crfm-helm 0.4.0__py3-none-any.whl → 0.5.10__py3-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.
- crfm_helm-0.5.10.dist-info/METADATA +369 -0
- crfm_helm-0.5.10.dist-info/RECORD +1008 -0
- {crfm_helm-0.4.0.dist-info → crfm_helm-0.5.10.dist-info}/WHEEL +1 -1
- helm/benchmark/adaptation/adapter_spec.py +80 -29
- helm/benchmark/adaptation/adapters/adapter.py +2 -2
- helm/benchmark/adaptation/adapters/adapter_factory.py +39 -28
- helm/benchmark/adaptation/adapters/binary_ranking_adapter.py +1 -1
- helm/benchmark/adaptation/adapters/chat_adapter.py +49 -0
- helm/benchmark/adaptation/adapters/ehr_instruction_adapter.py +108 -0
- helm/benchmark/adaptation/adapters/generation_adapter.py +2 -1
- helm/benchmark/adaptation/adapters/in_context_learning_adapter.py +24 -8
- helm/benchmark/adaptation/adapters/language_modeling_adapter.py +3 -4
- helm/benchmark/adaptation/adapters/multimodal/generation_multimodal_adapter.py +4 -2
- helm/benchmark/adaptation/adapters/multimodal/in_context_learning_multimodal_adapter.py +2 -1
- helm/benchmark/adaptation/adapters/multimodal/multimodal_prompt.py +7 -0
- helm/benchmark/adaptation/adapters/multimodal/multiple_choice_joint_multimodal_adapter.py +112 -0
- helm/benchmark/adaptation/adapters/multimodal/test_in_context_learning_multimodal_adapter.py +6 -3
- helm/benchmark/adaptation/adapters/multimodal/test_multimodal_prompt.py +3 -1
- helm/benchmark/adaptation/adapters/multiple_choice_calibrated_adapter.py +1 -1
- helm/benchmark/adaptation/adapters/multiple_choice_joint_adapter.py +18 -8
- helm/benchmark/adaptation/adapters/multiple_choice_joint_chain_of_thought_adapter.py +87 -0
- helm/benchmark/adaptation/adapters/multiple_choice_separate_adapter.py +1 -1
- helm/benchmark/adaptation/adapters/test_adapter.py +5 -4
- helm/benchmark/adaptation/adapters/test_generation_adapter.py +46 -22
- helm/benchmark/adaptation/adapters/test_language_modeling_adapter.py +17 -29
- helm/benchmark/adaptation/adapters/test_multiple_choice_joint_adapter.py +138 -16
- helm/benchmark/adaptation/common_adapter_specs.py +443 -0
- helm/benchmark/adaptation/prompt.py +1 -1
- helm/benchmark/adaptation/request_state.py +6 -1
- helm/benchmark/adaptation/scenario_state.py +6 -2
- helm/benchmark/annotation/aci_bench_annotator.py +84 -0
- helm/benchmark/annotation/air_bench_annotator.py +79 -0
- helm/benchmark/annotation/alrage_annotator.py +90 -0
- helm/benchmark/annotation/annotator.py +48 -0
- helm/benchmark/annotation/annotator_factory.py +50 -0
- helm/benchmark/annotation/anthropic_red_team_annotator.py +57 -0
- helm/benchmark/annotation/autobencher_capabilities_annotator.py +107 -0
- helm/benchmark/annotation/autobencher_safety_annotator.py +98 -0
- helm/benchmark/annotation/bigcodebench_annotator.py +108 -0
- helm/benchmark/annotation/bird_sql_annotator.py +58 -0
- helm/benchmark/annotation/call_center_annotator.py +258 -0
- helm/benchmark/annotation/chw_care_plan_annotator.py +82 -0
- helm/benchmark/annotation/czech_bank_qa_annotator.py +78 -0
- helm/benchmark/annotation/dischargeme_annotator.py +96 -0
- helm/benchmark/annotation/ehr_sql_annotator.py +87 -0
- helm/benchmark/annotation/financebench_annotator.py +79 -0
- helm/benchmark/annotation/harm_bench_annotator.py +55 -0
- helm/benchmark/annotation/helpdesk_call_summarization_annotator.py +131 -0
- helm/benchmark/annotation/image2struct/image_compiler_annotator.py +93 -0
- helm/benchmark/annotation/image2struct/latex_compiler_annotator.py +59 -0
- helm/benchmark/annotation/image2struct/lilypond_compiler_annotator.py +86 -0
- helm/benchmark/annotation/image2struct/webpage_compiler_annotator.py +132 -0
- helm/benchmark/annotation/live_qa_annotator.py +76 -0
- helm/benchmark/annotation/med_dialog_annotator.py +88 -0
- helm/benchmark/annotation/medalign_annotator.py +89 -0
- helm/benchmark/annotation/medi_qa_annotator.py +87 -0
- helm/benchmark/annotation/medication_qa_annotator.py +86 -0
- helm/benchmark/annotation/mental_health_annotator.py +87 -0
- helm/benchmark/annotation/mimic_bhc_annotator.py +89 -0
- helm/benchmark/annotation/mimic_rrs_annotator.py +89 -0
- helm/benchmark/annotation/model_as_judge.py +309 -0
- helm/benchmark/annotation/mtsamples_procedures_annotator.py +87 -0
- helm/benchmark/annotation/mtsamples_replicate_annotator.py +90 -0
- helm/benchmark/annotation/omni_math/gpt_evaluation_template.txt +152 -0
- helm/benchmark/annotation/omni_math/gpt_evaluation_zero_shot_template.txt +36 -0
- helm/benchmark/annotation/omni_math_annotator.py +131 -0
- helm/benchmark/annotation/simple_safety_tests_annotator.py +50 -0
- helm/benchmark/annotation/spider_annotator.py +18 -0
- helm/benchmark/annotation/starr_patient_instructions_annotator.py +87 -0
- helm/benchmark/annotation/test_annotator_factory.py +26 -0
- helm/benchmark/annotation/test_dummy_annotator.py +44 -0
- helm/benchmark/annotation/wildbench/eval_template.pairwise.v2.md +75 -0
- helm/benchmark/annotation/wildbench/eval_template.score.v2.md +66 -0
- helm/benchmark/annotation/wildbench_annotator.py +119 -0
- helm/benchmark/annotation/xstest_annotator.py +100 -0
- helm/benchmark/annotation_executor.py +144 -0
- helm/benchmark/augmentations/cleva_perturbation.py +9 -8
- helm/benchmark/augmentations/contraction_expansion_perturbation.py +2 -2
- helm/benchmark/augmentations/contrast_sets_perturbation.py +2 -2
- helm/benchmark/augmentations/data_augmenter.py +0 -2
- helm/benchmark/augmentations/dialect_perturbation.py +4 -5
- helm/benchmark/augmentations/extra_space_perturbation.py +2 -2
- helm/benchmark/augmentations/filler_words_perturbation.py +2 -2
- helm/benchmark/augmentations/gender_perturbation.py +3 -3
- helm/benchmark/augmentations/lowercase_perturbation.py +2 -2
- helm/benchmark/augmentations/mild_mix_perturbation.py +6 -6
- helm/benchmark/augmentations/misspelling_perturbation.py +2 -2
- helm/benchmark/augmentations/person_name_perturbation.py +4 -5
- helm/benchmark/augmentations/perturbation.py +26 -4
- helm/benchmark/augmentations/perturbation_description.py +1 -1
- helm/benchmark/augmentations/space_perturbation.py +2 -2
- helm/benchmark/augmentations/suffix_perturbation.py +29 -0
- helm/benchmark/augmentations/synonym_perturbation.py +4 -3
- helm/benchmark/augmentations/test_perturbation.py +56 -19
- helm/benchmark/augmentations/translate_perturbation.py +31 -0
- helm/benchmark/augmentations/typos_perturbation.py +2 -2
- helm/benchmark/config_registry.py +7 -1
- helm/benchmark/data_preprocessor.py +2 -2
- helm/benchmark/executor.py +54 -25
- helm/benchmark/huggingface_registration.py +28 -10
- helm/benchmark/metrics/air_bench_metrics.py +3212 -0
- helm/benchmark/metrics/alrage_metric.py +35 -0
- helm/benchmark/metrics/annotation_metrics.py +108 -0
- helm/benchmark/metrics/basic_metrics.py +437 -667
- helm/benchmark/metrics/bbq_metrics.py +17 -6
- helm/benchmark/metrics/bias_metrics.py +18 -9
- helm/benchmark/metrics/bias_word_lists.py +1 -1
- helm/benchmark/metrics/bigcodebench_metrics.py +25 -0
- helm/benchmark/metrics/bird_sql_metrics.py +28 -0
- helm/benchmark/metrics/classification_metrics.py +107 -22
- helm/benchmark/metrics/cleva_accuracy_metrics.py +8 -5
- helm/benchmark/metrics/cleva_harms_metrics.py +12 -11
- helm/benchmark/metrics/code_metrics.py +5 -5
- helm/benchmark/metrics/code_metrics_helper.py +11 -3
- 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 +125 -0
- helm/benchmark/metrics/common_metric_specs.py +174 -0
- helm/benchmark/metrics/conv_fin_qa_calc_metrics.py +83 -0
- helm/benchmark/metrics/copyright_metrics.py +5 -5
- helm/benchmark/metrics/czech_bank_qa_metrics.py +29 -0
- helm/benchmark/metrics/decodingtrust_fairness_metrics.py +72 -0
- helm/benchmark/metrics/decodingtrust_ood_knowledge_metrics.py +66 -0
- helm/benchmark/metrics/decodingtrust_privacy_metrics.py +101 -0
- helm/benchmark/metrics/decodingtrust_stereotype_bias_metrics.py +202 -0
- helm/benchmark/metrics/disinformation_metrics.py +8 -114
- helm/benchmark/metrics/dry_run_metrics.py +35 -6
- helm/benchmark/metrics/efficiency_metrics.py +287 -0
- helm/benchmark/metrics/ehr_sql_metrics.py +159 -0
- helm/benchmark/metrics/evaluate_instances_metric.py +59 -0
- helm/benchmark/metrics/evaluate_reference_metrics.py +831 -0
- helm/benchmark/metrics/fin_qa_metrics.py +60 -0
- helm/benchmark/metrics/fin_qa_metrics_helper.py +398 -0
- helm/benchmark/metrics/gpqa_chain_of_thought_metric.py +115 -0
- helm/benchmark/metrics/gpt4_audio_critique_metrics.py +167 -0
- helm/benchmark/metrics/gpt4_audio_refusal_metrics.py +145 -0
- helm/benchmark/metrics/gpt4v_originality_critique_metrics.py +126 -0
- helm/benchmark/metrics/helpdesk_call_summarization_metrics.py +48 -0
- helm/benchmark/metrics/ifeval/instructions.py +1574 -0
- helm/benchmark/metrics/ifeval/instructions_registry.py +182 -0
- helm/benchmark/metrics/ifeval/instructions_registry.pyi +3 -0
- helm/benchmark/metrics/ifeval/instructions_util.py +153 -0
- helm/benchmark/metrics/ifeval_metrics.py +67 -0
- helm/benchmark/metrics/image_generation/aesthetics_metrics.py +54 -0
- helm/benchmark/metrics/image_generation/aesthetics_scorer.py +66 -0
- helm/benchmark/metrics/image_generation/clip_score_metrics.py +84 -0
- helm/benchmark/metrics/image_generation/denoised_runtime_metric.py +42 -0
- helm/benchmark/metrics/image_generation/detection_metrics.py +57 -0
- helm/benchmark/metrics/image_generation/detectors/base_detector.py +8 -0
- helm/benchmark/metrics/image_generation/detectors/vitdet.py +178 -0
- helm/benchmark/metrics/image_generation/efficiency_metrics.py +41 -0
- helm/benchmark/metrics/image_generation/fidelity_metrics.py +168 -0
- helm/benchmark/metrics/image_generation/fractal_dimension/__init__.py +0 -0
- helm/benchmark/metrics/image_generation/fractal_dimension/fractal_dimension_util.py +63 -0
- helm/benchmark/metrics/image_generation/fractal_dimension/test_fractal_dimension_util.py +33 -0
- helm/benchmark/metrics/image_generation/fractal_dimension_metric.py +50 -0
- helm/benchmark/metrics/image_generation/gender_metrics.py +58 -0
- helm/benchmark/metrics/image_generation/image_critique_metrics.py +284 -0
- helm/benchmark/metrics/image_generation/lpips_metrics.py +82 -0
- helm/benchmark/metrics/image_generation/multi_scale_ssim_metrics.py +82 -0
- helm/benchmark/metrics/image_generation/nsfw_detector.py +96 -0
- helm/benchmark/metrics/image_generation/nsfw_metrics.py +103 -0
- helm/benchmark/metrics/image_generation/nudity_metrics.py +38 -0
- helm/benchmark/metrics/image_generation/photorealism_critique_metrics.py +153 -0
- helm/benchmark/metrics/image_generation/psnr_metrics.py +78 -0
- helm/benchmark/metrics/image_generation/q16/__init__.py +0 -0
- helm/benchmark/metrics/image_generation/q16/q16_toxicity_detector.py +90 -0
- helm/benchmark/metrics/image_generation/q16/test_q16.py +20 -0
- helm/benchmark/metrics/image_generation/q16_toxicity_metrics.py +48 -0
- helm/benchmark/metrics/image_generation/skin_tone_metrics.py +164 -0
- helm/benchmark/metrics/image_generation/uiqi_metrics.py +92 -0
- helm/benchmark/metrics/image_generation/watermark/__init__.py +0 -0
- helm/benchmark/metrics/image_generation/watermark/test_watermark_detector.py +16 -0
- helm/benchmark/metrics/image_generation/watermark/watermark_detector.py +87 -0
- helm/benchmark/metrics/image_generation/watermark_metrics.py +48 -0
- helm/benchmark/metrics/instruction_following_critique_metrics.py +48 -5
- helm/benchmark/metrics/kpi_edgar_metrics.py +142 -0
