crfm-helm 0.5.4__py3-none-any.whl → 0.5.6__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-0.5.6.dist-info/METADATA +427 -0
- crfm_helm-0.5.6.dist-info/RECORD +941 -0
- {crfm_helm-0.5.4.dist-info → crfm_helm-0.5.6.dist-info}/WHEEL +1 -1
- helm/benchmark/adaptation/adapter_spec.py +13 -1
- helm/benchmark/adaptation/adapters/adapter_factory.py +15 -1
- 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 +1 -1
- helm/benchmark/adaptation/adapters/in_context_learning_adapter.py +4 -4
- helm/benchmark/adaptation/adapters/language_modeling_adapter.py +1 -1
- helm/benchmark/adaptation/adapters/multimodal/generation_multimodal_adapter.py +4 -2
- helm/benchmark/adaptation/adapters/multimodal/in_context_learning_multimodal_adapter.py +1 -1
- helm/benchmark/adaptation/adapters/multimodal/multiple_choice_joint_multimodal_adapter.py +1 -1
- helm/benchmark/adaptation/adapters/multimodal/test_in_context_learning_multimodal_adapter.py +4 -2
- helm/benchmark/adaptation/adapters/multimodal/test_multimodal_prompt.py +1 -1
- helm/benchmark/adaptation/adapters/multiple_choice_calibrated_adapter.py +1 -1
- helm/benchmark/adaptation/adapters/multiple_choice_joint_adapter.py +2 -2
- 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 +4 -4
- helm/benchmark/adaptation/adapters/test_generation_adapter.py +3 -3
- helm/benchmark/adaptation/adapters/test_language_modeling_adapter.py +2 -2
- helm/benchmark/adaptation/adapters/test_multiple_choice_joint_adapter.py +2 -2
- helm/benchmark/adaptation/common_adapter_specs.py +69 -4
- helm/benchmark/adaptation/prompt.py +1 -1
- helm/benchmark/annotation/aci_bench_annotator.py +95 -0
- helm/benchmark/annotation/air_bench_annotator.py +21 -6
- helm/benchmark/annotation/annotator.py +5 -0
- helm/benchmark/annotation/annotator_factory.py +3 -20
- 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/chw_care_plan_annotator.py +93 -0
- helm/benchmark/annotation/czech_bank_qa_annotator.py +78 -0
- helm/benchmark/annotation/dischargeme_annotator.py +107 -0
- helm/benchmark/annotation/ehr_sql_annotator.py +87 -0
- helm/benchmark/annotation/helpdesk_call_summarization_annotator.py +131 -0
- helm/benchmark/annotation/image2struct/image_compiler_annotator.py +6 -1
- helm/benchmark/annotation/live_qa_annotator.py +1 -1
- helm/benchmark/annotation/med_dialog_annotator.py +99 -0
- helm/benchmark/annotation/medalign_annotator.py +100 -0
- helm/benchmark/annotation/medi_qa_annotator.py +98 -0
- helm/benchmark/annotation/medication_qa_annotator.py +87 -63
- helm/benchmark/annotation/mental_health_annotator.py +98 -0
- helm/benchmark/annotation/mimic_bhc_annotator.py +100 -0
- helm/benchmark/annotation/mimic_rrs_annotator.py +100 -0
- helm/benchmark/annotation/model_as_judge.py +214 -6
- helm/benchmark/annotation/mtsamples_procedures_annotator.py +98 -0
- helm/benchmark/annotation/mtsamples_replicate_annotator.py +101 -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/spider_annotator.py +18 -0
- helm/benchmark/annotation/starr_patient_instructions_annotator.py +98 -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_executor.py +35 -15
- 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/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 +2 -2
- 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 +1 -1
- helm/benchmark/augmentations/space_perturbation.py +2 -2
- helm/benchmark/augmentations/suffix_perturbation.py +2 -2
- helm/benchmark/augmentations/synonym_perturbation.py +4 -3
- helm/benchmark/augmentations/test_perturbation.py +16 -13
- helm/benchmark/augmentations/translate_perturbation.py +2 -2
- helm/benchmark/augmentations/typos_perturbation.py +2 -2
- helm/benchmark/data_preprocessor.py +2 -2
- helm/benchmark/executor.py +11 -12
- helm/benchmark/huggingface_registration.py +2 -7
- helm/benchmark/metrics/aci_bench_metrics.py +14 -0
- helm/benchmark/metrics/basic_metrics.py +6 -6
- helm/benchmark/metrics/bbq_metrics.py +2 -2
- helm/benchmark/metrics/bias_metrics.py +12 -3
- 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/chw_care_plan_metrics.py +14 -0
- helm/benchmark/metrics/classification_metrics.py +76 -12
- helm/benchmark/metrics/cleva_harms_metrics.py +10 -9
- helm/benchmark/metrics/code_metrics.py +5 -5
- helm/benchmark/metrics/comet_metric.py +125 -0
- helm/benchmark/metrics/common_metric_specs.py +9 -2
- helm/benchmark/metrics/conv_fin_qa_calc_metrics.py +72 -0
- helm/benchmark/metrics/copyright_metrics.py +4 -4
- helm/benchmark/metrics/czech_bank_qa_metrics.py +29 -0
- helm/benchmark/metrics/decodingtrust_fairness_metrics.py +2 -2
- helm/benchmark/metrics/decodingtrust_privacy_metrics.py +2 -2
- helm/benchmark/metrics/decodingtrust_stereotype_bias_metrics.py +2 -2
- helm/benchmark/metrics/dischargeme_metrics.py +14 -0
- helm/benchmark/metrics/disinformation_metrics.py +4 -4
- helm/benchmark/metrics/dry_run_metrics.py +5 -5
- helm/benchmark/metrics/efficiency_metrics.py +6 -6
- helm/benchmark/metrics/ehr_sql_metrics.py +103 -0
- helm/benchmark/metrics/evaluate_instances_metric.py +3 -3
- helm/benchmark/metrics/evaluate_reference_metrics.py +144 -16
- helm/benchmark/metrics/gpqa_chain_of_thought_metric.py +103 -0
- helm/benchmark/metrics/gpt4_audio_critique_metrics.py +167 -0
- helm/benchmark/metrics/gpt4_audio_refusal_metrics.py +145 -0
- helm/benchmark/metrics/helpdesk_call_summarization_metrics.py +36 -0
- helm/benchmark/metrics/ifeval/__init__.py +0 -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 +55 -0
- helm/benchmark/metrics/image_generation/aesthetics_metrics.py +1 -1
- helm/benchmark/metrics/image_generation/detection_metrics.py +1 -1
- helm/benchmark/metrics/image_generation/detectors/vitdet.py +1 -1
- helm/benchmark/metrics/image_generation/fractal_dimension/test_fractal_dimension_util.py +1 -1
- helm/benchmark/metrics/image_generation/fractal_dimension_metric.py +1 -1
- helm/benchmark/metrics/image_generation/nsfw_metrics.py +1 -1
- helm/benchmark/metrics/image_generation/q16/test_q16.py +3 -1
- helm/benchmark/metrics/image_generation/q16_toxicity_metrics.py +1 -1
- helm/benchmark/metrics/image_generation/skin_tone_metrics.py +2 -2
- helm/benchmark/metrics/image_generation/watermark/test_watermark_detector.py +1 -1
- helm/benchmark/metrics/image_generation/watermark_metrics.py +1 -1
- helm/benchmark/metrics/instruction_following_critique_metrics.py +4 -4
- helm/benchmark/metrics/kpi_edgar_metrics.py +121 -0
- helm/benchmark/metrics/language_modeling_metrics.py +4 -4
- helm/benchmark/metrics/llm_jury_metrics.py +46 -0
- helm/benchmark/metrics/machine_translation_metrics.py +2 -2
- helm/benchmark/metrics/med_dialog_metrics.py +14 -0
- helm/benchmark/metrics/medalign_metrics.py +14 -0
- helm/benchmark/metrics/medcalc_bench_metrics.py +124 -0
- helm/benchmark/metrics/medec_metrics.py +101 -0
- helm/benchmark/metrics/medi_qa_metrics.py +14 -0
- helm/benchmark/metrics/medication_qa_metrics.py +10 -19
- 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/mental_health_metrics.py +14 -0
- helm/benchmark/metrics/metric.py +3 -3
- helm/benchmark/metrics/metric_service.py +11 -11
- helm/benchmark/metrics/mimic_bhc_metrics.py +14 -0
- helm/benchmark/metrics/mimic_rrs_metrics.py +14 -0
- helm/benchmark/metrics/mimiciv_billing_code_metrics.py +96 -0
- helm/benchmark/metrics/mtsamples_procedures_metrics.py +14 -0
- helm/benchmark/metrics/mtsamples_replicate_metrics.py +14 -0
- helm/benchmark/metrics/nltk_helper.py +32 -0
- helm/benchmark/metrics/numeracy_metrics.py +4 -4
- helm/benchmark/metrics/omni_math_metrics.py +32 -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 +2 -2
- helm/benchmark/metrics/ranking_metrics.py +3 -3
- helm/benchmark/metrics/reference_metric.py +3 -3
- helm/benchmark/metrics/ruler_qa_metrics.py +34 -0
- helm/benchmark/metrics/{bhasa_metrics.py → seahelm_metrics.py} +3 -3
- helm/benchmark/metrics/seahelm_metrics_specs.py +10 -0
- helm/benchmark/metrics/spider_metrics.py +7 -0
- helm/benchmark/metrics/starr_patient_instructions_metrics.py +14 -0
- helm/benchmark/metrics/statistic.py +1 -1
- helm/benchmark/metrics/summac/model_summac.py +2 -3
- helm/benchmark/metrics/summarization_critique_metrics.py +4 -4
- helm/benchmark/metrics/summarization_metrics.py +20 -9
- helm/benchmark/metrics/test_bias_metrics.py +5 -1
- helm/benchmark/metrics/test_classification_metrics.py +140 -68
- helm/benchmark/metrics/test_evaluate_reference_metrics.py +15 -0
- helm/benchmark/metrics/test_metric.py +1 -1
- 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 +1 -1
- helm/benchmark/metrics/tokens/openai_token_cost_estimator.py +1 -1
- helm/benchmark/metrics/tokens/test_ai21_token_cost_estimator.py +1 -1
- helm/benchmark/metrics/tokens/test_openai_token_cost_estimator.py +3 -3
- helm/benchmark/metrics/toxicity_metrics.py +6 -6
- helm/benchmark/metrics/unitxt_metrics.py +7 -5
- helm/benchmark/metrics/vision_language/emd_utils.py +4 -2
- helm/benchmark/metrics/vision_language/image_metrics.py +1 -1
- helm/benchmark/metrics/vision_language/image_utils.py +2 -2
- helm/benchmark/metrics/wildbench_metrics.py +34 -0
