crfm-helm 0.4.0__py3-none-any.whl → 0.5.0__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.
- {crfm_helm-0.4.0.dist-info → crfm_helm-0.5.0.dist-info}/METADATA +134 -31
- crfm_helm-0.5.0.dist-info/RECORD +642 -0
- {crfm_helm-0.4.0.dist-info → crfm_helm-0.5.0.dist-info}/WHEEL +1 -1
- helm/benchmark/adaptation/adapter_spec.py +31 -3
- helm/benchmark/adaptation/adapters/adapter.py +2 -2
- helm/benchmark/adaptation/adapters/adapter_factory.py +24 -27
- helm/benchmark/adaptation/adapters/generation_adapter.py +1 -0
- helm/benchmark/adaptation/adapters/in_context_learning_adapter.py +20 -4
- helm/benchmark/adaptation/adapters/language_modeling_adapter.py +2 -3
- helm/benchmark/adaptation/adapters/multimodal/multiple_choice_joint_multimodal_adapter.py +104 -0
- helm/benchmark/adaptation/adapters/multimodal/test_in_context_learning_multimodal_adapter.py +2 -1
- helm/benchmark/adaptation/adapters/test_adapter.py +2 -1
- helm/benchmark/adaptation/adapters/test_generation_adapter.py +32 -8
- helm/benchmark/adaptation/adapters/test_language_modeling_adapter.py +7 -19
- helm/benchmark/adaptation/adapters/test_multiple_choice_joint_adapter.py +60 -6
- helm/benchmark/adaptation/common_adapter_specs.py +376 -0
- helm/benchmark/adaptation/request_state.py +6 -1
- helm/benchmark/adaptation/scenario_state.py +6 -2
- helm/benchmark/annotation/annotator.py +43 -0
- helm/benchmark/annotation/annotator_factory.py +61 -0
- helm/benchmark/annotation/image2structure/image_compiler_annotator.py +88 -0
- helm/benchmark/annotation/image2structure/latex_compiler_annotator.py +59 -0
- helm/benchmark/annotation/image2structure/lilypond_compiler_annotator.py +84 -0
- helm/benchmark/annotation/image2structure/webpage_compiler_annotator.py +132 -0
- helm/benchmark/annotation/test_annotator_factory.py +26 -0
- helm/benchmark/annotation/test_dummy_annotator.py +44 -0
- helm/benchmark/annotation_executor.py +124 -0
- helm/benchmark/augmentations/data_augmenter.py +0 -2
- helm/benchmark/augmentations/gender_perturbation.py +1 -1
- helm/benchmark/augmentations/perturbation.py +8 -2
- helm/benchmark/augmentations/perturbation_description.py +1 -1
- helm/benchmark/augmentations/suffix_perturbation.py +29 -0
- helm/benchmark/augmentations/test_perturbation.py +11 -7
- helm/benchmark/augmentations/translate_perturbation.py +30 -0
- helm/benchmark/config_registry.py +7 -1
- helm/benchmark/executor.py +46 -16
- helm/benchmark/huggingface_registration.py +20 -7
- helm/benchmark/metrics/basic_metrics.py +169 -664
- helm/benchmark/metrics/bbq_metrics.py +3 -4
- helm/benchmark/metrics/bias_metrics.py +6 -6
- helm/benchmark/metrics/classification_metrics.py +11 -8
- helm/benchmark/metrics/cleva_accuracy_metrics.py +8 -5
- helm/benchmark/metrics/cleva_harms_metrics.py +2 -2
- helm/benchmark/metrics/code_metrics_helper.py +0 -2
- helm/benchmark/metrics/common_metric_specs.py +167 -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 +4 -110
- helm/benchmark/metrics/dry_run_metrics.py +2 -2
- helm/benchmark/metrics/efficiency_metrics.py +206 -0
- helm/benchmark/metrics/evaluate_instances_metric.py +59 -0
- helm/benchmark/metrics/evaluate_reference_metrics.py +376 -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 +73 -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 +18 -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 +3 -1
- helm/benchmark/metrics/language_modeling_metrics.py +99 -0
- helm/benchmark/metrics/machine_translation_metrics.py +89 -0
- helm/benchmark/metrics/metric.py +93 -172
- helm/benchmark/metrics/metric_name.py +0 -1
- helm/benchmark/metrics/metric_service.py +16 -0
- helm/benchmark/metrics/paraphrase_generation_metrics.py +3 -4
- helm/benchmark/metrics/ranking_metrics.py +2 -2
- helm/benchmark/metrics/reference_metric.py +148 -0
- helm/benchmark/metrics/summac/model_summac.py +0 -2
- helm/benchmark/metrics/summarization_metrics.py +2 -2
- helm/benchmark/metrics/test_classification_metrics.py +8 -5
- helm/benchmark/metrics/test_disinformation_metrics.py +78 -0
- helm/benchmark/metrics/{test_basic_metrics.py → test_evaluate_reference_metrics.py} +5 -1
- helm/benchmark/metrics/test_metric.py +2 -2
- helm/benchmark/metrics/tokens/gooseai_token_cost_estimator.py +10 -2
- helm/benchmark/metrics/toxicity_metrics.py +1 -1
- helm/benchmark/metrics/toxicity_utils.py +23 -0
- helm/benchmark/metrics/unitxt_metrics.py +81 -0
- helm/benchmark/metrics/vision_language/__init__.py +0 -0
- helm/benchmark/metrics/vision_language/emd_utils.py +341 -0
- helm/benchmark/metrics/vision_language/image_metrics.py +450 -0
- helm/benchmark/metrics/vision_language/image_utils.py +100 -0
- helm/benchmark/model_deployment_registry.py +74 -0
- helm/benchmark/model_metadata_registry.py +36 -0
- helm/benchmark/multi_gpu_runner.py +133 -0
- helm/benchmark/presentation/create_plots.py +8 -7
- helm/benchmark/presentation/run_display.py +26 -10
- helm/benchmark/presentation/schema.py +15 -40
- helm/benchmark/presentation/summarize.py +119 -79
- helm/benchmark/presentation/table.py +8 -8
- helm/benchmark/presentation/test_contamination.py +2 -2
- helm/benchmark/presentation/test_run_entry.py +1 -2
- helm/benchmark/presentation/test_summarize.py +3 -3
- helm/benchmark/run.py +54 -26
- helm/benchmark/run_expander.py +214 -16
- helm/benchmark/run_spec.py +93 -0
- helm/benchmark/run_spec_factory.py +162 -0