- helm/benchmark/metrics/language_modeling_metrics.py +111 -0
- helm/benchmark/metrics/live_qa_metrics.py +35 -0
- helm/benchmark/metrics/llm_jury_metrics.py +58 -0
- helm/benchmark/metrics/lmkt_metric_specs.py +12 -0
- helm/benchmark/metrics/lmkt_metrics.py +47 -0
- helm/benchmark/metrics/machine_translation_metrics.py +89 -0
- helm/benchmark/metrics/medcalc_bench_metrics.py +137 -0
- helm/benchmark/metrics/medec_metrics.py +124 -0
- helm/benchmark/metrics/melt_bias_metric.py +234 -0
- helm/benchmark/metrics/melt_bias_word_lists.py +1367 -0
- helm/benchmark/metrics/melt_metric_specs.py +43 -0
- helm/benchmark/metrics/melt_toxicity_metric.py +107 -0
- helm/benchmark/metrics/metric.py +121 -175
- helm/benchmark/metrics/metric_name.py +0 -1
- helm/benchmark/metrics/metric_service.py +23 -7
- helm/benchmark/metrics/mimiciv_billing_code_metrics.py +127 -0
- helm/benchmark/metrics/nltk_helper.py +32 -0
- helm/benchmark/metrics/omni_math_metrics.py +44 -0
- helm/benchmark/metrics/openai_mrcr_metrics.py +52 -0
- helm/benchmark/metrics/output_processing_metric.py +60 -0
- helm/benchmark/metrics/output_processors.py +15 -0
- helm/benchmark/metrics/paraphrase_generation_metrics.py +5 -6
- helm/benchmark/metrics/prometheus_vision_critique_metrics.py +185 -0
- helm/benchmark/metrics/ranking_metrics.py +5 -5
- helm/benchmark/metrics/reference_metric.py +148 -0
- helm/benchmark/metrics/reka_vibe_critique_metrics.py +158 -0
- helm/benchmark/metrics/ruler_qa_metrics.py +34 -0
- helm/benchmark/metrics/safety_metrics.py +91 -0
- helm/benchmark/metrics/seahelm_metrics.py +201 -0
- helm/benchmark/metrics/seahelm_metrics_specs.py +10 -0
- helm/benchmark/metrics/spider_metrics.py +7 -0
- helm/benchmark/metrics/statistic.py +1 -1
- helm/benchmark/metrics/summac/model_summac.py +8 -11
- helm/benchmark/metrics/summarization_critique_metrics.py +4 -4
- helm/benchmark/metrics/summarization_metrics.py +150 -11
- helm/benchmark/metrics/test_bias_metrics.py +5 -1
- helm/benchmark/metrics/test_classification_metrics.py +145 -70
- helm/benchmark/metrics/test_disinformation_metrics.py +78 -0
- helm/benchmark/metrics/{test_basic_metrics.py → test_evaluate_reference_metrics.py} +20 -1
- helm/benchmark/metrics/test_metric.py +3 -3
- helm/benchmark/metrics/test_statistic.py +2 -2
- helm/benchmark/metrics/tokens/ai21_token_cost_estimator.py +1 -1
- helm/benchmark/metrics/tokens/auto_token_cost_estimator.py +6 -6
- helm/benchmark/metrics/tokens/cohere_token_cost_estimator.py +1 -1
- helm/benchmark/metrics/tokens/free_token_cost_estimator.py +1 -1
- helm/benchmark/metrics/tokens/gooseai_token_cost_estimator.py +11 -3
- helm/benchmark/metrics/tokens/openai_token_cost_estimator.py +1 -1
- helm/benchmark/metrics/tokens/test_ai21_token_cost_estimator.py +3 -3
- helm/benchmark/metrics/tokens/test_openai_token_cost_estimator.py +7 -7
- helm/benchmark/metrics/toxicity_metrics.py +37 -7
- helm/benchmark/metrics/toxicity_utils.py +23 -0
- helm/benchmark/metrics/ultra_suite_asr_classification_metrics.py +52 -0
- helm/benchmark/metrics/unitxt_metrics.py +107 -0
- helm/benchmark/metrics/vision_language/__init__.py +0 -0
- helm/benchmark/metrics/vision_language/emd_utils.py +347 -0
- helm/benchmark/metrics/vision_language/image_metrics.py +537 -0
- helm/benchmark/metrics/vision_language/image_utils.py +100 -0
- helm/benchmark/metrics/wildbench_metrics.py +54 -0
- helm/benchmark/model_deployment_registry.py +69 -5
- helm/benchmark/model_metadata_registry.py +58 -2
- helm/benchmark/multi_gpu_runner.py +133 -0
- helm/benchmark/presentation/contamination.py +3 -3
- helm/benchmark/presentation/create_plots.py +51 -20
- helm/benchmark/presentation/run_display.py +51 -12
- helm/benchmark/presentation/run_entry.py +2 -2
- helm/benchmark/presentation/schema.py +83 -66
- helm/benchmark/presentation/summarize.py +483 -388
- helm/benchmark/presentation/table.py +8 -8
- helm/benchmark/presentation/taxonomy_info.py +20 -0
- helm/benchmark/presentation/test_contamination.py +2 -2
- helm/benchmark/presentation/test_create_plots.py +4 -1
- helm/benchmark/presentation/test_run_entry.py +2 -2
- helm/benchmark/presentation/test_schema.py +11 -0
- helm/benchmark/presentation/test_summarize.py +148 -6
- helm/benchmark/presentation/torr_robustness_summarizer.py +178 -0
- helm/benchmark/reeval_run.py +202 -0
- helm/benchmark/reeval_runner.py +355 -0
- helm/benchmark/run.py +151 -87
- helm/benchmark/run_expander.py +418 -33
- helm/benchmark/run_spec.py +93 -0
- helm/benchmark/run_spec_factory.py +180 -0
- helm/benchmark/run_specs/__init__.py +0 -0
- helm/benchmark/run_specs/air_bench_run_specs.py +58 -0
- helm/benchmark/run_specs/arabic_run_specs.py +197 -0
- helm/benchmark/run_specs/audio_run_specs.py +657 -0
- helm/benchmark/run_specs/bluex_run_specs.py +40 -0
- helm/benchmark/run_specs/call_center_run_specs.py +201 -0
- helm/benchmark/run_specs/capabilities_run_specs.py +308 -0
- helm/benchmark/run_specs/classic_run_specs.py +1393 -0
- helm/benchmark/run_specs/cleva_run_specs.py +277 -0
- helm/benchmark/run_specs/codeinsights_run_specs.py +192 -0
- helm/benchmark/run_specs/decodingtrust_run_specs.py +316 -0
- helm/benchmark/run_specs/enem_challenge_specs.py +31 -0
- helm/benchmark/run_specs/enterprise_run_specs.py +280 -0
- helm/benchmark/run_specs/experimental_run_specs.py +224 -0
- helm/benchmark/run_specs/finance_run_specs.py +114 -0
- helm/benchmark/run_specs/healthqa_br_run_specs.py +40 -0
- helm/benchmark/run_specs/heim_run_specs.py +625 -0
- helm/benchmark/run_specs/imdb_ptbr_run_specs.py +30 -0
- helm/benchmark/run_specs/instruction_following_run_specs.py +129 -0
- helm/benchmark/run_specs/lite_run_specs.py +307 -0
- helm/benchmark/run_specs/lmkt_run_specs.py +144 -0
- helm/benchmark/run_specs/long_context_run_specs.py +188 -0
- 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 +1570 -0
- helm/benchmark/run_specs/melt_run_specs.py +783 -0
- helm/benchmark/run_specs/mmlu_clinical_afr_run_specs.py +49 -0
- helm/benchmark/run_specs/multilingual_run_specs.py +50 -0
- helm/benchmark/run_specs/oab_exams_specs.py +32 -0
- helm/benchmark/run_specs/safety_run_specs.py +191 -0
- helm/benchmark/run_specs/seahelm_run_specs.py +652 -0
- helm/benchmark/run_specs/simple_run_specs.py +104 -0
- helm/benchmark/run_specs/speech_disorder_audio_run_specs.py +167 -0
- helm/benchmark/run_specs/sql_run_specs.py +54 -0
- helm/benchmark/run_specs/tweetsentbr_run_specs.py +32 -0
- helm/benchmark/run_specs/unitxt_run_specs.py +51 -0
- helm/benchmark/run_specs/vlm_run_specs.py +1057 -0
- helm/benchmark/run_specs/winogrande_afr_run_specs.py +47 -0
- helm/benchmark/runner.py +63 -62
- helm/benchmark/runner_config_registry.py +21 -0
- helm/benchmark/scenarios/aci_bench_scenario.py +149 -0
- helm/benchmark/scenarios/air_bench_scenario.py +76 -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 +27 -3
- helm/benchmark/scenarios/anthropic_red_team_scenario.py +82 -0
- 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/__init__.py +0 -0
- helm/benchmark/scenarios/audio_language/air_bench_chat_scenario.py +130 -0
- helm/benchmark/scenarios/audio_language/air_bench_foundation_scenario.py +154 -0
- helm/benchmark/scenarios/audio_language/ami_scenario.py +96 -0
- helm/benchmark/scenarios/audio_language/audio_mnist_scenario.py +62 -0
- helm/benchmark/scenarios/audio_language/audio_pairs_scenario.py +62 -0
- helm/benchmark/scenarios/audio_language/audiocaps_scenario.py +59 -0
- helm/benchmark/scenarios/audio_language/casual_conversations2_scenario.py +152 -0
- helm/benchmark/scenarios/audio_language/common_voice_15_scenario.py +99 -0
- helm/benchmark/scenarios/audio_language/corebench_scenario.py +77 -0
- helm/benchmark/scenarios/audio_language/covost2_scenario.py +163 -0
- helm/benchmark/scenarios/audio_language/fleurs_fairness_scenario.py +83 -0
- helm/benchmark/scenarios/audio_language/fleurs_scenario.py +312 -0
- helm/benchmark/scenarios/audio_language/iemocap_audio_scenario.py +83 -0
- helm/benchmark/scenarios/audio_language/librispeech_fairness_scenario.py +96 -0
- helm/benchmark/scenarios/audio_language/librispeech_scenario.py +80 -0
- helm/benchmark/scenarios/audio_language/meld_audio_scenario.py +113 -0
- helm/benchmark/scenarios/audio_language/multilingual_librispeech_scenario.py +80 -0
- helm/benchmark/scenarios/audio_language/mustard_scenario.py +142 -0
- helm/benchmark/scenarios/audio_language/mutox_scenario.py +254 -0
- helm/benchmark/scenarios/audio_language/parade_scenario.py +97 -0
- helm/benchmark/scenarios/audio_language/speech_robust_bench_scenario.py +124 -0
- helm/benchmark/scenarios/audio_language/ultra_suite_asr_classification_scenario.py +74 -0
- helm/benchmark/scenarios/audio_language/ultra_suite_asr_transcription_scenario.py +70 -0
- helm/benchmark/scenarios/audio_language/ultra_suite_classification_scenario.py +79 -0
- helm/benchmark/scenarios/audio_language/ultra_suite_disorder_breakdown_scenario.py +78 -0
- helm/benchmark/scenarios/audio_language/ultra_suite_disorder_symptoms_scenario.py +78 -0
- helm/benchmark/scenarios/audio_language/vocal_sound_scenario.py +83 -0
- helm/benchmark/scenarios/audio_language/voice_jailbreak_attacks_scenario.py +87 -0
- helm/benchmark/scenarios/audio_language/voxceleb2_scenario.py +105 -0
- helm/benchmark/scenarios/autobencher_capabilities_scenario.py +68 -0
- helm/benchmark/scenarios/autobencher_safety_scenario.py +51 -0
- helm/benchmark/scenarios/babi_qa_scenario.py +16 -1
- helm/benchmark/scenarios/banking77_scenario.py +77 -0
- helm/benchmark/scenarios/bbq_scenario.py +17 -2
- helm/benchmark/scenarios/best_chatgpt_prompts.yaml +473 -0
- helm/benchmark/scenarios/big_bench_scenario.py +11 -1
- helm/benchmark/scenarios/bigcodebench_scenario.py +58 -0
- helm/benchmark/scenarios/bird_sql_scenario.py +112 -0
- helm/benchmark/scenarios/bird_sql_scenario_helper.py +118 -0
- helm/benchmark/scenarios/blimp_scenario.py +1 -1
- helm/benchmark/scenarios/bluex_scenario.py +70 -0
- helm/benchmark/scenarios/bold_scenario.py +18 -3
- helm/benchmark/scenarios/boolq_scenario.py +21 -1
- helm/benchmark/scenarios/call_center_scenario.py +84 -0
- helm/benchmark/scenarios/casehold_scenario.py +79 -0
- helm/benchmark/scenarios/chw_care_plan_scenario.py +129 -0
- helm/benchmark/scenarios/ci_mcqa_scenario.py +80 -0
- helm/benchmark/scenarios/civil_comments_scenario.py +14 -1
- helm/benchmark/scenarios/clear_scenario.py +180 -0
- helm/benchmark/scenarios/cleva_scenario.py +482 -3
- helm/benchmark/scenarios/code_scenario.py +46 -4
- 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 +33 -1
- helm/benchmark/scenarios/compositional_instructions.yaml +70 -0
- helm/benchmark/scenarios/conv_fin_qa_calc_scenario.py +118 -0
- helm/benchmark/scenarios/copyright_scenario.py +35 -1
- helm/benchmark/scenarios/covid_dialog_scenario.py +10 -1
- helm/benchmark/scenarios/cti_to_mitre_scenario.py +261 -0
- helm/benchmark/scenarios/custom_mcqa_scenario.py +1 -1
- helm/benchmark/scenarios/czech_bank_qa_scenario.py +148 -0
- helm/benchmark/scenarios/decodingtrust_adv_demonstration_scenario.py +190 -0
- helm/benchmark/scenarios/decodingtrust_adv_robustness_scenario.py +143 -0
- helm/benchmark/scenarios/decodingtrust_fairness_scenario.py +98 -0
- helm/benchmark/scenarios/decodingtrust_machine_ethics_scenario.py +344 -0
- helm/benchmark/scenarios/decodingtrust_ood_robustness_scenario.py +217 -0
- helm/benchmark/scenarios/decodingtrust_privacy_scenario.py +571 -0
- helm/benchmark/scenarios/decodingtrust_stereotype_bias_scenario.py +80 -0