- helm/benchmark/model_deployment_registry.py +6 -8
- helm/benchmark/model_metadata_registry.py +16 -0
- helm/benchmark/presentation/contamination.py +3 -3
- helm/benchmark/presentation/create_plots.py +33 -12
- helm/benchmark/presentation/run_display.py +13 -0
- helm/benchmark/presentation/schema.py +2 -1
- helm/benchmark/presentation/summarize.py +97 -67
- 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 +86 -90
- helm/benchmark/run_expander.py +90 -9
- helm/benchmark/run_spec_factory.py +13 -0
- helm/benchmark/run_specs/air_bench_run_specs.py +21 -3
- helm/benchmark/run_specs/audio_run_specs.py +657 -0
- helm/benchmark/run_specs/call_center_run_specs.py +49 -0
- helm/benchmark/run_specs/capabilities_run_specs.py +308 -0
- helm/benchmark/run_specs/classic_run_specs.py +1 -69
- 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 +142 -3
- helm/benchmark/run_specs/imdb_ptbr_run_specs.py +30 -0
- helm/benchmark/run_specs/lite_run_specs.py +2 -2
- helm/benchmark/run_specs/long_context_run_specs.py +141 -0
- helm/benchmark/run_specs/medhelm_run_specs.py +1260 -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/oab_exams_specs.py +32 -0
- helm/benchmark/run_specs/safety_run_specs.py +37 -0
- helm/benchmark/run_specs/{bhasa_run_specs.py → seahelm_run_specs.py} +44 -44
- helm/benchmark/run_specs/speech_disorder_audio_run_specs.py +169 -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 +14 -5
- helm/benchmark/run_specs/vlm_run_specs.py +103 -2
- helm/benchmark/run_specs/winogrande_afr_run_specs.py +47 -0
- helm/benchmark/runner.py +5 -5
- helm/benchmark/scenarios/aci_bench_scenario.py +126 -0
- helm/benchmark/scenarios/air_bench_scenario.py +6 -1
- helm/benchmark/scenarios/anthropic_hh_rlhf_scenario.py +5 -3
- helm/benchmark/scenarios/anthropic_red_team_scenario.py +1 -1
- 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.py +103 -0
- helm/benchmark/scenarios/audio_language/ultra_suite_classification_scenario.py +110 -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 +109 -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 +1 -1
- helm/benchmark/scenarios/banking77_scenario.py +6 -1
- helm/benchmark/scenarios/bbq_scenario.py +1 -1
- helm/benchmark/scenarios/big_bench_scenario.py +11 -1
- helm/benchmark/scenarios/bigcodebench_scenario.py +58 -0
- helm/benchmark/scenarios/bird_sql_scenario.py +94 -0
- helm/benchmark/scenarios/bird_sql_scenario_helper.py +118 -0
- helm/benchmark/scenarios/blimp_scenario.py +1 -1
- helm/benchmark/scenarios/bold_scenario.py +1 -1
- helm/benchmark/scenarios/boolq_scenario.py +1 -1
- helm/benchmark/scenarios/casehold_scenario.py +79 -0
- helm/benchmark/scenarios/chw_care_plan_scenario.py +106 -0
- helm/benchmark/scenarios/civil_comments_scenario.py +1 -1
- helm/benchmark/scenarios/clear_scenario.py +157 -0
- helm/benchmark/scenarios/cleva_scenario.py +2 -2
- helm/benchmark/scenarios/code_scenario.py +17 -4
- helm/benchmark/scenarios/commonsense_scenario.py +1 -1
- helm/benchmark/scenarios/conv_fin_qa_calc_scenario.py +97 -0
- helm/benchmark/scenarios/copyright_scenario.py +1 -1
- helm/benchmark/scenarios/covid_dialog_scenario.py +10 -1
- helm/benchmark/scenarios/cti_to_mitre_scenario.py +240 -0
- helm/benchmark/scenarios/custom_mcqa_scenario.py +1 -1
- helm/benchmark/scenarios/czech_bank_qa_scenario.py +130 -0
- helm/benchmark/scenarios/decodingtrust_adv_demonstration_scenario.py +1 -1
- helm/benchmark/scenarios/decodingtrust_privacy_scenario.py +1 -1
- helm/benchmark/scenarios/decodingtrust_stereotype_bias_scenario.py +1 -1
- helm/benchmark/scenarios/decodingtrust_toxicity_prompts_scenario.py +1 -1
- helm/benchmark/scenarios/dialogue_scenarios.py +13 -2
- helm/benchmark/scenarios/dischargeme_scenario.py +172 -0
- helm/benchmark/scenarios/disinformation_scenario.py +10 -1
- helm/benchmark/scenarios/dyck_language_scenario.py +10 -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 +1519 -0
- helm/benchmark/scenarios/enem_challenge_scenario.py +58 -0
- helm/benchmark/scenarios/entity_data_imputation_scenario.py +11 -1
- helm/benchmark/scenarios/entity_matching_scenario.py +12 -2
- helm/benchmark/scenarios/financial_phrasebank_scenario.py +94 -0
- helm/benchmark/scenarios/gold_commodity_news_scenario.py +124 -0
- helm/benchmark/scenarios/gpqa_scenario.py +80 -0
- helm/benchmark/scenarios/grammar.py +2 -2
- helm/benchmark/scenarios/grammar_scenario.py +2 -2
- helm/benchmark/scenarios/gsm_scenario.py +10 -1
- helm/benchmark/scenarios/harm_bench_gcg_transfer_scenario.py +50 -0
- helm/benchmark/scenarios/harm_bench_scenario.py +1 -1
- helm/benchmark/scenarios/headqa_scenario.py +136 -0
- helm/benchmark/scenarios/helpdesk_call_summarization_scenario.py +37 -0
- helm/benchmark/scenarios/ice_scenario.py +8 -4
- helm/benchmark/scenarios/ifeval_scenario.py +53 -0
- helm/benchmark/scenarios/imdb_ptbr_scenario.py +60 -0
- helm/benchmark/scenarios/imdb_scenario.py +11 -2
- helm/benchmark/scenarios/infinite_bench_en_qa_scenario.py +85 -0
- helm/benchmark/scenarios/infinite_bench_en_sum_scenario.py +79 -0
- helm/benchmark/scenarios/interactive_qa_mmlu_scenario.py +2 -2
- helm/benchmark/scenarios/koala_scenario.py +1 -1
- helm/benchmark/scenarios/kpi_edgar_scenario.py +151 -0
- helm/benchmark/scenarios/legal_contract_summarization_scenario.py +129 -0
- helm/benchmark/scenarios/legal_opinion_sentiment_classification_scenario.py +77 -0
- helm/benchmark/scenarios/legal_summarization_scenario.py +11 -1
- helm/benchmark/scenarios/legal_support_scenario.py +11 -1
- helm/benchmark/scenarios/legalbench_scenario.py +22 -3
- helm/benchmark/scenarios/lex_glue_scenario.py +12 -2
- helm/benchmark/scenarios/lextreme_scenario.py +11 -1
- helm/benchmark/scenarios/live_qa_scenario.py +1 -1
- helm/benchmark/scenarios/lm_entry_scenario.py +1 -1
- helm/benchmark/scenarios/lsat_qa_scenario.py +1 -1
- helm/benchmark/scenarios/math_scenario.py +9 -1
- helm/benchmark/scenarios/me_q_sum_scenario.py +10 -1
- helm/benchmark/scenarios/med_dialog_scenario.py +25 -22
- helm/benchmark/scenarios/med_mcqa_scenario.py +10 -1
- helm/benchmark/scenarios/med_paragraph_simplification_scenario.py +10 -1
- helm/benchmark/scenarios/med_qa_scenario.py +10 -1
- helm/benchmark/scenarios/medalign_scenario.py +94 -0
- helm/benchmark/scenarios/medalign_scenario_helper.py +432 -0
- helm/benchmark/scenarios/medbullets_scenario.py +145 -0
- helm/benchmark/scenarios/medcalc_bench_scenario.py +127 -0
- helm/benchmark/scenarios/medec_scenario.py +125 -0
- helm/benchmark/scenarios/medhallu_scenario.py +72 -0
- helm/benchmark/scenarios/medi_qa_scenario.py +111 -0
- helm/benchmark/scenarios/medication_qa_scenario.py +8 -2
- 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 +123 -0
- helm/benchmark/scenarios/mimic_bhc_scenario.py +103 -0
- helm/benchmark/scenarios/mimic_rrs_scenario.py +98 -0
- helm/benchmark/scenarios/mimiciv_billing_code_scenario.py +77 -0
- helm/benchmark/scenarios/mmlu_clinical_afr_scenario.py +74 -0
- helm/benchmark/scenarios/mmlu_pro_scenario.py +95 -0
- helm/benchmark/scenarios/mmlu_scenario.py +11 -1
- helm/benchmark/scenarios/msmarco_scenario.py +1 -1
- helm/benchmark/scenarios/mtsamples_procedures_scenario.py +144 -0
- helm/benchmark/scenarios/mtsamples_replicate_scenario.py +142 -0
- helm/benchmark/scenarios/n2c2_ct_matching_scenario.py +277 -0
- helm/benchmark/scenarios/narrativeqa_scenario.py +1 -1
- helm/benchmark/scenarios/natural_qa_scenario.py +1 -1
- helm/benchmark/scenarios/newsqa_scenario.py +1 -1
- helm/benchmark/scenarios/numeracy_scenario.py +12 -2
- helm/benchmark/scenarios/oab_exams_scenario.py +57 -0
- helm/benchmark/scenarios/omni_math_scenario.py +53 -0
- helm/benchmark/scenarios/open_assistant_scenario.py +11 -2
- helm/benchmark/scenarios/openai_mrcr_scenario.py +79 -0
- helm/benchmark/scenarios/opinions_qa_scenario.py +1 -1
- helm/benchmark/scenarios/pubmed_qa_scenario.py +59 -43
- helm/benchmark/scenarios/quac_scenario.py +10 -1
- helm/benchmark/scenarios/race_based_med_scenario.py +152 -0
- helm/benchmark/scenarios/raft_scenario.py +17 -2
- helm/benchmark/scenarios/real_toxicity_prompts_scenario.py +1 -1
- helm/benchmark/scenarios/ruler_qa_scenario_helper.py +171 -0
- helm/benchmark/scenarios/ruler_qa_scenarios.py +88 -0
- helm/benchmark/scenarios/scenario.py +9 -1
- helm/benchmark/scenarios/{bhasa_scenario.py → seahelm_scenario.py} +7 -2
- helm/benchmark/scenarios/self_instruct_scenario.py +1 -1
- helm/benchmark/scenarios/shc_bmt_scenario.py +75 -0
- helm/benchmark/scenarios/shc_cdi_scenario.py +75 -0
- helm/benchmark/scenarios/shc_conf_scenario.py +76 -0
- helm/benchmark/scenarios/shc_ent_scenario.py +77 -0
- helm/benchmark/scenarios/shc_gip_scenario.py +74 -0
- helm/benchmark/scenarios/shc_privacy_scenario.py +78 -0
- helm/benchmark/scenarios/shc_proxy_scenario.py +76 -0
- helm/benchmark/scenarios/shc_ptbm_scenario.py +81 -0
- helm/benchmark/scenarios/shc_sei_scenario.py +94 -0
- helm/benchmark/scenarios/shc_sequoia_scenario.py +77 -0
- helm/benchmark/scenarios/simple_safety_tests_scenario.py +1 -1
- helm/benchmark/scenarios/spider_scenario.py +91 -0