- helm/benchmark/run_specs/__init__.py +0 -0
- helm/benchmark/run_specs/classic_run_specs.py +1510 -0
- helm/benchmark/run_specs/cleva_run_specs.py +277 -0
- helm/benchmark/run_specs/decodingtrust_run_specs.py +314 -0
- helm/benchmark/run_specs/heim_run_specs.py +623 -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/simple_run_specs.py +104 -0
- helm/benchmark/run_specs/unitxt_run_specs.py +42 -0
- helm/benchmark/run_specs/vlm_run_specs.py +501 -0
- helm/benchmark/runner.py +51 -57
- helm/benchmark/runner_config_registry.py +21 -0
- helm/benchmark/scenarios/bbq_scenario.py +1 -1
- helm/benchmark/scenarios/bold_scenario.py +2 -2
- helm/benchmark/scenarios/code_scenario.py +1 -0
- helm/benchmark/scenarios/decodingtrust_adv_demonstration_scenario.py +169 -0
- helm/benchmark/scenarios/decodingtrust_adv_robustness_scenario.py +121 -0
- helm/benchmark/scenarios/decodingtrust_fairness_scenario.py +77 -0
- helm/benchmark/scenarios/decodingtrust_machine_ethics_scenario.py +324 -0
- helm/benchmark/scenarios/decodingtrust_ood_robustness_scenario.py +204 -0
- helm/benchmark/scenarios/decodingtrust_privacy_scenario.py +559 -0
- helm/benchmark/scenarios/decodingtrust_stereotype_bias_scenario.py +67 -0
- helm/benchmark/scenarios/decodingtrust_toxicity_prompts_scenario.py +78 -0
- helm/benchmark/scenarios/dialogue_scenarios.py +0 -1
- 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_scenario.py +0 -1
- helm/benchmark/scenarios/live_qa_scenario.py +94 -0
- helm/benchmark/scenarios/lm_entry_scenario.py +185 -0
- helm/benchmark/scenarios/math_scenario.py +19 -2
- helm/benchmark/scenarios/medication_qa_scenario.py +60 -0
- helm/benchmark/scenarios/numeracy_scenario.py +1 -1
- helm/benchmark/scenarios/opinions_qa_scenario.py +0 -4
- helm/benchmark/scenarios/scenario.py +4 -0
- helm/benchmark/scenarios/simple_scenarios.py +122 -1
- helm/benchmark/scenarios/test_math_scenario.py +6 -0
- helm/benchmark/scenarios/test_scenario.py +6 -3
- helm/benchmark/scenarios/test_simple_scenarios.py +50 -0
- helm/benchmark/scenarios/thai_exam_scenario.py +135 -0
- helm/benchmark/scenarios/unitxt_scenario.py +56 -0
- helm/benchmark/scenarios/verifiability_judgment_scenario.py +3 -1
- helm/benchmark/scenarios/vicuna_scenario.py +1 -1
- helm/benchmark/scenarios/vision_language/bingo_scenario.py +103 -0
- helm/benchmark/scenarios/vision_language/hateful_memes_scenario.py +92 -0
- helm/benchmark/scenarios/vision_language/heim_human_eval_scenario.py +113 -0
- helm/benchmark/scenarios/vision_language/image2structure/__init__.py +0 -0
- helm/benchmark/scenarios/vision_language/image2structure/chart2csv_scenario.py +55 -0
- helm/benchmark/scenarios/vision_language/image2structure/image2structure_scenario.py +214 -0
- helm/benchmark/scenarios/vision_language/image2structure/latex_scenario.py +25 -0
- helm/benchmark/scenarios/vision_language/image2structure/musicsheet_scenario.py +20 -0
- helm/benchmark/scenarios/vision_language/image2structure/utils_latex.py +347 -0
- helm/benchmark/scenarios/vision_language/image2structure/webpage/__init__.py +0 -0
- helm/benchmark/scenarios/vision_language/image2structure/webpage/driver.py +84 -0
- helm/benchmark/scenarios/vision_language/image2structure/webpage/jekyll_server.py +182 -0
- helm/benchmark/scenarios/vision_language/image2structure/webpage/utils.py +31 -0
- helm/benchmark/scenarios/vision_language/image2structure/webpage_scenario.py +225 -0
- helm/benchmark/scenarios/vision_language/mementos_scenario.py +124 -0
- helm/benchmark/scenarios/vision_language/mme_scenario.py +145 -0
- helm/benchmark/scenarios/vision_language/mmmu_scenario.py +187 -0
- helm/benchmark/scenarios/vision_language/multipanelvqa_scenario.py +169 -0
- helm/benchmark/scenarios/vision_language/pope_scenario.py +104 -0
- helm/benchmark/scenarios/vision_language/seed_bench_scenario.py +129 -0
- helm/benchmark/scenarios/vision_language/unicorn_scenario.py +108 -0
- helm/benchmark/scenarios/vision_language/viz_wiz_scenario.py +1 -2
- helm/benchmark/scenarios/vision_language/vqa_scenario.py +1 -1
- helm/benchmark/scenarios/wmt_14_scenario.py +1 -1
- helm/benchmark/server.py +24 -1
- helm/benchmark/slurm_runner.py +70 -49
- helm/benchmark/static/benchmarking.js +1 -1
- helm/benchmark/static/schema_classic.yaml +258 -1066
- helm/benchmark/static/schema_instruction_following.yaml +210 -0
- helm/benchmark/static/schema_lite.yaml +2 -227
- helm/benchmark/static/schema_mmlu.yaml +1507 -0
- helm/benchmark/static/schema_unitxt.yaml +428 -0
- helm/benchmark/static/schema_vlm.yaml +576 -0
- helm/benchmark/static_build/assets/01-694cb9b7.png +0 -0
- helm/benchmark/static_build/assets/ai21-0eb91ec3.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/crfm-logo-74391ab8.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/heim-logo-3e5e3aa4.png +0 -0
- helm/benchmark/static_build/assets/helm-logo-simple-2ed5400b.png +0 -0
- helm/benchmark/static_build/assets/helmhero-28e90f4d.png +0 -0
- helm/benchmark/static_build/assets/index-5088afcb.css +1 -0
- helm/benchmark/static_build/assets/index-d839df55.js +9 -0
- 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/react-d4a0b69b.js +85 -0
- helm/benchmark/static_build/assets/recharts-6d337683.js +97 -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/tremor-54a99cc4.js +10 -0
- helm/benchmark/static_build/assets/tsinghua-keg-97d4b395.png +0 -0