- helm/benchmark/scenarios/decodingtrust_toxicity_prompts_scenario.py +90 -0
- helm/benchmark/scenarios/dialogue_scenarios.py +13 -3
- helm/benchmark/scenarios/dischargeme_scenario.py +196 -0
- helm/benchmark/scenarios/disinformation_scenario.py +32 -1
- helm/benchmark/scenarios/dyck_language_scenario.py +25 -1
- helm/benchmark/scenarios/echr_judgment_classification_scenario.py +113 -0
- helm/benchmark/scenarios/ehr_sql_scenario.py +137 -0
- helm/benchmark/scenarios/ehrshot_scenario.py +1541 -0
- helm/benchmark/scenarios/enem_challenge_scenario.py +77 -0
- helm/benchmark/scenarios/entity_data_imputation_scenario.py +33 -3
- helm/benchmark/scenarios/entity_matching_scenario.py +26 -2
- helm/benchmark/scenarios/ewok_scenario.py +116 -0
- helm/benchmark/scenarios/exams_multilingual_scenario.py +115 -0
- helm/benchmark/scenarios/fin_qa_scenario.py +139 -0
- helm/benchmark/scenarios/financebench_scenario.py +74 -0
- helm/benchmark/scenarios/financial_phrasebank_scenario.py +115 -0
- helm/benchmark/scenarios/gold_commodity_news_scenario.py +145 -0
- helm/benchmark/scenarios/gpqa_scenario.py +98 -0
- helm/benchmark/scenarios/grammar.py +2 -2
- helm/benchmark/scenarios/grammar_scenario.py +21 -2
- helm/benchmark/scenarios/gsm_scenario.py +31 -1
- helm/benchmark/scenarios/harm_bench_gcg_transfer_scenario.py +61 -0
- helm/benchmark/scenarios/harm_bench_scenario.py +70 -0
- helm/benchmark/scenarios/headqa_scenario.py +158 -0
- helm/benchmark/scenarios/healthqa_br_scenario.py +80 -0
- helm/benchmark/scenarios/helpdesk_call_summarization_scenario.py +50 -0
- helm/benchmark/scenarios/ice_scenario.py +28 -4
- helm/benchmark/scenarios/ifeval_scenario.py +71 -0
- helm/benchmark/scenarios/image_generation/__init__.py +0 -0
- helm/benchmark/scenarios/image_generation/common_syntactic_processes_scenario.py +105 -0
- helm/benchmark/scenarios/image_generation/cub200_scenario.py +95 -0
- helm/benchmark/scenarios/image_generation/daily_dalle_scenario.py +124 -0
- helm/benchmark/scenarios/image_generation/demographic_stereotypes_scenario.py +82 -0
- helm/benchmark/scenarios/image_generation/detection_scenario.py +83 -0
- helm/benchmark/scenarios/image_generation/draw_bench_scenario.py +74 -0
- helm/benchmark/scenarios/image_generation/i2p_scenario.py +57 -0
- helm/benchmark/scenarios/image_generation/landing_page_scenario.py +46 -0
- helm/benchmark/scenarios/image_generation/logos_scenario.py +223 -0
- helm/benchmark/scenarios/image_generation/magazine_cover_scenario.py +91 -0
- helm/benchmark/scenarios/image_generation/mental_disorders_scenario.py +46 -0
- helm/benchmark/scenarios/image_generation/mscoco_scenario.py +91 -0
- helm/benchmark/scenarios/image_generation/paint_skills_scenario.py +72 -0
- helm/benchmark/scenarios/image_generation/parti_prompts_scenario.py +94 -0
- helm/benchmark/scenarios/image_generation/radiology_scenario.py +42 -0
- helm/benchmark/scenarios/image_generation/relational_understanding_scenario.py +52 -0
- helm/benchmark/scenarios/image_generation/time_most_significant_historical_figures_scenario.py +124 -0
- helm/benchmark/scenarios/image_generation/winoground_scenario.py +62 -0
- helm/benchmark/scenarios/imdb_ptbr_scenario.py +60 -0
- helm/benchmark/scenarios/imdb_scenario.py +26 -3
- helm/benchmark/scenarios/infinite_bench_en_mc_scenario.py +111 -0
- helm/benchmark/scenarios/infinite_bench_en_qa_scenario.py +85 -0
- helm/benchmark/scenarios/infinite_bench_en_sum_scenario.py +98 -0
- helm/benchmark/scenarios/interactive_qa_mmlu_scenario.py +2 -2
- helm/benchmark/scenarios/koala_scenario.py +21 -1
- helm/benchmark/scenarios/kpi_edgar_scenario.py +172 -0
- helm/benchmark/scenarios/legal_contract_summarization_scenario.py +149 -0
- helm/benchmark/scenarios/legal_opinion_sentiment_classification_scenario.py +77 -0
- helm/benchmark/scenarios/legal_summarization_scenario.py +61 -1
- helm/benchmark/scenarios/legal_support_scenario.py +24 -1
- helm/benchmark/scenarios/legalbench_scenario.py +45 -3
- helm/benchmark/scenarios/lex_glue_scenario.py +23 -2
- helm/benchmark/scenarios/lextreme_scenario.py +22 -1
- helm/benchmark/scenarios/live_qa_scenario.py +94 -0
- helm/benchmark/scenarios/lm_entry_scenario.py +185 -0
- helm/benchmark/scenarios/lmkt_scenarios.py +288 -0
- helm/benchmark/scenarios/lsat_qa_scenario.py +15 -1
- helm/benchmark/scenarios/madinah_qa_scenario.py +73 -0
- helm/benchmark/scenarios/math_scenario.py +81 -22
- helm/benchmark/scenarios/mbzuai_human_translated_arabic_mmlu.py +68 -0
- helm/benchmark/scenarios/me_q_sum_scenario.py +10 -1
- helm/benchmark/scenarios/med_dialog_scenario.py +56 -22
- helm/benchmark/scenarios/med_mcqa_scenario.py +24 -1
- helm/benchmark/scenarios/med_paragraph_simplification_scenario.py +10 -1
- helm/benchmark/scenarios/med_qa_scenario.py +30 -1
- helm/benchmark/scenarios/medalign_scenario.py +117 -0
- helm/benchmark/scenarios/medalign_scenario_helper.py +326 -0
- helm/benchmark/scenarios/medbullets_scenario.py +167 -0
- helm/benchmark/scenarios/medcalc_bench_scenario.py +149 -0
- helm/benchmark/scenarios/medec_scenario.py +148 -0
- helm/benchmark/scenarios/medhallu_scenario.py +95 -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 +134 -0
- helm/benchmark/scenarios/medication_qa_scenario.py +96 -0
- helm/benchmark/scenarios/melt_ir_scenario.py +171 -0
- helm/benchmark/scenarios/melt_knowledge_scenario.py +246 -0
- helm/benchmark/scenarios/melt_lm_scenarios.py +252 -0
- helm/benchmark/scenarios/melt_scenarios.py +793 -0
- helm/benchmark/scenarios/melt_srn_scenario.py +342 -0
- helm/benchmark/scenarios/melt_synthetic_reasoning_scenario.py +222 -0
- helm/benchmark/scenarios/melt_translation_scenario.py +152 -0
- helm/benchmark/scenarios/mental_health_scenario.py +146 -0
- helm/benchmark/scenarios/mimic_bhc_scenario.py +127 -0
- helm/benchmark/scenarios/mimic_rrs_scenario.py +121 -0
- helm/benchmark/scenarios/mimiciv_billing_code_scenario.py +99 -0
- helm/benchmark/scenarios/mmlu_clinical_afr_scenario.py +74 -0
- helm/benchmark/scenarios/mmlu_pro_scenario.py +113 -0
- helm/benchmark/scenarios/mmlu_scenario.py +32 -1
- helm/benchmark/scenarios/mmmlu_scenario.py +85 -0
- helm/benchmark/scenarios/msmarco_scenario.py +31 -1
- helm/benchmark/scenarios/mtsamples_procedures_scenario.py +166 -0
- helm/benchmark/scenarios/mtsamples_replicate_scenario.py +164 -0
- helm/benchmark/scenarios/n2c2_ct_matching_scenario.py +297 -0
- helm/benchmark/scenarios/narrativeqa_scenario.py +20 -1
- helm/benchmark/scenarios/natural_qa_scenario.py +33 -1
- helm/benchmark/scenarios/newsqa_scenario.py +1 -1
- helm/benchmark/scenarios/oab_exams_scenario.py +57 -0
- helm/benchmark/scenarios/omni_math_scenario.py +71 -0
- helm/benchmark/scenarios/open_assistant_scenario.py +33 -2
- helm/benchmark/scenarios/openai_mrcr_scenario.py +94 -0
- helm/benchmark/scenarios/opinions_qa_scenario.py +1 -5
- helm/benchmark/scenarios/pubmed_qa_scenario.py +81 -43
- helm/benchmark/scenarios/quac_scenario.py +24 -1
- helm/benchmark/scenarios/race_based_med_scenario.py +175 -0
- helm/benchmark/scenarios/raft_scenario.py +33 -3
- helm/benchmark/scenarios/real_toxicity_prompts_scenario.py +14 -1
- helm/benchmark/scenarios/ruler_qa_scenario_helper.py +171 -0
- helm/benchmark/scenarios/ruler_qa_scenarios.py +128 -0
- helm/benchmark/scenarios/scenario.py +44 -1
- helm/benchmark/scenarios/seahelm_scenario.py +2295 -0
- helm/benchmark/scenarios/self_instruct_scenario.py +29 -1
- helm/benchmark/scenarios/shc_bmt_scenario.py +97 -0
- helm/benchmark/scenarios/shc_cdi_scenario.py +95 -0
- helm/benchmark/scenarios/shc_conf_scenario.py +99 -0
- helm/benchmark/scenarios/shc_ent_scenario.py +98 -0
- helm/benchmark/scenarios/shc_gip_scenario.py +94 -0
- helm/benchmark/scenarios/shc_privacy_scenario.py +100 -0
- helm/benchmark/scenarios/shc_proxy_scenario.py +98 -0
- helm/benchmark/scenarios/shc_ptbm_scenario.py +104 -0
- helm/benchmark/scenarios/shc_sei_scenario.py +94 -0
- helm/benchmark/scenarios/shc_sequoia_scenario.py +98 -0
- helm/benchmark/scenarios/simple_safety_tests_scenario.py +44 -0
- helm/benchmark/scenarios/simple_scenarios.py +122 -1
- helm/benchmark/scenarios/situation_prompts.yaml +49 -0
- helm/benchmark/scenarios/spider_scenario.py +109 -0
- helm/benchmark/scenarios/starr_patient_instructions_scenario.py +119 -0
- helm/benchmark/scenarios/summarization_scenario.py +48 -1
- helm/benchmark/scenarios/sumosum_scenario.py +157 -0
- helm/benchmark/scenarios/synthetic_efficiency_scenario.py +22 -1
- helm/benchmark/scenarios/synthetic_reasoning_natural_scenario.py +24 -1
- helm/benchmark/scenarios/synthetic_reasoning_scenario.py +11 -1
- helm/benchmark/scenarios/test_air_bench_scenario.py +27 -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_bigcodebench_scenario.py +26 -0
- helm/benchmark/scenarios/test_bluex_scenario.py +59 -0
- helm/benchmark/scenarios/test_commonsense_scenario.py +21 -0
- helm/benchmark/scenarios/test_czech_bank_qa_scenario.py +18 -0
- helm/benchmark/scenarios/test_enem_challenge_scenario.py +53 -0
- helm/benchmark/scenarios/test_ewok_scenario.py +29 -0
- helm/benchmark/scenarios/test_exams_multilingual_scenario.py +29 -0
- helm/benchmark/scenarios/test_financebench_scenario.py +26 -0
- helm/benchmark/scenarios/test_gold_commodity_news_scenario.py +18 -0
- helm/benchmark/scenarios/test_gpqa_scenario.py +44 -0
- helm/benchmark/scenarios/test_gsm_scenario.py +31 -0
- helm/benchmark/scenarios/test_healtha_br_scenario.py +57 -0
- helm/benchmark/scenarios/test_ifeval_scenario.py +36 -0
- helm/benchmark/scenarios/test_imdb_ptbr_scenario.py +27 -0
- helm/benchmark/scenarios/test_infinite_bench_en_qa_scenario.py +18 -0
- helm/benchmark/scenarios/test_infinite_bench_en_sum_scenario.py +31 -0
- helm/benchmark/scenarios/test_legalbench_scenario.py +30 -0
- helm/benchmark/scenarios/test_math_scenario.py +4 -3
- helm/benchmark/scenarios/test_med_qa_scenario.py +30 -0
- helm/benchmark/scenarios/test_mmlu_clinical_afr_scenario.py +21 -0
- helm/benchmark/scenarios/test_mmlu_pro_scenario.py +53 -0
- helm/benchmark/scenarios/test_mmlu_scenario.py +33 -0
- helm/benchmark/scenarios/test_narrativeqa_scenario.py +73 -0
- helm/benchmark/scenarios/test_oab_exams_scenario.py +51 -0
- helm/benchmark/scenarios/test_omni_math_scenario.py +27 -0
- helm/benchmark/scenarios/test_scenario.py +6 -3
- helm/benchmark/scenarios/test_simple_scenarios.py +50 -0
- helm/benchmark/scenarios/test_tweetsentbr_scenario.py +24 -0
- helm/benchmark/scenarios/test_wildbench_scenario.py +15 -0
- helm/benchmark/scenarios/test_winogrande_afr_scenario.py +19 -0
- helm/benchmark/scenarios/thai_exam_scenario.py +239 -0
- helm/benchmark/scenarios/the_pile_scenario.py +13 -1
- helm/benchmark/scenarios/truthful_qa_scenario.py +26 -2
- helm/benchmark/scenarios/tweetsentbr_scenario.py +66 -0
- helm/benchmark/scenarios/twitter_aae_scenario.py +20 -1
- helm/benchmark/scenarios/unitxt_scenario.py +62 -0
- helm/benchmark/scenarios/verifiability_judgment_scenario.py +4 -2
- helm/benchmark/scenarios/vicuna_scenario.py +22 -2
- helm/benchmark/scenarios/vision_language/a_okvqa_scenario.py +83 -0
- helm/benchmark/scenarios/vision_language/bingo_scenario.py +103 -0
- helm/benchmark/scenarios/vision_language/blink_scenario.py +140 -0
- helm/benchmark/scenarios/vision_language/crossmodal_3600_scenario.py +135 -0
- helm/benchmark/scenarios/vision_language/exams_v_scenario.py +104 -0
- helm/benchmark/scenarios/vision_language/fair_face_scenario.py +136 -0