- helm/benchmark/scenarios/starr_patient_instructions_scenario.py +97 -0
- helm/benchmark/scenarios/summarization_scenario.py +11 -1
- helm/benchmark/scenarios/sumosum_scenario.py +157 -0
- helm/benchmark/scenarios/synthetic_efficiency_scenario.py +1 -1
- helm/benchmark/scenarios/synthetic_reasoning_natural_scenario.py +11 -1
- helm/benchmark/scenarios/synthetic_reasoning_scenario.py +11 -1
- helm/benchmark/scenarios/test_bigcodebench_scenario.py +26 -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 +6 -2
- helm/benchmark/scenarios/test_gold_commodity_news_scenario.py +18 -0
- helm/benchmark/scenarios/test_gpqa_scenario.py +44 -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_math_scenario.py +1 -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_oab_exams_scenario.py +51 -0
- helm/benchmark/scenarios/test_omni_math_scenario.py +27 -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 +10 -1
- helm/benchmark/scenarios/the_pile_scenario.py +1 -1
- helm/benchmark/scenarios/truthful_qa_scenario.py +12 -2
- helm/benchmark/scenarios/tweetsentbr_scenario.py +66 -0
- helm/benchmark/scenarios/twitter_aae_scenario.py +1 -1
- helm/benchmark/scenarios/unitxt_scenario.py +8 -2
- helm/benchmark/scenarios/verifiability_judgment_scenario.py +1 -1
- helm/benchmark/scenarios/vicuna_scenario.py +1 -1
- helm/benchmark/scenarios/vision_language/blink_scenario.py +140 -0
- helm/benchmark/scenarios/vision_language/mm_star_scenario.py +95 -0
- helm/benchmark/scenarios/vision_language/msr_vtt_scenario.py +75 -0
- helm/benchmark/scenarios/vision_language/vqa_rad_scenario.py +88 -0
- helm/benchmark/scenarios/wikifact_scenario.py +11 -1
- helm/benchmark/scenarios/wikitext_103_scenario.py +1 -1
- helm/benchmark/scenarios/wildbench_scenario.py +83 -0
- helm/benchmark/scenarios/winogrande_afr_scenario.py +78 -0
- helm/benchmark/scenarios/wmt_14_scenario.py +14 -2
- helm/benchmark/scenarios/xstest_scenario.py +1 -1
- helm/benchmark/server.py +13 -1
- helm/benchmark/slurm_runner.py +1 -1
- helm/benchmark/static/schema_audio.yaml +763 -0
- helm/benchmark/static/schema_autobencher.yaml +150 -0
- helm/benchmark/static/schema_call_center.yaml +97 -60
- helm/benchmark/static/{schema_medical.yaml → schema_capabilities.yaml} +100 -101
- helm/benchmark/static/schema_czech_bank.yaml +148 -0
- helm/benchmark/static/schema_enem_challenge.yaml +146 -0
- helm/benchmark/static/schema_enterprise.yaml +319 -0
- helm/benchmark/static/schema_finance.yaml +14 -12
- helm/benchmark/static/schema_heim.yaml +1389 -0
- helm/benchmark/static/schema_long_context.yaml +283 -0
- helm/benchmark/static/schema_medhelm.yaml +1140 -0
- helm/benchmark/static/schema_melt.yaml +1257 -0
- helm/benchmark/static/schema_mmlu_winogrande_afr.yaml +1045 -0
- helm/benchmark/static/schema_safety.yaml +18 -1
- helm/benchmark/static/{schema_bhasa.yaml → schema_seahelm.yaml} +30 -16
- 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_tables.yaml → schema_torr.yaml} +169 -36
- helm/benchmark/static/schema_tweetsentbr.yaml +146 -0
- helm/benchmark/static/schema_vhelm.yaml +129 -56
- helm/benchmark/static/schema_video.yaml +219 -0
- helm/benchmark/static_build/assets/helm-safety-2907a7b6.png +0 -0
- helm/benchmark/static_build/assets/index-94295e78.js +10 -0
- helm/benchmark/static_build/assets/index-b9779128.css +1 -0
- helm/benchmark/static_build/assets/medhelm-overview-eac29843.png +0 -0
- helm/benchmark/static_build/assets/medhelm-v1-overview-3ddfcd65.png +0 -0
- helm/benchmark/static_build/assets/{react-d4a0b69b.js → react-f82877fd.js} +1 -1
- helm/benchmark/static_build/assets/{recharts-6d337683.js → recharts-4037aff0.js} +1 -1
- helm/benchmark/static_build/assets/{tremor-54a99cc4.js → tremor-38a10867.js} +2 -2
- helm/benchmark/static_build/config.js +1 -1
- helm/benchmark/static_build/index.html +6 -6
- helm/benchmark/window_services/default_window_service.py +1 -1
- helm/benchmark/window_services/encoder_decoder_window_service.py +4 -4
- helm/benchmark/window_services/ice_window_service.py +1 -1
- helm/benchmark/window_services/image_generation/lexica_search_window_service.py +1 -1
- helm/benchmark/window_services/image_generation/openai_dalle_window_service.py +1 -1
- helm/benchmark/window_services/local_window_service.py +2 -2
- helm/benchmark/window_services/test_anthropic_window_service.py +3 -3
- helm/benchmark/window_services/test_bloom_window_service.py +3 -3
- helm/benchmark/window_services/test_gpt2_window_service.py +7 -2
- helm/benchmark/window_services/test_gpt4_window_service.py +8 -3
- helm/benchmark/window_services/test_gptj_window_service.py +8 -3
- helm/benchmark/window_services/test_gptneox_window_service.py +3 -3
- helm/benchmark/window_services/test_openai_window_service.py +8 -3
- helm/benchmark/window_services/test_opt_window_service.py +3 -3
- helm/benchmark/window_services/test_palmyra_window_service.py +3 -3
- helm/benchmark/window_services/test_t0pp_window_service.py +3 -3
- helm/benchmark/window_services/test_t511b_window_service.py +3 -3
- helm/benchmark/window_services/test_ul2_window_service.py +3 -3
- helm/benchmark/window_services/test_utils.py +4 -5
- helm/benchmark/window_services/test_yalm_window_service.py +3 -3
- helm/benchmark/window_services/tokenizer_service.py +7 -8
- helm/benchmark/window_services/yalm_window_service.py +1 -1
- helm/clients/ai21_client.py +3 -3
- helm/clients/aleph_alpha_client.py +1 -1
- helm/clients/anthropic_client.py +69 -29
- helm/clients/audio_language/__init__.py +0 -0
- helm/clients/audio_language/diva_llama_client.py +120 -0
- helm/clients/audio_language/llama_omni_client.py +198 -0
- helm/clients/audio_language/qwen2_5_omni_client.py +197 -0
- helm/clients/audio_language/qwen2_audiolm_client.py +190 -0
- helm/clients/audio_language/qwen_audiolm_client.py +152 -0
- helm/clients/audio_language/test.py +62 -0
- helm/clients/auto_client.py +4 -2
- helm/clients/azure_openai_client.py +55 -0
- helm/clients/bedrock_client.py +203 -7
- helm/clients/bedrock_utils.py +33 -0
- helm/clients/client.py +7 -7
- helm/clients/clip_scorers/clip_scorer.py +1 -1
- helm/clients/clip_scorers/multilingual_clip_scorer.py +1 -1
- helm/clients/cohere_client.py +3 -3
- helm/clients/google_client.py +1 -1
- helm/clients/grok_client.py +36 -0
- helm/clients/http_model_client.py +1 -1
- helm/clients/huggingface_client.py +52 -21
- helm/clients/huggingface_pipeline_client.py +138 -0
- helm/clients/ibm_client.py +267 -0
- helm/clients/image_generation/adobe_vision_client.py +1 -1
- helm/clients/image_generation/aleph_alpha_image_generation_client.py +1 -1
- helm/clients/image_generation/cogview2/sr_pipeline/__init__.py +3 -3
- helm/clients/image_generation/cogview2/sr_pipeline/direct_sr.py +5 -2
- helm/clients/image_generation/cogview2/sr_pipeline/iterative_sr.py +5 -2
- helm/clients/image_generation/cogview2/sr_pipeline/sr_group.py +2 -2
- helm/clients/image_generation/cogview2_client.py +1 -1
- helm/clients/image_generation/dalle2_client.py +1 -1
- helm/clients/image_generation/dalle3_client.py +2 -2
- helm/clients/image_generation/dalle_mini/__init__.py +1 -1
- helm/clients/image_generation/dalle_mini/data.py +1 -1
- helm/clients/image_generation/dalle_mini/model/__init__.py +5 -5
- helm/clients/image_generation/dalle_mini/model/configuration.py +2 -2
- helm/clients/image_generation/dalle_mini/model/modeling.py +3 -3
- helm/clients/image_generation/dalle_mini/model/processor.py +5 -5
- helm/clients/image_generation/dalle_mini/model/tokenizer.py +2 -2
- helm/clients/image_generation/dalle_mini/vqgan_jax/__init__.py +1 -1
- helm/clients/image_generation/dalle_mini/vqgan_jax/convert_pt_model_to_jax.py +2 -2
- helm/clients/image_generation/dalle_mini/vqgan_jax/modeling_flax_vqgan.py +1 -1
- helm/clients/image_generation/dalle_mini_client.py +1 -1
- helm/clients/image_generation/deep_floyd_client.py +1 -1
- helm/clients/image_generation/huggingface_diffusers_client.py +1 -1
- helm/clients/image_generation/lexica_client.py +1 -1
- helm/clients/image_generation/mindalle/models/__init__.py +6 -6
- helm/clients/image_generation/mindalle/models/stage1/vqgan.py +1 -1
- helm/clients/image_generation/mindalle/models/stage2/transformer.py +1 -1
- helm/clients/image_generation/mindalle/utils/__init__.py +3 -3
- helm/clients/image_generation/mindalle_client.py +1 -1
- helm/clients/image_generation/together_image_generation_client.py +1 -1
- helm/clients/lit_gpt_client.py +2 -2
- helm/clients/mistral_client.py +62 -18
- helm/clients/nvidia_nim_client.py +0 -3
- helm/clients/openai_client.py +308 -43
- helm/clients/openai_responses_client.py +174 -0
- helm/clients/palmyra_client.py +3 -9
- helm/clients/reka_client.py +3 -3
- 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 +93 -0
- helm/clients/stanfordhealthcare_openai_client.py +62 -0
- helm/clients/stanfordhealthcare_shc_openai_client.py +42 -0
- helm/clients/test_client.py +1 -1
- helm/clients/test_together_client.py +6 -1
- helm/clients/together_client.py +76 -9
- helm/clients/upstage_client.py +23 -0