- helm/benchmark/static_build/assets/vhelm-framework-cde7618a.png +0 -0
- helm/benchmark/static_build/assets/vhelm-model-6d812526.png +0 -0
- helm/benchmark/static_build/assets/yandex-38e09d70.png +0 -0
- helm/benchmark/static_build/config.js +4 -0
- helm/benchmark/static_build/index.html +20 -0
- helm/benchmark/test_data_preprocessor.py +3 -3
- helm/benchmark/test_model_deployment_definition.py +14 -16
- helm/benchmark/test_run_expander.py +1 -1
- helm/benchmark/window_services/ai21_window_service.py +22 -33
- helm/benchmark/window_services/cohere_window_service.py +1 -63
- helm/benchmark/window_services/default_window_service.py +2 -44
- helm/benchmark/window_services/encoder_decoder_window_service.py +0 -11
- helm/benchmark/window_services/ice_window_service.py +0 -34
- helm/benchmark/window_services/image_generation/__init__.py +0 -0
- helm/benchmark/window_services/image_generation/clip_window_service.py +15 -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 +21 -4
- helm/benchmark/window_services/test_anthropic_window_service.py +2 -1
- helm/benchmark/window_services/test_bloom_window_service.py +2 -1
- helm/benchmark/window_services/test_cohere_window_service.py +2 -1
- helm/benchmark/window_services/test_flan_t5_window_service.py +2 -1
- helm/benchmark/window_services/test_gpt2_window_service.py +2 -2
- helm/benchmark/window_services/test_gpt4_window_service.py +2 -1
- helm/benchmark/window_services/test_gptj_window_service.py +3 -2
- helm/benchmark/window_services/test_gptneox_window_service.py +3 -2
- helm/benchmark/window_services/test_ice_window_service.py +2 -1
- helm/benchmark/window_services/test_openai_window_service.py +2 -1
- helm/benchmark/window_services/test_opt_window_service.py +3 -2
- helm/benchmark/window_services/test_palmyra_window_service.py +2 -1
- helm/benchmark/window_services/test_t0pp_window_service.py +2 -1
- helm/benchmark/window_services/test_t511b_window_service.py +2 -1
- helm/benchmark/window_services/test_ul2_window_service.py +2 -1
- helm/benchmark/window_services/test_utils.py +3 -2
- helm/benchmark/window_services/test_yalm_window_service.py +2 -1
- 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 +0 -27
- helm/clients/__init__.py +0 -0
- helm/{proxy/clients → clients}/ai21_client.py +3 -9
- helm/clients/aleph_alpha_client.py +112 -0
- helm/{proxy/clients → clients}/anthropic_client.py +203 -18
- helm/{proxy/clients → clients}/auto_client.py +59 -31
- helm/clients/bedrock_client.py +128 -0
- helm/clients/bedrock_utils.py +72 -0
- helm/{proxy/clients → clients}/client.py +65 -7
- 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 +4 -11
- helm/clients/gcs_client.py +82 -0
- helm/{proxy/clients → clients}/google_client.py +5 -5
- helm/clients/google_translate_client.py +35 -0
- helm/{proxy/clients → clients}/http_model_client.py +5 -7
- helm/{proxy/clients → clients}/huggingface_client.py +43 -64
- helm/clients/image_generation/__init__.py +0 -0
- helm/clients/image_generation/adobe_vision_client.py +78 -0
- helm/clients/image_generation/aleph_alpha_image_generation_client.py +98 -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 +96 -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 +141 -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 +191 -0
- helm/clients/image_generation/dalle2_client.py +192 -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 +190 -0
- helm/clients/image_generation/deep_floyd_client.py +78 -0
- helm/clients/image_generation/huggingface_diffusers_client.py +249 -0
- helm/clients/image_generation/image_generation_client_utils.py +9 -0
- helm/clients/image_generation/lexica_client.py +86 -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 +115 -0
- helm/clients/image_generation/nudity_check_client.py +64 -0
- helm/clients/image_generation/together_image_generation_client.py +111 -0
- helm/{proxy/clients → clients}/lit_gpt_client.py +4 -4
- helm/{proxy/clients → clients}/megatron_client.py +5 -5
- helm/clients/mistral_client.py +134 -0
- helm/clients/moderation_api_client.py +109 -0
- helm/clients/open_lm_client.py +43 -0
- helm/clients/openai_client.py +302 -0
- helm/{proxy/clients → clients}/palmyra_client.py +6 -8
- helm/{proxy/clients → clients}/perspective_api_client.py +7 -8
- helm/clients/simple_client.py +64 -0
- helm/{proxy/clients → clients}/test_auto_client.py +13 -15
- helm/clients/test_client.py +100 -0
- helm/{proxy/clients → clients}/test_huggingface_client.py +15 -16
- helm/clients/test_simple_client.py +19 -0
- helm/{proxy/clients → clients}/test_together_client.py +20 -8
- helm/{proxy/clients → clients}/together_client.py +12 -72
- helm/clients/vertexai_client.py +391 -0
- helm/clients/vision_language/__init__.py +0 -0
- helm/clients/vision_language/huggingface_vlm_client.py +104 -0
- helm/{proxy/clients → clients}/vision_language/idefics_client.py +53 -48
- 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/qwen_vlm_client.py +171 -0
- helm/clients/vllm_client.py +46 -0
- helm/common/cache.py +16 -4
- helm/common/cache_backend_config.py +47 -0
- helm/common/clip_score_request.py +41 -0
- 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 +1 -1
- helm/common/image_generation_parameters.py +25 -0
- helm/common/images_utils.py +24 -1
- helm/common/key_value_store.py +35 -4
- helm/common/media_object.py +13 -0