- helm/benchmark/scenarios/vision_language/flickr30k_scenario.py +74 -0
- helm/benchmark/scenarios/vision_language/gqa_scenario.py +91 -0
- helm/benchmark/scenarios/vision_language/hateful_memes_scenario.py +94 -0
- helm/benchmark/scenarios/vision_language/heim_human_eval_scenario.py +113 -0
- helm/benchmark/scenarios/vision_language/image2struct/__init__.py +0 -0
- helm/benchmark/scenarios/vision_language/image2struct/chart2csv_scenario.py +55 -0
- helm/benchmark/scenarios/vision_language/image2struct/image2struct_scenario.py +225 -0
- helm/benchmark/scenarios/vision_language/image2struct/latex_scenario.py +21 -0
- helm/benchmark/scenarios/vision_language/image2struct/musicsheet_scenario.py +16 -0
- helm/benchmark/scenarios/vision_language/image2struct/utils_latex.py +339 -0
- helm/benchmark/scenarios/vision_language/image2struct/webpage/__init__.py +0 -0
- helm/benchmark/scenarios/vision_language/image2struct/webpage/driver.py +84 -0
- helm/benchmark/scenarios/vision_language/image2struct/webpage/jekyll_server.py +182 -0
- helm/benchmark/scenarios/vision_language/image2struct/webpage/utils.py +31 -0
- helm/benchmark/scenarios/vision_language/image2struct/webpage_scenario.py +256 -0
- helm/benchmark/scenarios/vision_language/math_vista_scenario.py +117 -0
- helm/benchmark/scenarios/vision_language/mementos_scenario.py +124 -0
- helm/benchmark/scenarios/vision_language/mm_safety_bench_scenario.py +103 -0
- helm/benchmark/scenarios/vision_language/mm_star_scenario.py +95 -0
- helm/benchmark/scenarios/vision_language/mme_scenario.py +148 -0
- helm/benchmark/scenarios/vision_language/mmmu_scenario.py +187 -0
- helm/benchmark/scenarios/vision_language/mscoco_captioning_scenario.py +92 -0
- helm/benchmark/scenarios/vision_language/mscoco_categorization_scenario.py +117 -0
- helm/benchmark/scenarios/vision_language/msr_vtt_scenario.py +75 -0
- helm/benchmark/scenarios/vision_language/multipanelvqa_scenario.py +169 -0
- helm/benchmark/scenarios/vision_language/originality_scenario.py +35 -0
- helm/benchmark/scenarios/vision_language/pairs_scenario.py +247 -0
- helm/benchmark/scenarios/vision_language/pope_scenario.py +105 -0
- helm/benchmark/scenarios/vision_language/real_world_qa_scenario.py +57 -0
- helm/benchmark/scenarios/vision_language/seed_bench_scenario.py +131 -0
- helm/benchmark/scenarios/vision_language/unicorn_scenario.py +108 -0
- helm/benchmark/scenarios/vision_language/vibe_eval_scenario.py +98 -0
- helm/benchmark/scenarios/vision_language/viz_wiz_scenario.py +4 -5
- helm/benchmark/scenarios/vision_language/vqa_rad_scenario.py +88 -0
- helm/benchmark/scenarios/vision_language/vqa_scenario.py +8 -4
- helm/benchmark/scenarios/wikifact_scenario.py +31 -1
- helm/benchmark/scenarios/wikitext_103_scenario.py +1 -1
- helm/benchmark/scenarios/wildbench_scenario.py +101 -0
- helm/benchmark/scenarios/winogrande_afr_scenario.py +78 -0
- helm/benchmark/scenarios/wmt_14_scenario.py +33 -2
- helm/benchmark/scenarios/xstest_scenario.py +35 -0
- helm/benchmark/server.py +32 -2
- helm/benchmark/slurm_jobs.py +1 -2
- helm/benchmark/slurm_runner.py +78 -50
- helm/benchmark/static/schema_air_bench.yaml +3149 -0
- helm/benchmark/static/schema_arabic.yaml +271 -0
- helm/benchmark/static/schema_audio.yaml +763 -0
- helm/benchmark/static/schema_autobencher.yaml +150 -0
- helm/benchmark/static/schema_call_center.yaml +269 -0
- helm/benchmark/static/schema_capabilities.yaml +254 -0
- helm/benchmark/static/schema_classic.yaml +259 -1140
- helm/benchmark/static/schema_cleva.yaml +768 -0
- helm/benchmark/static/schema_czech_bank.yaml +148 -0
- helm/benchmark/static/schema_decodingtrust.yaml +444 -0
- helm/benchmark/static/schema_enem_challenge.yaml +146 -0
- helm/benchmark/static/schema_enterprise.yaml +319 -0
- helm/benchmark/static/schema_ewok.yaml +367 -0
- helm/benchmark/static/schema_finance.yaml +191 -0
- helm/benchmark/static/schema_heim.yaml +1389 -0
- helm/benchmark/static/schema_image2struct.yaml +588 -0
- helm/benchmark/static/schema_instruction_following.yaml +161 -0
- helm/benchmark/static/schema_legal.yaml +566 -0
- helm/benchmark/static/schema_lite.yaml +3 -286
- helm/benchmark/static/schema_long_context.yaml +282 -0
- helm/benchmark/static/schema_medhelm.yaml +1176 -0
- helm/benchmark/static/schema_melt.yaml +1257 -0
- helm/benchmark/static/schema_mmlu.yaml +1449 -0
- helm/benchmark/static/schema_mmlu_winogrande_afr.yaml +1045 -0
- helm/benchmark/static/schema_safety.yaml +283 -0
- helm/benchmark/static/schema_seahelm.yaml +723 -0
- helm/benchmark/static/schema_slp.yaml +219 -0
- helm/benchmark/static/schema_slphelm.yaml +162 -0
- helm/benchmark/static/schema_social_audio.yaml +224 -0
- helm/benchmark/static/schema_sql.yaml +171 -0
- helm/benchmark/static/schema_thai.yaml +244 -0
- helm/benchmark/static/schema_torr.yaml +474 -0
- helm/benchmark/static/schema_tweetsentbr.yaml +146 -0
- helm/benchmark/static/schema_unitxt.yaml +370 -0
- helm/benchmark/static/schema_vhelm.yaml +933 -0
- helm/benchmark/static/schema_vhelm_lite.yaml +109 -0
- helm/benchmark/static/schema_video.yaml +219 -0
- helm/benchmark/static_build/assets/air-overview-DpBbyagA.png +0 -0
- helm/benchmark/static_build/assets/audio-table-Dn5NMMeJ.png +0 -0
- helm/benchmark/static_build/assets/heim-logo-BJtQlEbV.png +0 -0
- helm/benchmark/static_build/assets/helm-safety-COfndXuS.png +0 -0
- helm/benchmark/static_build/assets/helmhero-D9TvmJsp.png +0 -0
- helm/benchmark/static_build/assets/index-oIeiQW2g.css +1 -0
- helm/benchmark/static_build/assets/index-qOFpOyHb.js +10 -0
- helm/benchmark/static_build/assets/medhelm-overview-CND0EIsy.png +0 -0
- helm/benchmark/static_build/assets/medhelm-v1-overview-Cu2tphBB.png +0 -0
- helm/benchmark/static_build/assets/overview-BwypNWnk.png +0 -0
- helm/benchmark/static_build/assets/process-flow-DWDJC733.png +0 -0
- helm/benchmark/static_build/assets/react-BteFIppM.js +85 -0
- helm/benchmark/static_build/assets/recharts-DxuQtTOs.js +97 -0
- helm/benchmark/static_build/assets/tremor-DR4fE7ko.js +10 -0
- helm/benchmark/static_build/assets/vhelm-aspects-NiDQofvP.png +0 -0
- helm/benchmark/static_build/assets/vhelm-framework-NxJE4fdA.png +0 -0
- helm/benchmark/static_build/assets/vhelm-model-ypCL5Yvq.png +0 -0
- helm/benchmark/static_build/config.js +4 -0
- helm/benchmark/static_build/index.html +19 -0
- helm/benchmark/test_data_preprocessor.py +3 -3
- helm/benchmark/test_run_expander.py +1 -1
- helm/benchmark/window_services/default_window_service.py +3 -45
- helm/benchmark/window_services/encoder_decoder_window_service.py +4 -15
- helm/benchmark/window_services/ice_window_service.py +1 -35
- helm/benchmark/window_services/image_generation/__init__.py +0 -0
- helm/benchmark/window_services/image_generation/clip_window_service.py +13 -0
- helm/benchmark/window_services/image_generation/lexica_search_window_service.py +9 -0
- helm/benchmark/window_services/image_generation/openai_dalle_window_service.py +9 -0
- helm/benchmark/window_services/image_generation/test_clip_window_service.py +29 -0
- helm/benchmark/window_services/image_generation/test_openai_dalle_window_service.py +30 -0
- helm/benchmark/window_services/local_window_service.py +22 -5
- helm/benchmark/window_services/test_anthropic_window_service.py +5 -4
- helm/benchmark/window_services/test_bloom_window_service.py +5 -4
- helm/benchmark/window_services/test_flan_t5_window_service.py +2 -1
- helm/benchmark/window_services/test_gpt2_window_service.py +9 -4
- helm/benchmark/window_services/test_gpt4_window_service.py +10 -4
- helm/benchmark/window_services/test_gptj_window_service.py +11 -5
- helm/benchmark/window_services/test_gptneox_window_service.py +6 -5
- helm/benchmark/window_services/test_openai_window_service.py +18 -12
- helm/benchmark/window_services/test_opt_window_service.py +6 -5
- helm/benchmark/window_services/test_palmyra_window_service.py +5 -4
- helm/benchmark/window_services/test_t0pp_window_service.py +5 -4
- helm/benchmark/window_services/test_t511b_window_service.py +5 -4
- helm/benchmark/window_services/test_ul2_window_service.py +5 -4
- helm/benchmark/window_services/test_utils.py +6 -6
- helm/benchmark/window_services/test_yalm_window_service.py +5 -4
- helm/benchmark/window_services/tokenizer_service.py +7 -13
- helm/benchmark/window_services/window_service.py +42 -0
- helm/benchmark/window_services/window_service_factory.py +4 -1
- helm/benchmark/window_services/yalm_window_service.py +1 -28
- helm/clients/__init__.py +0 -0
- helm/{proxy/clients → clients}/ai21_client.py +78 -12
- helm/clients/aleph_alpha_client.py +114 -0
- helm/{proxy/clients → clients}/anthropic_client.py +304 -21
- helm/clients/audio_language/__init__.py +0 -0
- helm/clients/audio_language/diva_llama_client.py +122 -0
- 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/llama_omni_client.py +199 -0
- helm/clients/audio_language/qwen2_5_omni_client.py +210 -0
- helm/clients/audio_language/qwen2_audiolm_client.py +191 -0
- helm/clients/audio_language/qwen_audiolm_client.py +153 -0
- 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/audio_language/test.py +62 -0
- helm/{proxy/clients → clients}/auto_client.py +72 -31
- helm/clients/azure_openai_client.py +55 -0
- helm/clients/bedrock_client.py +381 -0
- helm/clients/bedrock_utils.py +105 -0
- helm/{proxy/clients → clients}/client.py +92 -17
- helm/clients/clip_score_client.py +49 -0
- helm/clients/clip_scorers/__init__.py +0 -0
- helm/clients/clip_scorers/base_clip_scorer.py +18 -0
- helm/clients/clip_scorers/clip_scorer.py +50 -0
- helm/clients/clip_scorers/multilingual_clip_scorer.py +50 -0
- helm/{proxy/clients → clients}/cohere_client.py +105 -14
- helm/clients/dspy_client.py +135 -0
- helm/clients/gcs_client.py +82 -0
- helm/{proxy/clients → clients}/google_client.py +8 -6
- helm/clients/google_translate_client.py +35 -0
- helm/clients/grok_client.py +36 -0
- helm/{proxy/clients → clients}/http_model_client.py +8 -8
- helm/{proxy/clients → clients}/huggingface_client.py +157 -86
- helm/clients/huggingface_pipeline_client.py +138 -0
- helm/clients/ibm_client.py +269 -0
- helm/clients/image_generation/__init__.py +0 -0
- helm/clients/image_generation/adobe_vision_client.py +80 -0
- helm/clients/image_generation/aleph_alpha_image_generation_client.py +100 -0
- helm/clients/image_generation/cogview2/__init__.py +0 -0
- helm/clients/image_generation/cogview2/coglm_strategy.py +96 -0
- helm/clients/image_generation/cogview2/coglm_utils.py +82 -0
- helm/clients/image_generation/cogview2/sr_pipeline/__init__.py +15 -0
- helm/clients/image_generation/cogview2/sr_pipeline/direct_sr.py +99 -0
- helm/clients/image_generation/cogview2/sr_pipeline/dsr_model.py +254 -0
- helm/clients/image_generation/cogview2/sr_pipeline/dsr_sampling.py +190 -0
- helm/clients/image_generation/cogview2/sr_pipeline/iterative_sr.py +144 -0
- helm/clients/image_generation/cogview2/sr_pipeline/itersr_model.py +269 -0
- helm/clients/image_generation/cogview2/sr_pipeline/itersr_sampling.py +120 -0
- helm/clients/image_generation/cogview2/sr_pipeline/sr_group.py +42 -0
- helm/clients/image_generation/cogview2_client.py +192 -0
- helm/clients/image_generation/dalle2_client.py +194 -0
- helm/clients/image_generation/dalle3_client.py +108 -0
- helm/clients/image_generation/dalle_mini/__init__.py +3 -0
- helm/clients/image_generation/dalle_mini/data.py +442 -0
- helm/clients/image_generation/dalle_mini/model/__init__.py +5 -0
- helm/clients/image_generation/dalle_mini/model/configuration.py +175 -0