- helm/clients/vertexai_client.py +45 -13
- helm/clients/vision_language/huggingface_vision2seq_client.py +6 -4
- helm/clients/vision_language/huggingface_vlm_client.py +2 -2
- helm/clients/vision_language/idefics_client.py +6 -2
- helm/clients/vision_language/open_flamingo/__init__.py +2 -2
- helm/clients/vision_language/open_flamingo/src/factory.py +3 -3
- helm/clients/vision_language/open_flamingo/src/flamingo.py +2 -2
- helm/clients/vision_language/open_flamingo/src/flamingo_lm.py +2 -2
- helm/clients/vision_language/paligemma_client.py +2 -2
- helm/clients/vision_language/qwen2_vlm_client.py +188 -0
- helm/clients/vision_language/qwen_vlm_client.py +7 -5
- helm/clients/vllm_client.py +4 -6
- helm/clients/writer_client.py +102 -0
- helm/clients/yi_client.py +0 -3
- helm/common/audio_utils.py +111 -0
- helm/common/context.py +80 -0
- helm/common/credentials_utils.py +5 -5
- helm/common/file_caches/local_file_cache.py +1 -1
- helm/common/file_caches/test_local_file_cache.py +1 -1
- helm/common/general.py +9 -2
- helm/common/hierarchical_logger.py +46 -3
- helm/common/images_utils.py +2 -2
- helm/common/local_context.py +140 -0
- helm/common/media_object.py +2 -2
- helm/common/multimodal_request_utils.py +26 -0
- helm/common/reeval_parameters.py +12 -0
- helm/common/remote_context.py +61 -0
- helm/common/request.py +14 -2
- helm/common/response_format.py +18 -0
- helm/common/test_media_object.py +1 -1
- helm/config/model_deployments.yaml +1792 -28
- helm/config/model_metadata.yaml +1606 -51
- helm/config/tokenizer_configs.yaml +521 -4
- helm/proxy/cli.py +5 -3
- helm/proxy/critique/mechanical_turk_utils.py +1 -1
- helm/proxy/example_queries.py +1 -1
- helm/proxy/server.py +11 -4
- helm/proxy/services/remote_service.py +1 -1
- helm/proxy/services/server_service.py +22 -86
- helm/proxy/services/test_remote_service.py +2 -2
- helm/proxy/services/test_service.py +1 -1
- helm/proxy/static/general.js +122 -0
- helm/proxy/static/help.html +99 -0
- helm/proxy/static/index.css +57 -0
- helm/proxy/static/index.html +40 -0
- helm/proxy/static/index.js +456 -0
- helm/proxy/static/info-icon.png +0 -0
- helm/proxy/test_retry.py +1 -1
- helm/proxy/token_counters/auto_token_counter.py +1 -1
- helm/tokenizers/aleph_alpha_tokenizer.py +1 -1
- helm/tokenizers/caching_tokenizer.py +2 -30
- helm/tokenizers/grok_tokenizer.py +53 -0
- helm/tokenizers/http_model_tokenizer.py +1 -1
- helm/tokenizers/huggingface_tokenizer.py +3 -3
- helm/tokenizers/lit_gpt_tokenizer.py +1 -1
- helm/tokenizers/test_anthropic_tokenizer.py +6 -2
- helm/tokenizers/test_grok_tokenizer.py +33 -0
- helm/tokenizers/test_huggingface_tokenizer.py +1 -1
- helm/tokenizers/test_yalm_tokenizer.py +1 -1
- helm/tokenizers/tiktoken_tokenizer.py +1 -1
- helm/tokenizers/tokenizer.py +3 -1
- helm/tokenizers/yalm_tokenizer.py +3 -3
- helm/tokenizers/yalm_tokenizer_data/test_yalm_tokenizer.py +1 -1
- crfm_helm-0.5.4.dist-info/METADATA +0 -350
- crfm_helm-0.5.4.dist-info/RECORD +0 -697
- helm/benchmark/metrics/bhasa_metrics_specs.py +0 -10
- helm/benchmark/static_build/assets/01-694cb9b7.png +0 -0
- helm/benchmark/static_build/assets/accenture-6f97eeda.png +0 -0
- helm/benchmark/static_build/assets/ai21-0eb91ec3.png +0 -0
- helm/benchmark/static_build/assets/aisingapore-6dfc9acf.png +0 -0
- helm/benchmark/static_build/assets/aleph-alpha-7ce10034.png +0 -0
- helm/benchmark/static_build/assets/anthropic-70d8bc39.png +0 -0
- helm/benchmark/static_build/assets/bigscience-7f0400c0.png +0 -0
- helm/benchmark/static_build/assets/cohere-3550c6cb.png +0 -0
- helm/benchmark/static_build/assets/cresta-9e22b983.png +0 -0
- helm/benchmark/static_build/assets/cuhk-8c5631e9.png +0 -0
- helm/benchmark/static_build/assets/eleutherai-b9451114.png +0 -0
- helm/benchmark/static_build/assets/google-06d997ad.png +0 -0
- helm/benchmark/static_build/assets/index-05c76bb1.css +0 -1
- helm/benchmark/static_build/assets/index-3ee38b3d.js +0 -10
- helm/benchmark/static_build/assets/meta-5580e9f1.png +0 -0
- helm/benchmark/static_build/assets/microsoft-f5ee5016.png +0 -0
- helm/benchmark/static_build/assets/mistral-18e1be23.png +0 -0
- helm/benchmark/static_build/assets/nvidia-86fa75c1.png +0 -0
- helm/benchmark/static_build/assets/openai-3f8653e4.png +0 -0
- helm/benchmark/static_build/assets/scb10x-204bd786.png +0 -0
- helm/benchmark/static_build/assets/tii-24de195c.png +0 -0
- helm/benchmark/static_build/assets/together-a665a35b.png +0 -0
- helm/benchmark/static_build/assets/tsinghua-keg-97d4b395.png +0 -0
- helm/benchmark/static_build/assets/wellsfargo-a86a6c4a.png +0 -0
- helm/benchmark/static_build/assets/yandex-38e09d70.png +0 -0
- helm/tokenizers/anthropic_tokenizer.py +0 -52
- {crfm_helm-0.5.4.dist-info → crfm_helm-0.5.6.dist-info}/entry_points.txt +0 -0
- {crfm_helm-0.5.4.dist-info → crfm_helm-0.5.6.dist-info/licenses}/LICENSE +0 -0
- {crfm_helm-0.5.4.dist-info → crfm_helm-0.5.6.dist-info}/top_level.txt +0 -0
|
@@ -0,0 +1,783 @@
|
|
|
1
|
+
from typing import Optional
|
|
2
|
+
|
|
3
|
+
import torch
|
|
4
|
+
|
|
5
|
+
from helm.benchmark.run_spec import RunSpec, run_spec_function
|
|
6
|
+
from helm.benchmark.scenarios.scenario import ScenarioSpec
|
|
7
|
+
from helm.benchmark.adaptation.adapters.binary_ranking_adapter import BinaryRankingAdapter
|
|
8
|
+
from helm.benchmark.adaptation.common_adapter_specs import (
|
|
9
|
+
get_generation_adapter_spec,
|
|
10
|
+
get_machine_translation_adapter_spec,
|
|
11
|
+
get_multiple_choice_adapter_spec,
|
|
12
|
+
get_ranking_binary_adapter_spec,
|
|
13
|
+
)
|
|
14
|
+
from helm.benchmark.adaptation.adapter_spec import (
|
|
15
|
+
ADAPT_GENERATION,
|
|
16
|
+
ADAPT_MULTIPLE_CHOICE_JOINT,
|
|
17
|
+
ADAPT_RANKING_BINARY,
|
|
18
|
+
AdapterSpec,
|
|
19
|
+
)
|
|
20
|
+
from helm.benchmark.metrics.common_metric_specs import (
|
|
21
|
+
get_exact_match_metric_specs,
|
|
22
|
+
get_f1_metric_specs,
|
|
23
|
+
get_summarization_metric_specs,
|
|
24
|
+
get_basic_metric_specs,
|
|
25
|
+
get_open_ended_generation_metric_specs,
|
|
26
|
+
get_classification_metric_specs,
|
|
27
|
+
get_basic_reference_metric_specs,
|
|
28
|
+
get_generic_metric_specs,
|
|
29
|
+
)
|
|
30
|
+
from helm.benchmark.metrics.melt_metric_specs import get_vietnamese_generative_harms_metric_specs
|
|
31
|
+
from helm.benchmark.metrics.metric import MetricSpec
|
|
32
|
+
|
|
33
|
+
|
|
34
|
+
@run_spec_function("melt_question_answering_mlqa")
|
|
35
|
+
def get_melt_question_answering_mlqa_spec(prompt_style: str = "normal") -> RunSpec:
|
|
36
|
+
scenario_spec = ScenarioSpec(class_name="helm.benchmark.scenarios.melt_scenarios.MELTQAMLQAScenario")
|
|
37
|
+
|
|
38
|
+
if prompt_style == "weak":
|
|
39
|
+
instruction = ""
|
|
40
|
+
elif prompt_style == "medium":
|
|
41
|
+
instruction = (
|
|
42
|
+
"Hãy trả lời câu hỏi bên dưới bằng tiếng Việt với các thông tin được cung cấp trong phần ngữ cảnh. "
|
|
43
|
+
'Nếu trong ngữ cảnh không có đủ thông tin, hãy trả lời "Tôi không biết".'
|
|
44
|
+
)
|
|
45
|
+
elif prompt_style == "normal":
|
|
46
|
+
instruction = (
|
|
47
|
+
"Bạn là một trợ lý hữu dụng sử dụng tiếng Việt, biết tôn trọng và thành thật. "
|
|
48
|
+
"Bạn luôn luôn trả lời các câu hỏi một cách có ích nhiều nhất có thể, nhưng đồng thời phải an toàn. "
|
|
49
|
+
"Câu trả lời của bạn không được bao gồm các ngôn từ độc hại, phân biệt chủng tộc, "
|
|
50
|
+
"phân biệt giới tính, nguy hiểm, nội dung vi phạm pháp luật. Làm ơn hãy chắc chắn câu trả lời "
|
|
51
|
+
"của bạn tự nhiên, tích cực và không thiên vị bất cứ cái gì. Nếu có câu hỏi không hợp lý hoặc không "
|
|
52
|
+
"rõ ràng thì hãy giải thích tại sao thay vì trả lời không đúng sự thật. Nếu bạn không biết câu trả lời "
|
|
53
|
+
"thì đừng chia sẻ thông tin sai sự thật."
|
|
54
|
+
)
|
|
55
|
+
else:
|
|
56
|
+
raise ValueError(f"Invalid prompt style: {prompt_style}")
|
|
57
|
+
|
|
58
|
+
adapter_spec = get_generation_adapter_spec(instructions=instruction, output_noun="Trả lời", max_tokens=128)
|
|
59
|
+
|
|
60
|
+
return RunSpec(
|
|
61
|
+
name=f"melt_question_answering_mlqa:prompt_style={prompt_style}",
|
|
62
|
+
scenario_spec=scenario_spec,
|
|
63
|
+
adapter_spec=adapter_spec,
|
|
64
|
+
metric_specs=get_exact_match_metric_specs()
|
|
65
|
+
+ get_f1_metric_specs()
|
|
66
|
+
+ get_vietnamese_generative_harms_metric_specs(include_generative_harms_metrics=True),
|
|
67
|
+
groups=["melt", "melt_question_answering_mlqa"],
|
|
68
|
+
)
|
|
69
|
+
|
|
70
|
+
|
|
71
|
+
@run_spec_function("melt_question_answering_xquad")
|
|
72
|
+
def get_melt_question_answering_xquad_spec(prompt_style: str = "normal") -> RunSpec:
|
|
73
|
+
scenario_spec = ScenarioSpec(class_name="helm.benchmark.scenarios.melt_scenarios.MELTQAXQuADScenario")
|
|
74
|
+
|
|
75
|
+
if prompt_style == "weak":
|
|
76
|
+
instruction = ""
|
|
77
|
+
elif prompt_style == "medium":
|
|
78
|
+
instruction = (
|
|
79
|
+
"Hãy trả lời câu hỏi bên dưới bằng tiếng Việt với các thông tin được cung cấp trong phần ngữ cảnh. "
|
|
80
|
+
'Nếu trong ngữ cảnh không có đủ thông tin, hãy trả lời "Tôi không biết".'
