 -0
helm/clients/audio_language/llama_omni/model/language_model/omni_speech2s_llama.py +190 -0
helm/clients/audio_language/llama_omni/model/language_model/omni_speech_llama.py +118 -0
helm/clients/audio_language/llama_omni/model/omni_speech_arch.py +249 -0
helm/clients/audio_language/llama_omni/model/speech_encoder/builder.py +9 -0
helm/clients/audio_language/llama_omni/model/speech_encoder/speech_encoder.py +27 -0
helm/clients/audio_language/llama_omni/model/speech_generator/builder.py +9 -0
helm/clients/audio_language/llama_omni/model/speech_generator/generation.py +622 -0
helm/clients/audio_language/llama_omni/model/speech_generator/speech_generator.py +104 -0
helm/clients/audio_language/llama_omni/model/speech_projector/builder.py +9 -0
helm/clients/audio_language/llama_omni/model/speech_projector/speech_projector.py +27 -0
helm/clients/audio_language/llama_omni/preprocess.py +295 -0
helm/clients/audio_language/llama_omni/utils.py +202 -0
helm/clients/audio_language/qwen2_5_omni_client.py +19 -7
helm/clients/audio_language/qwen_omni/configuration_qwen2_5_omni.py +519 -0
helm/clients/audio_language/qwen_omni/modeling_qwen2_5_omni.py +4308 -0
helm/clients/audio_language/qwen_omni/processing_qwen2_5_omni.py +270 -0
helm/clients/audio_language/qwen_omni/qwen2_5_omni_utils/__init__.py +0 -0
helm/clients/audio_language/qwen_omni/qwen2_5_omni_utils/v2_5/__init__.py +8 -0
helm/clients/audio_language/qwen_omni/qwen2_5_omni_utils/v2_5/audio_process.py +56 -0
helm/clients/audio_language/qwen_omni/qwen2_5_omni_utils/v2_5/vision_process.py +380 -0
helm/clients/huggingface_client.py +2 -2
helm/clients/image_generation/cogview2/sr_pipeline/dsr_model.py +1 -1
helm/clients/image_generation/mindalle/models/stage1/layers.py +2 -2
helm/clients/openai_client.py +33 -20
helm/clients/openai_responses_client.py +34 -8
helm/clients/openrouter_client.py +31 -0
helm/clients/test_huggingface_client.py +3 -3
helm/clients/test_openrouter_client.py +69 -0
helm/clients/together_client.py +48 -13
helm/clients/vertexai_client.py +19 -11
helm/clients/vllm_client.py +43 -7
helm/clients/vllm_granite_thinking_client.py +56 -0
helm/common/critique_request.py +0 -1
helm/common/hierarchical_logger.py +83 -34
helm/common/object_spec.py +23 -8
helm/common/test_logging.py +94 -0
helm/config/model_deployments.yaml +525 -172
helm/config/model_metadata.yaml +185 -10
helm/config/tokenizer_configs.yaml +100 -2
helm/proxy/cli.py +1 -1
helm/proxy/example_queries.py +8 -8
helm/proxy/retry.py +5 -0
helm/proxy/server.py +2 -1
helm/proxy/static/index.css +4 -0
helm/proxy/static/index.js +7 -1
helm/tokenizers/grok_tokenizer.py +2 -0
helm/benchmark/metrics/aci_bench_metrics.py +0 -14
helm/benchmark/metrics/chw_care_plan_metrics.py +0 -14
helm/benchmark/metrics/dischargeme_metrics.py +0 -14
helm/benchmark/metrics/med_dialog_metrics.py +0 -14
helm/benchmark/metrics/medalign_metrics.py +0 -14
helm/benchmark/metrics/medi_qa_metrics.py +0 -14
helm/benchmark/metrics/medication_qa_metrics.py +0 -14
helm/benchmark/metrics/mental_health_metrics.py +0 -14
helm/benchmark/metrics/mimic_bhc_metrics.py +0 -14
helm/benchmark/metrics/mimic_rrs_metrics.py +0 -14
helm/benchmark/metrics/mtsamples_procedures_metrics.py +0 -14
helm/benchmark/metrics/mtsamples_replicate_metrics.py +0 -14
helm/benchmark/metrics/numeracy_metrics.py +0 -72
helm/benchmark/metrics/starr_patient_instructions_metrics.py +0 -14
helm/benchmark/metrics/test_numeracy_metrics.py +0 -95
helm/benchmark/scenarios/numeracy_scenario.py +0 -794
helm/benchmark/static_build/assets/index-94295e78.js +0 -10
helm/benchmark/static_build/assets/index-b9779128.css +0 -1
{crfm_helm-0.5.6.dist-info → crfm_helm-0.5.8.dist-info}/WHEEL +0 -0
{crfm_helm-0.5.6.dist-info → crfm_helm-0.5.8.dist-info}/entry_points.txt +0 -0
{crfm_helm-0.5.6.dist-info → crfm_helm-0.5.8.dist-info}/licenses/LICENSE +0 -0
{crfm_helm-0.5.6.dist-info → crfm_helm-0.5.8.dist-info}/top_level.txt +0 -0

{crfm_helm-0.5.6.dist-info → crfm_helm-0.5.8.dist-info}/RECORD RENAMED Viewed

@@ -1,4 +1,4 @@
-crfm_helm-0.5.6.dist-info/licenses/LICENSE,sha256=bJiay7Nn5SHQ2n_4ZIT3AE0W1RGq4O7pxOApgBsaT64,11349
+crfm_helm-0.5.8.dist-info/licenses/LICENSE,sha256=bJiay7Nn5SHQ2n_4ZIT3AE0W1RGq4O7pxOApgBsaT64,11349
 helm/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 helm/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 helm/benchmark/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -7,25 +7,25 @@ helm/benchmark/config_registry.py,sha256=Cd25a8FHriUzAgvGGU5sBAPyhisdSIjdUJR4YbY
 helm/benchmark/data_preprocessor.py,sha256=wqGzAiLwOYa4v6TVPe6ayrnuzdNbmfjeiofRQiO2uso,2201
 helm/benchmark/executor.py,sha256=E7cF1vMXBn5eT1z5Le5ng4M9AaIMLjxfLgMmF1EfZy0,4843
 helm/benchmark/huggingface_registration.py,sha256=DAiHffNmo4H90rBfvQ_LHADtUCnCk6dfpI7Wbat1DZA,4389
-helm/benchmark/model_deployment_registry.py,sha256=zDpqsgjCvtesRan-z2TQA7G97g14UPgjG0Cbi9owWaY,9472
+helm/benchmark/model_deployment_registry.py,sha256=aPBkSr59jqx6ThFW-DYFhi3tPsLLhSKF5JC4-pxqLrk,9011
 helm/benchmark/model_metadata_registry.py,sha256=7XisV0an_edM8hvP8LSoCnTeUN2QLJrQknOCA6-OE7M,8841
 helm/benchmark/multi_gpu_runner.py,sha256=WmTKpVfcKXyiiPzrmxpbvQoZy0Ua8IyPgxB8r_3jrRw,4773
 helm/benchmark/reeval_run.py,sha256=vImL8JNhveEOftZbRQ6JAxF0L-XCKIwh65M6fIYo4RU,7198
 helm/benchmark/reeval_runner.py,sha256=bJPl7XVOVwK2fUA7voOVQYwVFEOfKVnrT2tbSGQzQY8,15584
-helm/benchmark/run.py,sha256=F65P6eG3S6dHDxRK8HMqDFGQjPBGIJouX80ANsHb0Y8,13806
-helm/benchmark/run_expander.py,sha256=hKFLpmq8W2KBl_mBf-ahHEbt67qZFgu-VxjvidOeQuE,56543
+helm/benchmark/run.py,sha256=ZyqkKnqkMqM2AH4HL6sH72H8-mrDWu0NW0piE7BY0HM,13973
+helm/benchmark/run_expander.py,sha256=IMPhg16Yd3diaFRLGYcLCXGO4L_B2WXW69oZP0fx6lE,56857
 helm/benchmark/run_spec.py,sha256=GiIU8iGO2FGYFDWIxt51CeNPsW7rM7BzDqH1KgEL1cg,3217
 helm/benchmark/run_spec_factory.py,sha256=Hxeft3fXoWNz9yGo-2nIfb5pd3GDWlwYWc6YYvAkTjM,7785
 helm/benchmark/runner.py,sha256=O-91eRRrNgE4_tlCVeLq9_0QsRfNELvaQT-KWtJw894,14618
 helm/benchmark/runner_config_registry.py,sha256=2gW5wBLkHdYb2WNbZulto06hTcto2ROvjy8HULw3jNM,515
 helm/benchmark/server.py,sha256=uphh9L0FQnVZVVoGx50MMb_jXh-uen6ouE3uDN5GKFE,6422
-helm/benchmark/slurm_jobs.py,sha256=eNCAoaWDfT0Wk32ZJRIGo-x8kgjhDPnPB4Xrvw_eLB0,3225
-helm/benchmark/slurm_runner.py,sha256=RjmwMqMdKwOzd9B2S6fkuSqB2UjybmiSRVjraiLtzgM,16567
+helm/benchmark/slurm_jobs.py,sha256=6m11gyMo-cA2dwxR2pBXv4tEds5Aok4YCQQyHRmPoPk,3164
+helm/benchmark/slurm_runner.py,sha256=T4vSoxwdRR8gqyL4S2sw_Le-9rv9BPC0BlOy88pwt70,16785
 helm/benchmark/test_data_preprocessor.py,sha256=_esdtkqyU_8Yp5ZOO7n1b-Y4Qc28wpD5drG-4Y4UhIM,2219
 helm/benchmark/test_run_expander.py,sha256=gLeHkNt_nLgbwEJiYxhwda-eKA3sJAxkYolCvgRN5TY,1163
 helm/benchmark/tokenizer_config_registry.py,sha256=ZOImg38ta0FXZYAWna6q7A5xrG2mU7Ofr-8j4EqGlUY,1585
 helm/benchmark/adaptation/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-helm/benchmark/adaptation/adapter_spec.py,sha256=WrDOvQoeV5Ciw2bmvtnz6HTCAEfjCHTYgfTZwRZzkN8,5680
+helm/benchmark/adaptation/adapter_spec.py,sha256=mfqU5lkvN2UOOUrldgTNq_u8iqRajagvzimyGWQhPQs,6054
 helm/benchmark/adaptation/common_adapter_specs.py,sha256=V8aYhQYuwohzwW0T_IU_ymGlxEwARKIiChLvwLKt-ew,12553
 helm/benchmark/adaptation/prompt.py,sha256=vPCFeKVUwpbnTe0IbphkyAKFkkM0YnEONfvjcb8Hj50,2158
 helm/benchmark/adaptation/request_state.py,sha256=WAPyubn35on-Ry7xKpXsVz3wYBMCMc_LidDOdcKxatI,3053
@@ -40,7 +40,7 @@ helm/benchmark/adaptation/adapters/generation_adapter.py,sha256=LI7uWpKIHvTUjGiy
 helm/benchmark/adaptation/adapters/in_context_learning_adapter.py,sha256=8LepCkI5b0MOL70pRPGb7vEH0KFMxIlpCQIVIzQT_vE,15030
 helm/benchmark/adaptation/adapters/language_modeling_adapter.py,sha256=u_GFEgg5wmpate-s5U5aMsmcHuFmreJcA8J0TO1kPCc,14907
 helm/benchmark/adaptation/adapters/multiple_choice_calibrated_adapter.py,sha256=-fY4mvzoGCCoR0HesT_xf2U2m2arVjgDuj59lm07_tg,1923
-helm/benchmark/adaptation/adapters/multiple_choice_joint_adapter.py,sha256=lzmHwvDOHWl9IWC3NTLGfJDbduXtK_zrS2_YoUQmdc8,4464
+helm/benchmark/adaptation/adapters/multiple_choice_joint_adapter.py,sha256=nOCuX9lFKb3BHpznhTwpNCO0YsZBNhcMYuFnsLT_u-s,4579
 helm/benchmark/adaptation/adapters/multiple_choice_joint_chain_of_thought_adapter.py,sha256=RV6B3i5juBbJCtPDWzSfma49YXeDq3vQAQ5xQwnH-cA,3282
 helm/benchmark/adaptation/adapters/multiple_choice_separate_adapter.py,sha256=hhH9ehK092j1WdUwrKYSy5PvNJ73gsIu6-5W8aLoYVI,2190
 helm/benchmark/adaptation/adapters/test_adapter.py,sha256=7Nr6kMK3JN0UjMjjZ6P1fsD5xhOeaqh0D1xI6LFKCos,641
@@ -51,12 +51,13 @@ helm/benchmark/adaptation/adapters/multimodal/__init__.py,sha256=47DEQpj8HBSa-_T
 helm/benchmark/adaptation/adapters/multimodal/generation_multimodal_adapter.py,sha256=MvE7YdIt8Y0nefXLskY9gPmXp7QWi2b8cqg8fxUpzbM,1980
 helm/benchmark/adaptation/adapters/multimodal/in_context_learning_multimodal_adapter.py,sha256=KXP9MzDdmUao3uVjPgZYKjZQ_LvGHgZvI-86o3E87xA,6404
 helm/benchmark/adaptation/adapters/multimodal/multimodal_prompt.py,sha256=jyL61UxBsIr68hUz-jtjBUnyB2HBp5ESNyECGp_Gf6Q,2129
-helm/benchmark/adaptation/adapters/multimodal/multiple_choice_joint_multimodal_adapter.py,sha256=ftwSOTPugDuw8vh2WaQDJb0tQAeWR7S7qtD4yE_nOt4,4804
+helm/benchmark/adaptation/adapters/multimodal/multiple_choice_joint_multimodal_adapter.py,sha256=GP2Fg1kW0-5jCkjgzVkhuN7YBQFyFgQpPTfpSgfbAvk,5178
 helm/benchmark/adaptation/adapters/multimodal/test_in_context_learning_multimodal_adapter.py,sha256=mjjyn9p31V-yt6S8BX7SvqvkQ56D9cKSff6d-daM6HM,10250
 helm/benchmark/adaptation/adapters/multimodal/test_multimodal_prompt.py,sha256=6nuz0Vn89A1mOedutsiq2SwTOG3qn8dUZTiaXhKffiw,3587
 helm/benchmark/annotation/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-helm/benchmark/annotation/aci_bench_annotator.py,sha256=SjXidlbpm5HOhdhNXg3HjabMEQvt3hq1iJ5GPajxt8M,3228
-helm/benchmark/annotation/air_bench_annotator.py,sha256=Xvqzf-f29dzLGuAMeNiQe_kSkMbXEN1_U1LwCAn6nJQ,3500
+helm/benchmark/annotation/aci_bench_annotator.py,sha256=aAzXqbjj_3bv0-ATCrFu4JvrsqORE5lkYpgxtXAEGSA,2777
+helm/benchmark/annotation/air_bench_annotator.py,sha256=CDyHVwD4eoymfLduJC5WvvhDX1DOgYBqgjvqBjoCfU8,3501
+helm/benchmark/annotation/alrage_annotator.py,sha256=3DcHbD8WXTg5PN3feipHTsFls0v5owMyb_rqpNWokls,3531
 helm/benchmark/annotation/annotator.py,sha256=__BkMVpAEpSs1pbwPK5sVWLdCAXnjsHcPYgmOqmNPu0,1843
 helm/benchmark/annotation/annotator_factory.py,sha256=8uo5uz1UpIVCHUd7CRvmy6b9XB1gspdHmgxH5UZMPVI,2335
 helm/benchmark/annotation/anthropic_red_team_annotator.py,sha256=4hob15m2k9e2A97E0aG9FstCbJ_oMM7-9y-nh2EaYqc,2395
@@ -65,28 +66,28 @@ helm/benchmark/annotation/autobencher_safety_annotator.py,sha256=w_xjZmY1zuLjVvV
 helm/benchmark/annotation/bigcodebench_annotator.py,sha256=CJG2pn1DeHJCp3yHETRquNIkCHfd6ZNuOiUjG1cQ_JY,4448
 helm/benchmark/annotation/bird_sql_annotator.py,sha256=FQDZs1-O1jfJOET0eDeU7lf5xLaiMPohC5BdmQ4XkzI,2436
 helm/benchmark/annotation/call_center_annotator.py,sha256=pTEjwfA4tgZhroFbamoQ8IO_D1O9r6k5GIlD50JEg5c,11601
-helm/benchmark/annotation/chw_care_plan_annotator.py,sha256=6ybNBvJi59i0cpAhI_fLwXoSnqhAH6m7Lo6ad_PufBs,2966
+helm/benchmark/annotation/chw_care_plan_annotator.py,sha256=R6Hexh20T6WBBRBhwLhQv_IQvW7Z55Pf9IYBCWxUTaQ,2517
 helm/benchmark/annotation/czech_bank_qa_annotator.py,sha256=YIH5g4zHe3BQF2Y-6uRVw7g9u_SPBncqBobdvZdIzyA,3096
-helm/benchmark/annotation/dischargeme_annotator.py,sha256=Z6xnUK1cNrFco9x0w8B_qhlLOEZrzXBwT6TKZPKoPBk,3676
+helm/benchmark/annotation/dischargeme_annotator.py,sha256=blP76BgwmbHDDDRdaaGwtTHfukCvXXLN72vjGj_LI_U,3225
 helm/benchmark/annotation/ehr_sql_annotator.py,sha256=Izpq0biZ9lkJOPk6NwTuv2wk8Bg88vj56BKZrY8XhT4,4021
 helm/benchmark/annotation/financebench_annotator.py,sha256=gNERLY35t2kcpayXGGrY4-pBs2jbEUomqElRYbb9nho,4150
 helm/benchmark/annotation/harm_bench_annotator.py,sha256=zhkWnV3qZgY-nvHgQRHGrrCMC7605JwFHesY7UC3ZnQ,2293
 helm/benchmark/annotation/helpdesk_call_summarization_annotator.py,sha256=I7TjpN502Sa-Z4uUKemJXSAdOiVA3MMO92YIAAXeDBg,6034
-helm/benchmark/annotation/live_qa_annotator.py,sha256=8DXsjwmeSyvC0kfp1uYds4cwpxqzF7FcskeZaXxXiOw,3552
-helm/benchmark/annotation/med_dialog_annotator.py,sha256=OVTFIlvdhcOr_hdK0tnrDes9hYdN1mDWFTp4GDYY7O0,3162
-helm/benchmark/annotation/medalign_annotator.py,sha256=8edAZh8oQgDKUT1bQ3Hp2NBE-QnBZ_-ZQjHkV7YKWhs,3240
-helm/benchmark/annotation/medi_qa_annotator.py,sha256=v8e6hkHZX1x9KtTedCnpCseh-Y72z5kUgUrXHWPUkX8,3074
-helm/benchmark/annotation/medication_qa_annotator.py,sha256=uZ3VpJ0nsDyF70_kn8kSSBPr4OlfiNdZC7q8wq_jJFE,3090
-helm/benchmark/annotation/mental_health_annotator.py,sha256=JwgSeXtwf4KFZxNtAxsnqdLJQSvP-F-ZoCcCWdasrMQ,3275
-helm/benchmark/annotation/mimic_bhc_annotator.py,sha256=pwwniNlu5VTa1ZdyO0KFcMFZcpqM5CjguujgSpEGslw,3174
-helm/benchmark/annotation/mimic_rrs_annotator.py,sha256=zABO1FJH9pOFhUe5vc2B-c14Hf5RsuU9jQAGiMg6G0I,3204
-helm/benchmark/annotation/model_as_judge.py,sha256=FIJOUzIhf2QpxqFf6hjgAM5hPEm0VlXzB-jiHJUrPDs,11985
-helm/benchmark/annotation/mtsamples_procedures_annotator.py,sha256=qqWHY2HfCwMP5GqvObS3JpMIYVs4yyITCsA1B7lcDks,3201