- helm/clients/image_generation/dalle_mini/model/modeling.py +1834 -0
- helm/clients/image_generation/dalle_mini/model/partitions.py +84 -0
- helm/clients/image_generation/dalle_mini/model/processor.py +63 -0
- helm/clients/image_generation/dalle_mini/model/text.py +251 -0
- helm/clients/image_generation/dalle_mini/model/tokenizer.py +9 -0
- helm/clients/image_generation/dalle_mini/model/utils.py +29 -0
- helm/clients/image_generation/dalle_mini/vqgan_jax/__init__.py +1 -0
- helm/clients/image_generation/dalle_mini/vqgan_jax/configuration_vqgan.py +40 -0
- helm/clients/image_generation/dalle_mini/vqgan_jax/convert_pt_model_to_jax.py +107 -0
- helm/clients/image_generation/dalle_mini/vqgan_jax/modeling_flax_vqgan.py +610 -0
- helm/clients/image_generation/dalle_mini_client.py +191 -0
- helm/clients/image_generation/deep_floyd_client.py +80 -0
- helm/clients/image_generation/huggingface_diffusers_client.py +250 -0
- helm/clients/image_generation/image_generation_client_utils.py +9 -0
- helm/clients/image_generation/lexica_client.py +88 -0
- helm/clients/image_generation/mindalle/__init__.py +0 -0
- helm/clients/image_generation/mindalle/models/__init__.py +216 -0
- helm/clients/image_generation/mindalle/models/stage1/__init__.py +0 -0
- helm/clients/image_generation/mindalle/models/stage1/layers.py +312 -0
- helm/clients/image_generation/mindalle/models/stage1/vqgan.py +103 -0
- helm/clients/image_generation/mindalle/models/stage2/__init__.py +0 -0
- helm/clients/image_generation/mindalle/models/stage2/layers.py +144 -0
- helm/clients/image_generation/mindalle/models/stage2/transformer.py +268 -0
- helm/clients/image_generation/mindalle/models/tokenizer.py +30 -0
- helm/clients/image_generation/mindalle/utils/__init__.py +3 -0
- helm/clients/image_generation/mindalle/utils/config.py +129 -0
- helm/clients/image_generation/mindalle/utils/sampling.py +149 -0
- helm/clients/image_generation/mindalle/utils/utils.py +89 -0
- helm/clients/image_generation/mindalle_client.py +116 -0
- helm/clients/image_generation/nudity_check_client.py +64 -0
- helm/clients/image_generation/together_image_generation_client.py +113 -0
- helm/{proxy/clients → clients}/lit_gpt_client.py +6 -6
- helm/{proxy/clients → clients}/megatron_client.py +7 -5
- helm/clients/mistral_client.py +180 -0
- helm/clients/moderation_api_client.py +111 -0
- helm/clients/nvidia_nim_client.py +32 -0
- helm/clients/open_lm_client.py +43 -0
- helm/clients/openai_client.py +604 -0
- helm/clients/openai_responses_client.py +200 -0
- helm/clients/openrouter_client.py +31 -0
- helm/{proxy/clients → clients}/palmyra_client.py +31 -14
- helm/{proxy/clients → clients}/perspective_api_client.py +18 -14
- helm/clients/reka_client.py +190 -0
- helm/clients/simple_client.py +64 -0
- helm/clients/stanfordhealthcare_azure_openai_client.py +58 -0
- helm/clients/stanfordhealthcare_claude_client.py +31 -0
- helm/clients/stanfordhealthcare_google_client.py +43 -0
- helm/clients/stanfordhealthcare_http_model_client.py +95 -0
- helm/clients/stanfordhealthcare_openai_client.py +62 -0
- helm/clients/stanfordhealthcare_shc_openai_client.py +42 -0
- helm/{proxy/clients → clients}/test_auto_client.py +13 -15
- helm/clients/test_client.py +98 -0
- helm/{proxy/clients → clients}/test_huggingface_client.py +31 -16
- helm/clients/test_openrouter_client.py +69 -0
- helm/clients/test_simple_client.py +19 -0
- helm/clients/test_together_client.py +184 -0
- helm/clients/together_client.py +599 -0
- helm/clients/upstage_client.py +23 -0
- helm/clients/vertexai_client.py +488 -0
- helm/clients/vision_language/__init__.py +0 -0
- helm/clients/vision_language/huggingface_vision2seq_client.py +148 -0
- helm/clients/vision_language/huggingface_vlm_client.py +114 -0
- helm/{proxy/clients → clients}/vision_language/idefics_client.py +61 -51
- helm/clients/vision_language/open_flamingo/__init__.py +2 -0
- helm/clients/vision_language/open_flamingo/src/__init__.py +0 -0
- helm/clients/vision_language/open_flamingo/src/factory.py +147 -0
- helm/clients/vision_language/open_flamingo/src/flamingo.py +337 -0
- helm/clients/vision_language/open_flamingo/src/flamingo_lm.py +155 -0
- helm/clients/vision_language/open_flamingo/src/helpers.py +267 -0
- helm/clients/vision_language/open_flamingo/src/utils.py +47 -0
- helm/clients/vision_language/open_flamingo_client.py +155 -0
- helm/clients/vision_language/paligemma_client.py +147 -0
- helm/clients/vision_language/palmyra_vision_client.py +101 -0
- helm/clients/vision_language/qwen2_vlm_client.py +189 -0
- helm/clients/vision_language/qwen_vlm_client.py +174 -0
- helm/clients/vllm_client.py +80 -0
- helm/clients/vllm_granite_thinking_client.py +56 -0
- helm/clients/writer_client.py +105 -0
- helm/clients/yi_client.py +28 -0
- helm/common/audio_utils.py +111 -0
- helm/common/cache.py +23 -33
- helm/common/cache_backend_config.py +47 -0
- helm/common/clip_score_request.py +41 -0
- helm/common/context.py +80 -0
- helm/common/credentials_utils.py +5 -5
- helm/common/critique_request.py +10 -2
- helm/common/file_caches/__init__.py +0 -0
- helm/common/file_caches/file_cache.py +16 -0
- helm/common/file_caches/local_file_cache.py +61 -0
- helm/common/file_caches/test_local_file_cache.py +25 -0
- helm/common/file_upload_request.py +27 -0
- helm/common/general.py +10 -3
- helm/common/hierarchical_logger.py +124 -12
- helm/common/image_generation_parameters.py +25 -0
- helm/common/images_utils.py +60 -5
- helm/common/key_value_store.py +41 -10
- helm/common/local_context.py +140 -0
- helm/common/media_object.py +14 -1
- helm/common/moderations_api_request.py +71 -0
- helm/common/mongo_key_value_store.py +8 -7
- helm/common/multimodal_request_utils.py +57 -0
- helm/common/nudity_check_request.py +29 -0
- helm/common/object_spec.py +23 -8
- helm/common/optional_dependencies.py +1 -1
- helm/common/reeval_parameters.py +12 -0
- helm/common/remote_context.py +61 -0
- helm/common/request.py +45 -19
- helm/common/response_format.py +18 -0
- helm/common/test_cache.py +1 -48
- helm/common/test_general.py +10 -0
- helm/common/test_logging.py +94 -0
- helm/common/test_media_object.py +1 -1
- helm/common/tokenization_request.py +1 -10
- helm/config/model_deployments.yaml +4713 -1005
- helm/config/model_metadata.yaml +4045 -255
- helm/config/tokenizer_configs.yaml +1091 -50
- helm/proxy/accounts.py +31 -4
- helm/proxy/cli.py +6 -4
- helm/proxy/critique/mechanical_turk_critique_importer.py +3 -0
- helm/proxy/critique/mechanical_turk_utils.py +1 -1
- helm/proxy/critique/model_critique_client.py +40 -10
- helm/proxy/example_queries.py +33 -28
- helm/proxy/retry.py +5 -0
- helm/proxy/server.py +82 -18
- helm/proxy/services/remote_service.py +32 -7
- helm/proxy/services/server_service.py +71 -69
- helm/proxy/services/service.py +30 -6
- helm/proxy/services/test_remote_service.py +6 -5
- helm/proxy/services/test_service.py +1 -13
- helm/proxy/static/help.html +99 -0
- helm/proxy/static/index.css +61 -0
- helm/proxy/static/index.html +40 -0
- helm/proxy/static/index.js +462 -0
- helm/proxy/test_accounts.py +32 -0
- helm/proxy/test_retry.py +1 -1
- helm/proxy/token_counters/auto_token_counter.py +37 -37
- helm/proxy/token_counters/test_auto_token_counter.py +164 -0
- helm/proxy/token_counters/token_counter.py +3 -5
- helm/tokenizers/__init__.py +0 -0
- helm/tokenizers/ai21_tokenizer.py +52 -0
- helm/{proxy/tokenizers → tokenizers}/aleph_alpha_tokenizer.py +1 -1
- helm/{proxy/tokenizers → tokenizers}/auto_tokenizer.py +9 -12
- helm/{proxy/tokenizers → tokenizers}/caching_tokenizer.py +2 -30
- helm/tokenizers/cohere_tokenizer.py +50 -0
- helm/tokenizers/grok_tokenizer.py +55 -0
- helm/{proxy/tokenizers → tokenizers}/http_model_tokenizer.py +4 -4
- helm/{proxy/tokenizers → tokenizers}/huggingface_tokenizer.py +44 -41
- helm/{proxy/tokenizers → tokenizers}/lit_gpt_tokenizer.py +1 -1
- helm/tokenizers/simple_tokenizer.py +33 -0
- helm/tokenizers/test_ai21_tokenizer.py +48 -0
- helm/{proxy/tokenizers → tokenizers}/test_anthropic_tokenizer.py +6 -2
- helm/tokenizers/test_cohere_tokenizer.py +39 -0
- helm/tokenizers/test_grok_tokenizer.py +33 -0
- helm/{proxy/tokenizers → tokenizers}/test_huggingface_tokenizer.py +9 -2
- helm/tokenizers/test_simple_tokenizer.py +33 -0
- helm/{proxy/tokenizers → tokenizers}/test_yalm_tokenizer.py +1 -1
- helm/{proxy/tokenizers → tokenizers}/tiktoken_tokenizer.py +1 -1
- helm/{proxy/tokenizers → tokenizers}/tokenizer.py +3 -1
- helm/{proxy/tokenizers → tokenizers}/vertexai_tokenizer.py +1 -1
- helm/{proxy/tokenizers → tokenizers}/yalm_tokenizer.py +8 -6
- helm/tokenizers/yalm_tokenizer_data/__init__.py +0 -0
- helm/{proxy/tokenizers → tokenizers}/yalm_tokenizer_data/test_yalm_tokenizer.py +1 -1
- helm/tokenizers/yalm_tokenizer_data/voc_100b.sp +0 -0
- helm/{proxy/tokenizers → tokenizers}/yalm_tokenizer_data/yalm_tokenizer.py +1 -1
- crfm_helm-0.4.0.dist-info/METADATA +0 -264
- crfm_helm-0.4.0.dist-info/RECORD +0 -397
- helm/benchmark/data_overlap/data_overlap_spec.py +0 -86
- helm/benchmark/data_overlap/export_scenario_text.py +0 -119
- helm/benchmark/data_overlap/light_scenario.py +0 -60
- helm/benchmark/metrics/numeracy_metrics.py +0 -72
- helm/benchmark/metrics/test_numeracy_metrics.py +0 -95
- helm/benchmark/run_specs.py +0 -2762
- helm/benchmark/scenarios/numeracy_scenario.py +0 -784
- helm/benchmark/static/benchmarking.css +0 -156
- helm/benchmark/static/benchmarking.js +0 -1705
- helm/benchmark/static/config.js +0 -3
- helm/benchmark/static/images/helm-logo.png +0 -0
- helm/benchmark/static/images/language-model-helm.png +0 -0
- helm/benchmark/static/images/organizations/ai21.png +0 -0
- helm/benchmark/static/images/organizations/anthropic.png +0 -0
- helm/benchmark/static/images/organizations/bigscience.png +0 -0
- helm/benchmark/static/images/organizations/cohere.png +0 -0
- helm/benchmark/static/images/organizations/eleutherai.png +0 -0
- helm/benchmark/static/images/organizations/google.png +0 -0
- helm/benchmark/static/images/organizations/meta.png +0 -0
- helm/benchmark/static/images/organizations/microsoft.png +0 -0
- helm/benchmark/static/images/organizations/nvidia.png +0 -0
- helm/benchmark/static/images/organizations/openai.png +0 -0
- helm/benchmark/static/images/organizations/together.png +0 -0
- helm/benchmark/static/images/organizations/tsinghua-keg.png +0 -0
- helm/benchmark/static/images/organizations/yandex.png +0 -0
- helm/benchmark/static/images/scenarios-by-metrics.png +0 -0
- helm/benchmark/static/images/taxonomy-scenarios.png +0 -0
- helm/benchmark/static/index.html +0 -68
- helm/benchmark/static/json-urls.js +0 -69
- helm/benchmark/static/plot-captions.js +0 -27
- helm/benchmark/static/utils.js +0 -285
- helm/benchmark/test_model_deployment_definition.py +0 -92
- helm/benchmark/test_model_properties.py +0 -1570
- helm/benchmark/vlm_run_specs.py +0 -97
- helm/benchmark/window_services/ai21_window_service.py +0 -258
- helm/benchmark/window_services/cohere_window_service.py +0 -163
- helm/benchmark/window_services/flan_t5_window_service.py +0 -29
- helm/benchmark/window_services/gpt2_window_service.py +0 -32
- helm/benchmark/window_services/huggingface_window_service.py +0 -60
- helm/benchmark/window_services/t0pp_window_service.py +0 -35
- helm/benchmark/window_services/t511b_window_service.py +0 -30
- helm/benchmark/window_services/test_ai21_window_service.py +0 -163
- helm/benchmark/window_services/test_cohere_window_service.py +0 -74