- helm/common/moderations_api_request.py +71 -0
- helm/common/mongo_key_value_store.py +3 -3
- helm/common/multimodal_request_utils.py +31 -0
- helm/common/nudity_check_request.py +29 -0
- helm/common/request.py +15 -17
- helm/common/test_general.py +6 -0
- helm/common/tokenization_request.py +1 -1
- helm/config/model_deployments.yaml +1069 -546
- helm/config/model_metadata.yaml +753 -31
- helm/config/tokenizer_configs.yaml +142 -43
- helm/proxy/accounts.py +31 -4
- helm/proxy/critique/mechanical_turk_critique_importer.py +3 -0
- helm/proxy/critique/model_critique_client.py +8 -6
- helm/proxy/example_queries.py +29 -17
- helm/proxy/server.py +70 -5
- helm/proxy/services/remote_service.py +31 -0
- helm/proxy/services/server_service.py +96 -16
- helm/proxy/services/service.py +30 -0
- helm/proxy/services/test_remote_service.py +4 -3
- helm/proxy/services/test_service.py +0 -12
- helm/proxy/test_accounts.py +32 -0
- 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/{proxy/tokenizers → tokenizers}/ai21_tokenizer.py +3 -3
- helm/{proxy/tokenizers → tokenizers}/anthropic_tokenizer.py +1 -1
- helm/{proxy/tokenizers → tokenizers}/auto_tokenizer.py +6 -9
- helm/{proxy/tokenizers → tokenizers}/cohere_tokenizer.py +1 -1
- helm/{proxy/tokenizers → tokenizers}/http_model_tokenizer.py +3 -3
- helm/{proxy/tokenizers → tokenizers}/huggingface_tokenizer.py +7 -26
- helm/tokenizers/simple_tokenizer.py +33 -0
- helm/{proxy/tokenizers → tokenizers}/test_anthropic_tokenizer.py +1 -1
- helm/{proxy/tokenizers → tokenizers}/test_huggingface_tokenizer.py +3 -0
- helm/tokenizers/test_simple_tokenizer.py +33 -0
- helm/{proxy/tokenizers → tokenizers}/vertexai_tokenizer.py +1 -1
- helm/{proxy/tokenizers → tokenizers}/yalm_tokenizer.py +5 -3
- helm/tokenizers/yalm_tokenizer_data/__init__.py +0 -0
- 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/RECORD +0 -397
- helm/benchmark/run_specs.py +0 -2762
- helm/benchmark/test_model_properties.py +0 -1570
- helm/benchmark/vlm_run_specs.py +0 -97
- 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_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/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/simple_tokenizer.py +0 -32
- {crfm_helm-0.4.0.dist-info → crfm_helm-0.5.0.dist-info}/LICENSE +0 -0
- {crfm_helm-0.4.0.dist-info → crfm_helm-0.5.0.dist-info}/entry_points.txt +0 -0
- {crfm_helm-0.4.0.dist-info → crfm_helm-0.5.0.dist-info}/top_level.txt +0 -0
- /helm/{proxy/clients → benchmark/annotation}/__init__.py +0 -0
- /helm/{proxy/clients/vision_language → benchmark/annotation/image2structure}/__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/{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/{proxy/tokenizers → tokenizers}/aleph_alpha_tokenizer.py +0 -0
- /helm/{proxy/tokenizers → tokenizers}/caching_tokenizer.py +0 -0
- /helm/{proxy/tokenizers → tokenizers}/ice_tokenizer.py +0 -0
- /helm/{proxy/tokenizers → tokenizers}/lit_gpt_tokenizer.py +0 -0
- /helm/{proxy/tokenizers → tokenizers}/test_ice_tokenizer.py +0 -0
- /helm/{proxy/tokenizers → tokenizers}/test_yalm_tokenizer.py +0 -0
- /helm/{proxy/tokenizers → tokenizers}/tiktoken_tokenizer.py +0 -0
- /helm/{proxy/tokenizers → tokenizers}/tokenizer.py +0 -0
- /helm/{proxy/tokenizers → tokenizers}/yalm_tokenizer_data/test_yalm_tokenizer.py +0 -0
|
@@ -10,129 +10,155 @@
|
|
|
10
10
|
|
|
11
11
|
tokenizer_configs:
|
|
12
12
|
|
|
13
|
-
- name: simple/
|
|
13
|
+
- name: simple/tokenizer1
|
|
14
14
|
tokenizer_spec:
|
|
15
|
-
class_name: "helm.
|
|
15
|
+
class_name: "helm.tokenizers.simple_tokenizer.SimpleTokenizer"
|
|
16
16
|
end_of_text_token: "</s>"
|
|
17
17
|
prefix_token: "<s>"
|
|
18
18
|
|
|
19
19
|
# AI21
|
|
20
20
|
- name: ai21/j1
|
|
21
21
|
tokenizer_spec:
|
|
22
|
-
class_name: "helm.
|
|
22
|
+
class_name: "helm.tokenizers.ai21_tokenizer.AI21Tokenizer"
|
|
23
23
|
end_of_text_token: " "
|
|
24
24
|
prefix_token: ""
|
|
25
25
|
|
|
26
26
|
# AlephAlpha
|
|
27
27
|
- name: AlephAlpha/luminous-base
|
|
28
28
|
tokenizer_spec:
|
|
29
|
-
class_name: "helm.
|
|
29
|
+
class_name: "helm.tokenizers.aleph_alpha_tokenizer.AlephAlphaTokenizer"
|
|
30
30
|
end_of_text_token: ""
|
|
31
31
|
prefix_token: ""
|
|
32
32
|
- name: AlephAlpha/luminous-extended
|
|
33
33
|
tokenizer_spec:
|
|
34
|
-
class_name: "helm.
|
|
34
|
+
class_name: "helm.tokenizers.aleph_alpha_tokenizer.AlephAlphaTokenizer"
|
|
35
35
|
end_of_text_token: ""
|
|
36
36
|
prefix_token: ""
|
|
37
37
|
- name: AlephAlpha/luminous-supreme
|
|
38
38
|
tokenizer_spec:
|
|
39
|
-
class_name: "helm.
|
|
39
|
+
class_name: "helm.tokenizers.aleph_alpha_tokenizer.AlephAlphaTokenizer"
|
|
40
40
|
end_of_text_token: ""
|
|
41
41
|
prefix_token: ""
|
|
42
42
|
- name: AlephAlpha/luminous-world
|
|
43
43
|
tokenizer_spec:
|
|
44
|
-
class_name: "helm.
|
|
44
|
+
class_name: "helm.tokenizers.aleph_alpha_tokenizer.AlephAlphaTokenizer"
|
|
45
45
|
end_of_text_token: ""
|
|
46
46
|
prefix_token: ""
|
|
47
47
|
|
|
48
48
|
# Anthropic
|
|
49
49
|
- name: anthropic/claude
|
|
50
50
|
tokenizer_spec:
|
|
51
|
-
class_name: "helm.
|
|
51
|
+
class_name: "helm.tokenizers.anthropic_tokenizer.AnthropicTokenizer"
|
|
52
52
|
end_of_text_token: "<|endoftext|>"
|
|
53
53
|
prefix_token: "<|endoftext|>"
|
|
54
54
|
|
|
55
55
|
# Bigcode
|
|
56
56
|
- name: bigcode/santacoder
|
|
57
57
|
tokenizer_spec:
|
|
58
|
-
class_name: "helm.