|
|
81
|
+
)
|
|
82
|
+
elif prompt_style == "normal":
|
|
83
|
+
instruction = (
|
|
84
|
+
"Bạn là một trợ lý hữu dụng sử dụng tiếng Việt, biết tôn trọng và thành thật. "
|
|
85
|
+
"Bạn luôn luôn trả lời các câu hỏi một cách có ích nhiều nhất có thể, nhưng đồng thời phải an toàn. "
|
|
86
|
+
"Câu trả lời của bạn không được bao gồm các ngôn từ độc hại, phân biệt chủng tộc, "
|
|
87
|
+
"phân biệt giới tính, nguy hiểm, nội dung vi phạm pháp luật. Làm ơn hãy chắc chắn câu trả lời "
|
|
88
|
+
"của bạn tự nhiên, tích cực và không thiên vị bất cứ cái gì. Nếu có câu hỏi không hợp lý hoặc không "
|
|
89
|
+
"rõ ràng thì hãy giải thích tại sao thay vì trả lời không đúng sự thật. Nếu bạn không biết câu trả lời "
|
|
90
|
+
"thì đừng chia sẻ thông tin sai sự thật."
|
|
91
|
+
)
|
|
92
|
+
else:
|
|
93
|
+
raise ValueError(f"Invalid prompt style: {prompt_style}")
|
|
94
|
+
|
|
95
|
+
adapter_spec = get_generation_adapter_spec(instructions=instruction, output_noun="Trả lời", max_tokens=128)
|
|
96
|
+
|
|
97
|
+
return RunSpec(
|
|
98
|
+
name=f"melt_question_answering_xquad:prompt_style={prompt_style},",
|
|
99
|
+
scenario_spec=scenario_spec,
|
|
100
|
+
adapter_spec=adapter_spec,
|
|
101
|
+
metric_specs=get_exact_match_metric_specs()
|
|
102
|
+
+ get_f1_metric_specs()
|
|
103
|
+
+ get_vietnamese_generative_harms_metric_specs(include_generative_harms_metrics=True),
|
|
104
|
+
groups=["melt", "melt_question_answering_xquad"],
|
|
105
|
+
)
|
|
106
|
+
|
|
107
|
+
|
|
108
|
+
@run_spec_function("melt_summarization_vietnews")
|
|
109
|
+
def get_melt_summarization_vietnews_spec(prompt_style: str = "normal", temperature: float = 1.0) -> RunSpec:
|
|
110
|
+
scenario_spec = ScenarioSpec(
|
|
111
|
+
class_name="helm.benchmark.scenarios.melt_scenarios.MELTSummarizationVietnewsScenario",
|
|
112
|
+
args={
|
|
113
|
+
"train_min_length": 64,
|
|
114
|
+
"train_max_length": 256,
|
|
115
|
+
"doc_max_length": 2048,
|
|
116
|
+
},
|
|
117
|
+
)
|
|
118
|
+
|
|
119
|
+
if prompt_style == "weak":
|
|
120
|
+
instruction = ""
|
|
121
|
+
elif prompt_style == "medium":
|
|
122
|
+
instruction = "Nhiệm vụ của bạn là tóm tắt đoạn văn bản sau, đưa ra câu trả lời là bản tóm tắt."
|
|
123
|
+
elif prompt_style == "normal":
|
|
124
|
+
instruction = (
|
|
125
|
+
"Bạn là một trợ lý hữu dụng, biết tôn trọng và thành thật. Bạn luôn luôn trả lời các câu hỏi một cách "
|
|
126
|
+
"có ích nhiều nhất có thể, nhưng đồng thời phải an toàn. Câu trả lời của bạn không được bao gồm "
|
|
127
|
+
"các ngôn từ độc hại, phân biệt chủng tộc, phân biệt giới tính, nguy hiểm, nội dung vi phạm pháp luật. "
|
|
128
|
+
"Nhiệm vụ của bạn là tóm tắt đoạn văn bản nằm trong triple backtick. Bài tóm tắt phải đầy đủ các thông tin "
|
|
129
|
+
"quan trọng, ngắn gọn và thu hút người đọc. Ngôn ngữ bạn phải sử dụng để tóm tắt là tiếng Việt."
|
|
130
|
+
)
|
|
131
|
+
else:
|
|
132
|
+
raise ValueError(f"Invalid prompt style: {prompt_style}")
|
|
133
|
+
|
|
134
|
+
adapter_spec = get_generation_adapter_spec(
|
|
135
|
+
instructions=instruction,
|
|
136
|
+
input_noun="Đoạn văn",
|
|
137
|
+
output_noun="Tóm tắt đoạn văn trên",
|
|
138
|
+
max_tokens=256,
|
|
139
|
+
temperature=temperature,
|
|
140
|
+
)
|
|
141
|
+
|
|
142
|
+
return RunSpec(
|
|
143
|
+
name=f"melt_summarization_vietnews:prompt_style={prompt_style},temperature={temperature}",
|
|
144
|
+
scenario_spec=scenario_spec,
|
|
145
|
+
adapter_spec=adapter_spec,
|
|
146
|
+
metric_specs=get_summarization_metric_specs(
|
|
147
|
+
{
|
|
148
|
+
"task": "summarization_vietnews",
|
|
149
|
+
"language": "vi",
|
|
150
|
+
"bertscore_model": "bert-base-multilingual-cased",
|
|
151
|
+
"rescale_with_baseline": False,
|
|
152
|
+
"device": "cuda" if torch.cuda.is_available() else "cpu",
|
|
153
|
+
}
|
|
154
|
+
)
|
|
155
|
+
+ get_vietnamese_generative_harms_metric_specs(include_generative_harms_metrics=True),
|
|
156
|
+
groups=["melt", "melt_summarization_vietnews"],
|
|
157
|
+
)
|
|
158
|
+
|
|
159
|
+
|
|
160
|
+
@run_spec_function("melt_summarization_wikilingua")
|
|
161
|
+
def get_melt_summarization_wikilingua_spec(prompt_style: str = "normal", temperature: float = 1.0) -> RunSpec:
|
|
162
|
+
scenario_spec = ScenarioSpec(
|
|
163
|
+
class_name="helm.benchmark.scenarios.melt_scenarios.MELTSummarizationWikilinguaScenario",
|
|
164
|
+
args={
|
|
165
|
+
"train_min_length": 64,
|
|
166
|
+
"train_max_length": 256,
|
|
167
|
+
"doc_max_length": 2048,
|
|
168
|
+
},
|
|
169
|
+
)
|
|
170
|
+
|
|
171
|
+
if prompt_style == "weak":
|
|
172
|
+
instruction = ""
|
|
173
|
+
elif prompt_style == "medium":
|
|
174
|
+
instruction = "Nhiệm vụ của bạn là tóm tắt đoạn văn bản sau, đưa ra câu trả lời là bản tóm tắt."
|
|
175
|
+
elif prompt_style == "normal":
|
|
176
|
+
instruction = (
|
|
177
|
+
"Bạn là một trợ lý hữu dụng, biết tôn trọng và thành thật. Bạn luôn luôn trả lời các câu hỏi một cách "
|
|
178
|
+
"có ích nhiều nhất có thể, nhưng đồng thời phải an toàn. Câu trả lời của bạn không được bao gồm "
|
|
179
|
+
"các ngôn từ độc hại, phân biệt chủng tộc, phân biệt giới tính, nguy hiểm, nội dung vi phạm pháp luật. "
|
|
180
|
+
"Nhiệm vụ của bạn là tóm tắt đoạn văn bản nằm trong triple backtick. Bài tóm tắt phải đầy đủ các thông tin "
|
|
181
|
+
"quan trọng, ngắn gọn và thu hút người đọc. Ngôn ngữ bạn phải sử dụng để tóm tắt là tiếng Việt."
|
|
182
|
+
)
|
|
183
|
+
else:
|
|
184
|
+
raise ValueError(f"Invalid prompt style: {prompt_style}")
|
|
185
|
+
|
|
186
|
+
adapter_spec = get_generation_adapter_spec(
|
|
187
|
+
instructions=instruction,
|
|
188
|
+
input_noun="Đoạn văn",
|
|
189
|
+
output_noun="Tóm tắt đoạn văn trên",
|
|
190
|
+
max_tokens=256,
|
|
191
|
+
temperature=temperature,
|
|
192
|
+
)
|
|
193
|
+
|
|
194
|
+
return RunSpec(
|
|
195
|
+
name=f"melt_summarization_wikilingua:prompt_style={prompt_style},temperature={temperature}",
|
|
196
|
+
scenario_spec=scenario_spec,
|
|
197
|
+
adapter_spec=adapter_spec,
|
|
198
|
+
metric_specs=get_summarization_metric_specs(
|
|
199
|
+
{
|
|
200
|
+
"task": "summarization_wikilingua",
|
|
201
|
+
"language": "vi",
|
|
202
|
+
"bertscore_model": "bert-base-multilingual-cased",
|
|
203
|
+
"rescale_with_baseline": False,
|
|
204
|
+
"device": "cuda" if torch.cuda.is_available() else "cpu",
|
|
205
|
+
}
|
|
206
|
+
)
|
|
207
|
+
+ get_vietnamese_generative_harms_metric_specs(include_generative_harms_metrics=True),
|
|
208
|
+
groups=["melt", "melt_summarization_wikilingua"],
|
|
209
|
+
)
|
|
210
|
+
|
|
211
|
+
|
|
212
|
+
@run_spec_function("melt_synthetic_reasoning")
|
|
213
|
+
def get_melt_synthetic_reasoning_spec(mode: str) -> RunSpec:
|
|
214
|
+
scenario_spec = ScenarioSpec(
|
|
215
|
+
class_name="helm.benchmark.scenarios.melt_synthetic_reasoning_scenario.MELTSyntheticReasoningScenario",
|
|
216
|
+
args={"mode": mode},
|
|
217
|
+
)
|
|
218
|
+
|
|
219
|
+
adapter_spec = get_generation_adapter_spec(
|
|
220
|
+
instructions="Hãy giải bài toán sau.",
|
|
221
|
+
input_noun="Bài toán",
|
|
222
|
+
output_noun="Lời giải",
|
|
223
|
+
max_train_instances=5,
|
|
224
|
+
stop_sequences=["\n"],
|
|
225
|
+
max_tokens=50, # answer upperbounded by 50 tokens
|
|
226
|
+
)
|
|
227
|
+
|
|
228
|
+
return RunSpec(
|
|
229
|
+
name=f"melt_synthetic_reasoning:mode={mode}",
|
|
230
|
+
scenario_spec=scenario_spec,
|
|
231
|
+
adapter_spec=adapter_spec,
|
|
232
|
+
metric_specs=get_exact_match_metric_specs()
|
|
233
|
+
+ get_vietnamese_generative_harms_metric_specs(include_generative_harms_metrics=True),
|
|
234
|
+
groups=["melt", "melt_synthetic_reasoning", f"melt_synthetic_reasoning_{mode}"],
|
|
235
|
+
)
|
|
236
|
+
|
|
237
|
+
|
|
238
|
+
@run_spec_function("melt_synthetic_reasoning_natural")
|
|
239
|
+
def get_melt_synthetic_reasoning_natural_spec(difficulty: str) -> RunSpec:
|
|
240
|
+
scenario_spec = ScenarioSpec(
|
|
241
|
+
class_name="helm.benchmark.scenarios.melt_srn_scenario.MELTSRNScenario",
|
|
242
|
+
args={"difficulty": difficulty},
|
|
243
|
+
)
|
|
244
|
+
|
|
245
|
+
adapter_spec = get_generation_adapter_spec(
|
|
246
|
+
instructions="Hãy giải quyết vấn đề sau.",
|
|
247
|
+
input_noun="Quy luật",
|
|
248
|
+
newline_after_input_noun=True,
|
|
249
|
+
output_noun=None,
|
|
250
|
+
max_train_instances=3, # limited by the context length
|
|
251
|
+
max_tokens=20,
|
|
252
|
+
)
|
|
253
|
+
srn_metric_specs = get_basic_metric_specs(["f1_set_match", "iou_set_match", "exact_set_match"])
|
|
254
|
+
|
|
255
|
+
return RunSpec(
|
|
256
|
+
name=f"melt_synthetic_reasoning_natural:difficulty={difficulty}",
|
|
257
|
+
scenario_spec=scenario_spec,
|
|
258
|
+
adapter_spec=adapter_spec,
|
|
259
|
+
metric_specs=srn_metric_specs
|
|
260
|
+
+ get_vietnamese_generative_harms_metric_specs(include_generative_harms_metrics=True),
|
|
261
|
+
groups=["melt", "melt_synthetic_reasoning", "melt_synthetic_reasoning_natural"],
|
|
262
|
+
)
|
|
263
|
+
|
|
264
|
+
|
|
265
|
+
@run_spec_function("melt_math")
|
|
266
|
+
def get_math_spec(
|
|
267
|
+
subject: str,
|
|
268
|
+
level: str,
|
|
269
|
+
use_official_examples: str = "False",
|
|
270
|
+
use_chain_of_thought: str = "False",
|
|
271
|
+
) -> RunSpec:
|
|
272
|
+
# Convert to bools and remove the str versions
|
|
273
|
+
use_official_examples_bool: bool = use_official_examples.lower() == "true"
|
|
274
|
+
use_chain_of_thought_bool: bool = use_chain_of_thought.lower() == "true"
|
|
275
|
+
del use_official_examples
|
|
276
|
+
del use_chain_of_thought
|
|
277
|
+
|
|
278
|
+
if use_chain_of_thought_bool:
|
|
279
|
+
assert not use_official_examples_bool, "Cannot use official examples when use_chain_of_thought is True."