-helm/benchmark/annotation/mtsamples_replicate_annotator.py,sha256=TUxNzJcItErsw0gw76hiKZAWeQTNHGHnC0qf-_CGeF0,3316
+helm/benchmark/annotation/live_qa_annotator.py,sha256=PSff59mU_t3ypmptYsYRKU3m1vMLF0dMyUySIOxBrPw,3553
+helm/benchmark/annotation/med_dialog_annotator.py,sha256=uGp8d74WGgOOiexpoKj5CMdr5jOvAnfe-ZLKGSHT6ng,2711
+helm/benchmark/annotation/medalign_annotator.py,sha256=glAPpVdIfebm39GhrBY3BE2hdofVBIBXUxPU3_qqZOw,2789
+helm/benchmark/annotation/medi_qa_annotator.py,sha256=bLXxXe-obPvud15sPrqp9i-wSq1QqguCPt_UJaXRz_I,2623
+helm/benchmark/annotation/medication_qa_annotator.py,sha256=98XU2VVSoQ8XlAkuVKWnNBOS76X_lIviq_A-nyrlqcw,2639
+helm/benchmark/annotation/mental_health_annotator.py,sha256=08b_XqgfSpIhutDUaaSgVRdiZB6metAQQ_WHF8U2-c0,2824
+helm/benchmark/annotation/mimic_bhc_annotator.py,sha256=a9AHMFY2shV4I2qVUfKnOvZFbmQjL5vPKsbytTBfU0A,2723
+helm/benchmark/annotation/mimic_rrs_annotator.py,sha256=eu9rZhRAXVbo0j7BP7vuAKwGkuwhTCvVRvJ4dPbcR4I,2753
+helm/benchmark/annotation/model_as_judge.py,sha256=eZZlyCrW6U9a8bHhaPrbV1AJ23q3uP0ho1NbVErGBXs,12160
+helm/benchmark/annotation/mtsamples_procedures_annotator.py,sha256=ZgJVtNpab3BrMs0ZXFW6L0CNp1Hcqfgv7FHP4rpxFPg,2750
+helm/benchmark/annotation/mtsamples_replicate_annotator.py,sha256=VtHiEGFZLUsd3zkgnSoti5itZnDPgERMPZlORkEp7ok,2865
 helm/benchmark/annotation/omni_math_annotator.py,sha256=PvZZb1oGw60qT-oHRIs93AZbh5wTbpsmD8BforudFhA,6144
 helm/benchmark/annotation/simple_safety_tests_annotator.py,sha256=if4S8MaENr1HZ42ZsOjDPXZ-kJ0p4l4B2j9m994RuxQ,2140
 helm/benchmark/annotation/spider_annotator.py,sha256=B48ylGg5J7xuTSUio7VztdXk3lI6ilMqrUvAD-ve0sE,621
-helm/benchmark/annotation/starr_patient_instructions_annotator.py,sha256=5jU-dK_0OvB_jXNLDZtQ5E3gaSUcAxFNzv6prA17eAg,3186
+helm/benchmark/annotation/starr_patient_instructions_annotator.py,sha256=Te9rQhcUV-T2I4oBCBzInAZW65EV3lv0LXLPgGzLd8c,2735
 helm/benchmark/annotation/test_annotator_factory.py,sha256=ifv5hxSbFe113AHeXLqTPkVJ-C2PW_gb9L3a0SHNi-M,986
 helm/benchmark/annotation/test_dummy_annotator.py,sha256=LfY1ErJDUJ7rD8JUy92RUDD1b91jUs4Nk8Gvope-Z98,1644
 helm/benchmark/annotation/wildbench_annotator.py,sha256=OXR59zdKw9W7v3Q_sFnt1cEPN3nOzQDVqSbh4jDbEUs,5457
@@ -126,74 +127,69 @@ helm/benchmark/efficiency_data/inference_denoised_runtimes.json,sha256=ios_dt-_8
 helm/benchmark/efficiency_data/inference_idealized_runtimes.json,sha256=5w7reeZc0yc4cjH8kJGxQQSoe8yaRVX2SSlSrx0QWFQ,12348
 helm/benchmark/efficiency_data/training_efficiency.json,sha256=aH2moiBLStOLVi8Ci2KTK5ZkWlTBLK-B3fRfNZwhoSg,9763
 helm/benchmark/metrics/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-helm/benchmark/metrics/aci_bench_metrics.py,sha256=fAuTm8Sr1vvyd7Tjcz9WWKrFkqrwCV-CiF6lqUO3dKU,442
-helm/benchmark/metrics/air_bench_metrics.py,sha256=VMNQDDEtz2CiK4U55lCHLz0b_DxHprTAZ1WtYtGXjcY,2282
+helm/benchmark/metrics/air_bench_metrics.py,sha256=WvfjjHLSE567Y7BC8tGlMINBwP-d1URRUZcMUF1yf1g,171277
+helm/benchmark/metrics/alrage_metric.py,sha256=4QHtL00aEIRYQx2QkDs5uldu7ZAkbFYMALH6DL9LSJg,1233
 helm/benchmark/metrics/annotation_metrics.py,sha256=JbXNleQsPJVF2uc1xXgUW2bzvJqwLPZyhnndqc6THv0,4268
-helm/benchmark/metrics/basic_metrics.py,sha256=d0iwYnwrbF7w7CFtazx8vPIsZnj51U2PVVoscCb-HJA,20495
+helm/benchmark/metrics/basic_metrics.py,sha256=3y1M0mFJL8FlkMkQWWs4ZV2NiriaMGydddbeY3F-vXk,30547
 helm/benchmark/metrics/bbq_metrics.py,sha256=GeZhSSJzqGD0e5EAiRHitIC3XtPICF7rDI6GfeYQc8E,6201
 helm/benchmark/metrics/bias_metrics.py,sha256=8qcInRJwQsuCI-lMC1umd-ZZaYvorUPrMjnuC6vSeb4,11602
 helm/benchmark/metrics/bias_word_lists.py,sha256=eyk6we2J4SW8ZaZxQUWLB7Yapn92uM5TCekhFB5vg-U,13908
 helm/benchmark/metrics/bigcodebench_metrics.py,sha256=JcPZrSiHR-kxT-MFM8zXqOs6wTC5Hus3TbxuHFQVZow,860
 helm/benchmark/metrics/bird_sql_metrics.py,sha256=ooCuXW5nPpRs_-4seCONQmn25DzTbcUgGXznXTK9y0Y,1153
-helm/benchmark/metrics/chw_care_plan_metrics.py,sha256=WOAdwuF4vusZhjaXSAB3r7PD_ZxeNmVu2oAmOqzVLtU,460
-helm/benchmark/metrics/classification_metrics.py,sha256=1Xa_bO4PqIAV2iZitE69kc4VKS4A7PloG5ElZAgvmh8,8851
+helm/benchmark/metrics/classification_metrics.py,sha256=CfkyMiiWo74VbIB7eEhNxIcPbGA_imbzETrAExqn5WM,9498
 helm/benchmark/metrics/cleva_accuracy_metrics.py,sha256=1eDxHxVk-JW1mF9SBcuplIefAoi_edUwKpp-XxYbmeU,2740
 helm/benchmark/metrics/cleva_harms_metrics.py,sha256=xVubv2pG3iinVs3namoVHWAmV9oUPywZwFB_0JGhP_w,11277
 helm/benchmark/metrics/cleva_metrics_helper.py,sha256=8UwiGhekUmp7DxYWU4rxqX2v3ewkg-O5-jOh49iOGmc,304
 helm/benchmark/metrics/code_metrics.py,sha256=SebQ5MXJe_phTiMfGMfhgYago-hwh_g9ctBWEHGqCnU,5230
 helm/benchmark/metrics/code_metrics_helper.py,sha256=UNai154RuhYRZM_YK-rveLct4Ui5iEBNPYmYdKq34Xs,22712
-helm/benchmark/metrics/comet_metric.py,sha256=qOvwE0ov1plb6SwwT3CbX1XuSo4GJ-M3iRe98yMiMaM,4797
+helm/benchmark/metrics/codeinsights_code_efficiency_metrics.py,sha256=biKk67r4ij3pK2L0OuGTJ4BAb8ig5tpGAV86uBD1qNs,7832
+helm/benchmark/metrics/codeinsights_code_evaluation_metrics.py,sha256=QrePgX-1UALQKs1dHMfOm1qoALvOU1pbLyC4JmcINx8,19083
+helm/benchmark/metrics/codeinsights_correct_code_metrics.py,sha256=CQs9HXh7P1vzkKWdpvugvttD_8ZF6W_QPp7_rhYFwsY,13873
+helm/benchmark/metrics/codeinsights_edge_case_metrics.py,sha256=B7EEELwwH67VxmgrTBSP25Etyb5XYIDuadfggMrHmcE,3866
+helm/benchmark/metrics/codeinsights_metric_specs.py,sha256=BkKWII9yTkChdZVsGeeeCbiWQDYvvcAKo0nxi_RTTUk,1798
+helm/benchmark/metrics/comet_metric.py,sha256=EJWZ9x8CGeDDQlfxYrY-np_NVJBt5gun0XLJvtpjXVI,4798
 helm/benchmark/metrics/common_metric_specs.py,sha256=JKqmO4ovBdfOYKC-00OSzOMv--g9NTCVfUHLaz-1Uns,6025
-helm/benchmark/metrics/conv_fin_qa_calc_metrics.py,sha256=Zrf6HyH_WNe7gGFgW0j8FJlX5KZvbk-05iX8QFPJDyU,2656
-helm/benchmark/metrics/copyright_metrics.py,sha256=_Lp7sKWgacY_13kFadNfnhrM2Ks8syBXnUW7zYuJkwo,7817
+helm/benchmark/metrics/conv_fin_qa_calc_metrics.py,sha256=F2bfg8XbjH3WOQ0O_c5S7UUxgpzu7AD5wRtNdNcJlUs,2997
+helm/benchmark/metrics/copyright_metrics.py,sha256=RYOWKFN97UCD2Vj51gzKGbnnY9wAq6KJgiRt2cecVfs,7824
 helm/benchmark/metrics/czech_bank_qa_metrics.py,sha256=bKoooK2T5v_fFKNbUnsuW6Mv9muAirJD5lTrzuHfpz8,1113
 helm/benchmark/metrics/decodingtrust_fairness_metrics.py,sha256=x66XP0iQGk4ThT7ddmrlLCA0XF4arRbQMDT42LHf2kE,3297
 helm/benchmark/metrics/decodingtrust_ood_knowledge_metrics.py,sha256=TxTkkWdx6d6ym0MirZTiucl_TWFdn4uJLnlTfLjQvgk,2925
 helm/benchmark/metrics/decodingtrust_privacy_metrics.py,sha256=OU7lka-hm6PubR5Gjj4uNyrqhjlfhe0mmjBCAz9vlRs,3456
-helm/benchmark/metrics/decodingtrust_stereotype_bias_metrics.py,sha256=BKDD3lblqT6Ebi5kEC4zbN_OvQwD1SdEtBv5Wf0kzWw,6460
-helm/benchmark/metrics/dischargeme_metrics.py,sha256=D8LI52E17hNSPDpEvb2tw1za4QWDE3p9xgx7Nm9l7_Y,454
+helm/benchmark/metrics/decodingtrust_stereotype_bias_metrics.py,sha256=bW4zafRyKFa__8fGrdiTPUu848ovNnvakLCfqcMrcHk,6461
 helm/benchmark/metrics/disinformation_metrics.py,sha256=5n8wgRBb6FaDjqe1nR3Cj9aS48esmMsIUq4KpBHoQoU,7870
-helm/benchmark/metrics/dry_run_metrics.py,sha256=Ss0lzf944HIbL1CX6QuJpGFPqOzhBT0qVWLNR1BoEjk,3784
-helm/benchmark/metrics/efficiency_metrics.py,sha256=SJqpA1d_GfBPl9H6moai8ra1GVe7tlaCfg3PeiWT54c,11845
-helm/benchmark/metrics/ehr_sql_metrics.py,sha256=YRjvPIty7zlyoyGD6wo3HYOz7y_PThySOZzVRJ38iww,4797
+helm/benchmark/metrics/dry_run_metrics.py,sha256=ouS6_8lESuCGSQgegN4xKKyoGr7Rb1K-dufHPT1fDwc,4886
+helm/benchmark/metrics/efficiency_metrics.py,sha256=VnM5PgxxK6UKk9MzPprnN_7d-t6xVlIgFMQYrFh8dwY,15262
+helm/benchmark/metrics/ehr_sql_metrics.py,sha256=yyz-2tsk4Fu6D5ELp3cbLaAWGjqtDGrUdvFvgHvxevg,7418
 helm/benchmark/metrics/evaluate_instances_metric.py,sha256=LGk1Dv_76Ak0YUlWKFTsOLEFiBSmcGVhNrbj_4zg9g4,2913
-helm/benchmark/metrics/evaluate_reference_metrics.py,sha256=t0251_2aA0CrXB8oUBKlPRgPl-xYjzdVhLcGjwuhOgo,19621
+helm/benchmark/metrics/evaluate_reference_metrics.py,sha256=T1AUnN1wYFrTBMLyys3AbvlArIenZwCPwHa_F7J9ODg,31476
 helm/benchmark/metrics/fin_qa_metrics.py,sha256=MtXxGMGYiCiwCD1CclBXPopzly-Tz3zJTrXJaHYTXn4,2470
 helm/benchmark/metrics/fin_qa_metrics_helper.py,sha256=sH5FIpsxxGUkXO21YGS2EtVsev1EdQ44lYoqFZPSSGo,11884
-helm/benchmark/metrics/gpqa_chain_of_thought_metric.py,sha256=HRRKkcTbCu5ScOVwmjzYaA7UAEGE_AJUZVOCDRuv4Po,4321
+helm/benchmark/metrics/gpqa_chain_of_thought_metric.py,sha256=Lkil9DRtO3NS3zr5Ef_qqGxZBL-ObCNpbKoJvMhCrb8,4762
 helm/benchmark/metrics/gpt4_audio_critique_metrics.py,sha256=L9tGFwvl1-Ew3MdInQ7KPa8OlI5YexIB2KuCYVYsuPY,7023
 helm/benchmark/metrics/gpt4_audio_refusal_metrics.py,sha256=vYPRJq-4uNhUWUWMrDkpHmfIBkhEyAgaMNEI6RKPP80,5896
 helm/benchmark/metrics/gpt4v_originality_critique_metrics.py,sha256=1m7IWy9vu66svnmdBRjZQI-2YsGYzH2vXZMptlRGM0Y,5654
-helm/benchmark/metrics/helpdesk_call_summarization_metrics.py,sha256=9-kB3NeBacI6nxs2oQ7Km_1SHyiz98UVZuR8PAlvCHM,1442
-helm/benchmark/metrics/ifeval_metrics.py,sha256=4_Vp9bNnrctKtv6xZ1RpvBstPAZPwv1xiohH-ogs99U,2565
-helm/benchmark/metrics/instruction_following_critique_metrics.py,sha256=RR9cMIG113oXUnBjU_denn7DaCGB11k1oGtQ5dQON3M,9874
-helm/benchmark/metrics/kpi_edgar_metrics.py,sha256=1GsW-nBz8TgP4wFIVEGA4_BhI17kihmk96zuLpD4NZc,4636
-helm/benchmark/metrics/language_modeling_metrics.py,sha256=yS7k8iFjxfkckSBA0RVA7VdOivSEBtNzCjczK6We7y0,4598
-helm/benchmark/metrics/live_qa_metrics.py,sha256=f2XFmQaohjQNqYqNg8NcDVavCzyP4cd8Cl8rLArn9EM,816
-helm/benchmark/metrics/llm_jury_metrics.py,sha256=yzAsdacyX0MFJy2qKIjhI0y7JvtflELpCh6R14wuCgk,1704
+helm/benchmark/metrics/helpdesk_call_summarization_metrics.py,sha256=5Z43F9ZI9OHBxeZENBGSE4fB1YTo1NKOquPt_Sw-F5s,1835
+helm/benchmark/metrics/ifeval_metrics.py,sha256=33IqTVdYlX9ZI6sR-FfFAKbVJ9tAGDNqZpLHS5yInio,3036
+helm/benchmark/metrics/instruction_following_critique_metrics.py,sha256=AK_ZpayimVZ9MxX8CJG-K1uPKo2j1dNJ_H9uSz1CWiY,11612
+helm/benchmark/metrics/kpi_edgar_metrics.py,sha256=rnvVlvFgWwaavaIu9n8iVlODhkk2g3liOiK7kwfGbN8,5474
+helm/benchmark/metrics/language_modeling_metrics.py,sha256=NK8vYLFyFAidDG8UXVkP242zbg_6W6EZ4xZPNbokGlw,5001
+helm/benchmark/metrics/live_qa_metrics.py,sha256=YGodrQ-b9ucQTK3ICKXRla5r26RR0wxC4iPOTcYrV1k,1195
+helm/benchmark/metrics/llm_jury_metrics.py,sha256=-5w8tFG4JE0cMcH3KS7xQ1z6mbdtDf7reCMz6u5vtag,2158
+helm/benchmark/metrics/lmkt_metric_specs.py,sha256=0Fa0xLjQDXwsRCE5VqGzEfb5ZdzKsDoSCwR_zHogFcc,376
+helm/benchmark/metrics/lmkt_metrics.py,sha256=GaZTfl-NQXa1YSzcJUGlZ5wZURH1CnJxGkPFBj8ydTQ,1856
 helm/benchmark/metrics/machine_translation_metrics.py,sha256=22vaGBCSw12uM1wmtDG-MBBZW8OiTZwNPaerjckdtDE,3860
-helm/benchmark/metrics/med_dialog_metrics.py,sha256=kzmrkQcmJ15zuOF9_Onk9N0oeNeyl9Rri1JEb1AqRT4,447
-helm/benchmark/metrics/medalign_metrics.py,sha256=q6l8p5Pie-H9pxhaA-lQkSOnliJWXr6zUeN8syEQ91Q,439
-helm/benchmark/metrics/medcalc_bench_metrics.py,sha256=9wZgg20-9QBNk0_XhuwR3LT940fqDPkCM4Kl0dPkbAs,5353
-helm/benchmark/metrics/medec_metrics.py,sha256=hNBOGX52G_QOmgTCp9LnIMrmGSRxbb5vgjxKU069TMQ,4152
-helm/benchmark/metrics/medi_qa_metrics.py,sha256=JWAEMuT0UXDZrb7qHn13W6W79ilbprk492V_9vWrB4s,432
-helm/benchmark/metrics/medication_qa_metrics.py,sha256=wit3nKNWpGFfgauu6Xye2IDTePAS0VHAQI_7OO9HR6M,462
+helm/benchmark/metrics/medcalc_bench_metrics.py,sha256=2viECYEj8y65_w5MPH295Z1OgLTNrgP_iMzzYSgc2hQ,5895
+helm/benchmark/metrics/medec_metrics.py,sha256=5z3HKZCEuQsOix-22PPzTHhWlYmjyHOAVFV-bgGUVJE,5137
 helm/benchmark/metrics/melt_bias_metric.py,sha256=mHDCkRGLD-0pyJA_depi_KX3sn7g7Bgd3_m0XdLQahY,11520
 helm/benchmark/metrics/melt_bias_word_lists.py,sha256=xA0araUdszAIOqfxiTi6MIJhKYwr_Gwsc1L9qinZx9U,27891
 helm/benchmark/metrics/melt_metric_specs.py,sha256=zaeV57LQEl8qK7be36NaojiUJlzmkoKY8JyOkOVuPqs,1619
-helm/benchmark/metrics/melt_toxicity_metric.py,sha256=8HxViwOJCAZ-luE_Br55xUfJn5XAVXg6lqcAUsP0GT8,4187
-helm/benchmark/metrics/mental_health_metrics.py,sha256=4HXCXl2GxFPn6wDzHptHeBTuP4BJVLUzEUKffpd5R_k,462
-helm/benchmark/metrics/metric.py,sha256=jqQyiKDq_pQv-ulGqfZI56ydRDQs3N3XhfHIPysUhrk,14311
+helm/benchmark/metrics/melt_toxicity_metric.py,sha256=ni6bb_QC51NM5jQpbFYLWtsQy3tNOLwQ_5b3PDV5vVk,4193
+helm/benchmark/metrics/metric.py,sha256=gF7KlWPoPIGUvbvqDeXagBNBZnl8rclh8JfgCPvuXvs,15065
 helm/benchmark/metrics/metric_name.py,sha256=POhgmUqqIWh_LjCbYpiKkzGqqChBLeW3FADy9u_FcWw,1354
 helm/benchmark/metrics/metric_service.py,sha256=bJaM7GisEgSWR3vPTcg7b67XF9X2K5viODacIgbGb24,1692
-helm/benchmark/metrics/mimic_bhc_metrics.py,sha256=da1YYrE8fL3YHeIJ9hf4WCKZtuj_8cksm3rJ24rcy70,442
-helm/benchmark/metrics/mimic_rrs_metrics.py,sha256=x3vSj1VG1UkNF3gbgJYDeA4z-crxfGIkK7iZo0xjq8c,442
-helm/benchmark/metrics/mimiciv_billing_code_metrics.py,sha256=Pu9efXoBrhsvxSeGHqwbUA5k365-pJTeXpMNhmcg0L0,3927
-helm/benchmark/metrics/mtsamples_procedures_metrics.py,sha256=XrddVk-gnc8jF8amCI1RBa_XTS9yEXD2Y9Ld9W7Q-m8,497
-helm/benchmark/metrics/mtsamples_replicate_metrics.py,sha256=rmH34aTX_wZWxLi4jrxf3sR1RIqNRF0QDANLRQUGhqM,492
+helm/benchmark/metrics/mimiciv_billing_code_metrics.py,sha256=3kypTnrkbdG-Dpdbg_A_WQYVx35ylvZFjh2-R5wvhSE,5347
 helm/benchmark/metrics/nltk_helper.py,sha256=QMEps-lqJZ_pCgvjlMf4BvC0pzDu3ez5jit5F4p8dAk,1313