|
|
58
|
+
class_name: "helm.tokenizers.huggingface_tokenizer.HuggingFaceTokenizer"
|
|
59
59
|
end_of_text_token: "<|endoftext|>"
|
|
60
60
|
prefix_token: "<|endoftext|>"
|
|
61
61
|
- name: bigcode/starcoder
|
|
62
62
|
tokenizer_spec:
|
|
63
|
-
class_name: "helm.
|
|
63
|
+
class_name: "helm.tokenizers.huggingface_tokenizer.HuggingFaceTokenizer"
|
|
64
64
|
end_of_text_token: "<|endoftext|>"
|
|
65
65
|
prefix_token: "<|endoftext|>"
|
|
66
66
|
|
|
67
67
|
# Bigscience
|
|
68
68
|
- name: bigscience/bloom
|
|
69
69
|
tokenizer_spec:
|
|
70
|
-
class_name: "helm.
|
|
70
|
+
class_name: "helm.tokenizers.huggingface_tokenizer.HuggingFaceTokenizer"
|
|
71
71
|
end_of_text_token: "</s>"
|
|
72
72
|
prefix_token: "</s>"
|
|
73
73
|
- name: bigscience/T0pp
|
|
74
74
|
tokenizer_spec:
|
|
75
|
-
class_name: "helm.
|
|
75
|
+
class_name: "helm.tokenizers.huggingface_tokenizer.HuggingFaceTokenizer"
|
|
76
76
|
end_of_text_token: "</s>"
|
|
77
77
|
prefix_token: ""
|
|
78
78
|
|
|
79
79
|
# Cohere
|
|
80
80
|
- name: cohere/cohere
|
|
81
81
|
tokenizer_spec:
|
|
82
|
-
class_name: "helm.
|
|
82
|
+
class_name: "helm.tokenizers.cohere_tokenizer.CohereTokenizer"
|
|
83
83
|
end_of_text_token: ""
|
|
84
84
|
prefix_token: ":"
|
|
85
85
|
|
|
86
|
+
# Databricks
|
|
87
|
+
- name: databricks/dbrx-instruct
|
|
88
|
+
tokenizer_spec:
|
|
89
|
+
class_name: "helm.tokenizers.huggingface_tokenizer.HuggingFaceTokenizer"
|
|
90
|
+
end_of_text_token: "<|endoftext|>"
|
|
91
|
+
prefix_token: "<|endoftext|>"
|
|
92
|
+
|
|
93
|
+
# DeepSeek
|
|
94
|
+
- name: deepseek-ai/deepseek-llm-67b-chat
|
|
95
|
+
tokenizer_spec:
|
|
96
|
+
class_name: "helm.tokenizers.huggingface_tokenizer.HuggingFaceTokenizer"
|
|
97
|
+
end_of_text_token: "<|end▁of▁sentence|>"
|
|
98
|
+
prefix_token: "<|begin▁of▁sentence|>"
|
|
99
|
+
|
|
86
100
|
# EleutherAI
|
|
87
101
|
- name: EleutherAI/gpt-j-6B
|
|
88
102
|
tokenizer_spec:
|
|
89
|
-
class_name: "helm.
|
|
103
|
+
class_name: "helm.tokenizers.huggingface_tokenizer.HuggingFaceTokenizer"
|
|
90
104
|
end_of_text_token: "<|endoftext|>"
|
|
91
105
|
prefix_token: "<|endoftext|>"
|
|
92
106
|
- name: EleutherAI/gpt-neox-20b
|
|
93
107
|
tokenizer_spec:
|
|
94
|
-
class_name: "helm.
|
|
108
|
+
class_name: "helm.tokenizers.huggingface_tokenizer.HuggingFaceTokenizer"
|
|
95
109
|
end_of_text_token: "<|endoftext|>"
|
|
96
110
|
prefix_token: "<|endoftext|>"
|
|
97
111
|
|
|
98
112
|
# Facebook
|
|
99
113
|
- name: facebook/opt-66b
|
|
100
114
|
tokenizer_spec:
|
|
101
|
-
class_name: "helm.
|
|
115
|
+
class_name: "helm.tokenizers.huggingface_tokenizer.HuggingFaceTokenizer"
|
|
102
116
|
end_of_text_token: "</s>"
|
|
103
117
|
prefix_token: "</s>"
|
|
104
118
|
|
|
105
119
|
# Google
|
|
106
120
|
- name: google/t5-11b
|
|
107
121
|
tokenizer_spec:
|
|
108
|
-
class_name: "helm.
|
|
122
|
+
class_name: "helm.tokenizers.huggingface_tokenizer.HuggingFaceTokenizer"
|
|
123
|
+
args:
|
|
124
|
+
pretrained_model_name_or_path: google-t5/t5-11b
|
|
109
125
|
end_of_text_token: "</s>"
|
|
110
126
|
prefix_token: ""
|
|
111
127
|
- name: google/flan-t5-xxl
|
|
112
128
|
tokenizer_spec:
|
|
113
|
-
class_name: "helm.
|
|
129
|
+
class_name: "helm.tokenizers.huggingface_tokenizer.HuggingFaceTokenizer"
|
|
114
130
|
end_of_text_token: "</s>"
|
|
115
131
|
prefix_token: ""
|
|
116
132
|
- name: google/ul2
|
|
117
133
|
tokenizer_spec:
|
|
118
|
-
class_name: "helm.
|
|
134
|
+
class_name: "helm.tokenizers.huggingface_tokenizer.HuggingFaceTokenizer"
|
|
119
135
|
end_of_text_token: "</s>"
|
|
120
136
|
prefix_token: ""
|
|
121
137
|
- name: google/mt5-base
|
|
122
138
|
tokenizer_spec:
|
|
123
|
-
class_name: "helm.
|
|
139
|
+
class_name: "helm.tokenizers.huggingface_tokenizer.HuggingFaceTokenizer"
|
|
124
140
|
end_of_text_token: "</s>"
|
|
125
141
|
prefix_token: ""
|
|
126
142
|
- name: google/text-bison@001
|
|
127
143
|
tokenizer_spec:
|
|
128
|
-
class_name: "helm.
|
|
144
|
+
class_name: "helm.tokenizers.vertexai_tokenizer.VertexAITokenizer"
|
|
145
|
+
end_of_text_token: "</s>"
|
|
146
|
+
prefix_token: ""
|
|
147
|
+
- name: google/text-bison@002
|
|
148
|
+
tokenizer_spec:
|
|
149
|
+
class_name: "helm.tokenizers.vertexai_tokenizer.VertexAITokenizer"
|
|
129
150
|
end_of_text_token: "</s>"
|
|
130
151
|
prefix_token: ""
|
|
131
152
|
- name: google/text-unicorn@001
|
|
132
153
|
tokenizer_spec:
|
|
133
|
-
class_name: "helm.