|
|
280
|
+
scenario_spec = ScenarioSpec(
|
|
281
|
+
class_name="helm.benchmark.scenarios.melt_scenarios.MELTMATHScenario",
|
|
282
|
+
args={
|
|
283
|
+
"subject": subject,
|
|
284
|
+
"level": level,
|
|
285
|
+
"use_official_examples": use_official_examples_bool,
|
|
286
|
+
"use_chain_of_thought": use_chain_of_thought_bool,
|
|
287
|
+
},
|
|
288
|
+
)
|
|
289
|
+
|
|
290
|
+
if use_chain_of_thought_bool: # Include the solution in the output as per https://arxiv.org/abs/2201.11903
|
|
291
|
+
output_prefix = "Lời giải: " # Don't include LaTeX '$' delimiters
|
|
292
|
+
output_suffix = "\n"
|
|
293
|
+
instance_prefix = "###\n" # Don't include LaTeX '$' delimiters
|
|
294
|
+
max_tokens = 400 # Increase the number of tokens to generate
|
|
295
|
+
stop_sequences = ["###"] # Break at the next instance; extraneous output will be stripped out
|
|
296
|
+
groups = ["melt_math_chain_of_thought"]
|
|
297
|
+
else:
|
|
298
|
+
output_prefix = "Lời giải: $"
|
|
299
|
+
output_suffix = "$\n"
|
|
300
|
+
instance_prefix = "###\n"
|
|
301
|
+
max_tokens = 20
|
|
302
|
+
stop_sequences = ["$"] # Break at the nearest LaTeX closing delimiter
|
|
303
|
+
groups = ["melt_math_regular"]
|
|
304
|
+
|
|
305
|
+
adapter_spec = AdapterSpec(
|
|
306
|
+
method=ADAPT_GENERATION,
|
|
307
|
+
instructions="Cho một bài toán, hãy tìm ra lời giải. Rút gọn câu trả lời của bạn càng nhiều càng tốt.\n",
|
|
308
|
+
max_train_instances=8,
|
|
309
|
+
num_outputs=1,
|
|
310
|
+
temperature=0.0,
|
|
311
|
+
stop_sequences=stop_sequences,
|
|
312
|
+
max_tokens=max_tokens,
|
|
313
|
+
input_prefix="Bài toán: ",
|
|
314
|
+
input_suffix="\n",
|
|
315
|
+
output_prefix=output_prefix,
|
|
316
|
+
output_suffix=output_suffix,
|
|
317
|
+
instance_prefix=instance_prefix,
|
|
318
|
+
)
|
|
319
|
+
|
|
320
|
+
return RunSpec(
|
|
321
|
+
name=f"melt_math:subject={subject},level={level},"
|
|
322
|
+
f"use_official_examples={use_official_examples_bool},use_chain_of_thought={use_chain_of_thought_bool}",
|
|
323
|
+
scenario_spec=scenario_spec,
|
|
324
|
+
adapter_spec=adapter_spec,
|
|
325
|
+
metric_specs=get_basic_metric_specs(
|
|
326
|
+
["math_equiv_chain_of_thought" if use_chain_of_thought_bool else "math_equiv"]
|
|
327
|
+
)
|
|
328
|
+
+ get_vietnamese_generative_harms_metric_specs(include_generative_harms_metrics=True),
|
|
329
|
+
groups=["melt", "melt_math"] + groups,
|
|
330
|
+
)
|
|
331
|
+
|
|
332
|
+
|
|
333
|
+
@run_spec_function("melt_translation_opus100")
|
|
334
|
+
def get_melt_translation_opus100_spec(language_pair: str, max_train_instances: int = 1) -> RunSpec:
|
|
335
|
+
FULL_LANGUAGE_NAMES = {
|
|
336
|
+
"vi": "Vietnamese",
|
|
337
|
+
"en": "English",
|
|
338
|
+
}
|
|
339
|
+
source_language, target_language = language_pair.split("-")
|
|
340
|
+
|
|
341
|
+
scenario_spec = ScenarioSpec(
|
|
342
|
+
class_name="helm.benchmark.scenarios.melt_translation_scenario.MELTTranslationOPUS100Scenario",
|
|
343
|
+
args={"source_language": source_language, "target_language": target_language},
|
|
344
|
+
)
|
|
345
|
+
|
|
346
|
+
adapter_spec = get_machine_translation_adapter_spec(
|
|
347
|
+
source_language=FULL_LANGUAGE_NAMES[source_language],
|
|
348
|
+
target_language=FULL_LANGUAGE_NAMES[target_language],
|
|
349
|
+
max_train_instances=max_train_instances,
|
|
350
|
+
)
|
|
351
|
+
|
|
352
|
+
return RunSpec(
|
|
353
|
+
name=(f"melt_translation_opus100:language_pair={language_pair}"),
|
|
354
|
+
scenario_spec=scenario_spec,
|
|
355
|
+
adapter_spec=adapter_spec,
|
|
356
|
+
metric_specs=get_open_ended_generation_metric_specs()
|
|
357
|
+
+ get_vietnamese_generative_harms_metric_specs(include_generative_harms_metrics=True),
|
|
358
|
+
groups=["melt", "melt_translation_opus100"],
|
|
359
|
+
)
|
|
360
|
+
|
|
361
|
+
|
|
362
|
+
@run_spec_function("melt_translation_phomt")
|
|
363
|
+
def get_melt_translation_phomt_spec(language_pair: str, max_train_instances: int = 1) -> RunSpec:
|
|
364
|
+
FULL_LANGUAGE_NAMES = {
|
|
365
|
+
"vi": "Vietnamese",
|
|
366
|
+
"en": "English",
|
|
367
|
+
}
|
|
368
|
+
source_language, target_language = language_pair.split("-")
|
|
369
|
+
|
|
370
|
+
scenario_spec = ScenarioSpec(
|
|
371
|
+
class_name="helm.benchmark.scenarios.melt_translation_scenario.MELTTranslationPhoMTScenario",
|
|
372
|
+
args={"source_language": source_language, "target_language": target_language},
|
|
373
|
+
)
|
|
374
|
+
|
|
375
|
+
adapter_spec = get_machine_translation_adapter_spec(
|
|
376
|
+
source_language=FULL_LANGUAGE_NAMES[source_language],
|
|
377
|
+
target_language=FULL_LANGUAGE_NAMES[target_language],
|
|
378
|
+
max_train_instances=max_train_instances,
|
|
379
|
+
)
|
|
380
|
+
|
|
381
|
+
return RunSpec(
|
|
382
|
+
name=(f"melt_translation_phomt:language_pair={language_pair}"),
|
|
383
|
+
scenario_spec=scenario_spec,
|
|
384
|
+
adapter_spec=adapter_spec,
|
|
385
|
+
metric_specs=get_open_ended_generation_metric_specs()
|
|
386
|
+
+ get_vietnamese_generative_harms_metric_specs(include_generative_harms_metrics=True),
|
|
387
|
+
groups=["melt", "melt_translation_phomt"],
|
|
388
|
+
)
|
|
389
|
+
|
|
390
|
+
|
|
391
|
+
@run_spec_function("melt_lm_mask_filling_mlqa")
|
|
392
|
+
def get_melt_lm_mask_filling_mlqaa_spec() -> RunSpec:
|
|
393
|
+
scenario_spec = ScenarioSpec(class_name="helm.benchmark.scenarios.melt_lm_scenarios.MELTLMMaskFillingMLQAScenario")
|
|
394
|
+
|
|
395
|
+
instruction = (
|
|
396
|
+
"Hãy hoàn thành câu sau bằng cách điền vào các vị trí trống được đánh dấu bằng [MASK]. "
|
|
397
|
+
"Chỉ trả lời bằng câu đã hoàn thành và không thêm gì khác."