-helm/benchmark/metrics/numeracy_metrics.py,sha256=3E-CMmB2wuGW5tLjmEm8wFMf85DJ1ZDUANfh84SQuP0,2906
-helm/benchmark/metrics/omni_math_metrics.py,sha256=Gqih87UrE93-a0hbRhTBkjmfGLNTkuKQGaG-sTQeuG8,1287
+helm/benchmark/metrics/omni_math_metrics.py,sha256=WF0cWpmJwduTdZw7c_O5QsXDNwet5GgHYV0Ww9PfKc8,1709
 helm/benchmark/metrics/openai_mrcr_metrics.py,sha256=TAop7G50FKaR-Jyo2EGLqmMOfJRmS2vNRDFiifa6mhg,2313
 helm/benchmark/metrics/output_processing_metric.py,sha256=ey9UBi2f3780OwFlp82ymzfjLR3MA2fpA9vW5R4W5TA,2581
 helm/benchmark/metrics/output_processors.py,sha256=ULZlDBOf6NupAXzDKBKyTDdgPZ5PSxOAlOYTbrQEek8,472
@@ -204,24 +200,22 @@ helm/benchmark/metrics/reference_metric.py,sha256=hseI7A16SOC8ymYZYFCL6nxnyxn0q9
 helm/benchmark/metrics/reka_vibe_critique_metrics.py,sha256=CwzzQ13bBT0r_o75TqFj2Zr0ST9vzQi74K_ezWTnLCU,6568
 helm/benchmark/metrics/ruler_qa_metrics.py,sha256=OuiA0ksByl0Tw1Oal7zbedhKjTrhJgQJDLXAgoTLXuc,1473
 helm/benchmark/metrics/safety_metrics.py,sha256=oARko_EwVnykBKYxi-w3ytKme4qcb1waz_0N2GKbSlg,3348
-helm/benchmark/metrics/seahelm_metrics.py,sha256=egRkeXnnb8Nqi9qJJMDXJRSl4NK6WvdUxAc_LffBips,6964
+helm/benchmark/metrics/seahelm_metrics.py,sha256=GlNoK1O7kcuiuEOJEgTsnrfK9TcGwH7-tPj6Qe6JV90,7493
 helm/benchmark/metrics/seahelm_metrics_specs.py,sha256=cx8p4kwTuEOWxZioK9CVoeTNJT0fZjxRy_6_EM9F394,452
 helm/benchmark/metrics/spider_metrics.py,sha256=RSrFJoA5SNcNxfmgVqCQixcSLrfJBYuVQw5jsfrc9Xg,189
-helm/benchmark/metrics/starr_patient_instructions_metrics.py,sha256=YHdTeIFdZxRbvqBnlWpAyIsWzZyWAjjDFuKOXhHYiSM,525
 helm/benchmark/metrics/statistic.py,sha256=ATuOm0jU3L-0ELiZaF2GVMNF22W66-rMvzxRtlfqcII,3446
 helm/benchmark/metrics/summarization_critique_metrics.py,sha256=-mki8-zvZx54dQg8X0BG2Y6wmfypQhkIuD_9ZjNBl78,4782
-helm/benchmark/metrics/summarization_metrics.py,sha256=FJCdGRmlCJX5A-AmbtpGGlGRfNgg5Z8Bo0d9yFiE33E,16876
+helm/benchmark/metrics/summarization_metrics.py,sha256=S99uhtvBtH0UQS-gDEuQLLTPYNG-dNUV1n3OnaOP7p8,22647
 helm/benchmark/metrics/test_bias_metrics.py,sha256=qEZsCULvwjVdIyfNgJSc2L7Xp9suKKW7L5OuQmGrwZ8,6393
 helm/benchmark/metrics/test_classification_metrics.py,sha256=CRDMGmVmzEUnNaM0C02qUTOU2AS11Mt2-GdEl89y7lw,9541
 helm/benchmark/metrics/test_disinformation_metrics.py,sha256=U3ZmS9s33oimTQbKO-7pgWeX_WiDB9chlOCtf_vslXw,2249
 helm/benchmark/metrics/test_evaluate_reference_metrics.py,sha256=B7xtDDWPAxF7d-vcUx_R51hFMae-DD52nUwbu_eWt6Y,1601
 helm/benchmark/metrics/test_metric.py,sha256=0sGlXE3_Al_VyKpOPBhQR_xT-XrcVgGepLpwut37DmA,771
-helm/benchmark/metrics/test_numeracy_metrics.py,sha256=ls1ZIHDePKpHMoqAbf4HmJ1SIBjLFuLIzGbfg6OiZvM,4162
 helm/benchmark/metrics/test_statistic.py,sha256=yK6m2BZ5UXWmb2D1cQzDH_2ELvrNDaR_lyzX4WoHw9Q,1273
-helm/benchmark/metrics/toxicity_metrics.py,sha256=ZLOzxDlMgbljl-9y6vT2ZgwdhsBZ4MfV-T66VpKk00U,4114
+helm/benchmark/metrics/toxicity_metrics.py,sha256=s5Ypodu4cBmIc_fCbbQ9kCqcvVJf-OQ6zAvb85r8Cv8,5509
 helm/benchmark/metrics/toxicity_utils.py,sha256=-bfittLtMkHyV5wu-hj6KVtaiNGgVIO5duUmThBlX8w,988
 helm/benchmark/metrics/unitxt_metrics.py,sha256=8fawxnrg0xsAe0xO2wbL7S_yisj8RzJnrn6xtk8C6q8,4852
-helm/benchmark/metrics/wildbench_metrics.py,sha256=sY7MNTzRlJJK3yph3rCijgbMaajtLyCCquThlsoE5wU,1380
+helm/benchmark/metrics/wildbench_metrics.py,sha256=THOguxE6GUun0zTr-pITXfQGEd664sScrfIzFGdNPXk,2163
 helm/benchmark/metrics/ifeval/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 helm/benchmark/metrics/ifeval/instructions.py,sha256=qNoa1vMPDNz6ORWfyMv_efwKZ4U5zkI-cf4aApyfSqU,53247
 helm/benchmark/metrics/ifeval/instructions_registry.py,sha256=NprvkRQz0QWaIpJsFp95CQCWsnuY_57ZSqFn2IISDP8,7555
@@ -230,7 +224,7 @@ helm/benchmark/metrics/ifeval/instructions_util.py,sha256=VhkJfZLCaHi094rZSoeQbo
 helm/benchmark/metrics/image_generation/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 helm/benchmark/metrics/image_generation/aesthetics_metrics.py,sha256=UqjBgAi1ylegvHBjALJ8vxINhHEqqr2fSvN9lXgyIZk,2140
 helm/benchmark/metrics/image_generation/aesthetics_scorer.py,sha256=ISdThDKMrx-SHQe69dCcr8qUrMCa_GsxX3BeZnd0WPA,2538
-helm/benchmark/metrics/image_generation/clip_score_metrics.py,sha256=tUnAoew24jjjbjPaoE2-4iyRTq6YNW8Xfk1p5JWZkAU,3338
+helm/benchmark/metrics/image_generation/clip_score_metrics.py,sha256=0B2WCTP5LDHDbWGoMW2mKnnImHt-QYEU2QzqYf4HxjQ,3812
 helm/benchmark/metrics/image_generation/denoised_runtime_metric.py,sha256=Nom_yw15ePU7wUuV2DFHpLnEAqaZQjlkW9LowRElOAI,1646
 helm/benchmark/metrics/image_generation/detection_metrics.py,sha256=mfYoPbLCmqWxqMSXbcX6TM0niNnpCeipcHImuV3mZ3c,2160
 helm/benchmark/metrics/image_generation/efficiency_metrics.py,sha256=neeNJNtHAVUMWqr5rvRIRlPKl225cXUGCURLB0z-rKQ,1459
@@ -253,7 +247,7 @@ helm/benchmark/metrics/image_generation/detectors/__init__.py,sha256=47DEQpj8HBS
 helm/benchmark/metrics/image_generation/detectors/base_detector.py,sha256=e4c8vPfioGzl2ftYzWOFIBDJcZJxBmpjU13n4fXaSvY,226
 helm/benchmark/metrics/image_generation/detectors/vitdet.py,sha256=kxXS8uNBC0pQ7LatuN85CXU8pJHZn0pJXY0rOLd_39g,7526
 helm/benchmark/metrics/image_generation/fractal_dimension/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-helm/benchmark/metrics/image_generation/fractal_dimension/fractal_dimension_util.py,sha256=vFO6s8QHo6Pt1QfbOKAI0m3mJrc0BeH1Hcf7u2uWMIk,2116
+helm/benchmark/metrics/image_generation/fractal_dimension/fractal_dimension_util.py,sha256=NwE85dtiVSlCJc50E57pkckgnCiKBsW0nF3cqgc2EUo,2128
 helm/benchmark/metrics/image_generation/fractal_dimension/test_fractal_dimension_util.py,sha256=5qKL-gHnEVmzSDW2GKDq6Uox_EJMDLe0QA55Nrl4H6s,1472
 helm/benchmark/metrics/image_generation/q16/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 helm/benchmark/metrics/image_generation/q16/q16_toxicity_detector.py,sha256=8Y5h-6RMjivm50RnNbNwV7wCug4RhKT5g8R_YeEp54I,3467
@@ -262,7 +256,7 @@ helm/benchmark/metrics/image_generation/watermark/__init__.py,sha256=47DEQpj8HBS
 helm/benchmark/metrics/image_generation/watermark/test_watermark_detector.py,sha256=Ir4u8blJWTRtEBogb6u22qCy3JXAIzvx-Th6dSBLfdw,698
 helm/benchmark/metrics/image_generation/watermark/watermark_detector.py,sha256=w6WnTc6t6zx0W0gTjgedXC9OO5dq5iWpx9UcnioKml4,3641
 helm/benchmark/metrics/summac/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-helm/benchmark/metrics/summac/model_summac.py,sha256=82S9BpPJENr_jiY-cNubECEhniu5Y3Arzv7AXK93PmE,17442
+helm/benchmark/metrics/summac/model_summac.py,sha256=zheAPIJAz5MH6GU1gXpWSc9Q9gouhNzYx92PDd5PUXU,17447
 helm/benchmark/metrics/summac/utils_misc.py,sha256=7_Q1c72cKt8PWtxn8u4R8nB53HK6_JF2nP8bBXYNk-A,1485
 helm/benchmark/metrics/tokens/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 helm/benchmark/metrics/tokens/ai21_token_cost_estimator.py,sha256=XDZGK8h84F2w_pK8Zjko8ssKZmVxKFqTOuHL0mLBzMY,694
@@ -280,216 +274,253 @@ helm/benchmark/metrics/vision_language/image_metrics.py,sha256=RgKAn7ftl4KCZ86V3
 helm/benchmark/metrics/vision_language/image_utils.py,sha256=xwtydR8-s23cJacIGXDXL_pUhAqi6O5CbhM4XNEFlDo,3787
 helm/benchmark/presentation/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 helm/benchmark/presentation/contamination.py,sha256=07IuIP92vfuI0GwfeNC-i_NZUlF8N1azzagC19YHOMQ,2802
-helm/benchmark/presentation/create_plots.py,sha256=m51mFsYD51Y1rbEQgwTbKZjCI3xQir437WyOS5z5k64,28916
+helm/benchmark/presentation/create_plots.py,sha256=bM6UNzH0Bx8Bv2iKcyMoYp7IwfCZSQob-w_XOOI6r1M,29090
 helm/benchmark/presentation/run_display.py,sha256=LmY2HES4dU94kRYuUxt-c9LTMDN6MU5CspWTF6rZwDo,12419
 helm/benchmark/presentation/run_entry.py,sha256=J1QgLOP99N7N4bs7nzXWxyU3pOd-a1j8xwL9ag1nP_Y,1158
-helm/benchmark/presentation/schema.py,sha256=gYlMysq_rIzQTE9I1K3mIC1fFjBdDe1yHqgwb4EIciU,10989
-helm/benchmark/presentation/summarize.py,sha256=Xk5FJRnWz7xAbPu6JQ96TJ6Fvb1-xWUGBdfetrTsmbA,59882
+helm/benchmark/presentation/schema.py,sha256=AMGmEwqxkHoZFkOKD-UVZ8aXwgbafG6KYASsWo6YEw8,11005
+helm/benchmark/presentation/summarize.py,sha256=m3RSw6ogUFasdeZ8xSUh4wKV-nYzVi3iQv-KrrwtDFM,67828
 helm/benchmark/presentation/table.py,sha256=-foH1BIfMiD6YvpwoGJ910CH7Hib-_pYtHH1hE8zwNc,2904
+helm/benchmark/presentation/taxonomy_info.py,sha256=pPIFOicis9H1sWeXApfsHHcqZpus1ezukxLQO7Lj2Vg,473
 helm/benchmark/presentation/test_contamination.py,sha256=RlihBOF6vx2tKEj6_EMnJojTYoStx0FUeJSLT1bdf8w,509
-helm/benchmark/presentation/test_create_plots.py,sha256=5PPPegMTdBZurxyyUxI4rN13AVsjV3eQrwFqlobJ8UA,1286
+helm/benchmark/presentation/test_create_plots.py,sha256=1FrJZnPW-5QUQKt_pf4y47uDha4B8wHyY1o5hqhKWhc,1293
 helm/benchmark/presentation/test_run_entry.py,sha256=4n484sSYT0gQ4WVt67Fs3ctKa4vi97hI32O5XXxGY1o,794
 helm/benchmark/presentation/test_schema.py,sha256=6mq6CeAOLW2Kxi1lX_ZW8QCVqVR73XImR8ylcRGFkBE,378
 helm/benchmark/presentation/test_summarize.py,sha256=GzZNwBDybpstzl6wT0Rgqn75N9iCNrUIzrdjOfUolu0,6317
 helm/benchmark/presentation/torr_robustness_summarizer.py,sha256=SmMOZWCQ-KaJBp78otwvAeE1btWignyWalaQ8QG87r4,8242
 helm/benchmark/run_specs/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 helm/benchmark/run_specs/air_bench_run_specs.py,sha256=K86SqpINMBOiLIpuHz-jwlQL3SrH6n6WbqjD90i4LQQ,2231
+helm/benchmark/run_specs/arabic_run_specs.py,sha256=fPAI9GCV_D0BHPcLGSNZN45sAO2d449Gb54iHW1nocc,7399
 helm/benchmark/run_specs/audio_run_specs.py,sha256=baJz5LZiwWZP3KD0hluKgpidtswzdorQnshX0CoqKAc,23383
+helm/benchmark/run_specs/bluex_run_specs.py,sha256=jwrH33YeXqoAex11071XMUwTCKNkoJTQQS7iNoJDLmg,1797
 helm/benchmark/run_specs/call_center_run_specs.py,sha256=QhRQw91WblB9UaB319XNCO5K8PX8Riiza41Ym-1CcRU,7044
 helm/benchmark/run_specs/capabilities_run_specs.py,sha256=sbqhIj4AoujV45erwoVK61lWdlkjg4qssmGlu0eSr1U,12067
-helm/benchmark/run_specs/classic_run_specs.py,sha256=1NYeYIwC2F7EjkPEPxNoFb3Ap6BUcUJK_hxBKq4lzt0,56144
+helm/benchmark/run_specs/classic_run_specs.py,sha256=fe98HhzMkfloKpOZbi_mIMp1Hi-clv22rgWT-EdS0e4,53743
 helm/benchmark/run_specs/cleva_run_specs.py,sha256=lEIHEqQY3Efx-sl2Z6Rq9Qq_1HEWHqFYuUkZbGvq66s,13387
+helm/benchmark/run_specs/codeinsights_run_specs.py,sha256=lz3yysrPjCIiObzrIkRjJsWzkABh9qIXn-o7FSqZPl0,9207
 helm/benchmark/run_specs/decodingtrust_run_specs.py,sha256=7slILDS9f0_Z0y-Pz5xEspoGQUmOCOI2K2r4XWUVsm8,14428
 helm/benchmark/run_specs/enem_challenge_specs.py,sha256=5UWeP2bsnwCHMMXI3DFRMUPKcnJ9_EL01qPUthbWIvE,1351
 helm/benchmark/run_specs/enterprise_run_specs.py,sha256=ul2YMPpvThOmi7yIc6xR3W0rtE-8tUIaIzuhGlMg2rY,9598
 helm/benchmark/run_specs/experimental_run_specs.py,sha256=tIgAdK3cm4t6ZBGkcPcPkxx0XAslKShYA1i3QxWVJEY,7675
 helm/benchmark/run_specs/finance_run_specs.py,sha256=5mwb7GbAcSLVZiumqCiAr9dr8qBYApkEt5Oben5CFXs,4371
-helm/benchmark/run_specs/heim_run_specs.py,sha256=Pt1eVbzvwZ5EXq8WB2b3XYw62SWYN_i1P_H3oE4i8KY,22096
+helm/benchmark/run_specs/healthqa_br_run_specs.py,sha256=515pDZf8rTpvebPmhr9pqY2c08Ey_OtWIGsFDVVcQqI,1416
+helm/benchmark/run_specs/heim_run_specs.py,sha256=9uOB_eW5bQqoP9eYRaJ2bcigPg75pQLQnyQ67fG9wHo,22226
 helm/benchmark/run_specs/imdb_ptbr_run_specs.py,sha256=nkW5A_xeD5kCKeJVxsL8RFS8r3UpP_WCcwSdMh2s850,1215
 helm/benchmark/run_specs/instruction_following_run_specs.py,sha256=GElJhgbQhlZMYSAM4YyGcYq0pqycR32kBCoHqG6m-ZY,4177
 helm/benchmark/run_specs/lite_run_specs.py,sha256=8OkL9g3wQBG96g0ijGZ9L1Trb59b7VPDyYMqvA3hXfE,11129
-helm/benchmark/run_specs/long_context_run_specs.py,sha256=mxgFgjdHnatOif4-xmTicGmpr4U720mfkhPIigeTrGQ,4773
-helm/benchmark/run_specs/medhelm_run_specs.py,sha256=--KgkjVwKt4uyiTebalrbeGV4FB-jGqPciYjFZED7zA,43407
+helm/benchmark/run_specs/lmkt_run_specs.py,sha256=tNZvlA4mXUX-NBC9enRR90qFLeh8SNGFq701rXmXc18,5376
+helm/benchmark/run_specs/long_context_run_specs.py,sha256=wn7yY9rMIBJY30SN-275qg9U49aGPUl4hVZphKYFkBI,6442
+helm/benchmark/run_specs/medhelm_run_specs.py,sha256=bi7sGIx5I7KQXAF_Uj6n_O_DFNgtc496unrVh7UuLcQ,53256
 helm/benchmark/run_specs/melt_run_specs.py,sha256=729MkALud2wG07yulx9zqAzejdXW_eVGkfF5cQWeGGY,32031
 helm/benchmark/run_specs/mmlu_clinical_afr_run_specs.py,sha256=kenpGGMK1XXaNtvNXsshPvdvN9ubv1sOfaPdjFM4obA,2034
+helm/benchmark/run_specs/multilingual_run_specs.py,sha256=umf8e6ZDgRXiU0G_BPoovj1UZ_dxyrXtIQ7i9WC6USg,2296
 helm/benchmark/run_specs/oab_exams_specs.py,sha256=ws7Vppo_zJvxKqQ_sNhm9N7-5eQbX2CBkcDI5c_sRG4,1658
 helm/benchmark/run_specs/safety_run_specs.py,sha256=3X6tYaq2SlRsZs9q6SCtBUgjNEpOwUtV6M7iY2Kowm0,6807
 helm/benchmark/run_specs/seahelm_run_specs.py,sha256=R3mg4_OoaRizZ5n0FHcUQpJLny3j-ulBlHzOyF0a0Ok,23904
 helm/benchmark/run_specs/simple_run_specs.py,sha256=0kK_e8U4JUWZ6wO4N-GPFRE1iGT4ilvSMUGfirvpIE0,3837
-helm/benchmark/run_specs/speech_disorder_audio_run_specs.py,sha256=FvigS7LXxKkg9ipTaIPXDN47qFk__Vrv47hb46_cR3Y,7441
+helm/benchmark/run_specs/speech_disorder_audio_run_specs.py,sha256=P1mxSu7ErjiK0ARbRmbIzFwYO3fC-6QpsZQeez4U3qI,7346
 helm/benchmark/run_specs/sql_run_specs.py,sha256=JWCICELKi81m11MggyR6CJNl3vpWPwk4kr8DZSsWvj4,1965
 helm/benchmark/run_specs/tweetsentbr_run_specs.py,sha256=qogc-fb83Rh1DooKKaskhak52ycvu8DAnhabw9rc7yA,1129
 helm/benchmark/run_specs/unitxt_run_specs.py,sha256=4Vbsq0MPpSe4cIJOXzeVpMm60N9Qafa2R85X5BeFQew,1873