|
|
154
|
+
class_name: "helm.tokenizers.vertexai_tokenizer.VertexAITokenizer"
|
|
134
155
|
end_of_text_token: "</s>"
|
|
135
156
|
prefix_token: ""
|
|
157
|
+
- name: google/gemma-2b
|
|
158
|
+
tokenizer_spec:
|
|
159
|
+
class_name: "helm.tokenizers.huggingface_tokenizer.HuggingFaceTokenizer"
|
|
160
|
+
end_of_text_token: "<eos>"
|
|
161
|
+
prefix_token: "<bos>"
|
|
136
162
|
|
|
137
163
|
# Hf-internal-testing
|
|
138
164
|
|
|
@@ -140,43 +166,51 @@ tokenizer_configs:
|
|
|
140
166
|
# https://huggingface.co/docs/transformers/main/en/model_doc/llama#transformers.LlamaTokenizerFast.example
|
|
141
167
|
- name: hf-internal-testing/llama-tokenizer
|
|
142
168
|
tokenizer_spec:
|
|
143
|
-
class_name: "helm.
|
|
169
|
+
class_name: "helm.tokenizers.huggingface_tokenizer.HuggingFaceTokenizer"
|
|
144
170
|
end_of_text_token: "</s>"
|
|
145
171
|
prefix_token: "<s>"
|
|
146
172
|
|
|
147
173
|
# HuggingFaceM4
|
|
148
174
|
- name: HuggingFaceM4/idefics-9b
|
|
149
175
|
tokenizer_spec:
|
|
150
|
-
class_name: "helm.
|
|
176
|
+
class_name: "helm.tokenizers.huggingface_tokenizer.HuggingFaceTokenizer"
|
|
151
177
|
end_of_text_token: "</s>"
|
|
152
178
|
prefix_token: "<s>"
|
|
153
179
|
- name: HuggingFaceM4/idefics-9b-instruct
|
|
154
180
|
tokenizer_spec:
|
|
155
|
-
class_name: "helm.
|
|
181
|
+
class_name: "helm.tokenizers.huggingface_tokenizer.HuggingFaceTokenizer"
|
|
156
182
|
end_of_text_token: "</s>"
|
|
157
183
|
prefix_token: "<s>"
|
|
158
184
|
- name: HuggingFaceM4/idefics-80b
|
|
159
185
|
tokenizer_spec:
|
|
160
|
-
class_name: "helm.
|
|
186
|
+
class_name: "helm.tokenizers.huggingface_tokenizer.HuggingFaceTokenizer"
|
|
161
187
|
end_of_text_token: "</s>"
|
|
162
188
|
prefix_token: "<s>"
|
|
163
189
|
- name: HuggingFaceM4/idefics-80b-instruct
|
|
164
190
|
tokenizer_spec:
|
|
165
|
-
class_name: "helm.
|
|
191
|
+
class_name: "helm.tokenizers.huggingface_tokenizer.HuggingFaceTokenizer"
|
|
166
192
|
end_of_text_token: "</s>"
|
|
167
193
|
prefix_token: "<s>"
|
|
194
|
+
|
|
195
|
+
- name: anas-awadalla/mpt-7b
|
|
196
|
+
tokenizer_spec:
|
|
197
|
+
class_name: "helm.tokenizers.huggingface_tokenizer.HuggingFaceTokenizer"
|
|
198
|
+
end_of_text_token: "<|endoftext|>"
|
|
199
|
+
prefix_token: ""
|
|
168
200
|
|
|
169
201
|
# Huggingface
|
|
170
202
|
- name: huggingface/gpt2
|
|
171
203
|
tokenizer_spec:
|
|
172
|
-
class_name: "helm.
|
|
204
|
+
class_name: "helm.tokenizers.huggingface_tokenizer.HuggingFaceTokenizer"
|
|
205
|
+
args:
|
|
206
|
+
pretrained_model_name_or_path: openai-community/gpt2
|
|
173
207
|
end_of_text_token: "<|endoftext|>"
|
|
174
208
|
prefix_token: "<|endoftext|>"
|
|
175
209
|
|
|
176
210
|
# Lighting AI
|
|
177
211
|
- name: lightningai/lit-gpt
|
|
178
212
|
tokenizer_spec:
|
|
179
|
-
class_name: "helm.
|
|
213
|
+
class_name: "helm.tokenizers.lit_gpt_tokenizer.LitGPTTokenizer"
|
|
180
214
|
end_of_text_token: "<|endoftext|>"
|
|
181
215
|
prefix_token: "<|endoftext|>"
|
|
182
216
|
|
|
@@ -194,70 +228,135 @@ tokenizer_configs:
|
|
|
194
228
|
# 'https://huggingface.co/models'
|
|
195
229
|
- name: meta-llama/Llama-2-7b-hf
|
|
196
230
|
tokenizer_spec:
|
|
197
|
-
class_name: "helm.
|
|
231
|
+
class_name: "helm.tokenizers.huggingface_tokenizer.HuggingFaceTokenizer"
|
|
198
232
|
end_of_text_token: "</s>"
|
|
199
233
|
prefix_token: "<s>"
|
|
200
234
|
|
|
235
|
+
- name: meta/llama-3-8b
|
|
236
|
+
tokenizer_spec:
|
|
237
|
+
class_name: "helm.tokenizers.huggingface_tokenizer.HuggingFaceTokenizer"
|
|
238
|
+
args:
|
|
239
|
+
pretrained_model_name_or_path: meta-llama/Meta-Llama-3-8B
|
|
240
|
+
prefix_token: "<|begin_of_text|>"
|
|
241
|
+
end_of_text_token: "<|end_of_text|>"
|
|
201
242
|
|
|
202
243
|
# 01-ai
|
|
203
244
|
- name: 01-ai/Yi-6B
|
|
204
245
|
tokenizer_spec:
|
|
205
|
-
class_name: "helm.