- helm/benchmark/window_services/test_cohere_window_service_utils.py +0 -8328
- helm/benchmark/window_services/test_ice_window_service.py +0 -326
- helm/benchmark/window_services/test_mt_nlg_window_service.py +0 -48
- helm/benchmark/window_services/ul2_window_service.py +0 -30
- helm/benchmark/window_services/wider_ai21_window_service.py +0 -24
- helm/common/cache_utils.py +0 -14
- helm/proxy/clients/aleph_alpha_client.py +0 -95
- helm/proxy/clients/goose_ai_client.py +0 -99
- helm/proxy/clients/microsoft_client.py +0 -180
- helm/proxy/clients/openai_client.py +0 -206
- helm/proxy/clients/simple_client.py +0 -60
- helm/proxy/clients/test_client.py +0 -49
- helm/proxy/clients/test_together_client.py +0 -97
- helm/proxy/clients/together_client.py +0 -334
- helm/proxy/clients/vertexai_client.py +0 -115
- helm/proxy/token_counters/ai21_token_counter.py +0 -20
- helm/proxy/token_counters/cohere_token_counter.py +0 -13
- helm/proxy/token_counters/free_token_counter.py +0 -12
- helm/proxy/token_counters/gooseai_token_counter.py +0 -24
- helm/proxy/token_counters/openai_token_counter.py +0 -22
- helm/proxy/token_counters/test_ai21_token_counter.py +0 -88
- helm/proxy/token_counters/test_openai_token_counter.py +0 -81
- helm/proxy/tokenizers/ai21_tokenizer.py +0 -60
- helm/proxy/tokenizers/anthropic_tokenizer.py +0 -52
- helm/proxy/tokenizers/cohere_tokenizer.py +0 -83
- helm/proxy/tokenizers/ice_tokenizer.py +0 -30
- helm/proxy/tokenizers/simple_tokenizer.py +0 -32
- helm/proxy/tokenizers/test_ice_tokenizer.py +0 -57
- {crfm_helm-0.4.0.dist-info → crfm_helm-0.5.10.dist-info}/entry_points.txt +0 -0
- {crfm_helm-0.4.0.dist-info → crfm_helm-0.5.10.dist-info/licenses}/LICENSE +0 -0
- {crfm_helm-0.4.0.dist-info → crfm_helm-0.5.10.dist-info}/top_level.txt +0 -0
- /helm/benchmark/{data_overlap → annotation}/__init__.py +0 -0
- /helm/{proxy/clients → benchmark/annotation/image2struct}/__init__.py +0 -0
- /helm/{proxy/clients/vision_language → benchmark/metrics/ifeval}/__init__.py +0 -0
- /helm/{proxy/tokenizers → benchmark/metrics/image_generation}/__init__.py +0 -0
- /helm/{proxy/tokenizers/yalm_tokenizer_data → benchmark/metrics/image_generation/detectors}/__init__.py +0 -0
- /helm/benchmark/{static/images/crfm-logo.png → static_build/assets/crfm-logo-Du4T1uWZ.png} +0 -0
- /helm/benchmark/{static/images/helm-logo-simple.png → static_build/assets/helm-logo-simple-DzOhNN41.png} +0 -0
- /helm/{proxy/clients → clients}/ai21_utils.py +0 -0
- /helm/{proxy/clients → clients}/cohere_utils.py +0 -0
- /helm/{proxy/clients → clients}/lit_gpt_generate.py +0 -0
- /helm/{proxy/clients → clients}/toxicity_classifier_client.py +0 -0
- /helm/{benchmark → proxy}/static/general.js +0 -0
- /helm/{benchmark → proxy}/static/info-icon.png +0 -0
|
@@ -0,0 +1,25 @@
|
|
|
1
|
+
import os
|
|
2
|
+
import shutil
|
|
3
|
+
import tempfile
|
|
4
|
+
import unittest
|
|
5
|
+
|
|
6
|
+
from helm.common.file_caches.local_file_cache import LocalFileCache
|
|
7
|
+
|
|
8
|
+
|
|
9
|
+
class TestLocalFileCache(unittest.TestCase):
|
|
10
|
+
def setup_method(self, _):
|
|
11
|
+
self.path: str = tempfile.mkdtemp()
|
|
12
|
+
|
|
13
|
+
def teardown_method(self, _):
|
|
14
|
+
shutil.rmtree(self.path)
|
|
15
|
+
|
|
16
|
+
def test_get(self):
|
|
17
|
+
cache = LocalFileCache(self.path, file_extension="txt")
|
|
18
|
+
file_path1: str = cache.store(lambda: "hello.".encode())
|
|
19
|
+
|
|
20
|
+
# Verify the contents of the file
|
|
21
|
+
with open(file_path1, "r") as f:
|
|
22
|
+
assert f.read() == "hello."
|
|
23
|
+
|
|
24
|
+
cache.store(lambda: "bye.".encode())
|
|
25
|
+
assert len(os.listdir(self.path)) == 2
|
|
@@ -0,0 +1,27 @@
|
|
|
1
|
+
from dataclasses import dataclass
|
|
2
|
+
from typing import Optional
|
|
3
|
+
|
|
4
|
+
|
|
5
|
+
@dataclass(frozen=True)
|
|
6
|
+
class FileUploadRequest:
|
|
7
|
+
"""Uploads a file at `path`."""
|
|
8
|
+
|
|
9
|
+
# Path of the file to upload
|
|
10
|
+
path: str
|
|
11
|
+
|
|
12
|
+
|
|
13
|
+
@dataclass(frozen=True)
|
|
14
|
+
class FileUploadResult:
|
|
15
|
+
"""Result after sending a `FileUploadRequest`."""
|
|
16
|
+
|
|
17
|
+
# Whether the request was successful
|
|
18
|
+
success: bool
|
|
19
|
+
|
|
20
|
+
# Whether the request was cached
|
|
21
|
+
cached: bool
|
|
22
|
+
|
|
23
|
+
# URL of the uploaded file
|
|
24
|
+
url: str
|
|
25
|
+
|
|
26
|
+
# If `success` is false, what was the error?
|
|
27
|
+
error: Optional[str] = None
|
helm/common/general.py
CHANGED
|
@@ -42,6 +42,13 @@ def ensure_directory_exists(path: str):
|
|
|
42
42
|
os.makedirs(path, exist_ok=True)
|
|
43
43
|
|
|
44
44
|
|
|
45
|
+
def check_file_exists(path: str, msg: Optional[str] = None):
|
|
46
|
+
"""Checks that `path` exists, raises FileNotFoundError if it doesn't."""
|
|
47
|
+
if not os.path.exists(path):
|
|
48
|
+
error_msg = msg if msg else f"Required file not found: {path}"
|
|
49
|
+
raise FileNotFoundError(error_msg)
|
|
50
|
+
|
|
51
|
+
|
|
45
52
|
def parse_hocon(text: str):
|
|
46
53
|
"""Parse `text` (in HOCON format) into a dict-like object."""
|
|
47
54
|
return pyhocon.ConfigFactory.parse_string(text)
|
|
@@ -63,7 +70,7 @@ def shell(args: List[str]):
|
|
|
63
70
|
hlog(f"Executing: {cmd}")
|
|
64
71
|
exit_code = subprocess.call(args)
|
|
65
72
|
if exit_code != 0:
|
|
66
|
-
|
|
73
|
+
raise Exception(f"Failed with exit code {exit_code}: {cmd}")
|
|
67
74
|
|
|
68
75
|
|
|
69
76
|
@htrack(None)
|
|
@@ -156,7 +163,7 @@ def format_split(split: str) -> str:
|
|
|
156
163
|
|
|
157
164
|
|
|
158
165
|
def asdict_without_nones(obj: Any) -> Dict[str, Any]:
|
|
159
|
-
if not is_dataclass(obj):
|
|
166
|
+
if not is_dataclass(obj) or isinstance(obj, type):
|
|
160
167
|
raise ValueError(f"Expected dataclass, got '{obj}'")
|
|
161
168
|
return asdict(obj, dict_factory=lambda x: {k: v for (k, v) in x if v is not None})
|
|
162
169
|
|
|
@@ -178,7 +185,7 @@ def binarize_dict(d: Dict[str, int]) -> Dict[str, int]:
|
|
|
178
185
|
|
|
179
186
|
def serialize(obj: Any) -> List[str]:
|
|
180
187
|
"""Takes in a dataclass and outputs all of its fields and values in a list."""
|
|
181
|
-
if not is_dataclass(obj):
|
|
188
|
+
if not is_dataclass(obj) or isinstance(obj, type):
|
|
182
189
|
raise ValueError(f"Expected dataclass, got '{obj}'")
|
|
183
190
|
return [f"{key}: {json.dumps(value)}" for key, value in asdict(obj).items()]
|
|
184
191
|
|
|
@@ -1,6 +1,11 @@
|
|
|
1
|
+
import logging
|
|
2
|
+
import logging.config
|
|
3
|
+
import yaml
|
|
4
|
+
import os
|
|
1
5
|
import sys
|
|
2
6
|
import time
|
|
3
7
|
from typing import Any, Callable, List, Optional
|
|
8
|
+
from colorlog import ColoredFormatter
|
|
4
9
|
|
|
5
10
|
|
|
6
11
|
class HierarchicalLogger(object):
|
|
@@ -20,24 +25,53 @@ class HierarchicalLogger(object):
|
|
|
20
25
|
} [0s]
|
|
21
26
|
"""
|
|
22
27
|
|
|
28
|
+
# Far too much effort to unwind every call to hlog to go via logging,
|
|
29
|
+
# And is a terrible idea to inspect the stack every time hlog is called
|
|
30
|
+
# to figure out the caller,
|
|
31
|
+
# So just log everything under "helm".
|
|
32
|
+
logger = logging.getLogger("helm")
|
|
33
|
+
|
|
23
34
|
def __init__(self) -> None:
|
|
24
35
|
self.start_times: List[float] = []
|
|
25
36
|
|
|
26
37
|
def indent(self) -> str:
|
|
27
38
|
return " " * len(self.start_times)
|
|
28
39
|
|
|
29
|
-
def track_begin(self, x: Any) -> None:
|
|
30
|
-
|
|
40
|
+
def track_begin(self, x: Any, **kwargs) -> None:
|
|
41
|
+
kwargs["stacklevel"] = kwargs.get("stacklevel", 1) + 1
|
|
42
|
+
self.logger.info(self.indent() + str(x) + " {", **kwargs)
|
|
31
43
|
sys.stdout.flush()
|
|
32
44
|
self.start_times.append(time.time())
|
|
33
45
|
|
|
34
|
-
def track_end(self) -> None:
|
|
46
|
+
def track_end(self, **kwargs) -> None:
|
|
47
|
+
kwargs["stacklevel"] = kwargs.get("stacklevel", 1) + 1
|
|
35
48
|
t = time.time() - self.start_times.pop()
|
|
36
|
-
|
|
49
|
+
self.logger.info(self.indent() + "} [%s]" % (format_time(t)), **kwargs)
|
|
50
|
+
sys.stdout.flush()
|
|
51
|
+
|
|
52
|
+
def log(self, x: Any, **kwargs) -> None:
|
|
53
|
+
kwargs["stacklevel"] = kwargs.get("stacklevel", 1) + 1
|
|
54
|
+
self.logger.info(self.indent() + str(x), **kwargs)
|
|
37
55
|
sys.stdout.flush()
|
|
38
56
|
|
|
39
|
-
def
|
|
40
|
-
|
|
57
|
+
def debug(self, x: Any, **kwargs) -> None:
|
|
58
|
+
kwargs["stacklevel"] = kwargs.get("stacklevel", 1) + 1
|
|
59
|
+
self.logger.debug(self.indent() + str(x), **kwargs)
|
|
60
|
+
sys.stdout.flush()
|
|
61
|
+
|
|
62
|
+
def warn(self, x: Any, **kwargs) -> None:
|
|
63
|
+
kwargs["stacklevel"] = kwargs.get("stacklevel", 1) + 1
|
|
64
|
+
self.logger.warning(self.indent() + str(x), **kwargs)
|
|
65
|
+
sys.stdout.flush()
|
|
66
|
+
|
|
67
|
+
def error(self, x: Any, **kwargs) -> None:
|
|
68
|
+
kwargs["stacklevel"] = kwargs.get("stacklevel", 1) + 1
|
|
69
|
+
self.logger.error(self.indent() + str(x), **kwargs)
|
|
70
|
+
sys.stdout.flush()
|
|
71
|
+
|
|
72
|
+
def exception(self, x: Any, **kwargs) -> None:
|
|
73
|
+
kwargs["stacklevel"] = kwargs.get("stacklevel", 1) + 1
|
|
74
|
+
self.logger.exception(self.indent() + str(x), **kwargs)
|
|
41
75
|
sys.stdout.flush()
|
|
42
76
|
|
|
43
77
|
|
|
@@ -57,19 +91,41 @@ singleton = HierarchicalLogger()
|
|
|
57
91
|
# Exposed public methods
|
|
58
92
|
|
|
59
93
|
|
|
60
|
-
def
|
|
61
|
-
|
|
94
|
+
def hdebug(x: Any, **kwargs) -> None:
|
|
95
|
+
kwargs["stacklevel"] = kwargs.get("stacklevel", 1) + 1
|
|
96
|
+
singleton.debug(x, **kwargs)
|
|
97
|
+
|
|
98
|
+
|
|
99
|
+
def hlog(x: Any, **kwargs) -> None:
|
|
100
|
+
kwargs["stacklevel"] = kwargs.get("stacklevel", 1) + 1
|
|
101
|
+
singleton.log(x, **kwargs)
|
|
102
|
+
|
|
103
|
+
|
|
104
|
+
def hwarn(x: Any, **kwargs) -> None:
|
|
105
|
+
kwargs["stacklevel"] = kwargs.get("stacklevel", 1) + 1
|
|
106
|
+
singleton.warn(x, **kwargs)
|
|
107
|
+
|
|
108
|
+
|
|
109
|
+
def herror(x: Any, **kwargs) -> None:
|
|
110
|
+
kwargs["stacklevel"] = kwargs.get("stacklevel", 1) + 1
|
|
111
|
+
singleton.error(x, **kwargs)
|
|
112
|
+
|
|
113
|
+
|
|
114
|
+
def hexception(x: Any, **kwargs) -> None:
|
|
115
|
+
kwargs["stacklevel"] = kwargs.get("stacklevel", 1) + 1
|
|
116
|
+
singleton.exception(x, **kwargs)
|
|
62
117
|
|
|
63
118
|
|
|
64
119
|
class htrack_block:
|
|
65
|
-
def __init__(self, x: Any) -> None:
|
|
120
|
+
def __init__(self, x: Any, stacklevel=1) -> None:
|
|
121
|
+
self._stacklevel = stacklevel + 1
|
|
66
122
|
self.x = x
|
|
67
123
|
|
|
68
124
|
def __enter__(self) -> None:
|
|
69
|
-
singleton.track_begin(self.x)
|
|
125
|
+
singleton.track_begin(self.x, stacklevel=self._stacklevel)
|
|
70
126
|
|
|
71
127
|
def __exit__(self, tpe: Any, value: Any, callback: Any) -> None:
|
|
72
|
-
singleton.track_end()
|
|
128
|
+
singleton.track_end(stacklevel=self._stacklevel)
|
|
73
129
|
|
|
74
130
|
|
|
75
131
|
class htrack:
|
|
@@ -100,7 +156,63 @@ class htrack:
|
|
|
100
156
|
description = description.replace("$" + k, str(v))
|
|
101
157
|
else:
|
|
102
158
|
description = ""
|
|
103
|
-
with htrack_block(parent + fn.__name__ + description):
|
|
159
|
+
with htrack_block(parent + fn.__name__ + description, stacklevel=2):
|
|
104
160
|
return fn(*args, **kwargs)
|
|
105
161
|
|
|
106
162
|
return wrapper
|
|
163
|
+
|
|
164
|
+
|
|
165
|
+
def setup_default_logging(config_path: Optional[str] = None):
|
|
166
|
+
"""
|
|
167
|
+
Setup Python logging for HELM
|
|
168
|
+
|
|
169
|
+
Priority:
|
|
170
|
+
1. External config file (YAML or JSON).
|
|
171
|
+
2. ENV var LOG_LEVEL.
|
|
172
|
+
3. a default logger to STDOUT
|
|
173
|
+
"""
|
|
174
|
+
logger = logging.getLogger("helm")
|
|
175
|
+
logger.propagate = False
|
|
176
|
+
|
|
177
|
+
if config_path and os.path.exists(config_path):
|
|
178
|
+
with open(config_path, "r") as f:
|
|
179
|
+
config = yaml.safe_load(f)
|
|
180
|
+
logging.config.dictConfig(config)
|
|
181
|
+
hdebug("setup custom HELM logging")
|
|
182
|
+
return
|
|
183
|
+
|
|
184
|
+
log_level = (os.getenv("HELM_LOG_LEVEL") or os.getenv("LOG_LEVEL") or "INFO").upper()
|
|
185
|
+
try:
|
|
186
|
+
logger.setLevel(getattr(logging, log_level))
|
|
187
|
+
except AttributeError:
|
|
188
|
+
logger.setLevel(logging.INFO)
|
|
189
|
+
|
|
190
|
+
# Set formatter
|
|
191
|
+
formatter: Optional[logging.Formatter] = None
|
|
192
|
+
if sys.stdout.isatty():
|
|
193
|
+
try:
|
|
194
|
+
formatter = ColoredFormatter(
|
|
195
|
+
"%(bold_black)s%(asctime)s%(reset)s %(log_color)s%(levelname)-8s%(reset)s %(message)s",
|
|
196
|
+
datefmt="%Y-%m-%dT%H:%M:%S",
|
|
197
|
+
reset=True,
|
|
198
|
+
log_colors={
|
|
199
|
+
"DEBUG": "cyan",
|
|
200
|
+
"INFO": "green",
|
|
201
|
+
"WARNING": "yellow",
|
|
202
|
+
"ERROR": "red",
|
|
203
|
+
"CRITICAL": "red,bg_white",
|
|
204
|
+
},
|
|
205
|
+
style="%",
|
|
206
|
+
)
|
|
207
|
+
except ImportError:
|
|
208
|
+
pass
|
|
209
|
+
|
|
210
|
+
if formatter is None:
|
|
211
|
+
# fallback
|
|
212
|
+
formatter = logging.Formatter("%(asctime)s %(levelname)-8s %(message)s")
|
|
213
|
+
|
|
214
|
+
# Add default stdout handler
|
|
215
|
+
handler = logging.StreamHandler(sys.stdout)
|
|
216
|
+
handler.setFormatter(formatter)
|
|
217
|
+
logger.addHandler(handler)
|
|
218
|
+
hdebug("setup default HELM logging")
|
|
@@ -0,0 +1,25 @@
|
|
|
1
|
+
from dataclasses import dataclass
|
|
2
|
+
from typing import Optional
|
|
3
|
+
|
|
4
|
+
|
|
5
|
+
@dataclass(frozen=True)
|
|
6
|
+
class ImageGenerationParameters:
|
|
7
|
+
"""
|
|
8
|
+
Parameters for image generation.
|
|
9
|
+
"""
|
|
10
|
+
|
|
11
|
+
output_image_width: Optional[int] = None
|
|
12
|
+
"""Width of the generated image. The model will generate images with the model's
|
|
13
|
+
default dimensions when unspecified."""
|
|
14
|
+
|
|
15
|
+
output_image_height: Optional[int] = None
|
|
16
|
+
"""Height of the generated image. The model will generate images with the model's
|
|
17
|
+
default dimensions when unspecified."""
|
|
18
|
+
|
|
19
|
+
guidance_scale: Optional[float] = None
|
|
20
|
+
"""A non-negative number determining how much importance is given to the prompt
|
|
21
|
+
when generating images. Higher values will generate images that follow more
|
|
22
|
+
closely to the prompt. Currently only for diffusion models."""
|
|
23
|
+
|
|
24
|
+
diffusion_denoising_steps: Optional[int] = None
|
|
25
|
+
"""The number of denoising steps for diffusion models."""
|
helm/common/images_utils.py
CHANGED
|
@@ -1,10 +1,16 @@
|
|
|
1
|
+
from hashlib import md5
|
|
1
2
|
import base64
|
|
2
3
|
import io
|
|
4
|
+
import os
|
|
5
|
+
|
|
3
6
|
import requests
|
|
4
7
|
import shutil
|
|
5
|
-
from typing import Optional
|
|
8
|
+
from typing import List, Optional, Tuple
|
|
9
|
+
from urllib.request import urlopen
|
|
10
|
+
|
|
11
|
+
import numpy as np
|
|
6
12
|
|
|
7
|
-
from .general import is_url
|
|
13
|
+
from helm.common.general import is_url
|
|
8
14
|
from helm.common.optional_dependencies import handle_module_not_found_error
|
|
9
15
|
|
|
10
16
|
try:
|
|
@@ -19,12 +25,18 @@ def open_image(image_location: str) -> Image.Image:
|
|
|
19
25
|
"""
|
|
20
26
|
image: Image.Image
|
|
21
27
|
if is_url(image_location):
|
|
22
|
-
image = Image.open(requests.get(image_location, stream=True).raw)
|
|
28
|
+
image = Image.open(requests.get(image_location, stream=True).raw) # type: ignore
|
|
23
29
|
else:
|
|
24
30
|
image = Image.open(image_location)
|
|
25
31
|
return image.convert("RGB")
|
|
26
32
|
|
|
27
33
|
|
|
34
|
+
def get_dimensions(image_location: str) -> Tuple[int, int]:
|
|
35
|
+
"""Returns the dimensions of the image."""
|
|
36
|
+
image: Image.Image = open_image(image_location)
|
|
37
|
+
return image.size
|
|
38
|
+
|
|
39
|
+
|
|
28
40
|
def encode_base64(image_location: str, format="JPEG") -> str:
|
|
29
41
|
"""Returns the base64 representation of an image file."""
|
|
30
42
|
image_file = io.BytesIO()
|
|
@@ -33,7 +45,12 @@ def encode_base64(image_location: str, format="JPEG") -> str:
|
|
|
33
45
|
return base64.b64encode(image_file.getvalue()).decode("ascii")
|
|
34
46
|
|
|
35
47
|
|
|
36
|
-
def
|
|
48
|
+
def generate_hash(image: Image.Image) -> str:
|
|
49
|
+
"""Generates a hash for the image."""
|
|
50
|
+
return md5(image.tobytes()).hexdigest()
|
|
51
|
+
|
|
52
|
+
|
|
53
|
+
def copy_image(src: str, dest: str, width: Optional[int] = None, height: Optional[int] = None) -> None:
|
|
37
54
|
"""
|
|
38
55
|
Copies the image file from `src` path to `dest` path. If dimensions `width` and `height`
|
|
39
56
|
are specified, resizes the image before copying. `src` can be a URL.
|
|
@@ -41,7 +58,45 @@ def copy_image(src: str, dest: str, width: Optional[int] = None, height: Optiona
|
|
|
41
58
|
if (width is not None and height is not None) or is_url(src):
|
|
42
59
|
image = open_image(src)
|
|
43
60
|
if width is not None and height is not None:
|
|
44
|
-
image = image.resize((width, height), Image.
|
|
61
|
+
image = image.resize((width, height), Image.Resampling.LANCZOS)
|
|
45
62
|
image.save(dest)
|
|
46
63
|
else:
|
|
47
64
|
shutil.copy(src, dest)
|
|
65
|
+
|
|
66
|
+
|
|
67
|
+
def resize_image_to_max_file_size(src: str, dest: str, max_size_in_bytes: int, step=10):
|
|
68
|
+
# Open an image file
|
|
69
|
+
with Image.open(src) as img:
|
|
70
|
+
width, height = img.size
|
|
71
|
+
|
|
72
|
+
# Reduce dimensions iteratively until the file size is under the limit
|
|
73
|
+
while True:
|
|
74
|
+
# Save the image temporarily to check the file size
|
|
75
|
+
img.save(dest, quality=95) # Start with high quality
|
|
76
|
+
if os.path.getsize(dest) < max_size_in_bytes:
|
|
77
|
+
break
|
|
78
|
+
|
|
79
|
+
# Reduce dimensions
|
|
80
|
+
width -= step
|
|
81
|
+
height -= step
|
|
82
|
+
img = img.resize((width, height), Image.Resampling.LANCZOS)
|
|
83
|
+
|
|
84
|
+
|
|
85
|
+
def is_blacked_out_image(image_location: str) -> bool:
|
|
86
|
+
"""Returns True if the image is all black. False otherwise."""
|
|
87
|
+
try:
|
|
88
|
+
import cv2
|
|
89
|
+
except ModuleNotFoundError as e:
|
|
90
|
+
handle_module_not_found_error(e, ["heim"])
|
|
91
|
+
|
|
92
|
+
if is_url(image_location):
|
|
93
|
+
arr = np.asarray(bytearray(urlopen(image_location).read()), dtype=np.uint8)
|
|
94
|
+
image = cv2.imdecode(arr, -1)
|
|
95
|
+
else:
|
|
96
|
+
image = cv2.imread(image_location, 0)
|
|
97
|
+
return cv2.countNonZero(image) == 0
|
|
98
|
+
|
|
99
|
+
|
|
100
|
+
def filter_blacked_out_images(image_locations: List[str]) -> List[str]:
|
|
101
|
+
"""Returns a list of image locations that are not blacked out."""
|
|
102
|
+
return [image_location for image_location in image_locations if not is_blacked_out_image(image_location)]
|
helm/common/key_value_store.py
CHANGED
|
@@ -1,12 +1,12 @@
|
|
|
1
1
|
from abc import abstractmethod
|
|
2
2
|
import contextlib
|
|
3
3
|
import json
|
|
4
|
-
from typing import Dict, Generator, Iterable, Optional, Tuple
|
|
4
|
+
from typing import Dict, Generator, Iterable, Mapping, Optional, Tuple
|
|
5
5
|
|
|
6
6
|
from sqlitedict import SqliteDict
|
|
7
7
|
|
|
8
8
|
|
|
9
|
-
def request_to_key(request:
|
|
9
|
+
def request_to_key(request: Mapping) -> str:
|
|
10
10
|
"""Normalize a `request` into a `key` so that we can hash using it."""
|
|
11
11
|
return json.dumps(request, sort_keys=True)
|
|
12
12
|
|
|
@@ -15,11 +15,11 @@ class KeyValueStore(contextlib.AbstractContextManager):
|
|
|
15
15
|
"""Key value store that persists writes."""
|
|
16
16
|
|
|
17
17
|
@abstractmethod
|
|
18
|
-
def contains(self, key:
|
|
18
|
+
def contains(self, key: Mapping) -> bool:
|
|
19
19
|
pass
|
|
20
20
|
|
|
21
21
|
@abstractmethod
|
|
22
|
-
def get(self, key:
|
|
22
|
+
def get(self, key: Mapping) -> Optional[Dict]:
|
|
23
23
|
pass
|
|
24
24
|
|
|
25
25
|
@abstractmethod
|
|
@@ -27,7 +27,7 @@ class KeyValueStore(contextlib.AbstractContextManager):
|
|
|
27
27
|
pass
|
|
28
28
|
|
|
29
29
|
@abstractmethod
|
|
30
|
-
def put(self, key:
|
|
30
|
+
def put(self, key: Mapping, value: Dict) -> None:
|
|
31
31
|
pass
|
|
32
32
|
|
|
33
33
|
@abstractmethod
|
|
@@ -35,7 +35,7 @@ class KeyValueStore(contextlib.AbstractContextManager):
|
|
|
35
35
|
pass
|
|
36
36
|
|
|
37
37
|
@abstractmethod
|
|
38
|
-
def remove(self, key:
|
|
38
|
+
def remove(self, key: Mapping) -> None:
|
|
39
39
|
pass
|
|
40
40
|
|
|
41
41
|
|
|
@@ -53,10 +53,10 @@ class SqliteKeyValueStore(KeyValueStore):
|
|
|
53
53
|
def __exit__(self, exc_type, exc_value, traceback) -> None:
|
|
54
54
|
self._sqlite_dict.__exit__(exc_type, exc_value, traceback)
|
|
55
55
|
|
|
56
|
-
def contains(self, key:
|
|
56
|
+
def contains(self, key: Mapping) -> bool:
|
|
57
57
|
return request_to_key(key) in self._sqlite_dict
|
|
58
58
|
|
|
59
|
-
def get(self, key:
|
|
59
|
+
def get(self, key: Mapping) -> Optional[Dict]:
|
|
60
60
|
key_string = request_to_key(key)
|
|
61
61
|
result = self._sqlite_dict.get(key_string)
|
|
62
62
|
if result is not None:
|
|
@@ -68,7 +68,7 @@ class SqliteKeyValueStore(KeyValueStore):
|
|
|
68
68
|
for key, value in self._sqlite_dict.items():
|
|
69
69
|
yield (key, value)
|
|
70
70
|
|
|
71
|
-
def put(self, key:
|
|
71
|
+
def put(self, key: Mapping, value: Dict) -> None:
|
|
72
72
|
key_string = request_to_key(key)
|
|
73
73
|
self._sqlite_dict[key_string] = value
|
|
74
74
|
self._sqlite_dict.commit()
|
|
@@ -77,6 +77,37 @@ class SqliteKeyValueStore(KeyValueStore):
|
|
|
77
77
|
for key, value in pairs:
|
|
78
78
|
self.put(key, value)
|
|
79
79
|
|
|
80
|
-
def remove(self, key:
|
|
80
|
+
def remove(self, key: Mapping) -> None:
|
|
81
81
|
del self._sqlite_dict[key]
|
|
82
82
|
self._sqlite_dict.commit()
|
|
83
|
+
|
|
84
|
+
|
|
85
|
+
class BlackHoleKeyValueStore(KeyValueStore):
|
|
86
|
+
"""Key value store that discards all data."""