 helm/benchmark/run_specs/vlm_run_specs.py,sha256=v-eWuDYc8u5HO46isLONPfAWv5zdA1ZOQrdyOvX3vlU,37512
 helm/benchmark/run_specs/winogrande_afr_run_specs.py,sha256=dhOm8z6Q_ZpnzYKrsS0nEbRQPWs_phkXxmL5pxCJzQA,1853
+helm/benchmark/run_specs/medhelm/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+helm/benchmark/run_specs/medhelm/benchmark_config.py,sha256=O1D5N4q1QwzrI1ioAQK815cch6hNoJoaIzzAlJo6GXk,7860
 helm/benchmark/scenarios/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-helm/benchmark/scenarios/aci_bench_scenario.py,sha256=W8h7eWz9mjR0kRAffKWSnA1Fs8t2l83sPyW8fjPOxWQ,5670
-helm/benchmark/scenarios/air_bench_scenario.py,sha256=B6_WMowLFe4gWfnoFA_yrHe0kagbIkZabEnK4kGGqSU,1884
-helm/benchmark/scenarios/anthropic_hh_rlhf_scenario.py,sha256=jnUGbppDGEsbe5xoJjmv7nW_RvwPIYm6cwSULeqk2Fk,5133
+helm/benchmark/scenarios/aci_bench_scenario.py,sha256=ry22AJdd3lvQuEFdzNf6wXzMyPFn46b0kScrYdpj-nA,6783
+helm/benchmark/scenarios/air_bench_scenario.py,sha256=Ufcpxm5KaXHI2FfK4tdQsURaCSdcWNcXVaNmYkE4bo4,2820
+helm/benchmark/scenarios/alghafa_scenario.py,sha256=FJXO3W6qYzCgLJMSiJEhpddNcFyR3N5Brh8pATW_9GM,5217
+helm/benchmark/scenarios/alrage_scenario.py,sha256=MN-gMQboAaJCasYNg_rLJVgcrk5KZ1WCBN9R_lyRrhE,1499
+helm/benchmark/scenarios/anthropic_hh_rlhf_scenario.py,sha256=EzS8td1lJE1yxEwFtuwTbjHtHm1hGIaur93BKAL_Hm4,6212
 helm/benchmark/scenarios/anthropic_red_team_scenario.py,sha256=_OWE33eVRaZI0gmfP7bLd572uOi_6jb39z_J6nkcvfg,3182
+helm/benchmark/scenarios/arabic_exams_scenario.py,sha256=hv28A2pM66ejrO6oFOgmCx3JIP_nqwdUYvIsfGc0Kew,5359
+helm/benchmark/scenarios/arabic_mmlu_scenario.py,sha256=xMRWPA16Wn8ONgAeyyHOB95X2SQca7tKUpUP8L5ZNJc,3018
+helm/benchmark/scenarios/aratrust_scenario.py,sha256=ismiWLm1M6JmBgVZ0SoVglaOyFbAlyOHsSsiAv8Np8Y,3125
 helm/benchmark/scenarios/autobencher_capabilities_scenario.py,sha256=fOCHumFWZa4OJZcTZefJiJbdWsb3zjQnWLJYd10Cctw,2496
 helm/benchmark/scenarios/autobencher_safety_scenario.py,sha256=MFt3f5baN5r-FmzWZfUChGR1mX_PUB_5hxoINac_Whs,1854
-helm/benchmark/scenarios/babi_qa_scenario.py,sha256=SyM6RP4v08B1PjumkdQnuKrM9L8SyK0bXbx-LgmyTPo,5067
+helm/benchmark/scenarios/babi_qa_scenario.py,sha256=CAmh3GfFjB9Xsuh9K-PUu-2xIFTV0v0YNgWbSuv09Y0,5711
 helm/benchmark/scenarios/banking77_scenario.py,sha256=dtiM-Q_pMDWqkLi-hgl0tH-aGuDdgHkXgweE1JqrPYs,1883
-helm/benchmark/scenarios/bbq_scenario.py,sha256=2A7MX6iMAZHuPpH9RePi9rVBeFRmGdiE6GlqZ5uNdAM,9603
+helm/benchmark/scenarios/bbq_scenario.py,sha256=mVfxztgLI9sFwOYntx0dxElm8RmOb7XQYS9DOfgYjkI,10360
+helm/benchmark/scenarios/best_chatgpt_prompts.yaml,sha256=KZdXj4KUbkwFzgIEXVakMpZLTqJ7rldxNuXVDIdlk-A,31304
 helm/benchmark/scenarios/big_bench_scenario.py,sha256=g1TLoDTYQAe-efzQnV9J5UBCaUfN1jWTTjTd-ZJQmVQ,8146
 helm/benchmark/scenarios/bigcodebench_scenario.py,sha256=zQLv91uwfGAR9N4jm_iBUmYOVFj9cL14Nj8aqoCqUM0,2004
 helm/benchmark/scenarios/bird_sql_scenario.py,sha256=n5elzanKEX9YclAl2l1y33aCjihTmaw1VF_ZsAU5IaM,3613
 helm/benchmark/scenarios/bird_sql_scenario_helper.py,sha256=FIwPk-dwfTY-8gDXeAiTZbfbS0Oe1OuWRlYiJOhZwk4,4664
 helm/benchmark/scenarios/blimp_scenario.py,sha256=9Ge3QKRgtVHpWy7aehZVKiO6JrsxK7zrEdtqAb4zxtQ,6284
-helm/benchmark/scenarios/bold_scenario.py,sha256=iE9drB9IeXfRn3xvLnaQi3-nJAp-bV1RE0GJGnp9dJc,4130
-helm/benchmark/scenarios/boolq_scenario.py,sha256=wPETIu5jcI4jgP5GoFa_xi4SsvHtS9gxQ5TD8neHmdk,8037
+helm/benchmark/scenarios/bluex_scenario.py,sha256=K4ob5_rd1hTOzlPJjuEvujcOdt_Ybgxj3jqj_BYjA9o,2599
+helm/benchmark/scenarios/bold_scenario.py,sha256=MsXwUiJgZgFyVxh-E5gAagi4aPGicDe2C0xct5lQYwA,4882
+helm/benchmark/scenarios/boolq_scenario.py,sha256=qQyJ0BdljChX9U_eEETdFyWLCSQvI0D4NrY6zOCXPh8,8824
 helm/benchmark/scenarios/call_center_scenario.py,sha256=19J2N57WnUkPMGRRbJyZak8YCeMTRwD3BRK1SArQlL0,3037
 helm/benchmark/scenarios/casehold_scenario.py,sha256=QSe0D3KQJhlTOo6kM9OHwdKy6NlclsFGRVCAB3mTG7s,3174
-helm/benchmark/scenarios/chw_care_plan_scenario.py,sha256=BbEjDqa4C5wpdil5jIb1nzj16CCZ29hKoZVsfapSfho,4005
+helm/benchmark/scenarios/chw_care_plan_scenario.py,sha256=PE4vbj0y39674UIIdH6mgUwSKe4wW_XqRrNsksrwQRs,5104
 helm/benchmark/scenarios/ci_mcqa_scenario.py,sha256=slZZT74QI3OMQAgT-ybcR_xVcRDoopXw6mMu4iy3XCY,3074
-helm/benchmark/scenarios/civil_comments_scenario.py,sha256=pnZU2U_cYFYOJmlmwTehHU5oLIPx_Yg8Ayxinroh4IQ,4875
-helm/benchmark/scenarios/clear_scenario.py,sha256=yGdPxWO6vY4JHNa4xywtvD-9lOn6s5cr3njpZyFA0D0,6183
-helm/benchmark/scenarios/cleva_scenario.py,sha256=yPIiToKow76YMc0EDYeqQEPx-9a_6Bm3w4S1IsRRV5E,57987
-helm/benchmark/scenarios/code_scenario.py,sha256=lSbZWw67ie9osOjXDZukj3EEZGa3L6TrMvTg--IbuxE,12520
+helm/benchmark/scenarios/civil_comments_scenario.py,sha256=N1ZmQyKXkRjRXKPTyEHOpbDhBkjcY8WyHPKMWaBl2qo,5481
+helm/benchmark/scenarios/clear_scenario.py,sha256=cLFlcWKUT1Uy6bYDnAjf1ySR06mK16NhN1AtsaEBZs0,7226
+helm/benchmark/scenarios/cleva_scenario.py,sha256=WQDiDCVo6bhtI926_p3uvr1WhIAkBU1gLNLA5viEwMw,78127
+helm/benchmark/scenarios/code_scenario.py,sha256=tdki0m59NzN4YOm1pMfaSkUP5uUDeTNMqUAB84p5QGI,13953
 helm/benchmark/scenarios/code_scenario_apps_pinned_file_order.py,sha256=KC-5MQ-d8Nn46aDN4FaPxmd6yk1DtVUmVR-CIZsNCp4,1738
 helm/benchmark/scenarios/code_scenario_helper.py,sha256=TnXAlY-wdAFwIDylFItf0z7HOu93WD6dNThwzZYe330,5904
-helm/benchmark/scenarios/commonsense_scenario.py,sha256=yZ6n9aqOi7UWY3q4uTDNc2JRNZxaBZPIp7n_Snt_8g8,9511
-helm/benchmark/scenarios/conv_fin_qa_calc_scenario.py,sha256=gKEwqHDD8KlKmW8z3xAxSIGmALTXrRRPcoDUzbv_IXg,3854
-helm/benchmark/scenarios/copyright_scenario.py,sha256=FHzUYEabj-BTKl90fgq7jSCq5_Yf9cO9MA9djn50B1Q,3697
+helm/benchmark/scenarios/codeinsights_code_efficiency_scenario.py,sha256=PK4wtuBXs4cPPwOoGfhBA4J4cGLQYC_MvRWuvWrkrv8,9068
+helm/benchmark/scenarios/codeinsights_correct_code_scenario.py,sha256=7BpcezugYHleSuG8hreHe5oXpm3bxoxQ4RCnx6rjKbU,3734
+helm/benchmark/scenarios/codeinsights_edge_case_scenario.py,sha256=csTwe-mv1f6Tyvnj9uZ0SYuj1GRVvgjzukV28gIhNpk,8703
+helm/benchmark/scenarios/codeinsights_student_coding_scenario.py,sha256=wc5Fefn4jpCw03dQ6WswCztJ8AO5j0Vrn6omcOVUq2k,7409
+helm/benchmark/scenarios/codeinsights_student_mistake_scenario.py,sha256=qX3yckZdMojYhiwvokvEuQpRXOzmN2zmzKjQb96Ljg8,9651
+helm/benchmark/scenarios/commonsense_scenario.py,sha256=VN6nNZZpz9a1IC-tW5MvqztaW71f2zsV8lq-A34p3iE,10696
+helm/benchmark/scenarios/compositional_instructions.yaml,sha256=mPsFzPU6uaAD0xghzv-QD5Wk4uhoLY2sF3Fw_lunAsI,1822
+helm/benchmark/scenarios/conv_fin_qa_calc_scenario.py,sha256=sR3UzObloLUzgjNwTbSHLGGkeA0g9-Aq_utpBPT2u_4,4757
+helm/benchmark/scenarios/copyright_scenario.py,sha256=GWRCJdLlnWZcz6ztB5XIASGMPNd2o8EZNR2GueP8xuc,5035
 helm/benchmark/scenarios/covid_dialog_scenario.py,sha256=Vnxfn6EKwN-KR1vH-x46YHUC5jf7UAOv7zsnXVHYmZY,4032
-helm/benchmark/scenarios/cti_to_mitre_scenario.py,sha256=pfHAteKXLNUrhKyAYk6m9j-d7iuEgz58o15xukp_GFQ,10260
+helm/benchmark/scenarios/cti_to_mitre_scenario.py,sha256=FM6ty-JSFTDqdKLzfwgfhl3zV2oh_DWjRw4qI4-IrI0,11169
 helm/benchmark/scenarios/custom_mcqa_scenario.py,sha256=rgdHsSh8QknlcdGfZQ4VvqBUMLfTTHaNolCv4QgWHzE,1939
 helm/benchmark/scenarios/czech_bank_qa_scenario.py,sha256=ZBfkUYlIa-BagRVBf97RoyLfEloAjnM0RPv5wmEWueQ,4406
-helm/benchmark/scenarios/decodingtrust_adv_demonstration_scenario.py,sha256=vOUE5-rj_Wr6m7n76knte-kCMsphb-SSq9LraYf-Dh8,7933
-helm/benchmark/scenarios/decodingtrust_adv_robustness_scenario.py,sha256=9qo3l44aby1EfQqohh1M2DVtHXqY1fuvj1XT3_n4hBk,5588
-helm/benchmark/scenarios/decodingtrust_fairness_scenario.py,sha256=rAOZnFSxO3ENOvcNz688P_f3Y7NzdwiWgoYTNvAaw3A,2866
-helm/benchmark/scenarios/decodingtrust_machine_ethics_scenario.py,sha256=qhzqW614WnsiyN7TiHUdZY_NpEdW_iMO0AMrLK8DmK0,14116
-helm/benchmark/scenarios/decodingtrust_ood_robustness_scenario.py,sha256=RSigvRdqjeFTwFfXNmslz8zyAGSmLf6UtBDA4NrQBCo,8304
-helm/benchmark/scenarios/decodingtrust_privacy_scenario.py,sha256=zaXn4sRPUEZiqPoudiDT1xHMV2DaiEXOOTz3qB5q7Go,20143
-helm/benchmark/scenarios/decodingtrust_stereotype_bias_scenario.py,sha256=NjutVTOVVze-IJniRFecz8gqh_BUpuJG3-BUboTGKRw,2933
-helm/benchmark/scenarios/decodingtrust_toxicity_prompts_scenario.py,sha256=EYKoXDWMesbY5dCNY-N0eYMRL0rjEfGsuS_TkeD3Suo,2952
+helm/benchmark/scenarios/decodingtrust_adv_demonstration_scenario.py,sha256=pZK3dbKKNfNOHvNaGMkN9pjFznu4raNyLe4fWkxNHSo,8604
+helm/benchmark/scenarios/decodingtrust_adv_robustness_scenario.py,sha256=hBKRRYIHegOrhIo_i7-1RPtbxmuhXcg29DkUIep0x_o,6304
+helm/benchmark/scenarios/decodingtrust_fairness_scenario.py,sha256=KzBz8nkrvPUTw5WmEoivtl0lLJ-mORek-IVKYmct2Pk,3460
+helm/benchmark/scenarios/decodingtrust_machine_ethics_scenario.py,sha256=OvJ3pfxbxtJRxeSfeK-uoYFZ4ZIDSqE7ZbqZBuO93DE,14743
+helm/benchmark/scenarios/decodingtrust_ood_robustness_scenario.py,sha256=zWhQWEE9Aa1O9ASLE5IAw55lzNLJ1ifGsBKZFk-jiXM,8942
+helm/benchmark/scenarios/decodingtrust_privacy_scenario.py,sha256=XO--1Rxb6kyLDRUQw-GhzLG-aTagVyN7ktWriAbBTAE,20756
+helm/benchmark/scenarios/decodingtrust_stereotype_bias_scenario.py,sha256=vIkAgy4LysSSIm553bnts3CEN6NVIDKr3xeGkZ2GNyk,3520
+helm/benchmark/scenarios/decodingtrust_toxicity_prompts_scenario.py,sha256=5l0lRRNNJ8nAb1R4bMxq3lakMF-P3XFvVpnT1PrwMms,3556
 helm/benchmark/scenarios/dialogue_scenarios.py,sha256=yXCMZegzlgL0CXTY1W5lXdkFFHicUvq4z7_284MfRpw,5778
-helm/benchmark/scenarios/dischargeme_scenario.py,sha256=rBzagg0JVVN3o0VUfmHy2cN7gutV_RAJAo5Fa_El0GY,7842
-helm/benchmark/scenarios/disinformation_scenario.py,sha256=0T7LhXguzBP645Fruc2udfTaMuy7XGtOEMJKpFMIFRk,8565
-helm/benchmark/scenarios/dyck_language_scenario.py,sha256=hygFPTcICGUEPwjtxULLKBSbuBOXLYpozIgiGcT__W0,9379
+helm/benchmark/scenarios/dischargeme_scenario.py,sha256=WTlqFnM76DFVGOUSLWv-g--vHWR71UWZ9VFXoEec3fo,9026
+helm/benchmark/scenarios/disinformation_scenario.py,sha256=lq9Aj-DDpPJeFVk99wXEd2Qv3kahiBe9c8-RoBieCDM,9581
+helm/benchmark/scenarios/dyck_language_scenario.py,sha256=HZEXetj5BkXrNJbAvg9HidrkxDgi2UUGIAVphNiN-jg,10052
 helm/benchmark/scenarios/echr_judgment_classification_scenario.py,sha256=IqODoUY1-zJD1KW4Qkg3VwJcUeeLgGUKThr62bW-wx8,4915
 helm/benchmark/scenarios/ehr_sql_scenario.py,sha256=Gm7Kw_TSUUxHW8ns-2e4E_tTBVX7h6Ta273VOpkMCQ8,5480
-helm/benchmark/scenarios/ehrshot_scenario.py,sha256=MWcTejCtwohBPbZYWei_WNZ-Hdnhml7ovTVbJAgUetU,67770
-helm/benchmark/scenarios/enem_challenge_scenario.py,sha256=sxYXKvf-mGNqctTkemwI9rrA_Rg2xA8mz3_W3TIfzUE,2147
-helm/benchmark/scenarios/entity_data_imputation_scenario.py,sha256=4V426oOuexGg59q0djHCTQjQmqYgyLT191Z5fayubmU,6681
-helm/benchmark/scenarios/entity_matching_scenario.py,sha256=kzzDaoVikL2P7Z-17EkLIVR_W7IHcNVerUts2oXDKLA,7111
+helm/benchmark/scenarios/ehrshot_scenario.py,sha256=OzZrgi-UZrMH70ZnHSeUWPCOesUue5vxPqnNOaN45dE,68830
+helm/benchmark/scenarios/enem_challenge_scenario.py,sha256=gceJqjxX-RxvOqPDANEwOrbHwKxtddpMz-FcsBfby0k,2854
+helm/benchmark/scenarios/entity_data_imputation_scenario.py,sha256=03Ju45Sju2r4A_Peq2EsOyg5Ik99lMUv-6X--ejB9fk,7332
+helm/benchmark/scenarios/entity_matching_scenario.py,sha256=83F017FPFED_106IOawJN1jdY6IfREGJPNRvCokKGNk,7761
 helm/benchmark/scenarios/entity_matching_scenario_fixed_random_state.py,sha256=TklbX7Kx4y-estV-YHUbI5O08q2qCZRrOmX9D3gZS9c,2193
 helm/benchmark/scenarios/ewok_scenario.py,sha256=vrbJg9vakAxE6n-1jURUcwb-ihrsYoY9e32BpnEGDaQ,4684
+helm/benchmark/scenarios/exams_multilingual_scenario.py,sha256=c9zMGGL8EbCeNogTm-88g_5wWUiX1Zr7z_tsyjUq2h0,5404
 helm/benchmark/scenarios/fin_qa_scenario.py,sha256=Dm_kGOivaxiKVhcqFgN8pRPs1eqm2LdBZxWy0yFhFuE,5958
 helm/benchmark/scenarios/financebench_scenario.py,sha256=cHMljdg0_9HA3FbwcwwMt3DR9rxl0jkyFN9jNrUStSE,1956
-helm/benchmark/scenarios/financial_phrasebank_scenario.py,sha256=dMTfI9MRHKXnECsXOIY8xvX6w5vAPEIa6A7TYyIu2Fw,4457
-helm/benchmark/scenarios/gold_commodity_news_scenario.py,sha256=-O4ilLwNcycmpQG5h_5WtQP7yJEr4mjWjKBe2eNP0uY,4806
-helm/benchmark/scenarios/gpqa_scenario.py,sha256=369E0JvaR12EcgcEFKKRcDw1iztt4sb8ghIsk9Brzi4,2884
+helm/benchmark/scenarios/financial_phrasebank_scenario.py,sha256=I7eoymZfxu4gky3YjyLnZgaFIJcMkprxQxiCLM7wJV8,5455
+helm/benchmark/scenarios/gold_commodity_news_scenario.py,sha256=Qw8OJzvp12716GRW5kIxxX--f92OFRcaP0oEy-gakjM,5674
+helm/benchmark/scenarios/gpqa_scenario.py,sha256=MsMsBqgxz6jKt2-ys98XAslGWkxZgzpYOws0b9e4Uj8,3520
 helm/benchmark/scenarios/grammar.py,sha256=58tQYKPj013V9jIpW7fXUqZBLuboqEi_WLlDjx74spM,5590
-helm/benchmark/scenarios/grammar_scenario.py,sha256=Hz59gp5ivH3tIP5UAcHZbnk8pBX6GhIABSQlG33gIRI,1502
-helm/benchmark/scenarios/gsm_scenario.py,sha256=QIj0QK5ncF31ES0GUlxbdBk6SIiJJnj5wzamj0do0tQ,2674
+helm/benchmark/scenarios/grammar_scenario.py,sha256=c3ATPkHM0WkA9QZEf2VNfThhuEUXD448uOuW6CAeVFw,2309
+helm/benchmark/scenarios/gsm_scenario.py,sha256=S_rD8uZsajgqyaJGNMpqYvshYYIW9hMV9N2udbI1Ax8,3405
 helm/benchmark/scenarios/harm_bench_gcg_transfer_scenario.py,sha256=8_ShEuOoEGu7iRE2b0tgi-cfBrCPF9k1L-Pgb__n3Bg,2005
 helm/benchmark/scenarios/harm_bench_scenario.py,sha256=CBo_AfbtHTlvJdsiquP0EDTKApVmDZc7EW0VTENNAfQ,2478
-helm/benchmark/scenarios/headqa_scenario.py,sha256=m6Kqt16JeqA1-OLJvmBPZzhVOVt7O6rbJGAwG9C7FZs,5658