|
|
398
|
+
)
|
|
399
|
+
adapter_spec = get_generation_adapter_spec(
|
|
400
|
+
instructions=instruction,
|
|
401
|
+
input_noun="Câu có chỗ trống",
|
|
402
|
+
output_noun="Câu đã hoàn thành",
|
|
403
|
+
num_outputs=1,
|
|
404
|
+
max_tokens=1024,
|
|
405
|
+
temperature=0.0,
|
|
406
|
+
)
|
|
407
|
+
|
|
408
|
+
return RunSpec(
|
|
409
|
+
name="melt_lm_mask_filling_mlqa",
|
|
410
|
+
scenario_spec=scenario_spec,
|
|
411
|
+
adapter_spec=adapter_spec,
|
|
412
|
+
metric_specs=get_exact_match_metric_specs() + get_f1_metric_specs(),
|
|
413
|
+
groups=["melt", "melt_lm_mask_filling_mlqa"],
|
|
414
|
+
)
|
|
415
|
+
|
|
416
|
+
|
|
417
|
+
@run_spec_function("melt_lm_spelling_correction_vsec")
|
|
418
|
+
def get_melt_lm_spelling_correction_vsec_spec() -> RunSpec:
|
|
419
|
+
scenario_spec = ScenarioSpec(
|
|
420
|
+
class_name="helm.benchmark.scenarios.melt_lm_scenarios.MELTLMSpellingCorrectionVSECScenario"
|
|
421
|
+
)
|
|
422
|
+
|
|
423
|
+
instruction = "Hãy sửa lỗi chính tả trong câu sau. Chỉ trả lời bằng câu đã sửa đúng chính tả và không thêm gì khác."
|
|
424
|
+
adapter_spec = get_generation_adapter_spec(
|
|
425
|
+
instructions=instruction,
|
|
426
|
+
input_noun="Câu có lỗi",
|
|
427
|
+
output_noun="Câu đã sửa",
|
|
428
|
+
num_outputs=1,
|
|
429
|
+
max_tokens=1024,
|
|
430
|
+
temperature=0.0,
|
|
431
|
+
)
|
|
432
|
+
|
|
433
|
+
return RunSpec(
|
|
434
|
+
name="melt_lm_spelling_correction_vsec",
|
|
435
|
+
scenario_spec=scenario_spec,
|
|
436
|
+
adapter_spec=adapter_spec,
|
|
437
|
+
metric_specs=get_exact_match_metric_specs() + get_f1_metric_specs(),
|
|
438
|
+
groups=["melt", "melt_lm_spelling_correction_vsec"],
|
|
439
|
+
)
|
|
440
|
+
|
|
441
|
+
|
|
442
|
+
@run_spec_function("melt_text_classification_vsmec")
|
|
443
|
+
def get_melt_text_classification_vsmec_spec() -> RunSpec:
|
|
444
|
+
scenario_spec = ScenarioSpec(
|
|
445
|
+
class_name="helm.benchmark.scenarios.melt_scenarios.MELTTextClassificationVSMECScenario"
|
|
446
|
+
)
|
|
447
|
+
|
|
448
|
+
instruction = (
|
|
449
|
+
"Hãy phân loại cảm xúc của bình luận sau vào một trong các nhóm: "
|
|
450
|
+
"sadness, surprise, disgust, fear, anger, enjoyment, other."
|
|
451
|
+
)
|
|
452
|
+
|
|
453
|
+
adapter_spec = get_generation_adapter_spec(
|
|
454
|
+
instructions=instruction,
|
|
455
|
+
input_noun="Bình luận",
|
|
456
|
+
output_noun="Loại cảm xúc",
|
|
457
|
+
max_tokens=50,
|
|
458
|
+
multi_label=False,
|
|
459
|
+
)
|
|
460
|
+
|
|
461
|
+
return RunSpec(
|
|
462
|
+
name="melt_text_classification_vsmec",
|
|
463
|
+
scenario_spec=scenario_spec,
|
|
464
|
+
adapter_spec=adapter_spec,
|
|
465
|
+
metric_specs=get_exact_match_metric_specs()
|
|
466
|
+
+ get_classification_metric_specs(
|
|
467
|
+
labels=["sadness", "surprise", "disgust", "fear", "anger", "enjoyment", "other"]
|
|
468
|
+
),
|
|
469
|
+
groups=["melt", "melt_text_classification_vsmec"],
|
|
470
|
+
)
|
|
471
|
+
|
|
472
|
+
|
|
473
|
+
@run_spec_function("melt_text_classification_phoatis")
|
|
474
|
+
def get_melt_text_classification_phoatis_spec() -> RunSpec:
|
|
475
|
+
scenario_spec = ScenarioSpec(
|
|
476
|
+
class_name="helm.benchmark.scenarios.melt_scenarios.MELTTextClassificationPhoATISScenario"
|
|
477
|
+
)
|
|
478
|
+
|
|
479
|
+
instruction = (
|
|
480
|
+
"Hãy phân loại yêu cầu của khách hàng vào trong các nhóm sau: "
|
|
481
|
+
"flight, airfare, ground_service, day_name, meal, airport, airline, flight_time, city, "
|
|
482
|
+
"ground_fare, quantity, abbreviation, distance, aircraft, capacity, flight_no, restriction. "
|
|
483
|
+
"Yêu cầu của khách hàng có thể thuộc tối đa 2 loại và phân biệt nhau bằng dấu phẩy."
|
|
484
|
+
)
|
|
485
|
+
|
|
486
|
+
adapter_spec = get_generation_adapter_spec(
|
|
487
|
+
instructions=instruction,
|
|
488
|
+
input_noun="Yêu cầu của khách hàng",
|
|
489
|
+
output_noun="Loại yêu cầu",
|
|
490
|
+
max_tokens=50,
|
|
491
|
+
multi_label=True,
|
|
492
|
+
)
|
|
493
|
+
|
|
494
|
+
return RunSpec(
|
|
495
|
+
name="melt_text_classification_phoatis",
|
|
496
|
+
scenario_spec=scenario_spec,
|
|
497
|
+
adapter_spec=adapter_spec,
|
|
498
|
+
metric_specs=get_exact_match_metric_specs()
|
|
499
|
+
+ get_classification_metric_specs(
|
|
500
|
+
delimiter=",",
|
|
501
|
+
labels=[
|
|
502
|
+
"flight",
|
|
503
|
+
"airfare",
|
|
504
|
+
"ground_service",
|
|
505
|
+
"day_name",
|
|
506
|
+
"meal",
|
|
507
|
+
"airport",
|
|
508
|
+
"airline",
|
|
509
|
+
"flight_time",
|
|
510
|
+
"city",
|
|
511
|
+
"ground_fare",
|
|
512
|
+
"quantity",
|
|
513
|
+
"abbreviation",
|
|
514
|
+
"distance",
|
|
515
|
+
"aircraft",
|
|
516
|
+
"capacity",
|
|
517
|
+
"flight_no",
|
|
518
|
+
"restriction",
|
|
519
|
+
],
|
|
520
|
+
),
|
|
521
|
+
groups=["melt", "melt_text_classification_phoatis"],
|
|
522
|
+
)
|
|
523
|
+
|
|
524
|
+
|
|
525
|
+
@run_spec_function("melt_sentiment_analysis_vlsp")
|
|
526
|
+
def get_melt_sentiment_analysis_vlsp_spec() -> RunSpec:
|
|
527
|
+
scenario_spec = ScenarioSpec(
|
|
528
|
+
class_name="helm.benchmark.scenarios.melt_scenarios.MELTTSentimentAnalysisVLSPScenario"
|
|
529
|
+
)
|
|
530
|
+
|
|
531
|
+
instruction = "Hãy phân tích quan điểm của nhận xét sau vào một trong các nhóm: negative, neutral, positive."
|
|
532
|
+
|
|
533
|
+
adapter_spec = get_generation_adapter_spec(
|
|
534
|
+
instructions=instruction,
|
|
535
|
+
input_noun="Nhận xét",
|
|
536
|
+
output_noun="Quan điểm",
|
|
537
|
+
max_tokens=50,
|
|
538
|
+
multi_label=False,
|
|
539
|
+
)
|
|
540
|
+
|
|
541
|
+
return RunSpec(
|
|
542
|
+
name="melt_sentiment_analysis_vlsp",
|
|
543
|
+
scenario_spec=scenario_spec,
|
|
544
|
+
adapter_spec=adapter_spec,
|
|
545
|
+
metric_specs=get_exact_match_metric_specs()
|
|
546
|
+
+ get_classification_metric_specs(labels=["negative", "neutral", "positive"]),
|
|
547
|
+
groups=["melt", "melt_sentiment_analysis_vlsp"],
|
|
548
|
+
)
|
|
549
|
+
|
|
550
|
+
|
|
551
|
+
@run_spec_function("melt_sentiment_analysis_vsfc")
|
|
552
|
+
def get_melt_sentiment_analysis_vsfc_spec() -> RunSpec:
|
|
553
|
+
scenario_spec = ScenarioSpec(
|
|
554
|
+
class_name="helm.benchmark.scenarios.melt_scenarios.MELTTSentimentAnalysisVSFCScenario"
|
|
555
|
+
)
|
|
556
|
+
|
|
557
|
+
instruction = "Hãy phân tích quan điểm của nhận xét sau vào một trong các nhóm: negative, neutral, positive."
|
|
558
|
+
|
|
559
|
+
adapter_spec = get_generation_adapter_spec(
|
|
560
|
+
instructions=instruction,
|
|
561
|
+
input_noun="Nhận xét",
|
|
562
|
+
output_noun="Quan điểm",
|
|
563
|
+
max_tokens=50,
|
|
564
|
+
multi_label=False,
|
|
565
|
+
)
|
|
566
|
+
|
|
567
|
+
return RunSpec(
|
|
568
|
+
name="melt_sentiment_analysis_vsfc",
|
|
569
|
+
scenario_spec=scenario_spec,
|
|
570
|
+
adapter_spec=adapter_spec,
|
|
571
|
+
metric_specs=get_exact_match_metric_specs()
|
|
572
|
+
+ get_classification_metric_specs(labels=["negative", "neutral", "positive"]),
|
|
573
|
+
groups=["melt", "melt_sentiment_analysis_vsfc"],
|
|
574
|
+
)
|
|
575
|
+
|
|
576
|
+
|
|
577
|
+
@run_spec_function("melt_knowledge_zalo")
|
|
578
|
+
def get_melt_knowledge_zalo_spec() -> RunSpec:
|
|
579
|
+
scenario_spec = ScenarioSpec(
|
|
580
|
+
class_name="helm.benchmark.scenarios.melt_knowledge_scenario.MELTKnowledgeZaloScenario"
|
|
581
|
+
)
|
|
582
|
+
|
|
583
|
+
instruction = (
|
|
584
|
+
"Hãy trả lời câu hỏi bên dưới bằng cách sử dụng các kiến thức thông thường trong cuộc sống. "
|
|
585
|
+
"Nếu bạn không biết câu trả lời, hãy trả lời 'không có đáp án'."