|
|
246
|
+
class_name: "helm.tokenizers.huggingface_tokenizer.HuggingFaceTokenizer"
|
|
206
247
|
end_of_text_token: "</s>"
|
|
207
248
|
prefix_token: "<s>"
|
|
208
249
|
|
|
209
|
-
|
|
210
|
-
|
|
211
|
-
|
|
212
|
-
|
|
250
|
+
|
|
251
|
+
# Allen Institute for AI
|
|
252
|
+
# The allenai/olmo-7b requires Python 3.9 or newer.
|
|
253
|
+
# To use the allenai/olmo-7b tokenizer, run `pip install crfm-helm[allenai]` first.
|
|
254
|
+
- name: allenai/olmo-7b
|
|
255
|
+
tokenizer_spec:
|
|
256
|
+
class_name: "helm.tokenizers.huggingface_tokenizer.HuggingFaceTokenizer"
|
|
257
|
+
args:
|
|
258
|
+
trust_remote_code: true
|
|
213
259
|
end_of_text_token: "<|endoftext|>"
|
|
214
|
-
prefix_token: "
|
|
260
|
+
prefix_token: ""
|
|
261
|
+
|
|
262
|
+
|
|
263
|
+
# Microsoft
|
|
264
|
+
- name: microsoft/phi-2
|
|
265
|
+
tokenizer_spec:
|
|
266
|
+
class_name: "helm.tokenizers.huggingface_tokenizer.HuggingFaceTokenizer"
|
|
267
|
+
end_of_text_token: "<|endoftext|>"
|
|
268
|
+
prefix_token: "<|endoftext|>"
|
|
215
269
|
|
|
216
270
|
# Mistralai
|
|
217
271
|
- name: mistralai/Mistral-7B-v0.1
|
|
218
272
|
tokenizer_spec:
|
|
219
|
-
class_name: "helm.
|
|
273
|
+
class_name: "helm.tokenizers.huggingface_tokenizer.HuggingFaceTokenizer"
|
|
220
274
|
end_of_text_token: "</s>"
|
|
221
275
|
prefix_token: "<s>"
|
|
222
276
|
|
|
223
277
|
# Neurips
|
|
224
278
|
- name: neurips/local
|
|
225
279
|
tokenizer_spec:
|
|
226
|
-
class_name: "helm.
|
|
280
|
+
class_name: "helm.tokenizers.http_model_tokenizer.HTTPModelTokenizer"
|
|
227
281
|
end_of_text_token: "<|endoftext|>"
|
|
228
282
|
prefix_token: "<|endoftext|>"
|
|
229
283
|
|
|
230
284
|
# Openai
|
|
231
285
|
- name: openai/cl100k_base
|
|
232
286
|
tokenizer_spec:
|
|
233
|
-
class_name: "helm.
|
|
287
|
+
class_name: "helm.tokenizers.tiktoken_tokenizer.TiktokenTokenizer"
|
|
234
288
|
end_of_text_token: "<|endoftext|>"
|
|
235
289
|
prefix_token: "<|endoftext|>"
|
|
236
290
|
|
|
291
|
+
- name: openai/clip-vit-large-patch14
|
|
292
|
+
tokenizer_spec:
|
|
293
|
+
class_name: "helm.tokenizers.huggingface_tokenizer.HuggingFaceTokenizer"
|
|
294
|
+
end_of_text_token: ""
|
|
295
|
+
prefix_token: ""
|
|
296
|
+
|
|
297
|
+
- name: qwen/qwen-7b
|
|
298
|
+
tokenizer_spec:
|
|
299
|
+
class_name: "helm.tokenizers.huggingface_tokenizer.HuggingFaceTokenizer"
|
|
300
|
+
args:
|
|
301
|
+
pretrained_model_name_or_path: Qwen/Qwen-7B
|
|
302
|
+
trust_remote_code: true
|
|
303
|
+
end_of_text_token: "<|endoftext|>"
|
|
304
|
+
prefix_token: ""
|
|
305
|
+
|
|
306
|
+
- name: qwen/qwen1.5-7b
|
|
307
|
+
tokenizer_spec:
|
|
308
|
+
class_name: "helm.tokenizers.huggingface_tokenizer.HuggingFaceTokenizer"
|
|
309
|
+
args:
|
|
310
|
+
pretrained_model_name_or_path: Qwen/Qwen1.5-7B
|
|
311
|
+
end_of_text_token: "<|endoftext|>"
|
|
312
|
+
prefix_token: ""
|
|
313
|
+
|
|
314
|
+
- name: qwen/qwen-vl
|
|
315
|
+
tokenizer_spec:
|
|
316
|
+
class_name: "helm.tokenizers.huggingface_tokenizer.HuggingFaceTokenizer"
|
|
317
|
+
args:
|
|
318
|
+
pretrained_model_name_or_path: Qwen/Qwen-VL
|
|
319
|
+
trust_remote_code: true
|
|
320
|
+
# Source: https://github.com/QwenLM/Qwen-VL
|
|
321
|
+
end_of_text_token: "<|endoftext|>"
|
|
322
|
+
prefix_token: ""
|
|
323
|
+
|
|
324
|
+
- name: qwen/qwen-vl-chat
|
|
325
|
+
tokenizer_spec:
|
|
326
|
+
class_name: "helm.tokenizers.huggingface_tokenizer.HuggingFaceTokenizer"
|
|
327
|
+
args:
|
|
328
|
+
pretrained_model_name_or_path: Qwen/Qwen-VL-Chat
|
|
329
|
+
trust_remote_code: true
|
|
330
|
+
# Source: https://github.com/QwenLM/Qwen-VL
|
|
331
|
+
end_of_text_token: "<|endoftext|>"
|
|
332
|
+
prefix_token: ""
|
|
333
|
+
|
|
237
334
|
# Tiiuae
|
|
238
335
|
- name: tiiuae/falcon-7b
|
|
239
336
|
tokenizer_spec:
|
|
240
|
-
class_name: "helm.
|
|
337
|
+
class_name: "helm.tokenizers.huggingface_tokenizer.HuggingFaceTokenizer"
|
|
241
338
|
end_of_text_token: "<|endoftext|>"
|
|
242
339
|
prefix_token: ""
|
|
243
340
|
|
|
244
341
|
# TsinghuaKEG
|
|
245
342
|
- name: TsinghuaKEG/ice
|
|
246
343
|
tokenizer_spec:
|
|
247
|
-
class_name: "helm.
|
|
344
|
+
class_name: "helm.tokenizers.ice_tokenizer.ICETokenizer"
|
|
248
345
|
end_of_text_token: "</s>"
|
|
249
346
|
prefix_token: ""
|
|
250
347
|
|
|
251
348
|
# Writer
|
|
252
349
|
- name: writer/gpt2
|
|
253
350
|
tokenizer_spec:
|
|
254
|
-
class_name: "helm.
|
|
351
|
+
class_name: "helm.tokenizers.huggingface_tokenizer.HuggingFaceTokenizer"
|
|
352
|
+
args:
|
|
353
|
+
pretrained_model_name_or_path: openai-community/gpt2
|
|
255
354
|
end_of_text_token: ""
|
|
256
355
|
prefix_token: ""
|
|
257
356
|
|
|
258
357
|
# Yandex
|
|
259
358
|
- name: Yandex/yalm
|
|
260
359
|
tokenizer_spec:
|
|
261
|
-
class_name: "helm.