-helm/benchmark/scenarios/helpdesk_call_summarization_scenario.py,sha256=iv1khpdiWW0Z7lshyWOhhjRfYFdAU6etN8X5EDEQCrc,1302
-helm/benchmark/scenarios/ice_scenario.py,sha256=NCbeqvpDFIIG7kSCrJrS-Z9S3iG2THZ7HpAqghpi_y4,16725
+helm/benchmark/scenarios/headqa_scenario.py,sha256=0hJewHkF9IKQfW6NUJ0DPjlwQmr7N90a2eSXrBQiFNA,6635
+helm/benchmark/scenarios/healthqa_br_scenario.py,sha256=YneXTfp8V6k8rYCF3BTX6bxN2ASxdG3qrBr7uH_IFWc,3406
+helm/benchmark/scenarios/helpdesk_call_summarization_scenario.py,sha256=5R9En7lTNirZCVsMNqNB2metw0dIEPa9usoFB9W11B4,1855
+helm/benchmark/scenarios/ice_scenario.py,sha256=tEkXqRtvtXaoC6JfbJOcY0E8xWyYKGMOvsSYJGjM_9Q,17674
 helm/benchmark/scenarios/ice_scenario_pinned_file_order.py,sha256=fuirubIdi-rkJMfSd7YoDdBX2q0f5K7GGTN4XVapAUY,1613
-helm/benchmark/scenarios/ifeval_scenario.py,sha256=SYn9itpFG0tlWSayf6v0P8bRgdtc-BmOV1dF-4TEm-0,1675
+helm/benchmark/scenarios/ifeval_scenario.py,sha256=v2Q1uYCd5i1jO4_gcIlTrbZdPZ27tJrCXi9e0sqcm8s,2308
 helm/benchmark/scenarios/imdb_ptbr_scenario.py,sha256=laq9UwyvBvZZuo54rf-8SdKTLrMdDHTdGWJ4TdC8Eng,2340
-helm/benchmark/scenarios/imdb_scenario.py,sha256=qHXd-QIXTCBq8rWW3N5I2Rvg6Pz9v1zFhZkwc73w9io,6259
+helm/benchmark/scenarios/imdb_scenario.py,sha256=H9iHmKK-q4a5edSMcS166f1fjkNbOS5BEIgR3md3k7M,6887
 helm/benchmark/scenarios/imdb_scenario_pinned_file_order.py,sha256=fjW0Gkzg2Y3IAbtYJ3KC7MueWd9U8h0tlcBCqxYmRrM,1621
-helm/benchmark/scenarios/infinite_bench_en_qa_scenario.py,sha256=RxK5T6Nu_KP3rLMaKkJWiI_3Sqpskgqwgn4Zj95lEvI,2854
+helm/benchmark/scenarios/infinite_bench_en_mc_scenario.py,sha256=JRTLaQc3PDpYeX9ewGnBteT9jXeaGbmJ1VzYGT8TsXI,3067
+helm/benchmark/scenarios/infinite_bench_en_qa_scenario.py,sha256=5fJHFonb7Ko7exHFtoUtvHar_7PhK2HjW9uDlU8Ljj0,2872
 helm/benchmark/scenarios/infinite_bench_en_sum_scenario.py,sha256=6z3VlcucrwK2B30artWiSpo-mOTr9tiwYV6Fu8XD0VY,2657
 helm/benchmark/scenarios/interactive_qa_mmlu_scenario.py,sha256=F-gDO6r4GPBJTLirhF5noRaV0edvoIT7tiIDlovBFfE,2253
-helm/benchmark/scenarios/koala_scenario.py,sha256=A5M6SD7Jjg7r9QlbHCtMaydBe-wpOtB6oc6gFXuZ47o,1389
-helm/benchmark/scenarios/kpi_edgar_scenario.py,sha256=23rZM3IA-phf2VnuPY9QWd64scE6eaJks49apDUNfic,6355
-helm/benchmark/scenarios/legal_contract_summarization_scenario.py,sha256=xjw3iKRf8P50Wo58n7ssnFiWHR2QFehzHlZhh9P1XKs,5374
+helm/benchmark/scenarios/koala_scenario.py,sha256=h-dTHQrNVoi7p7sTXZDqWcpjlznfUgxNrgr4nW8Hrk0,2212
+helm/benchmark/scenarios/kpi_edgar_scenario.py,sha256=DE8efUmcPW5R62tZ46Rdsjv-EQs4lXm403O5XxM9heQ,7303
+helm/benchmark/scenarios/legal_contract_summarization_scenario.py,sha256=JTm4Zkwqed7PijdeHzSbQ2l4YZY037OYF_fbnKmlpKg,6185
 helm/benchmark/scenarios/legal_opinion_sentiment_classification_scenario.py,sha256=q_iezJo23_HNNoIXYT4cLYCbwNzLYJx6uvxgPSE5bQA,2804
-helm/benchmark/scenarios/legal_summarization_scenario.py,sha256=BFK524H7uLfz_ZURuRS7KrhzRCP-WyhIcOgdcBrsldA,8709
-helm/benchmark/scenarios/legal_support_scenario.py,sha256=Ty93M8yTr_VEHomJ_36htqqBDZZKWI7PHtXA4qkSJGk,3990
-helm/benchmark/scenarios/legalbench_scenario.py,sha256=Yfyouxb4ir16zlBea77Xzjc0BlcYPkiXoBoVI38FXwM,4779
-helm/benchmark/scenarios/lex_glue_scenario.py,sha256=-3fsSjTXjgRN96Hl4GzDIMB_dlxSR9NR0ATUb-CiU3w,10357
-helm/benchmark/scenarios/lextreme_scenario.py,sha256=gVTHoMYX6Q_Itt5rOVO5lYmqWfAtuuf63CnKAF8b_ak,20461
+helm/benchmark/scenarios/legal_summarization_scenario.py,sha256=0DraJdQebbl8tv7S3WmLos98wnQFGJOzY6suGPoxR40,10954
+helm/benchmark/scenarios/legal_support_scenario.py,sha256=cM98PnIAfjQzciUYGtgHqHYnWIdbdJfh3uy6uEIo488,4567
+helm/benchmark/scenarios/legalbench_scenario.py,sha256=K_KjDH3Rk1AM6pXLRedo-6o2rEw9OIk3porlCr3IGvQ,5623
+helm/benchmark/scenarios/lex_glue_scenario.py,sha256=H7f3F7gK7bgf6FXvqXGTQrecTE6RtZaitIKmwQLksck,10736
+helm/benchmark/scenarios/lextreme_scenario.py,sha256=dR5UUIymth3J3RInoNybygZg0rNZ-8wwzVHneuTTOGE,20843
 helm/benchmark/scenarios/live_qa_scenario.py,sha256=TnWaOPOcA4U1_8JdahQOUZ9KBj0MpMf4BcK2TDBl3BE,3666
 helm/benchmark/scenarios/lm_entry_scenario.py,sha256=kQTnj5gKJmDxCgynmzQOmghwNySpna7aTY7K7RPD2x4,9109
-helm/benchmark/scenarios/lsat_qa_scenario.py,sha256=2VUJ36vHUZp6fZuLfRIuPSsU_K6Z3Im2ums06sZENqo,6153
-helm/benchmark/scenarios/math_scenario.py,sha256=UtNj0UaCxt0RjM-uwD_Evm7SjKnvMlfCt6K0HQOAVC0,14377
+helm/benchmark/scenarios/lmkt_scenarios.py,sha256=K51CdOZqMOMOozUmADjrJuNCpUtXVEZwcOeIY-EZrwM,11162
+helm/benchmark/scenarios/lsat_qa_scenario.py,sha256=ZtheFEcsBMSqGIPw5UPOO_b3v93mPFar1yqxVnsLq4E,6785
+helm/benchmark/scenarios/madinah_qa_scenario.py,sha256=W7YEQTHyNWUJD8sKFmXU9e-ubzvleWQs7Cj_1zdq2bk,2482
+helm/benchmark/scenarios/math_scenario.py,sha256=p9tsdNsiYFtuG89cMByZYn60QjWzEsnCO21OHPr4DJo,16034
+helm/benchmark/scenarios/mbzuai_human_translated_arabic_mmlu.py,sha256=Gtc9DgV2bLPIDngROmizTWQHbTftnwVodi9CYT0_P2A,2146
 helm/benchmark/scenarios/me_q_sum_scenario.py,sha256=7DOqQmO70BpDeJy_S4fJ5i2UcCH8tunxzjFgTIim9bQ,4062
-helm/benchmark/scenarios/med_dialog_scenario.py,sha256=AE10W1UWhOrgKUnz7e2brKSaQR1WJkQUcPoo4s6n0Fs,7553
-helm/benchmark/scenarios/med_mcqa_scenario.py,sha256=XEipvuIA-QoyZrtlm8nnaPuyZzdDaeTskAhnseD3Q68,5096
+helm/benchmark/scenarios/med_dialog_scenario.py,sha256=MKDlZLJEUq1nDRzlkHlpTWOxHwgghWMXcQvHJcM2LP0,8615
+helm/benchmark/scenarios/med_mcqa_scenario.py,sha256=tvF6d6e4WQi_mUIlZoLQvbOpVIfHR4nyMVVR8z4AkAE,5752
 helm/benchmark/scenarios/med_paragraph_simplification_scenario.py,sha256=0Z1JrizLygjd9v_LLFMk8uZ805IWjJPvg-ZvPVhtMm4,7652
-helm/benchmark/scenarios/med_qa_scenario.py,sha256=m0W-FgFi58psLglZyQy_ouMQIDP-2j3aL7uInkdVtms,4478
-helm/benchmark/scenarios/medalign_scenario.py,sha256=mhd8REXpPwxftH48-KKb0ZURJ1mdOlvPRmvN4g4M9Ho,3383
-helm/benchmark/scenarios/medalign_scenario_helper.py,sha256=itxQxG0igEr-8PY3cXmUafM45bqxtov-iHEIy_ZuQYQ,15612
-helm/benchmark/scenarios/medbullets_scenario.py,sha256=8O0UsPWw-ESkrgiuWz4f8gR99jH5-wS5HtCKYwZ1ycs,6713
-helm/benchmark/scenarios/medcalc_bench_scenario.py,sha256=vwmEQZ119tOVeZtl6Zt-nXKwkA8Qt4WRiH2HogIkV0w,5560
-helm/benchmark/scenarios/medec_scenario.py,sha256=Lo7iVkek7C9omJ5LX-C83pA_Q5OrAfdNhJY4rslJWTQ,5270
-helm/benchmark/scenarios/medhallu_scenario.py,sha256=d4HlEi1cQtvh1a39jvIHezDDmjuIEsSPdqDLLkDTzw4,2544
-helm/benchmark/scenarios/medi_qa_scenario.py,sha256=FmXI3UwfbL8zinFPtSyTyw4X5VIe2d32HAg93vbXR94,4118
-helm/benchmark/scenarios/medication_qa_scenario.py,sha256=StQmfHTYi8pZLP9FMPzyS-VB9gilZS0XBme7MzAL2QA,2583
+helm/benchmark/scenarios/med_qa_scenario.py,sha256=uW8FOEQhMw6k0WF_LKlH0oFTQVS9D_9MHXvVTNwDC7k,5140
+helm/benchmark/scenarios/medalign_scenario.py,sha256=5ALak5Hq2XQbqwTF3fQYKg-QPtL_vjY7J1UsMm9SOFk,4481
+helm/benchmark/scenarios/medalign_scenario_helper.py,sha256=fKXJFVLGnLcZKRBLsbjJA6YA4WqMaQAjkEU-i6YzSTQ,11626
+helm/benchmark/scenarios/medbullets_scenario.py,sha256=oMqnF3Ri9dghEWpGQYzfcTnYGMK5b2cJNVpJoqdtdUo,7694
+helm/benchmark/scenarios/medcalc_bench_scenario.py,sha256=EDeeBKmbosUaMo3dg2MNVs_Cb_ws6WfnBYk15_B3lkU,6608
+helm/benchmark/scenarios/medec_scenario.py,sha256=sLx6tcFXcvhDIThGNVi-425znECAn5pkUgRk83CM-Q8,6343
+helm/benchmark/scenarios/medhallu_scenario.py,sha256=0EgeIxGuYMyBzM8xIOF4WcxfCOVqCp-oOuZe4Ai-CRM,3660
+helm/benchmark/scenarios/medhelm_configurable_scenario.py,sha256=vxvvAaIFW4cWaMez1xbEOZBh6S2wEH6Ws8KcGpnaZbs,3852
+helm/benchmark/scenarios/medi_qa_scenario.py,sha256=KXHQIliik9Cihaw2_M6GW5QdmHBeGoPc-0tnTw-_M5w,5224
+helm/benchmark/scenarios/medication_qa_scenario.py,sha256=uyYxtCm_dX9Jt6X-3ha2gAUyxF55wKn3_k95g7VAzHQ,3636
 helm/benchmark/scenarios/melt_ir_scenario.py,sha256=d88DEGKVJZCeGnbrXrQZO_W4VJeqW8XNaYc8wIUiJtA,5978
 helm/benchmark/scenarios/melt_knowledge_scenario.py,sha256=FDG4OGYEV6Ac40VC7KAeikzbFKAK2XXFhH1-QUTw8jo,7923
 helm/benchmark/scenarios/melt_lm_scenarios.py,sha256=kSm0lRRixhnXctMprPnzi09PLOmgfs-C7TAW3QI8RmE,8969
-helm/benchmark/scenarios/melt_scenarios.py,sha256=Zg_Uyq-e9Y-Er4IpWU1o29YC07Q9rOxxhokPyKq57Ik,30140
+helm/benchmark/scenarios/melt_scenarios.py,sha256=_WShDpmPaKrujGbZcazCqleDn0TKDhFg1h-vu3ieS8E,30144
 helm/benchmark/scenarios/melt_srn_scenario.py,sha256=EQSOZIXbfvVWCJMJ4H2e_CiBz6wc8THJndnbK2WwTHM,14674
 helm/benchmark/scenarios/melt_synthetic_reasoning_scenario.py,sha256=ptMQWgNn6R-XpAVAAjutSdZg_9ZUqG6fVotzAgeead4,7945
 helm/benchmark/scenarios/melt_translation_scenario.py,sha256=j9YrY60DQHZz4m1MJZaGLzyI6FERlHRx2wy9auyAVB8,5415
-helm/benchmark/scenarios/mental_health_scenario.py,sha256=O1Lfd0MxqawLZLKUDSynaqqbaGHRjDglmePIqepnJI4,4961
-helm/benchmark/scenarios/mimic_bhc_scenario.py,sha256=PGa0Nvbad_wH3qRSMPHgg9CgicOi7n25qLDnEucXapo,4097
-helm/benchmark/scenarios/mimic_rrs_scenario.py,sha256=bxwVWjE_z4I_Nk5eD78g3QAGyjpsNg7DVWpkp8IGWXM,3841
-helm/benchmark/scenarios/mimiciv_billing_code_scenario.py,sha256=tZBUZEaUMZvfSlsU6hcPs-pxQ0kDIL6qebGd7JmpDbk,2699
+helm/benchmark/scenarios/mental_health_scenario.py,sha256=dwirS093vIdS1VG5yKqUw863TJoCF_keO-pr7ysTIxA,6066
+helm/benchmark/scenarios/mimic_bhc_scenario.py,sha256=OIDB-f8wyn0ApsPqwpP11yJEpEtSpf3aYc6VVap6Jr8,5275
+helm/benchmark/scenarios/mimic_rrs_scenario.py,sha256=pG_NK1Et0QZosQAOLAxbciyNSq_wIdOT7hkXsBb4mTg,4902
+helm/benchmark/scenarios/mimiciv_billing_code_scenario.py,sha256=KRl1lYX-ITWTGxWS_NNQ0o3I4E__jlzNDhAYvI1by7g,3749
 helm/benchmark/scenarios/mmlu_clinical_afr_scenario.py,sha256=-OkPMRyB7aO6QBFwoTl6a2rpzcoHeEl84tqz7k9kpCM,2982
-helm/benchmark/scenarios/mmlu_pro_scenario.py,sha256=pwpp0wqNhsGc8v2V11aUyEWbwdkmIm-42N676j1T3Ws,4031
-helm/benchmark/scenarios/mmlu_scenario.py,sha256=_5cX2uI7CxD7K_GvO3MD8CRJLuN4EzS2o_EFvbrfjSU,3855
-helm/benchmark/scenarios/msmarco_scenario.py,sha256=-l7_rIMQjMWcpTyn6dGotmNJ5XxN_Ze8dEJyv5ftWFA,34050
-helm/benchmark/scenarios/mtsamples_procedures_scenario.py,sha256=13pXjs9lFduM-QL03mpM10hU0iA8Vr2jJG2FVBQdKOI,5577
-helm/benchmark/scenarios/mtsamples_replicate_scenario.py,sha256=RlyWrlI9e5MLsGbkQWpO2WRsIOZJi39xHskOIBypHdo,5399
-helm/benchmark/scenarios/n2c2_ct_matching_scenario.py,sha256=-Et7hJnQJOGl1U9Xdb5mLckYTpU_Ve1sCe450M-5haw,13513
-helm/benchmark/scenarios/narrativeqa_scenario.py,sha256=MiSq0UnUllJxHFU2gO7m4vr_vmulavJxc4ruZhsAt2U,5632
-helm/benchmark/scenarios/natural_qa_scenario.py,sha256=g-fP8L1lXs7zwNVQOc0ZUnbYkCyElQtLVt5fe5dtvSE,12564
+helm/benchmark/scenarios/mmlu_pro_scenario.py,sha256=2FVL-6Umn0BufFpJ0e405q1ZgeeP8Np1kCvsE61GaOE,4686
+helm/benchmark/scenarios/mmlu_scenario.py,sha256=P68i3gBlvVwjItZhLimtM6-zVGv3cYitSPH8ARwnkEk,4610
+helm/benchmark/scenarios/mmmlu_scenario.py,sha256=CyOISLOsXF9IEYGfeqWyYYkWGvrUvGivlWSJ5ttN9qY,2762
+helm/benchmark/scenarios/msmarco_scenario.py,sha256=p9YNL5oTa9isCGVvmqHHVofKmiwitjPQd28ElXmRAN4,35601
+helm/benchmark/scenarios/mtsamples_procedures_scenario.py,sha256=gtVSZxrs321tOolyD0gOoLzc0--uTc--3_HdlBVIuHo,6607
+helm/benchmark/scenarios/mtsamples_replicate_scenario.py,sha256=FIdI509nn0LN9opC4yJ8UsvWmh6-KECUMZF88duIEq0,6395
+helm/benchmark/scenarios/n2c2_ct_matching_scenario.py,sha256=o7RydazvQkYK90epvuXsdEyE02fmpsDEwS6253fNptk,14365
+helm/benchmark/scenarios/narrativeqa_scenario.py,sha256=XBGq3_gz1vaMhVX17RWF7mhXaSlKsv-_-JWCyHDkGWA,6428
+helm/benchmark/scenarios/natural_qa_scenario.py,sha256=3wkXvYm7m0Isxv2EW6SIuIEwZEV2lihsSLQZaANsKZo,14017
 helm/benchmark/scenarios/newsqa_scenario.py,sha256=G25VYaLrV_JyyoT0jpzJ6p4l5qsOydm8rlzTvSptNKQ,7284
-helm/benchmark/scenarios/numeracy_scenario.py,sha256=lgTGzZc81RyL8iB4K67PAHbyYz6BM2ieub8RSFi2aRc,30895
 helm/benchmark/scenarios/oab_exams_scenario.py,sha256=vbjUzQP0zU4ckvMbsk4lh24NddVWbUAtfWmsq1h24_w,2101
-helm/benchmark/scenarios/omni_math_scenario.py,sha256=5qb2cO-Ibb3kDbwYvkzsoU_aOsoKV3ROLgZbi83OyGU,1955
-helm/benchmark/scenarios/open_assistant_scenario.py,sha256=zd8T6eLOlYMZiFyKrRjc-EPwk5_KpbBedAcKDbZ-TdI,5609
+helm/benchmark/scenarios/omni_math_scenario.py,sha256=nB2miRRQ-cWwhpqUkypOZibYugD56wZ299nxE5bty9Q,2582
+helm/benchmark/scenarios/open_assistant_scenario.py,sha256=Z9eyaaHGRtFZTogIkOe1Pr6d70lqSe80tMsNPWR_jog,6577
 helm/benchmark/scenarios/openai_mrcr_scenario.py,sha256=XbO8Wpjjq2e8OsC2s_ZScV4TcZg3hlpVGy56hgxXY9w,3253
 helm/benchmark/scenarios/opinions_qa_scenario.py,sha256=JK39tq306tKe0RDBDLz1AfAdZwNjK_Ng-rHvu6bTRY4,7395
-helm/benchmark/scenarios/pubmed_qa_scenario.py,sha256=73D9D_q4Zw904qfd3tVPPhHxpGN4IZcWjlA6ZHEfp2s,8070
-helm/benchmark/scenarios/quac_scenario.py,sha256=RpJpOPbvhB0jv3R91Odc20LcNyZsny9J4IF24GNEygQ,6689
-helm/benchmark/scenarios/race_based_med_scenario.py,sha256=vZB43jtM47PWrl9L4HYOf1i7orpscKcHX01m0oVmk2g,5778
-helm/benchmark/scenarios/raft_scenario.py,sha256=Yk56dUMqDGXpp6SxoGWhyxa4lAIniSQfivjkoPqMuFA,4644
-helm/benchmark/scenarios/real_toxicity_prompts_scenario.py,sha256=zpQthgDi-AyEgOUFO5F0qaWCctLEI5WGHBEGlPEVpqc,2424
+helm/benchmark/scenarios/pubmed_qa_scenario.py,sha256=Z8gR19kiTIugBTvBj6g9LiBXicfAxZ1AFh_GF_axgQc,9043
+helm/benchmark/scenarios/quac_scenario.py,sha256=y5bm1LXHIICqPIkWOg3sibnH_sC15b2zYUfT-_Y0V4E,7349
+helm/benchmark/scenarios/race_based_med_scenario.py,sha256=pyeOUjWlQ30WgNr48BuV7kP7fhqZljLfizbTfWjyV_k,6862
+helm/benchmark/scenarios/raft_scenario.py,sha256=BQ-faIiWBuUYmHTMCRbI8XpymtWvKK8DN6oNejjNi7M,5443