|
|
586
|
+
)
|
|
587
|
+
|
|
588
|
+
adapter_spec = get_generation_adapter_spec(
|
|
589
|
+
instructions=instruction,
|
|
590
|
+
output_noun="Trả lời",
|
|
591
|
+
max_tokens=128,
|
|
592
|
+
)
|
|
593
|
+
|
|
594
|
+
return RunSpec(
|
|
595
|
+
name="melt_knowledge_zalo",
|
|
596
|
+
scenario_spec=scenario_spec,
|
|
597
|
+
adapter_spec=adapter_spec,
|
|
598
|
+
metric_specs=get_exact_match_metric_specs() + get_f1_metric_specs(),
|
|
599
|
+
groups=["melt", "melt_knowledge_zalo"],
|
|
600
|
+
)
|
|
601
|
+
|
|
602
|
+
|
|
603
|
+
@run_spec_function("melt_knowledge_vimmrc")
|
|
604
|
+
def get_melt_knowledge_vimmrc_spec(randomize_order: bool = False) -> RunSpec:
|
|
605
|
+
scenario_spec = ScenarioSpec(
|
|
606
|
+
class_name="helm.benchmark.scenarios.melt_knowledge_scenario.MELTKnowledgeViMMRCScenario",
|
|
607
|
+
args={"randomize_order": randomize_order},
|
|
608
|
+
)
|
|
609
|
+
|
|
610
|
+
adapter_spec = get_multiple_choice_adapter_spec(
|
|
611
|
+
method=ADAPT_MULTIPLE_CHOICE_JOINT,
|
|
612
|
+
instructions="Sau đây là các câu hỏi trắc nghiệm (có đáp án).",
|
|
613
|
+
input_noun=None,
|
|
614
|
+
output_noun="Trả lời",
|
|
615
|
+
)
|
|
616
|
+
|
|
617
|
+
return RunSpec(
|
|
618
|
+
name="melt_knowledge_vimmrc",
|
|
619
|
+
scenario_spec=scenario_spec,
|
|
620
|
+
adapter_spec=adapter_spec,
|
|
621
|
+
metric_specs=get_exact_match_metric_specs(),
|
|
622
|
+
groups=["melt", "melt_knowledge_vimmrc"],
|
|
623
|
+
)
|
|
624
|
+
|
|
625
|
+
|
|
626
|
+
@run_spec_function("melt_toxicity_detection_vihsd")
|
|
627
|
+
def get_melt_toxicity_detection_vihsd_spec() -> RunSpec:
|
|
628
|
+
scenario_spec = ScenarioSpec(
|
|
629
|
+
class_name="helm.benchmark.scenarios.melt_scenarios.MELTToxicityDetectionViHSDScenario"
|
|
630
|
+
)
|
|
631
|
+
|
|
632
|
+
instruction = (
|
|
633
|
+
"Hãy phân loại độc hại của bình luận sau vào một trong các nhóm:\n"
|
|
634
|
+
"- clean: Không độc hại\n"
|
|
635
|
+
"- offensive: Ngôn từ xúc phạm\n"
|
|
636
|
+
"- hate: Ngôn từ thù địch\n"
|
|
637
|
+
"Chỉ trả lời một trong các nhóm trên mà không cần giải thích thêm."
|
|
638
|
+
)
|
|
639
|
+
|
|
640
|
+
adapter_spec = get_generation_adapter_spec(
|
|
641
|
+
instructions=instruction,
|
|
642
|
+
input_noun="Bình luận",
|
|
643
|
+
output_noun="Phân loại độc hại",
|
|
644
|
+
max_tokens=50,
|
|
645
|
+
multi_label=False,
|
|
646
|
+
)
|
|
647
|
+
|
|
648
|
+
return RunSpec(
|
|
649
|
+
name="melt_toxicity_detection_vihsd",
|
|
650
|
+
scenario_spec=scenario_spec,
|
|
651
|
+
adapter_spec=adapter_spec,
|
|
652
|
+
metric_specs=get_exact_match_metric_specs()
|
|
653
|
+
+ get_classification_metric_specs(labels=["clean", "offensive", "hate"]),
|
|
654
|
+
groups=["melt", "melt_toxicity_detection_vihsd"],
|
|
655
|
+
)
|
|
656
|
+
|
|
657
|
+
|
|
658
|
+
@run_spec_function("melt_toxicity_detection_victsd")
|
|
659
|
+
def get_melt_toxicity_detection_victsd_spec() -> RunSpec:
|
|
660
|
+
scenario_spec = ScenarioSpec(
|
|
661
|
+
class_name="helm.benchmark.scenarios.melt_scenarios.MELTToxicityDetectionViCTSDScenario"
|
|
662
|
+
)
|
|
663
|
+
|
|
664
|
+
instruction = (
|
|
665
|
+
"Hãy phân loại độc hại của bình luận sau vào một trong các nhóm:\n"
|
|
666
|
+
"- clean: Không độc hại\n"
|
|
667
|
+
"- toxic: Ngôn từ độc hại\n"
|
|
668
|
+
"Chỉ trả lời một trong các nhóm trên mà không cần giải thích thêm."
|
|
669
|
+
)
|
|
670
|
+
|
|
671
|
+
adapter_spec = get_generation_adapter_spec(
|
|
672
|
+
instructions=instruction,
|
|
673
|
+
input_noun="Bình luận",
|
|
674
|
+
output_noun="Phân loại độc hại",
|
|
675
|
+
max_tokens=50,
|
|
676
|
+
multi_label=True,
|
|
677
|
+
)
|
|
678
|
+
|
|
679
|
+
return RunSpec(
|
|
680
|
+
name="melt_toxicity_detection_victsd",
|
|
681
|
+
scenario_spec=scenario_spec,
|
|
682
|
+
adapter_spec=adapter_spec,
|
|
683
|
+
metric_specs=get_exact_match_metric_specs() + get_classification_metric_specs(labels=["clean", "toxic"]),
|
|
684
|
+
groups=["melt", "melt_toxicity_detection_victsd"],
|
|
685
|
+
)
|
|
686
|
+
|
|
687
|
+
|
|
688
|
+
@run_spec_function("melt_information_retrieval_mmarco")
|
|
689
|
+
def get_melt_information_retrieval_mmarco_spec(valid_topk: Optional[int] = None) -> RunSpec:
|
|
690
|
+
from helm.benchmark.scenarios.msmarco_scenario import MSMARCOScenario
|
|
691
|
+
|
|
692
|
+
valid_topk = None if valid_topk is None else int(valid_topk)
|
|
693
|
+
scenario_spec = ScenarioSpec(
|
|
694
|
+
class_name="helm.benchmark.scenarios.melt_ir_scenario.MELTInformationRetrievalMMARCOScenario",
|
|
695
|
+
args={"valid_topk": valid_topk},
|
|
696
|
+
)
|
|
697
|
+
|
|
698
|
+
adapter_spec: AdapterSpec = get_ranking_binary_adapter_spec(
|
|
699
|
+
document_noun="Đoạn văn",
|
|
700
|
+
query_noun="Câu hỏi",
|
|
701
|
+
output_prefix="Đoạn văn này có trả lời được câu hỏi không?",
|
|
702
|
+
output_noun="Trả lời",
|
|
703
|
+
stop_sequences=["\n"],
|
|
704
|
+
)
|
|
705
|
+
|
|
706
|
+
# Names of the measures we want to compute.
|
|
707
|
+
measure_names = MSMARCOScenario.MEASURE_NAMES["regular"]
|
|
708
|
+
multiple_relevance_values = set(MSMARCOScenario.GOLD_RELATIONS["regular"]) != {1}
|
|
709
|
+
|
|
710
|
+
metric_specs = (
|
|
711
|
+
[
|
|
712
|
+
MetricSpec(
|
|
713
|
+
class_name="helm.benchmark.metrics.ranking_metrics.RankingMetric",
|
|
714
|
+
args={
|
|
715
|
+
"method": ADAPT_RANKING_BINARY,
|
|
716
|
+
"measure_names": measure_names,
|
|
717
|
+
"correct_output": BinaryRankingAdapter.RANKING_CORRECT_LABEL,
|
|
718
|
+
"wrong_output": BinaryRankingAdapter.RANKING_WRONG_LABEL,
|
|
719
|
+
"rank": valid_topk,
|
|
720
|
+
"multiple_relevance_values": multiple_relevance_values,
|
|
721
|
+
},
|
|
722
|
+
),
|
|
723
|
+
]
|
|
724
|
+
+ get_basic_reference_metric_specs()
|
|
725
|
+
+ get_generic_metric_specs()
|
|
726
|
+
)
|
|
727
|
+
|
|
728
|
+
return RunSpec(
|
|
729
|
+
name=f"melt_information_retrieval_mmarco:valid_topk={valid_topk}",
|
|
730
|
+
scenario_spec=scenario_spec,
|
|
731
|
+
adapter_spec=adapter_spec,
|
|
732
|
+
metric_specs=metric_specs,
|
|
733
|
+
groups=["melt", "melt_information_retrieval_mmarco"],
|
|
734
|
+
)
|
|
735
|
+
|
|
736
|
+
|
|
737
|
+
@run_spec_function("melt_information_retrieval_mrobust")
|
|
738
|
+
def get_melt_information_retrieval_mrobust_spec(valid_topk: Optional[int] = None) -> RunSpec:
|
|
739
|
+
from helm.benchmark.scenarios.msmarco_scenario import MSMARCOScenario
|
|
740
|
+
|
|
741
|
+
valid_topk = None if valid_topk is None else int(valid_topk)
|
|
742
|
+
scenario_spec = ScenarioSpec(
|
|
743
|
+
class_name="helm.benchmark.scenarios.melt_ir_scenario.MELTInformationRetrievalMRobustScenario",
|
|
744
|
+
args={"valid_topk": valid_topk},
|
|
745
|
+
)
|
|
746
|
+
|
|
747
|
+
adapter_spec: AdapterSpec = get_ranking_binary_adapter_spec(
|
|
748
|
+
document_noun="Đoạn văn",
|
|
749
|
+
query_noun="Câu hỏi",
|
|
750
|
+
output_prefix="Đoạn văn này có trả lời được câu hỏi không?",
|
|
751
|
+
output_noun="Trả lời",
|
|
752
|
+
stop_sequences=["\n"],
|
|
753
|
+
)
|
|
754
|
+
|
|
755
|
+
# Names of the measures we want to compute.
|
|
756
|
+
measure_names = MSMARCOScenario.MEASURE_NAMES["trec"]
|
|
757
|
+
multiple_relevance_values = set(MSMARCOScenario.GOLD_RELATIONS["trec"]) != {1}
|
|
758
|
+
|
|
759
|
+
metric_specs = (
|
|
760
|
+
[
|
|
761
|
+
MetricSpec(
|
|
762
|
+
class_name="helm.benchmark.metrics.ranking_metrics.RankingMetric",
|
|
763
|
+
args={
|
|
764
|
+
"method": ADAPT_RANKING_BINARY,
|
|
765
|
+
"measure_names": measure_names,
|
|
766
|
+
"correct_output": BinaryRankingAdapter.RANKING_CORRECT_LABEL,
|
|
767
|
+
"wrong_output": BinaryRankingAdapter.RANKING_WRONG_LABEL,
|
|
768
|
+
"rank": valid_topk,
|
|
769
|
+
"multiple_relevance_values": multiple_relevance_values,
|
|
770
|
+
},
|
|
771
|
+
),
|
|
772
|
+
]
|
|
773
|
+
+ get_basic_reference_metric_specs()
|
|
774
|
+
+ get_generic_metric_specs()
|
|
775
|
+
)
|
|
776
|
+
|
|
777
|
+
return RunSpec(
|
|
778
|
+
name=f"melt_information_retrieval_mrobust:valid_topk={valid_topk}",
|
|
779
|
+
scenario_spec=scenario_spec,
|
|
780
|
+
adapter_spec=adapter_spec,
|
|
781
|
+
metric_specs=metric_specs,
|
|
782
|
+
groups=["melt", "melt_information_retrieval_mrobust"],
|
|
783
|
+
)
|