|
|
87
|
+
|
|
88
|
+
def __enter__(self) -> "BlackHoleKeyValueStore":
|
|
89
|
+
return self
|
|
90
|
+
|
|
91
|
+
def __exit__(self, exc_type, exc_value, traceback) -> None:
|
|
92
|
+
pass
|
|
93
|
+
|
|
94
|
+
def contains(self, key: Mapping) -> bool:
|
|
95
|
+
return False
|
|
96
|
+
|
|
97
|
+
def get(self, key: Mapping) -> Optional[Dict]:
|
|
98
|
+
return None
|
|
99
|
+
|
|
100
|
+
def get_all(self) -> Generator[Tuple[Dict, Dict], None, None]:
|
|
101
|
+
# Return an empty generator.
|
|
102
|
+
# See: https://stackoverflow.com/a/13243870
|
|
103
|
+
return
|
|
104
|
+
yield
|
|
105
|
+
|
|
106
|
+
def put(self, key: Mapping, value: Dict) -> None:
|
|
107
|
+
return None
|
|
108
|
+
|
|
109
|
+
def multi_put(self, pairs: Iterable[Tuple[Dict, Dict]]) -> None:
|
|
110
|
+
return None
|
|
111
|
+
|
|
112
|
+
def remove(self, key: Mapping) -> None:
|
|
113
|
+
return None
|
|
@@ -0,0 +1,140 @@
|
|
|
1
|
+
import dataclasses
|
|
2
|
+
import os
|
|
3
|
+
from typing import Optional
|
|
4
|
+
|
|
5
|
+
from helm.common.context import Context
|
|
6
|
+
from helm.common.cache import CacheConfig
|
|
7
|
+
from helm.common.cache_backend_config import CacheBackendConfig, BlackHoleCacheBackendConfig
|
|
8
|
+
from helm.common.critique_request import CritiqueRequest, CritiqueRequestResult
|
|
9
|
+
from helm.common.moderations_api_request import ModerationAPIRequest, ModerationAPIRequestResult
|
|
10
|
+
from helm.common.clip_score_request import CLIPScoreRequest, CLIPScoreResult
|
|
11
|
+
from helm.common.nudity_check_request import NudityCheckRequest, NudityCheckResult
|
|
12
|
+
from helm.common.file_upload_request import FileUploadRequest, FileUploadResult
|
|
13
|
+
from helm.common.general import ensure_directory_exists, parse_hocon, get_credentials
|
|
14
|
+
from helm.common.perspective_api_request import PerspectiveAPIRequest, PerspectiveAPIRequestResult
|
|
15
|
+
from helm.common.tokenization_request import (
|
|
16
|
+
TokenizationRequest,
|
|
17
|
+
TokenizationRequestResult,
|
|
18
|
+
DecodeRequest,
|
|
19
|
+
DecodeRequestResult,
|
|
20
|
+
)
|
|
21
|
+
from helm.common.request import Request, RequestResult
|
|
22
|
+
from helm.clients.auto_client import AutoClient
|
|
23
|
+
from helm.clients.moderation_api_client import ModerationAPIClient
|
|
24
|
+
from helm.clients.image_generation.nudity_check_client import NudityCheckClient
|
|
25
|
+
from helm.clients.gcs_client import GCSClient
|
|
26
|
+
from helm.clients.clip_score_client import CLIPScoreClient
|
|
27
|
+
from helm.clients.toxicity_classifier_client import ToxicityClassifierClient
|
|
28
|
+
from helm.proxy.example_queries import example_queries
|
|
29
|
+
from helm.benchmark.model_metadata_registry import ALL_MODELS_METADATA
|
|
30
|
+
from helm.proxy.query import Query, QueryResult
|
|
31
|
+
from helm.proxy.retry import retry_request
|
|
32
|
+
from helm.tokenizers.auto_tokenizer import AutoTokenizer
|
|
33
|
+
from helm.proxy.services.service import (
|
|
34
|
+
CACHE_DIR,
|
|
35
|
+
GeneralInfo,
|
|
36
|
+
VERSION,
|
|
37
|
+
expand_environments,
|
|
38
|
+
synthesize_request,
|
|
39
|
+
)
|
|
40
|
+
|
|
41
|
+
|
|
42
|
+
class LocalContext(Context):
|
|
43
|
+
"""
|
|
44
|
+
Main class that supports various functionality for the server.
|
|
45
|
+
"""
|
|
46
|
+
|
|
47
|
+
def __init__(
|
|
48
|
+
self,
|
|
49
|
+
base_path: str = "prod_env",
|
|
50
|
+
cache_backend_config: CacheBackendConfig = BlackHoleCacheBackendConfig(),
|
|
51
|
+
):
|
|
52
|
+
ensure_directory_exists(base_path)
|
|
53
|
+
client_file_storage_path = os.path.join(base_path, CACHE_DIR)
|
|
54
|
+
ensure_directory_exists(client_file_storage_path)
|
|
55
|
+
|
|
56
|
+
credentials = get_credentials(base_path)
|
|
57
|
+
|
|
58
|
+
self.cache_backend_config = cache_backend_config
|
|
59
|
+
self.client = AutoClient(credentials, client_file_storage_path, cache_backend_config)
|
|
60
|
+
self.tokenizer = AutoTokenizer(credentials, cache_backend_config)
|
|
61
|
+
|
|
62
|
+
# Lazily instantiate the following clients
|
|
63
|
+
self.moderation_api_client: Optional[ModerationAPIClient] = None
|
|
64
|
+
self.toxicity_classifier_client: Optional[ToxicityClassifierClient] = None
|
|
65
|
+
self.perspective_api_client: Optional[ToxicityClassifierClient] = None
|
|
66
|
+
self.nudity_check_client: Optional[NudityCheckClient] = None
|
|
67
|
+
self.clip_score_client: Optional[CLIPScoreClient] = None
|
|
68
|
+
self.gcs_client: Optional[GCSClient] = None
|
|
69
|
+
|
|
70
|
+
def get_general_info(self) -> GeneralInfo:
|
|
71
|
+
# Can't send release_dates in ModelMetadata bacause dates cannot be round-tripped to and from JSON easily.
|
|
72
|
+
# TODO(#2158): Either fix this or delete get_general_info.
|
|
73
|
+
all_models = [dataclasses.replace(model_metadata, release_date=None) for model_metadata in ALL_MODELS_METADATA]
|
|
74
|
+
return GeneralInfo(version=VERSION, example_queries=example_queries, all_models=all_models)
|
|
75
|
+
|
|
76
|
+
def expand_query(self, query: Query) -> QueryResult:
|
|
77
|
+
"""Turn the `query` into requests."""
|
|
78
|
+
prompt = query.prompt
|
|
79
|
+
settings = query.settings
|
|
80
|
+
environments = parse_hocon(query.environments)
|
|
81
|
+
requests = []
|
|
82
|
+
for environment in expand_environments(environments):
|
|
83
|
+
request = synthesize_request(prompt, settings, environment)
|
|
84
|
+
requests.append(request)
|
|
85
|
+
return QueryResult(requests=requests)
|
|
86
|
+
|
|
87
|
+
def make_request(self, request: Request) -> RequestResult:
|
|
88
|
+
"""Actually make a request to an API."""
|
|
89
|
+
return self.client.make_request(request)
|
|
90
|
+
|
|
91
|
+
def tokenize(self, request: TokenizationRequest) -> TokenizationRequestResult:
|
|
92
|
+
return self.tokenizer.tokenize(request)
|
|
93
|
+
|
|
94
|
+
def decode(self, request: DecodeRequest) -> DecodeRequestResult:
|
|
95
|
+
return self.tokenizer.decode(request)
|
|
96
|
+
|
|
97
|
+
def upload(self, request: FileUploadRequest) -> FileUploadResult:
|
|
98
|
+
if not self.gcs_client:
|
|
99
|
+
self.gcs_client = self.client.get_gcs_client()
|
|
100
|
+
|
|
101
|
+
assert self.gcs_client
|
|
102
|
+
return self.gcs_client.upload(request)
|
|
103
|
+
|
|
104
|
+
def check_nudity(self, request: NudityCheckRequest) -> NudityCheckResult:
|
|
105
|
+
if not self.nudity_check_client:
|
|
106
|
+
self.nudity_check_client = self.client.get_nudity_check_client()
|
|
107
|
+
|
|
108
|
+
assert self.nudity_check_client
|
|
109
|
+
return self.nudity_check_client.check_nudity(request)
|
|
110
|
+
|
|
111
|
+
def compute_clip_score(self, request: CLIPScoreRequest) -> CLIPScoreResult:
|
|
112
|
+
if not self.clip_score_client:
|
|
113
|
+
self.clip_score_client = self.client.get_clip_score_client()
|
|
114
|
+
|
|
115
|
+
assert self.clip_score_client
|
|
116
|
+
return self.clip_score_client.compute_score(request)
|
|
117
|
+
|
|
118
|
+
def get_toxicity_scores(self, request: PerspectiveAPIRequest) -> PerspectiveAPIRequestResult:
|
|
119
|
+
@retry_request
|
|
120
|
+
def get_toxicity_scores_with_retry(request: PerspectiveAPIRequest) -> PerspectiveAPIRequestResult:
|
|
121
|
+
if not self.toxicity_classifier_client:
|
|
122
|
+
self.toxicity_classifier_client = self.client.get_toxicity_classifier_client()
|
|
123
|
+
return self.toxicity_classifier_client.get_toxicity_scores(request)
|
|
124
|
+
|
|
125
|
+
return get_toxicity_scores_with_retry(request)
|
|
126
|
+
|
|
127
|
+
def get_moderation_results(self, request: ModerationAPIRequest) -> ModerationAPIRequestResult:
|
|
128
|
+
@retry_request
|
|
129
|
+
def get_moderation_results_with_retry(request: ModerationAPIRequest) -> ModerationAPIRequestResult:
|
|
130
|
+
if not self.moderation_api_client:
|
|
131
|
+
self.moderation_api_client = self.client.get_moderation_api_client()
|
|
132
|
+
return self.moderation_api_client.get_moderation_results(request)
|
|
133
|
+
|
|
134
|
+
return get_moderation_results_with_retry(request)
|
|
135
|
+
|
|
136
|
+
def make_critique_request(self, request: CritiqueRequest) -> CritiqueRequestResult:
|
|
137
|
+
return self.client.get_critique_client().make_critique_request(request)
|
|
138
|
+
|
|
139
|
+
def get_cache_config(self, shard_name: str) -> CacheConfig:
|
|
140
|
+
return self.cache_backend_config.get_cache_config(shard_name)
|
helm/common/media_object.py
CHANGED
|
@@ -2,9 +2,10 @@ import os
|
|
|
2
2
|
import urllib
|
|
3
3
|
from copy import deepcopy
|
|
4
4
|
from dataclasses import dataclass, field, replace
|
|
5
|
-
from typing import List, Optional
|
|
5
|
+
from typing import Any, Dict, List, Optional
|
|
6
6
|
|
|
7
7
|
|
|
8
|
+
IMAGE_TYPE = "image"
|
|
8
9
|
TEXT_TYPE = "text"
|
|
9
10
|
|
|
10
11
|
|
|
@@ -26,6 +27,10 @@ class MediaObject:
|
|
|
26
27
|
location: Optional[str] = None
|
|
27
28
|
"""When the media object is a file, specify the location of the media object, which can be a local path or URL."""
|
|
28
29
|
|
|
30
|
+
def to_dict(self) -> Dict[str, Any]:
|
|
31
|
+
"""Converts the media object to a dictionary."""
|
|
32
|
+
return {key: value for key, value in self.__dict__.items() if value is not None}
|
|
33
|
+
|
|
29
34
|
@property
|
|
30
35
|
def type(self) -> str:
|
|
31
36
|
"""The MIME type of the media object."""
|
|
@@ -113,6 +118,14 @@ class MultimediaObject:
|
|
|
113
118
|
"""
|
|
114
119
|
return MultimediaObject(media_objects=self.media_objects + other.media_objects)
|
|
115
120
|
|
|
121
|
+
@property
|
|
122
|
+
def size(self) -> int:
|
|
123
|
+
"""
|
|
124
|
+
Get the number of `MediaObject`s in this multimodal content.
|
|
125
|
+
:return: The number of `MediaObject`s .
|
|
126
|
+
"""
|
|
127
|
+
return len(self.media_objects)
|
|
128
|
+
|
|
116
129
|
@property
|
|
117
130
|
def text(self) -> str:
|
|
118
131
|
"""
|