+helm/benchmark/scenarios/real_toxicity_prompts_scenario.py,sha256=USsjBVzoL-Bgq8B2clQvl3d-g4XlOlt8gvBje9VD7Dk,3077
 helm/benchmark/scenarios/ruler_qa_scenario_helper.py,sha256=jgVf1D4eTSxwxQsW0GBou5hfSo2dnlEJvHpVJqk3BxM,6327
 helm/benchmark/scenarios/ruler_qa_scenarios.py,sha256=Dy0INRMzxSiIs9Pm3fa0hYodN-W--WPSv4kcmeQhucM,3270
-helm/benchmark/scenarios/scenario.py,sha256=kSy7tmtFeC6-QSEsBuvlrMTA1PB6fOY9jycMld-vBVM,8592
-helm/benchmark/scenarios/seahelm_scenario.py,sha256=GA46ShNGUjVdMLK0ZbN4vPuGEWFQsDPJXEGHQbs1qf8,78150
-helm/benchmark/scenarios/self_instruct_scenario.py,sha256=3Kvi3pLL6eGOEezjoQoGv9c1UxKiRVlFmILKzqst4pI,2309
-helm/benchmark/scenarios/shc_bmt_scenario.py,sha256=wF_sD61IZ4RDznBVQ1HYbGh3Vc2qjbcBuU0jdmp1aD8,2803
-helm/benchmark/scenarios/shc_cdi_scenario.py,sha256=5aVEiRgFCutEWW9yMcJBxEo11FlwW0SiZTaOyXY6ioc,2693
-helm/benchmark/scenarios/shc_conf_scenario.py,sha256=3LDB2pT6yi-ubSooGAD_0Ao7sYLo_MMAHNfm5Ux9Yvk,2889
-helm/benchmark/scenarios/shc_ent_scenario.py,sha256=PS_O_keZ5s5_nSKxAC1k_WV2W8umEbyyKmlFtxvaReI,2855
-helm/benchmark/scenarios/shc_gip_scenario.py,sha256=cxMpMmS05QpZ4xW2eogPH1hcDv6GzA6UQoAi9OSFO_Q,2702
-helm/benchmark/scenarios/shc_privacy_scenario.py,sha256=dbQI_pDqXepV6EyxMUNumIpyQ8oDwnu37qyQ29rxZfY,2998
-helm/benchmark/scenarios/shc_proxy_scenario.py,sha256=edepzg5qrN_GKa7u1W0RRhkpmfUi2vFHCvI1ma205WQ,2908
-helm/benchmark/scenarios/shc_ptbm_scenario.py,sha256=QOQdz21s_YaRyGz-ciCPHH-fCy6hiGIrHUZz0SWPm5o,3391
+helm/benchmark/scenarios/scenario.py,sha256=6zYT0heGPh1HXmv9l2g360Y3CwcV4xjA6jUq5snNF5I,9482
+helm/benchmark/scenarios/seahelm_scenario.py,sha256=Pgw05ZT9NByV7GL0031vGImbhGOZPrHv8aOR5DmP7sA,94098
+helm/benchmark/scenarios/self_instruct_scenario.py,sha256=uPVclF96zh0P2VJ163nLa7XuTKlMKGaTDFN-6IcdbXQ,3164
+helm/benchmark/scenarios/shc_bmt_scenario.py,sha256=kLnoSmpNaPKUcHDPhS6sDP38TC0YII5dlvEKpiUZYKY,3787
+helm/benchmark/scenarios/shc_cdi_scenario.py,sha256=Fg6PKKLLtmVxuu8pTOAmmoRpPIlFhxWl4VzIUNr7w6Y,3519
+helm/benchmark/scenarios/shc_conf_scenario.py,sha256=605KB8lTHlJh44XwbkilKXXAfJQGD2XVnZJmFoaV4Vw,3948
+helm/benchmark/scenarios/shc_ent_scenario.py,sha256=Sr4E3z0keK69b0DIZ1QFISvG0TsEQ6S567h84eSEHcc,3737
+helm/benchmark/scenarios/shc_gip_scenario.py,sha256=MhQ4mdKMJOtcZJ0gKxoVCg2RVyM8OKfjW_EA3wna_2c,3564
+helm/benchmark/scenarios/shc_privacy_scenario.py,sha256=OTYdD5mifaEZeI84RF5fz3Q10M8cE74H0GR3a7QisAE,3974
+helm/benchmark/scenarios/shc_proxy_scenario.py,sha256=bM_qSCv5Qp_03TiDezgl1gUSSs49IZ_M1L4xZnMzToc,3915
+helm/benchmark/scenarios/shc_ptbm_scenario.py,sha256=BttMbH39uai4qg621W0ySAFX-UtoRLuyEi-f4bfSrFo,4461
 helm/benchmark/scenarios/shc_sei_scenario.py,sha256=pTcb7n97VkesyRuqUqe5JGed1jDsQEd19udciDras8E,4532
-helm/benchmark/scenarios/shc_sequoia_scenario.py,sha256=vjDyRZXP9UjkQzmA6u7SmKtMBuUwwn6KRQ4rT3vZqqc,2796
+helm/benchmark/scenarios/shc_sequoia_scenario.py,sha256=IPOuJ74AIWOLDVIQ5lNUjMswcU9zeB_gOXg-K9HLTO4,3703
 helm/benchmark/scenarios/simple_safety_tests_scenario.py,sha256=sjIHT5NZlHv_IcXr_15-pOiBUPKKwykyH-QpMfvrHAY,1247
 helm/benchmark/scenarios/simple_scenarios.py,sha256=ersSzp9bFEFfpJ-SNy368AuonwswLnuyA1n7FOgkw4U,6459
+helm/benchmark/scenarios/situation_prompts.yaml,sha256=nJA3X_I67PIpXgd7LTekWwEr5zn1ryqIHgvqCpAwoGQ,1790
 helm/benchmark/scenarios/spider_scenario.py,sha256=mhiV3XWGwpnIQkaHFM_rvZlrwE7nqS12-F9t1eB8kdI,3306
-helm/benchmark/scenarios/starr_patient_instructions_scenario.py,sha256=zdokiMy2Lrg5mS3V2QEakcZyJxIkqcoT5CqVCAtyoKU,4146
-helm/benchmark/scenarios/summarization_scenario.py,sha256=WZnqhMQED6UBmRjHSboygdenLecOqIhvgdYVXzy6Q-I,6912
+helm/benchmark/scenarios/starr_patient_instructions_scenario.py,sha256=ZiXGXeKelEm9NrFsHQS5ft1L4oL6a_IlAJm_flRv-Z4,5228
+helm/benchmark/scenarios/summarization_scenario.py,sha256=wry6hAO_YXk56gS79jJ6HP6VhrRjpExvEZSsl2vM910,8883
 helm/benchmark/scenarios/sumosum_scenario.py,sha256=HG3wrKj5alV0a2aKb_nau8bB4oKDtTOLtdf3bx8h7sw,7695
-helm/benchmark/scenarios/synthetic_efficiency_scenario.py,sha256=OaxEvT1H9VjOjBSw_yKs3dcYt33vFE_UARr-UIP9pBY,3120
-helm/benchmark/scenarios/synthetic_reasoning_natural_scenario.py,sha256=pt2Aln_dX1YMSl-9hV1HJmwW90MC3fWwGsMxZg-Q-UY,16391
+helm/benchmark/scenarios/synthetic_efficiency_scenario.py,sha256=HbCeVUzPm3miSZoIDivTcAkP-fwi6X4TnyaAx0jUumk,3737
+helm/benchmark/scenarios/synthetic_reasoning_natural_scenario.py,sha256=Ir8CVC0aD7Cy7H_ZKGMd1c0iLK-dWbkuMuUl2D7kcR4,17048
 helm/benchmark/scenarios/synthetic_reasoning_scenario.py,sha256=7STCSHiHGIQ2aaN_PwDE5jXUJ-qcu8PaS4pC-pbOceE,8410
 helm/benchmark/scenarios/test_air_bench_scenario.py,sha256=9o92CK57xxgPaA9Xt9uJPPie4Cxllzq-KbMt3G35UQ0,1320
+helm/benchmark/scenarios/test_alghafa_scenario.py,sha256=ARQyzjmEpX_FpN2QLnIB7P-ToAeMtE4dqsolzlq8KPQ,1696
+helm/benchmark/scenarios/test_alrage_scenario.py,sha256=9ofFc05Sy1mdfU9VgHdL_SNp8olJ4ComnZllkMU6itU,6711
+helm/benchmark/scenarios/test_arabic_exams_scenario.py,sha256=nD221WpNE3Ddy-VOdLQGYOHiYVBAcyJxeMc__lVNRLo,985
+helm/benchmark/scenarios/test_aratrust_scenario.py,sha256=6Ks4DA13gU4BAP46qKwPISkqIJw-RiZt4ZhyviXdrUY,918
 helm/benchmark/scenarios/test_bigcodebench_scenario.py,sha256=q9FWJsxLJoFaB3PSMLjI_-YyPoZYusOsMPwn6X6NKXw,1304
+helm/benchmark/scenarios/test_bluex_scenario.py,sha256=QCIqq0GRRrjb55lwLpBiEkDwSFzEZxBKbCQHvyYO_Fk,1928
 helm/benchmark/scenarios/test_commonsense_scenario.py,sha256=V5Mq4cxWqU6j1U3icfIuzcnCZsZO7NTKLQgF0lEpdyc,924
 helm/benchmark/scenarios/test_czech_bank_qa_scenario.py,sha256=bZNLEGu58iHmutGlSp-2uVC2931TO6Rxw7giqFh9RHY,828
 helm/benchmark/scenarios/test_enem_challenge_scenario.py,sha256=XfPkYaSwdGa63ToC_BLuVKTRSldWNBlKsZYK6CFzL3w,2000
 helm/benchmark/scenarios/test_ewok_scenario.py,sha256=WY2vqbHF1120ht4PER0uviKMb2jnoPM3ff4KwvwcU4I,1291
+helm/benchmark/scenarios/test_exams_multilingual_scenario.py,sha256=vHLTcEzo5SkZgy2yXYm1Sex641qkr4HQWmVsOrlCQ_s,1764
 helm/benchmark/scenarios/test_financebench_scenario.py,sha256=EFZLJXXBoyjlTiMQFaQ6MiYkve1lfQDjQWjn4BjqgAQ,1184
 helm/benchmark/scenarios/test_gold_commodity_news_scenario.py,sha256=RO0NcIkJuujdPVO6tDygmDxhZ5YlmIIYlhwx9LeXlQs,731
 helm/benchmark/scenarios/test_gpqa_scenario.py,sha256=QQJ_-nmujZBSmhBhikRUWznFJ4jHPbGDnUVCP_17poI,1884
 helm/benchmark/scenarios/test_grammar.py,sha256=sPlA36sHpThbXgnGlXyOuqHfDPe2epIafmzIeL0nkoU,1364
 helm/benchmark/scenarios/test_gsm_scenario.py,sha256=I-Sl8Sg8kmFd7u0zZbwbNmeFV1mQLuOHoQ1cQDDwovs,1123
+helm/benchmark/scenarios/test_healtha_br_scenario.py,sha256=YmhXK24MuTPyLFCkXXI7IlwwiiJxytAbONOEh6wSJWI,1935
 helm/benchmark/scenarios/test_ifeval_scenario.py,sha256=h3CBg13VKwyb1Xaddwg2GWOzAXz4stK5lXdQtHenAw0,1646
 helm/benchmark/scenarios/test_imdb_ptbr_scenario.py,sha256=8kfCkMRUMU7N4WIrWawFDoxaLB2iTvQ-sPj4RoE2Osg,887
 helm/benchmark/scenarios/test_infinite_bench_en_qa_scenario.py,sha256=qZE-fi1tdNOybpvEQZJUpq9fHsyrPW7NYqj_RTwsv2A,746
@@ -509,18 +540,18 @@ helm/benchmark/scenarios/test_tweetsentbr_scenario.py,sha256=V6ZsT405ltgC3pYXW-F
 helm/benchmark/scenarios/test_wildbench_scenario.py,sha256=pmQ87MNoGAXwAmPf0eoep5qf9hk6BPP2zzgzGuKXwzs,527
 helm/benchmark/scenarios/test_winogrande_afr_scenario.py,sha256=LZfE4J42BZ7OF3BvfKgMWuCHpdw4-LpWnFiKyrHGXp8,910
 helm/benchmark/scenarios/thai_exam_scenario.py,sha256=YjFsom1yiu-xBZ3SGenNuczVCwQcmyoITTMavGv-QEk,6069
-helm/benchmark/scenarios/the_pile_scenario.py,sha256=X3GWABiJ5cSoZzeNpgNUVAz7_A9SyM5MhgpJseKpZow,5019
-helm/benchmark/scenarios/truthful_qa_scenario.py,sha256=kUQ-Bpu1N1s525EP3pa7v3sp9Wybl0RuJv2pVu0pAGQ,6155
+helm/benchmark/scenarios/the_pile_scenario.py,sha256=Dz51JxxazqPiX_fk6viOav8hQ2n6Iw0LIPhouquu6aw,5632
+helm/benchmark/scenarios/truthful_qa_scenario.py,sha256=0U7q8E9XB0H9oSN3OzhfsiZ-8PJrYXCCC04dffjicB8,6822
 helm/benchmark/scenarios/tweetsentbr_scenario.py,sha256=ppugbPWd_3hHesLC52QbC-wUknctr9ZX4tmHefnPf6w,2879
-helm/benchmark/scenarios/twitter_aae_scenario.py,sha256=wnP-zH38J62zmbdeOLzdU-E3iclbQPApgEk4AGyhdoo,2120
+helm/benchmark/scenarios/twitter_aae_scenario.py,sha256=ydG8MvBF3v6TXHScMK0_-HPAhmPhMWh5G4foBEHDp84,2905
 helm/benchmark/scenarios/unitxt_scenario.py,sha256=uL8Gni-Uw_eIp9xKQefp4J7XtKSttjJHzJE4USyoC2U,1930
 helm/benchmark/scenarios/verifiability_judgment_scenario.py,sha256=2iCJplnxdR7NNKjhsLR5o51pL55Q0bcbjjWlvrk5lw4,6067
-helm/benchmark/scenarios/vicuna_scenario.py,sha256=RFLUXx4zTfVPl5nT5j_DZ9TuHzk216PQcktomXqqR50,1685
-helm/benchmark/scenarios/wikifact_scenario.py,sha256=tTIHk7-xEsi-CGTobcEdbsjVrtAXTZOeWXRVj0hOeWA,5856
+helm/benchmark/scenarios/vicuna_scenario.py,sha256=OWwbFkhgEMHd5YH2G3v2E_p22DmYmPVsDbKKhBbyTDY,2478
+helm/benchmark/scenarios/wikifact_scenario.py,sha256=AHHZz_trtGf8HRoCnE6vukqrTD_Of9XQ1GcrqyctgR0,6702
 helm/benchmark/scenarios/wikitext_103_scenario.py,sha256=k13TxITriwqoBrMzf-JzPKr5wHaC9M2A_HyxxBaASnk,3111
-helm/benchmark/scenarios/wildbench_scenario.py,sha256=Qd9b1SC9ZtY1spf1vVuuFXXVxSJ0FlmR_DP7mIvAs8I,2981
+helm/benchmark/scenarios/wildbench_scenario.py,sha256=dWJSqF06ZWAyZhaejNmrZ0Uu4Vlh5HMdTaMLNkMfe8Q,3668
 helm/benchmark/scenarios/winogrande_afr_scenario.py,sha256=3SOVyrQ8D7Wzz06uSbczDE-IN4sjKSEAJ7Po-_-O6qw,3131
-helm/benchmark/scenarios/wmt_14_scenario.py,sha256=1YYjz4x2RbYfJAXBTux9X30dxYTSC-YNngCCLhEiNfI,4646
+helm/benchmark/scenarios/wmt_14_scenario.py,sha256=TNIYBXnbuvaOcpfmKqRZF6-yta1pTZSLA4Fd_XHhjCY,5159
 helm/benchmark/scenarios/xstest_scenario.py,sha256=ndRNB5ApW4th5iltlmT9-Nfw9eTaVZQw5AMC4HZCI-k,1309
 helm/benchmark/scenarios/audio_language/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 helm/benchmark/scenarios/audio_language/air_bench_chat_scenario.py,sha256=NtTEHzmbeCicbjTRxPBUueZrBGOPwF6RVc2Yftc-VKs,5634
@@ -529,7 +560,7 @@ helm/benchmark/scenarios/audio_language/ami_scenario.py,sha256=SH4r2YyW2kQ8r6-nS
 helm/benchmark/scenarios/audio_language/audio_mnist_scenario.py,sha256=kiUngeoAVOXfuKgqo96RgK_volpJUPFziu-cYDqT8WM,2685
 helm/benchmark/scenarios/audio_language/audio_pairs_scenario.py,sha256=oLOeBGjQCa3hpzjhX2bNS6637VD9VF1KbSJri9BJ3PI,2698
 helm/benchmark/scenarios/audio_language/audiocaps_scenario.py,sha256=PkVqQM1zX6ecXYk-Pz4YWlST3Hnla8NyeBHbuHvhSlY,2447
-helm/benchmark/scenarios/audio_language/casual_conversations2_scenario.py,sha256=uoiB3mnkudRH_rY1qeZRgobYYZ0xDn93F1Mn6Avl24Q,6724
+helm/benchmark/scenarios/audio_language/casual_conversations2_scenario.py,sha256=4X_C68yoMKRUC3SuNHYK4_fcboOz-9gbjhbUK1g3VVY,6725
 helm/benchmark/scenarios/audio_language/common_voice_15_scenario.py,sha256=CbcoGPW65xXRRkrDthotDfoVn51ozANG9s3LCsjxkLA,3706
 helm/benchmark/scenarios/audio_language/corebench_scenario.py,sha256=R8RAUtdRAQcUAN0PFXybQUekdQFNtT8hXtoR1A1hMGk,3155
 helm/benchmark/scenarios/audio_language/covost2_scenario.py,sha256=3YiaQXuLGfths2XswRw30Vf26bO9jEW_kAj5wZQSOSI,5119
@@ -540,14 +571,15 @@ helm/benchmark/scenarios/audio_language/librispeech_fairness_scenario.py,sha256=
 helm/benchmark/scenarios/audio_language/librispeech_scenario.py,sha256=ogMXxnyTG05tCyJ2d4hiuiVsbQvf4TbndksYeaJXl1s,3475
 helm/benchmark/scenarios/audio_language/meld_audio_scenario.py,sha256=j1JFX0jGfcqX0QZBKSjYjDWo1jHJbW5Q9jHyOs6Kgls,4903
 helm/benchmark/scenarios/audio_language/multilingual_librispeech_scenario.py,sha256=Jo_-3zC226iKGT-ac0JNMhlEccazMMiHbomx_qU0rxg,3098
-helm/benchmark/scenarios/audio_language/mustard_scenario.py,sha256=9bpcvFtWq5Pd9i9X8iaY9jod3YcRqk88xnXfjwcNMoY,6130
+helm/benchmark/scenarios/audio_language/mustard_scenario.py,sha256=7YHgfSpua5OdEGPlmxoufwGXQjvGJMTlEWFiJ_ap5ME,6131
 helm/benchmark/scenarios/audio_language/mutox_scenario.py,sha256=bDCQbhsRDR6iQGNlCu_35kjmjGjuzjOIoraSncfOlOY,10277
 helm/benchmark/scenarios/audio_language/parade_scenario.py,sha256=UuOa5cSrHh5n3VF_SuJp4cy1MxlI3uEKHLrNEhGuyuw,4186
 helm/benchmark/scenarios/audio_language/speech_robust_bench_scenario.py,sha256=oN4vBkElVzjccaEK2JFqoXMCGFTTHD0gcYwSDhvHTpQ,5438
-helm/benchmark/scenarios/audio_language/ultra_suite_asr_classification.py,sha256=TfMTdQ_D4foKO4NRPXygDgdF0ST2LYiOcV3gXO3WEYE,3691
-helm/benchmark/scenarios/audio_language/ultra_suite_classification_scenario.py,sha256=OUPFMOpRCTLN0o_lo7JJ7oOHxp9VuwC0fz4abWVS7hA,4713
-helm/benchmark/scenarios/audio_language/ultra_suite_disorder_breakdown_scenario.py,sha256=7-M5HXNE-YDM44f6LO4aYKBeFQxa3PfvN7q4u4BBYxU,4089
-helm/benchmark/scenarios/audio_language/ultra_suite_disorder_symptoms_scenario.py,sha256=c36E2RkeSDumLZgN6dBGzGz1ltgPdcBSqx8XD0qNH-U,5078
+helm/benchmark/scenarios/audio_language/ultra_suite_asr_classification_scenario.py,sha256=2qzPYfn0YYzzOtffD50kQu_ePpFJj_sSW7Bq8ZS6M2g,3559
+helm/benchmark/scenarios/audio_language/ultra_suite_asr_transcription_scenario.py,sha256=TEyfAsas3ihN4b4bpGkbK_M_uDt39fVrL5k8vl2Cdyw,3389
+helm/benchmark/scenarios/audio_language/ultra_suite_classification_scenario.py,sha256=qPOP6eIEwxPKu6q5EzcrRmhMxMUQk5F9iq8zdJ1Ccrc,4819
+helm/benchmark/scenarios/audio_language/ultra_suite_disorder_breakdown_scenario.py,sha256=CGteDFCd31vbu_eg5oal1cnfjQ2J0Ty3C2HYyBLhI5M,4186
+helm/benchmark/scenarios/audio_language/ultra_suite_disorder_symptoms_scenario.py,sha256=sL93Q2ERzYiWcTOFEyvjUNbX0BgPdsyHKt6eTr51-Kc,5177
 helm/benchmark/scenarios/audio_language/vocal_sound_scenario.py,sha256=wkKyTCtx4isQSMufap_6DsNdGkHi7L8FQ2p7n58kKYI,3124
 helm/benchmark/scenarios/audio_language/voice_jailbreak_attacks_scenario.py,sha256=4M_gTWs4CoJ1Ce9dDFBTAe9dzSovpsve_sN1eco2V2A,3155
 helm/benchmark/scenarios/audio_language/voxceleb2_scenario.py,sha256=L04ee5bM5E0UNNmkwEzVwug4HJXQoIcVjujPgxtU2h0,4366
@@ -570,6 +602,8 @@ helm/benchmark/scenarios/image_generation/radiology_scenario.py,sha256=7JN8OYap8
 helm/benchmark/scenarios/image_generation/relational_understanding_scenario.py,sha256=DoabanZhd-2MHFDZeR9EoPit0T2TvbVwZGUR0RfJyW0,2362
 helm/benchmark/scenarios/image_generation/time_most_significant_historical_figures_scenario.py,sha256=IB4_GbzQjjXBp-551XZ6PTNUCRX1jLcGfB3bVFI5lo4,3547
 helm/benchmark/scenarios/image_generation/winoground_scenario.py,sha256=E2xPQNQzylDSmqLjjMkQB8D7A6g7bzqtSF4bXPgfVbI,2889
+helm/benchmark/scenarios/medhelm/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+helm/benchmark/scenarios/medhelm/judges.yaml,sha256=g_O6zVgOMSL4_f1yNz8muDuUUBzcsM8e5gpfe56eI4Y,663
 helm/benchmark/scenarios/vision_language/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 helm/benchmark/scenarios/vision_language/a_okvqa_scenario.py,sha256=s-sdEFVx2BgqDFTzuQCCQr4oXaYHUUeQpFgblcCU97I,3052
 helm/benchmark/scenarios/vision_language/bingo_scenario.py,sha256=6YlGGGZW04Oy5A1-UG8JrN6jwR5eBuzrQ5qAise88o4,4108
@@ -614,11 +648,12 @@ helm/benchmark/scenarios/vision_language/image2struct/webpage/jekyll_server.py,s
 helm/benchmark/scenarios/vision_language/image2struct/webpage/utils.py,sha256=UYe3PnxCKBYEbZTTEzdIoTY9gW7ZZAWmVISRIdItD-A,940
 helm/benchmark/static/contamination.yaml,sha256=rAfh1DqwyUcDtyzHPQ2QiUK5eY7QfuuRtBXpZMn4TeA,3171
 helm/benchmark/static/schema_air_bench.yaml,sha256=LapSMj3Ecl1Gp9XIwVCYfrerqS93GNErvp6oDnBCtgw,142378
+helm/benchmark/static/schema_arabic.yaml,sha256=Iui-4_M4tV45Xzs3bz0diI3UZwVAuaLAxD5uNhjurgs,8925
 helm/benchmark/static/schema_audio.yaml,sha256=lVslZX7JmFo0ZgLU4n6amrs9DK8y43Ux0I9QyDUG-14,29119
 helm/benchmark/static/schema_autobencher.yaml,sha256=yb-NkF5w5R2YOg7RIsadNHJ_5G7lG1gbcDVq_25luEk,5716
 helm/benchmark/static/schema_call_center.yaml,sha256=i30aFzWqdOJRyAHN8vAzyHEX1v95DEK0TI1SMKTN4TE,9106
 helm/benchmark/static/schema_capabilities.yaml,sha256=HHy0aafhOaqL0C4TZw2mMt1Dce2_wuN062ORNZIbwYg,8733
-helm/benchmark/static/schema_classic.yaml,sha256=sK3yVQCrk3Tn3Kmg9WITBmJZI7AKVjmIY0f3zgH_t0c,104611
+helm/benchmark/static/schema_classic.yaml,sha256=pRkfy6jrdslx5onmeCUdkRi9y2DQrcPIjVyZLJ7uKCs,104147
 helm/benchmark/static/schema_cleva.yaml,sha256=TDh-zcCzzTTs7bu0IWlY5dXYaTFhxly8sJIBGQdBvug,25401
 helm/benchmark/static/schema_czech_bank.yaml,sha256=jkTRQVmmbKkbB0zPH9AtYh6Lt33ymMInRBQnHE5lIOo,5462
 helm/benchmark/static/schema_decodingtrust.yaml,sha256=2VPxzcyKYea7mx-qmswyVRjPfVatjVH4Rs3OU82mgII,15670
@@ -631,13 +666,14 @@ helm/benchmark/static/schema_image2struct.yaml,sha256=cD1X99YcPI8BMAnNfDmXlM-FN0
 helm/benchmark/static/schema_instruction_following.yaml,sha256=mYLpMv-iNtsmrv9ewfN9ceDOBBg8nSxOWfc6ByATmIk,6056
 helm/benchmark/static/schema_legal.yaml,sha256=RpoFOuVSIowNgxlPn3UMfJC-68RFr3CGDciUGLPfVqc,28806
 helm/benchmark/static/schema_lite.yaml,sha256=rFSoG7zGPNOtKkJyGgOViWf5WJbMiJMAXrgmqCAi9X4,36611
-helm/benchmark/static/schema_long_context.yaml,sha256=0xcyw8WI4SiLM1QPnjhTM-1SMGIyA5IDwWKpJzfQt9g,10795
-helm/benchmark/static/schema_medhelm.yaml,sha256=84BrIengbq0m42ICWvyEWoYtdERR-8J8-8QbPOqUzvA,50747
+helm/benchmark/static/schema_long_context.yaml,sha256=p01u7yPN75ZNmJhQodCRJo4q4Zb4vBieHKYqp4fD9Jg,11520
+helm/benchmark/static/schema_medhelm.yaml,sha256=e3vVHdEXcS-joOUMUoIoFA3x9hEE__svDoajbjfqpLE,51793
 helm/benchmark/static/schema_melt.yaml,sha256=mmPqwDa26DVZXsRJkmKQSyD0OStvjlxaMoSPM25SpD4,47494
 helm/benchmark/static/schema_mmlu.yaml,sha256=KI3XnzEwBRpzfYGjP77yKL-hBklEg72D3vL0kVl1BeI,49666
 helm/benchmark/static/schema_mmlu_winogrande_afr.yaml,sha256=YIVYf-mOFPq82UVBdMhnCWNOr4sV8Oi3-ozOszJ2tWQ,40143
 helm/benchmark/static/schema_safety.yaml,sha256=7RfZDX4wr8Xr1BJ149ZwmplPzPkNL0-BKbEZuzUsl_0,9278
 helm/benchmark/static/schema_seahelm.yaml,sha256=9XF9Rlr7I-g-uW6R0LNh7Xg52Xs3_058QybXEiN-hnM,28296
+helm/benchmark/static/schema_slp.yaml,sha256=5AV2leKoSBZwP3rIfXcwiqqpXPQbyWjXKE5kU73IAt4,7122
 helm/benchmark/static/schema_slphelm.yaml,sha256=3avOfp-ZEmVRGei3_M_WX6cSP5hQjbfHsDr1XrjayMY,5294
 helm/benchmark/static/schema_social_audio.yaml,sha256=Nj3ORXDT4RHD52cyo1RHfueWwbhqp1qW06TaVJ2lUfE,8653
 helm/benchmark/static/schema_sql.yaml,sha256=8rRff6p_i1CsH7oDbUjau2qRWbLGspuM1Hy-g5pOQiU,6047
@@ -649,15 +685,15 @@ helm/benchmark/static/schema_vhelm.yaml,sha256=0slYep2eepUefgtK_m4iSS785sHdJzljm
 helm/benchmark/static/schema_vhelm_lite.yaml,sha256=4I68Em9q5wW8sFzj5GCJz8m49fBEuMyVmSZM0-wbfOk,4024
 helm/benchmark/static/schema_video.yaml,sha256=FkpI5Slc4w-ty4hns82ArXIvTdqppWDnkJSpIp74QN4,9713
 helm/benchmark/static_build/config.js,sha256=o98g6QSly1NAfqhYWbU4lEoZB4LEpIrePZtmimiuoXc,165
-helm/benchmark/static_build/index.html,sha256=kpJ5Riw0YUmOOo2lSyWPgWx5XOwxxiLvPmG3wHwn2tM,1178
+helm/benchmark/static_build/index.html,sha256=MRRycZym58h-5KW7aKyiqGxIpRB8DV5OHkND5JL5aDk,1178
 helm/benchmark/static_build/assets/air-overview-d2e6c49f.png,sha256=0ubEn4J0T51-jx7IlwjaEGSrofZWlW_e67MJw47Ujzg,733055
 helm/benchmark/static_build/assets/crfm-logo-74391ab8.png,sha256=dDkauL_wJR_Luu7L7pltphS3a9HSLjDkpVLa6C9vcA4,62712
 helm/benchmark/static_build/assets/heim-logo-3e5e3aa4.png,sha256=Pl46pKbC_TU3L6kZQ_3G-0wTseluAhIYwb3EqpdQAjQ,1344452
 helm/benchmark/static_build/assets/helm-logo-simple-2ed5400b.png,sha256=LtVAC4OgcWgMAob53rTrf7cRDu-O0z85ZOGGj9wR9hw,86133
 helm/benchmark/static_build/assets/helm-safety-2907a7b6.png,sha256=KQentq_1e3uGwiWMViAPxHu2XZ60gqFgovP3UWTyMmw,72312
 helm/benchmark/static_build/assets/helmhero-28e90f4d.png,sha256=KOkPTf-q28PdvGOBp1G5O4q1eWUJjuij3z2h_SUUf8s,55314
-helm/benchmark/static_build/assets/index-94295e78.js,sha256=yvo6hRwNE6Ns7NxJHOdVfUOhc8HsW8eZVadLMW0Wn0w,124386
-helm/benchmark/static_build/assets/index-b9779128.css,sha256=uXeRKCUzQAC32ofNoaK3-WC7kRWR--KnR6--1m9NdQA,491471
+helm/benchmark/static_build/assets/index-671a5e06.js,sha256=XEa85-IyP6ZeHfsWGoPno-Qj9pSxlnHsjLYmaqzdzqg,124954
+helm/benchmark/static_build/assets/index-9352595e.css,sha256=k1JZXkXPFsUerOZ37oDhxjcb1ypOFEdDogJUP6H-NAQ,491553
 helm/benchmark/static_build/assets/medhelm-overview-eac29843.png,sha256=6sKYQ79cN07-cUsnt-JPsdoVwUBWu5KxOaHWSdwjdgA,284408
 helm/benchmark/static_build/assets/medhelm-v1-overview-3ddfcd65.png,sha256=Pd_NZfAf1ZeU2BIGx9zNT6WmypZNP2bk5z5AxDkbwoU,270625
 helm/benchmark/static_build/assets/overview-74aea3d8.png,sha256=dK6j2Nn3j9O-FMUIVRT5HGBpR_GL78vrKi8oHdG1eaI,74685
@@ -694,7 +730,7 @@ helm/benchmark/window_services/window_service.py,sha256=y6BthPY1V-ugmYfaJElm5Wfy
 helm/benchmark/window_services/window_service_factory.py,sha256=T55F0Y2jiOYxUHHZxT4YX4fFXY5gfFhn56zIwUBhc7s,3423
 helm/benchmark/window_services/yalm_window_service.py,sha256=EwwCoMpr9WVLhCI7OI_7tmZHQfTUwn9FFWjbhIBFRfA,1089
 helm/benchmark/window_services/image_generation/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-helm/benchmark/window_services/image_generation/clip_window_service.py,sha256=2JHld8GiR_eIQyHMPSN8K2VOswmKJEPMPJLsxlLpU-Q,631
+helm/benchmark/window_services/image_generation/clip_window_service.py,sha256=bhCZXzQDm2fEDKEslWDzkwPihQgmZS0DLVo__Ll9aLI,605
 helm/benchmark/window_services/image_generation/lexica_search_window_service.py,sha256=uDCUclHvo8toxSTMztK3zG7Eb-hjueobGQaBqPqVJlk,454
 helm/benchmark/window_services/image_generation/openai_dalle_window_service.py,sha256=8U2qDrUB1QJHRy5STV5FywkeVm6qfNOaeVBkMQhyMGc,453
 helm/benchmark/window_services/image_generation/test_clip_window_service.py,sha256=domn2MRduHVAdruSUuGPDIGKyDrh-gFxW-fZaBYR7cg,1430
@@ -717,7 +753,7 @@ helm/clients/google_client.py,sha256=mIaUzK7GHCa9pqK1BEVhdt6dZsJfHv1Qdsf3I0Ayq8A
 helm/clients/google_translate_client.py,sha256=TgiQEscjOae58Ptgp9f4n0LXUtl1Jf6v9BI-Z1_wcuw,1304
 helm/clients/grok_client.py,sha256=SbVB6AduTwfElzUgEMnQW2kQUFVTCv4TpPPJvElQEe0,1127
 helm/clients/http_model_client.py,sha256=_F3_y2UWqbzESQdzV0FMEsECIKjporVSAW6iUQhJ35c,2818
-helm/clients/huggingface_client.py,sha256=FYrg8XoCHXi5eUWjS0S_n-eiva-Ri0g1oaaeT_ky-tE,17615
+helm/clients/huggingface_client.py,sha256=oWR4yNFk28nrnB3IoznrhcEuU0pZkNywP0E82z1-NGM,17671
 helm/clients/huggingface_pipeline_client.py,sha256=ivFTMNHBwwIUjkeOHkl-veZi5nNAjtnkYvneRFWs-6Q,6154
 helm/clients/ibm_client.py,sha256=4W4fbjnDNjXrP4gVwSfBHPus0QcqFOQzFvfaST1BE1Y,9701
 helm/clients/lit_gpt_client.py,sha256=pgLfSvusNpdj8F5DVxzQdHxTDRNX4RVt6unegao803U,6229
@@ -727,8 +763,9 @@ helm/clients/mistral_client.py,sha256=ceM8KLAcniAqK1BNVdUGzqy4av2SEEau6PVmPivxc0
 helm/clients/moderation_api_client.py,sha256=I5pYWRb2MmcLDYrScnC3P5N7OUFzQiVQ828_hf7zjM4,4719
 helm/clients/nvidia_nim_client.py,sha256=Z1UAqR2jHacIO_QGqQl1JUZ_82JiSPstBOtj6xURmQk,902
 helm/clients/open_lm_client.py,sha256=qFgYqlV_3UiW8WJKz66lLqRqg2jt1qtJ1bHMRAtBn40,1749
-helm/clients/openai_client.py,sha256=s62_qafDVbDu5pzIkfQsflIwRzc4sXkSiDNkmZz68Ow,27775
-helm/clients/openai_responses_client.py,sha256=zua7DZWLeOdpb1yY8YV10gmuGdqvvo_9YQPW3OIGPDU,7219
+helm/clients/openai_client.py,sha256=4Q4LVMqvPo-37MV_BhsMydpwmMLfo-2kftRZH9lGtZs,28538
+helm/clients/openai_responses_client.py,sha256=FhQcOcXNZc5AuDMh1KBD3ZoRdEREy73dIeFBjUg9YDo,8444
+helm/clients/openrouter_client.py,sha256=oK8gXBhBs1y0AriZ9tVp8kx5lSY7gUgQJv-mfywSTfI,980
 helm/clients/palmyra_client.py,sha256=4AaZcV2tPHU4HJ9FWSkOY8_C9ndEckH3PH715QxJQ8E,7086
 helm/clients/perspective_api_client.py,sha256=o_1FFTCrTny6AZ4EJTstX1H9t8SQSQ8dvhi321RTcL4,6105
 helm/clients/reka_client.py,sha256=hA0tq3Hc9669q2sYa4Jr5yWy2NAbvoFDnVqQ6vds62w,8334
@@ -741,23 +778,49 @@ helm/clients/stanfordhealthcare_openai_client.py,sha256=Qyl8voGz1hJPqT6g4PunMuN9
 helm/clients/stanfordhealthcare_shc_openai_client.py,sha256=V7K4KZaSjIiE0FkoY4qy6ifJ8pUiNa3vBcWiDsIwXFI,1343
 helm/clients/test_auto_client.py,sha256=bc-rsMJ8JM0MFnQ4B48hBJ1jL3RtRyVvmPwOgzF2mF8,3155
 helm/clients/test_client.py,sha256=T27UsIPWsbE1JK_8DN_DW9LkEcIGRbgDjio14YOIAb0,3854
-helm/clients/test_huggingface_client.py,sha256=x2NjMuIrinfUy0wQ1S6F5cYZVr09YfvN6LfhWmyGNAM,3388
+helm/clients/test_huggingface_client.py,sha256=8Shzrf1Pad1UsiUAdeOSqsTPQaay0CrWXmdNeIfrJ2Y,3418
+helm/clients/test_openrouter_client.py,sha256=gCzchJMQZi4kkgtpGe1Ma0xF2nsP1uDevJcqbprZ6RE,2414
 helm/clients/test_simple_client.py,sha256=G0JRQX69ypQN2VxhlNQXs5u2Tdtkcl_aeHqudDUVKi4,702
 helm/clients/test_together_client.py,sha256=kyBLu-2i4EJyuJm5ft0yg8W-H1IqmULRXggEbChuxdo,6178
-helm/clients/together_client.py,sha256=tgjMlWscrauLFfMxDenh14oEBfLWyP9XYhz--YlvKVw,24264
+helm/clients/together_client.py,sha256=kEa6z54zPWlcLHCb2g2PCxLRpdJ8aE9zvG5Yzkaeun4,25518
 helm/clients/toxicity_classifier_client.py,sha256=AI_FizxMurubTIyeceRdkixSnhWQbcD-oEEONj5ve7o,464
 helm/clients/upstage_client.py,sha256=iSL1G8G3jWSbrpacz4I0l6Lwc5T01fsLR-wZzF39ftM,679
-helm/clients/vertexai_client.py,sha256=PjMnz4u5YQdpIbfLLBFsrPuHCNrj0_fatf1rY89d-nQ,23113
-helm/clients/vllm_client.py,sha256=YLIxGoQ_ZXejA4nfVpmFE4tmHROEFxEbFsV8Ba25Eac,1658
+helm/clients/vertexai_client.py,sha256=Qm-EkbpXnwiwZzB592-FPBuSlxKIkVH7tWBFFvOBvCY,23631
+helm/clients/vllm_client.py,sha256=xmXf35WX2oOZhpQnRxeooXGshENySOHZCUQ1E4pbQbA,2647
+helm/clients/vllm_granite_thinking_client.py,sha256=fds2i8LUG78OJYke1uYdDy6XRFqE3rZgSornFjzu4Sk,2172
 helm/clients/writer_client.py,sha256=flKLeMbFkyGfNmv1ozZGU4dxNy-QF5bFJF0mGHqpU3c,4467
 helm/clients/yi_client.py,sha256=nC60d2HiUL2W59FTne9tWmZ9bGGY1OvI7Ob3Ng4wSPE,750
 helm/clients/audio_language/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 helm/clients/audio_language/diva_llama_client.py,sha256=Bvcf4wE7yMZlqETgKEMtCug8-2fQI8QCDdaGWSeQ2X8,4864
 helm/clients/audio_language/llama_omni_client.py,sha256=OCak716q97uEk9CBXQqnmUsbLFR-dddMzg5eyIZ4gzE,8718
-helm/clients/audio_language/qwen2_5_omni_client.py,sha256=lbv6Hr22p0ReyR1bnN-dR8BzdPgilvGES7G03of8BWA,9090
+helm/clients/audio_language/qwen2_5_omni_client.py,sha256=ftAVtOG0azvRQEcFjkSSBMU6SDk9Bi8WIks6o6UCbKQ,9684
 helm/clients/audio_language/qwen2_audiolm_client.py,sha256=s9eH8fnVgw5xV39b_8AGt6IyNN3q9Uhcx6HZVxt7TM8,8981
 helm/clients/audio_language/qwen_audiolm_client.py,sha256=RvYweXANEyzhHYDx38H10F0ZEFaL8kj7n7TZ-UrRmZs,6338
 helm/clients/audio_language/test.py,sha256=FrKpirOwJW1__E2egq4VPgsTrgiSHZHBwfUCvxNjC0o,1969
+helm/clients/audio_language/llama_omni/arguments.py,sha256=MxzZKE8sNsOe5eUse96gejOsmu_MfTJGiuOwR87xiSA,2334
+helm/clients/audio_language/llama_omni/constants.py,sha256=IjFS9EUI5p1DLtGcX0B1lSxESkxcx5dMbuMkMm1UaSs,183
+helm/clients/audio_language/llama_omni/conversation.py,sha256=SgoMEf1Roi_8ZxiIM6DXwY3ozw0ExOCYdFFX-5rRA0g,6881
+helm/clients/audio_language/llama_omni/preprocess.py,sha256=2-YA6czgO1Zr-C1ChHvqVEfYa8qHhHp6n1Lb1Uw67qg,10764
+helm/clients/audio_language/llama_omni/utils.py,sha256=GycpuTkNEZtMNG2ZTZ7cYVjPEilyC4o2itT9K9kwJFI,7556
+helm/clients/audio_language/llama_omni/model/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+helm/clients/audio_language/llama_omni/model/builder.py,sha256=-y7amgUyPMEMknVutSSb_W3Zsm09r3K7u08jgEMinYA,3875
+helm/clients/audio_language/llama_omni/model/omni_speech_arch.py,sha256=-Sgo9fEGHRBfkZrR63i3-uXZ19wkqYbGLqAiDqevRr0,11465
+helm/clients/audio_language/llama_omni/model/language_model/omni_speech2s_llama.py,sha256=CqtEURdHlk6_29iM8WZnsmd7DMrUcnULGD2U2inWIxw,8426
+helm/clients/audio_language/llama_omni/model/language_model/omni_speech_llama.py,sha256=ZmtQY7JT74O4OH78UYSuBnmxq5Hi4-86kEY5-svfU-M,4564
+helm/clients/audio_language/llama_omni/model/speech_encoder/builder.py,sha256=TwSVGfSOA5N82pB2_P4cElN7w_4w2XHBXr9qicluM2w,389
+helm/clients/audio_language/llama_omni/model/speech_encoder/speech_encoder.py,sha256=LF8znt1puoExQ87ovtoyc1-pzO9kWNqTu_CvUWr3nBE,965
+helm/clients/audio_language/llama_omni/model/speech_generator/builder.py,sha256=nIjOSYgJTrdnqDvy5jnYjMcHvpOirAyvMNLuUbnL9pY,358
+helm/clients/audio_language/llama_omni/model/speech_generator/generation.py,sha256=Rka7iVephHHj0z0mPPQLfe-3Tt_UsWbTI7VRevs1ek4,30644
+helm/clients/audio_language/llama_omni/model/speech_generator/speech_generator.py,sha256=mllXYemRl-laMRntRsKtak8SIWEbVfWk0EpxPqs-su0,4612
+helm/clients/audio_language/llama_omni/model/speech_projector/builder.py,sha256=rmzWg4yZIfGpYD7VhfSrRNN7t5U4xNq8TVugq0KLYWc,372
+helm/clients/audio_language/llama_omni/model/speech_projector/speech_projector.py,sha256=naunMdDZXzK8VpASZJYsY6TwvuxQn3Uw9r_MUouUG5k,950
+helm/clients/audio_language/qwen_omni/configuration_qwen2_5_omni.py,sha256=oIaVRv1KlFYPqbT1nPtATgTcVomfNvtHmxnIZ2wcTC4,19088
+helm/clients/audio_language/qwen_omni/modeling_qwen2_5_omni.py,sha256=s08H7EY_TzHqVk1b6DZv_gI4VVwP_ub_FwF6JJu0z-c,180552
+helm/clients/audio_language/qwen_omni/processing_qwen2_5_omni.py,sha256=n8by91xA1xTYz8BfsbYAwCL5G0x1FuLhSGDAP33Qyyw,12216
+helm/clients/audio_language/qwen_omni/qwen2_5_omni_utils/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+helm/clients/audio_language/qwen_omni/qwen2_5_omni_utils/v2_5/__init__.py,sha256=ZZ5I9X_p1-ttDbYsLBxImO_CxbC5LESLI8AAIe9kKv0,365
+helm/clients/audio_language/qwen_omni/qwen2_5_omni_utils/v2_5/audio_process.py,sha256=VKATc5W9kl0fo9TuU19MaXYSObGxX2V2Fo1NlD4GC4I,2516
+helm/clients/audio_language/qwen_omni/qwen2_5_omni_utils/v2_5/vision_process.py,sha256=TFvQvPiP0X8Zt-agQR84o75LUZp0uXDZAUqUl0vhPM8,14635
 helm/clients/clip_scorers/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 helm/clients/clip_scorers/base_clip_scorer.py,sha256=NfXe79g6M4Wype3Xf-oXxscaUgjbZLmy9dRnBaLiWwk,695
 helm/clients/clip_scorers/clip_scorer.py,sha256=5KzYTrGuy5zA8yHX6c67Is98HLkqQooWhioPxHNLJ7s,1932
@@ -781,7 +844,7 @@ helm/clients/image_generation/cogview2/coglm_strategy.py,sha256=P3NU3Z4jsj171PrH
 helm/clients/image_generation/cogview2/coglm_utils.py,sha256=EJPOEQJInCDVi2LHqkjEUsgw6GgVlLDrIptlT9cXk-Y,2900
 helm/clients/image_generation/cogview2/sr_pipeline/__init__.py,sha256=qWuNwKlcvGwEFcw5932wk_t0_baNwUILIJzQWJjgh2A,488
 helm/clients/image_generation/cogview2/sr_pipeline/direct_sr.py,sha256=1DwcUw9Tb563JpKpkPNIB5Ew1djozvPiGASShffiABk,3716
-helm/clients/image_generation/cogview2/sr_pipeline/dsr_model.py,sha256=xYn3acxU4BRdDeRjk98Vj0qq8qqty93kPCLdz-bOMKs,10818
+helm/clients/image_generation/cogview2/sr_pipeline/dsr_model.py,sha256=IUTvHpIaaYrH00CQZZX9L45JMRb-twYir99K7LLnOzQ,10819
 helm/clients/image_generation/cogview2/sr_pipeline/dsr_sampling.py,sha256=OonYjdtNKJo12cNb-t-gFHLXRFxItCXjKgS9YxWAI-k,7718
 helm/clients/image_generation/cogview2/sr_pipeline/iterative_sr.py,sha256=LSvAHRupsOqk3yb4GxyTsubRxrnPOEfObFym2j4eiKc,5120
 helm/clients/image_generation/cogview2/sr_pipeline/itersr_model.py,sha256=5D1QWyAcY0CpwITk7EBN6ylUtc7mvZaE9iHG628AqMQ,10390
@@ -805,7 +868,7 @@ helm/clients/image_generation/mindalle/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JC
 helm/clients/image_generation/mindalle/models/__init__.py,sha256=1UieFJ0LGinYSB-idy3atl-gFAmS_ouiiGX6TM2Mh-I,8372
 helm/clients/image_generation/mindalle/models/tokenizer.py,sha256=NFFdLUhoxEkv9SZqU3QIFk0ukaCcn6w_xFWQIRGhZJ4,1190
 helm/clients/image_generation/mindalle/models/stage1/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-helm/clients/image_generation/mindalle/models/stage1/layers.py,sha256=cg5c6KShCMbMmHFuzDBIG-WgIBBkDrG9XvXC1DxqowA,11044
+helm/clients/image_generation/mindalle/models/stage1/layers.py,sha256=Q-yZeB8ZIxwOdQaKpEeBVbwF9nXeQJ2xJhiD6KjqRi4,11046
 helm/clients/image_generation/mindalle/models/stage1/vqgan.py,sha256=KcarvKoMuPBpP0H8F8W67FogdvHaAQuo9jP3rFRxc5E,4035
 helm/clients/image_generation/mindalle/models/stage2/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 helm/clients/image_generation/mindalle/models/stage2/layers.py,sha256=LvDADun5nMaencaRT0pm-dq78xHpPPkpi8rlu7RLHco,5306
@@ -840,11 +903,11 @@ helm/common/codec.py,sha256=gTh6AwIQ0Bbul_QSnIO7eItwMZmYtnkIrG1jkc4GOL4,7100
 helm/common/concurrency.py,sha256=8THtHlCtXo5c8iCuz_UcBBdzZX6aiEALLc4u0M4SYL0,856
 helm/common/context.py,sha256=0U5KNNKLHiiqjb8JVq03mninagEp9zTzFKP0He8o7A8,2788
 helm/common/credentials_utils.py,sha256=BX_P6wUpLKA7Bg3Dztm7jVI2j4ls7H-h38UbmGMBt3A,1101
-helm/common/critique_request.py,sha256=yo4aRe-DEjudUmydthtpTj6LdhRXfZ3JZptxTkWzZ3U,3068
+helm/common/critique_request.py,sha256=DZhJ_sY2IMluOxz-FeHvuEkA2Ujsx65HXT__7T3UxGk,3005
 helm/common/file_upload_request.py,sha256=OZeAW1_zsiNdXnWDwNNvhPs0b48TUmW_e4kzzCYmyiY,543
 helm/common/general.py,sha256=TcdPXn_bgPFvXtFP2lJhncz4Q8SdTXnKOinHOTBsegw,12027
 helm/common/gpu_utils.py,sha256=pmLq6ipYNLEm28VxxSNeZuVt-gAw-WnYmBvxP1P1p6M,480
-helm/common/hierarchical_logger.py,sha256=KR5R7tjUJN-hTFdnfzEyfwAhvgTFH3JJCH-LSiilqLk,4192
+helm/common/hierarchical_logger.py,sha256=qIbhwh-dlCcnYG10qTSMxIMM7_Q9VJj8ymDqnWlseuo,6151
 helm/common/image_generation_parameters.py,sha256=nsbuk_-BlRMK6IwP5y6BnTXbTRTOcvZ6uLblL5VHLOo,916
 helm/common/images_utils.py,sha256=8BsN0fd8pc0rh_TSDvippWhTfwmJJXKNF2zqKLB8cps,3372
 helm/common/key_value_store.py,sha256=D9ZBORzZncf3zHQOP4AuNbQnV8cZpO_kqHY1mDRugqQ,3174
@@ -854,7 +917,7 @@ helm/common/moderations_api_request.py,sha256=3xTsErSsCr2PHD2jpdV1JglHaYHwP2Yqu2
 helm/common/mongo_key_value_store.py,sha256=G0TIWQcvwMjyXh4TnN6xJ462HKHUAZtQJJYQOrHK-K8,3887
 helm/common/multimodal_request_utils.py,sha256=n6HgTyHNqfGmU9qmVK-wxQzrkPZ5Wdh-lO_y_ln6VYc,2184
 helm/common/nudity_check_request.py,sha256=VMsujI_RBy5u_cGEk0teE4KyX1dL2Zt3Pb4U6LpBdSY,728
-helm/common/object_spec.py,sha256=_usgTDQULBF6_jy7C6m-9ZNVvNxbGoTE_CdGcSvBASU,4327
+helm/common/object_spec.py,sha256=sKcEdggqRa3a8TovHAS4lf1LaahOFInvMl5DUF4tE6c,5186
 helm/common/optional_dependencies.py,sha256=Qam3QCHff8tuXbS-fCw-MVe-pK18gSvHw-uQoXXxT7M,616
 helm/common/perspective_api_request.py,sha256=WAVwtajNVmi5XJNsPcorGEAVrqkpPSk-Kd3b0hJghbA,2427
 helm/common/reeval_parameters.py,sha256=exaEucXnSI8a076uq_qhO3CTBztMMRoRzL_7v1N4adE,300
@@ -864,6 +927,7 @@ helm/common/response_format.py,sha256=wIptA8FydZoRjMvO5SFIplgDXhwpZvZmFI-Bi-7mcG
 helm/common/test_cache.py,sha256=j19p-qzv_98X_TMW4b39ZHwSJ-MX3p91PrkYumarS6Y,4870
 helm/common/test_codec.py,sha256=igL--k-2DwAy0eoMr8D9Xs8MOjBoT0LutbMPzDlTNkM,5885
 helm/common/test_general.py,sha256=c8Lh0mK8I-SfcMprq909B6zWRBxSBngq2nNL1L6-cYA,1788
+helm/common/test_logging.py,sha256=tkb_QDPkKBfaEQ5Y8Xip9PgMYhqOFakcENqyzO5Mj2o,2681
 helm/common/test_media_object.py,sha256=SUWLfms_vkXNivRYM0ZT8AI3_2ru6GON5l-Hb-lk-t0,1661
 helm/common/tokenization_request.py,sha256=NND9ESiiDE0H8QRNpfHVjXS7MQfKKIwtVRKDIjPnnJM,3344
 helm/common/file_caches/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -871,16 +935,16 @@ helm/common/file_caches/file_cache.py,sha256=QfF1hlF8FQ-rcPn9Zyl6L0dOCokvYgd-dFq
 helm/common/file_caches/local_file_cache.py,sha256=NiXbat1BBGl5P27oERqSLFfhIHpYqA1IQrvE_N1sWR8,1944
 helm/common/file_caches/test_local_file_cache.py,sha256=ANb01ctUV-J4i1ab3l4uhg9Ce54U_56xq9Hayjt1WhQ,686
 helm/config/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-helm/config/model_deployments.yaml,sha256=ec7CZLii6mpJeNC93J4gMgh1YrkU6Fj2XpXJaes01xY,160890
-helm/config/model_metadata.yaml,sha256=JvvKKEePcGCQf_cHGQv-k_Yj4GmB71lvRY2Is176a9s,263155
-helm/config/tokenizer_configs.yaml,sha256=Xju6-GcWJD-nmS5U0dUgkOexHWVWCd-J59EiVufoOCs,37687
+helm/config/model_deployments.yaml,sha256=JGM4eLHXv3KgndTu2ZqnMH5rwvoXNvKAoTAnmfZDs7A,174425
+helm/config/model_metadata.yaml,sha256=8W9u04RugI_L6Kj3ipGqQlWLeXAd_FQwcw-2usKm5y4,274605
+helm/config/tokenizer_configs.yaml,sha256=KZ6nReCV6AoActBoQYfi9BH4eGYkSx4OmSa2gzWh0uo,41039
 helm/proxy/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 helm/proxy/accounts.py,sha256=gd5cKhKeqklf_cXCAISl65AUvZeD6afBNrs6WK3IBvQ,14764
-helm/proxy/cli.py,sha256=apG3ByfyMciZFXV5wX2177p1B5eqkxCY6VoRgwJ81Kk,8316
-helm/proxy/example_queries.py,sha256=EB2vVpAryOUAFiLrwsMiFz0zGl_UAQ8TJ9SkWngvsu4,4389
+helm/proxy/cli.py,sha256=kEDoHpisFO0EJ0Wfm1FLpJdP9sXk9j8WCILEq42RKb0,8317
+helm/proxy/example_queries.py,sha256=A4JKvLwkHQIprsgMFhGvruW1-Ud4YKNqwUWhv6iWfzw,4449
 helm/proxy/query.py,sha256=eftbiUICMh8QIHVs-7cLtv_rDXKeKdRPmwjLMu0TDxQ,645
-helm/proxy/retry.py,sha256=iLZmKATEJQa9jsSpOIx6YDRhmrA8G1Qm21cUxCuo2Ug,3490
-helm/proxy/server.py,sha256=Q4Mzts8mketktGVJ5AoOEA-_SGCue5QeOlK8dqPUuHI,10853
+helm/proxy/retry.py,sha256=o64BZsW2vwu2iewRA18wdsru2xC3eNBQ7WUw3IjC_5g,3698
+helm/proxy/server.py,sha256=PYG8oMb-lq8eGR3Kad2ZTudJxgY4QH4jVbyoOgjes7I,10904
 helm/proxy/test_accounts.py,sha256=Vs1iOzTPN29LosDAAEs6IagQ3PccvutrJTlR1qNIcj0,1146
 helm/proxy/test_retry.py,sha256=db0owyGTThmIMhYWU_Eh1U-AJvQ-Wa9j_kRmC9DNjOA,1059
 helm/proxy/critique/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -900,9 +964,9 @@ helm/proxy/services/test_remote_service.py,sha256=xzkyptctXw3y5d1fgbidBMyw8B4rIL
 helm/proxy/services/test_service.py,sha256=oDYen-71iwZ6YMNBVbVSdEFsH6GMvZYw5tS5Eg4YHjY,8987
 helm/proxy/static/general.js,sha256=qcsntanG5UMWK2vznSVAVFy9zd3BMc8DFfNa7KKezew,3053
 helm/proxy/static/help.html,sha256=2Rn_lGZspqrZhNfLQ4wIAvYO_BK9q67Q_AS2-3WsMpY,6231
-helm/proxy/static/index.css,sha256=1OBOJ87LhwI2PtpoIyZoGQbSxQK2dz2vxk8BVmAybWY,717
+helm/proxy/static/index.css,sha256=3z_JuWVuJFngWtHI4T5-EVyk4LyaCPDcSzlalvUYhmQ,754
 helm/proxy/static/index.html,sha256=nUJf_hwBPokqrm_hDZsVfHcJrnhZLYhkVSoLdGOocf8,2009
-helm/proxy/static/index.js,sha256=-OXgf2rUYI49vg4KhwdL2VygKgzAGoYHKngaWgMb4E0,14996
+helm/proxy/static/index.js,sha256=bCjx29j88UnfoeYL4jRYGaqg7fd6o8IePZ0sTl-HRy8,15292
 helm/proxy/static/info-icon.png,sha256=P-PW3Ek3NGiRAW5BXOjJRPBfMVqprjAqtQheGWu7zNI,3428
 helm/proxy/token_counters/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 helm/proxy/token_counters/auto_token_counter.py,sha256=Ag368Sb-eLQUMLW7lmWc2EOKN3kgkiCTsYnHNrsf9kw,2071
@@ -914,7 +978,7 @@ helm/tokenizers/aleph_alpha_tokenizer.py,sha256=Ofc5thTfW_eb5ztiU-y_0p6e2PIGbHMb
 helm/tokenizers/auto_tokenizer.py,sha256=Of-T-CFOhLAjjU45T1hnrEPG_k_hzPufuDE7FRAcSN8,4251
 helm/tokenizers/caching_tokenizer.py,sha256=BwcyVzG7vy3R2O0UgbNxNP2nN4wBnsvpG_9mXQuDYfw,7300
 helm/tokenizers/cohere_tokenizer.py,sha256=6WwHIt7SsICmYR2QQpwDJ7pfNF8VWrFHFxF5Kynq6aY,2116
-helm/tokenizers/grok_tokenizer.py,sha256=Ms7QFYNookeq29AIfHUIXfKhrpRrPOPsNs0zBzWdLKA,2084
+helm/tokenizers/grok_tokenizer.py,sha256=t_cl1BnjRNCW24mU3Z6eAMhh-86FnCcSo-jB2AhvlL4,2142
 helm/tokenizers/http_model_tokenizer.py,sha256=J5Myg6JVDNgHMN7XOHwGV3WrhilUZ9Sw_FrgO4frYuY,3124
 helm/tokenizers/huggingface_tokenizer.py,sha256=P2ri4n-SUWB9ShMlxlJ9kO-mPmbSTizMGwAf41JE5ds,8734
 helm/tokenizers/lit_gpt_tokenizer.py,sha256=0c6KDeLNHPd6h27SXQvkUfmrCSLYa1kQY1GqCHVfhvw,1675
@@ -934,8 +998,8 @@ helm/tokenizers/yalm_tokenizer_data/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQ
 helm/tokenizers/yalm_tokenizer_data/test_yalm_tokenizer.py,sha256=1ZcPL3srfk031LmA8bEdPcIraAPnHGiYi_CqTiJSTlc,904
 helm/tokenizers/yalm_tokenizer_data/voc_100b.sp,sha256=LmPD0_OIOXi8dWuNjXUYOSPhf8kPp2xhvK-g3bXcwrQ,2815034
 helm/tokenizers/yalm_tokenizer_data/yalm_tokenizer.py,sha256=kH5Qig1_6r_sKbAHinX7C83tqBUoTwbe-gGZCbGVkko,6389
-crfm_helm-0.5.6.dist-info/METADATA,sha256=QlR8qMFpWzt_gIs6aCdrEEUuOS5uCdg1kbRMoI7YGYc,23069
-crfm_helm-0.5.6.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-crfm_helm-0.5.6.dist-info/entry_points.txt,sha256=AvH9soAH3uey9xffisWewd0yrmPWGASC036jHd1SFyg,300
-crfm_helm-0.5.6.dist-info/top_level.txt,sha256=s9yl-XmuTId6n_W_xRjCS99MHTwPXOlkKxmTr8xZUNY,5
-crfm_helm-0.5.6.dist-info/RECORD,,
+crfm_helm-0.5.8.dist-info/METADATA,sha256=UCr1ojkpYEsbV8_KfuhviO1vhPRs0fnfz7ADVaqa32E,18414
+crfm_helm-0.5.8.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+crfm_helm-0.5.8.dist-info/entry_points.txt,sha256=AvH9soAH3uey9xffisWewd0yrmPWGASC036jHd1SFyg,300
+crfm_helm-0.5.8.dist-info/top_level.txt,sha256=s9yl-XmuTId6n_W_xRjCS99MHTwPXOlkKxmTr8xZUNY,5
+crfm_helm-0.5.8.dist-info/RECORD,,

crfm-helm 0.5.6__py3-none-any.whl → 0.5.8__py3-none-any.whl

Potentially problematic release.

crfm-helm 0.5.6py3-none-any.whl → 0.5.8py3-none-any.whl