|
|
360
|
+
class_name: "helm.tokenizers.yalm_tokenizer.YaLMTokenizer"
|
|
262
361
|
end_of_text_token: "</s>"
|
|
263
362
|
prefix_token: "</s>"
|
helm/proxy/accounts.py
CHANGED
|
@@ -23,6 +23,9 @@ DEFAULT_QUOTAS = {
|
|
|
23
23
|
"jurassic": {"daily": 10000},
|
|
24
24
|
"gooseai": {"daily": 10000},
|
|
25
25
|
"cohere": {"daily": 10000},
|
|
26
|
+
"dall_e": {"daily": 5}, # In terms of the number of generated images
|
|
27
|
+
"together_vision": {"daily": 30},
|
|
28
|
+
"simple": {"daily": 10000},
|
|
26
29
|
}
|
|
27
30
|
|
|
28
31
|
|
|
@@ -303,7 +306,7 @@ class Accounts:
|
|
|
303
306
|
model_group: str,
|
|
304
307
|
granularity: str,
|
|
305
308
|
compute_period: Callable[[], str],
|
|
306
|
-
):
|
|
309
|
+
) -> None:
|
|
307
310
|
"""Helper that checks the usage at a certain granularity (e.g., daily, monthly, total)."""
|
|
308
311
|
|
|
309
312
|
model_group_usages = account.usages.get(model_group)
|
|
@@ -321,14 +324,38 @@ class Accounts:
|
|
|
321
324
|
if not usage.can_use():
|
|
322
325
|
raise InsufficientQuotaError(f"{granularity} quota ({usage.quota}) for {model_group} already used up")
|
|
323
326
|
|
|
327
|
+
def check_non_empty_quota(
|
|
328
|
+
account: Account,
|
|
329
|
+
model_group: str,
|
|
330
|
+
) -> None:
|
|
331
|
+
"""Helper that checks that the account has quota at some granularity.
|
|
332
|
+
|
|
333
|
+
At each granularity, a quota of None means unlimited quota.
|
|
334
|
+
However, if the quota is None at every granularity, it means that there is no quota.
|
|
335
|
+
To enforce this rule, this helper raises a InsufficientQuotaError if the quota is None
|
|
336
|
+
at every granularity."""
|
|
337
|
+
model_group_usages = account.usages.get(model_group)
|
|
338
|
+
if model_group_usages is None:
|
|
339
|
+
raise InsufficientQuotaError(f"No quota for {model_group}")
|
|
340
|
+
if all(
|
|
341
|
+
[
|
|
342
|
+
granularity_usage.quota is None or granularity_usage.quota <= 0
|
|
343
|
+
for granularity_usage in model_group_usages.values()
|
|
344
|
+
]
|
|
345
|
+
):
|
|
346
|
+
raise InsufficientQuotaError(f"No quota for {model_group}")
|
|
347
|
+
|
|
324
348
|
if self.root_mode:
|
|
325
349
|
return
|
|
326
350
|
|
|
327
351
|
with SqliteDict(self.path) as cache:
|
|
328
352
|
account: Account = from_dict(Account, cache[api_key])
|
|
329
|
-
|
|
330
|
-
|
|
331
|
-
|
|
353
|
+
if account.is_admin:
|
|
354
|
+
return
|
|
355
|
+
granular_check_can_use(account, model_group, "daily", compute_daily_period)
|
|
356
|
+
granular_check_can_use(account, model_group, "monthly", compute_monthly_period)
|
|
357
|
+
granular_check_can_use(account, model_group, "total", compute_total_period)
|
|
358
|
+
check_non_empty_quota(account, model_group)
|
|
332
359
|
|
|
333
360
|
def use(self, api_key: str, model_group: str, delta: int):
|
|
334
361
|
"""
|
|
@@ -4,6 +4,7 @@ import os
|
|
|
4
4
|
from threading import Lock
|
|
5
5
|
from typing import Dict, List, Optional, Tuple, Union
|
|
6
6
|
import re
|
|
7
|
+
import sys
|
|
7
8
|
|
|
8
9
|
from helm.common.critique_request import (
|
|
9
10
|
CritiqueRequest,
|
|
@@ -15,6 +16,8 @@ from helm.common.critique_request import (
|
|
|
15
16
|
from helm.common.hierarchical_logger import hlog
|
|
16
17
|
from helm.proxy.critique.mechanical_turk_utils import replace_emoji_characters
|
|
17
18
|
|
|
19
|
+
csv.field_size_limit(sys.maxsize)
|
|
20
|
+
|
|
18
21
|
# A representation of fields that can be used as a dict key.
|
|
19
22
|
_CritiqueRequestKey = Tuple[Tuple[str, str], ...]
|
|
20
23
|
|
|
@@ -2,7 +2,7 @@ from typing import Dict, List, Union, Optional
|
|
|
2
2
|
import string
|
|
3
3
|
import dataclasses
|
|
4
4
|
|
|
5
|
-
from helm.benchmark.
|
|
5
|
+
from helm.benchmark.run_spec_factory import get_default_model_deployment_for_model
|
|
6
6
|
from helm.common.critique_request import (
|
|
7
7
|
CritiqueRequest,
|
|
8
8
|
CritiqueRequestResult,
|
|
@@ -12,8 +12,8 @@ from helm.common.critique_request import (
|
|
|
12
12
|
)
|
|
13
13
|
from helm.common.hierarchical_logger import hlog
|
|
14
14
|
from helm.common.optional_dependencies import handle_module_not_found_error
|
|
15
|
-
from helm.common.request import Request, RequestResult,
|
|
16
|
-
from helm.
|
|
15
|
+
from helm.common.request import Request, RequestResult, GeneratedOutput
|
|
16
|
+
from helm.clients.client import Client
|
|
17
17
|
from helm.proxy.critique.critique_client import CritiqueClient
|
|
18
18
|
|
|
19
19
|
|
|
@@ -114,7 +114,7 @@ class ModelCritiqueClient(CritiqueClient):
|
|
|
114
114
|
return answers
|
|
115
115
|
|
|
116
116
|
def _multiple_choice_completion_to_answer(
|
|
117
|
-
self, question: CritiqueQuestionTemplate, completion:
|
|
117
|
+
self, question: CritiqueQuestionTemplate, completion: GeneratedOutput
|
|
118
118
|
) -> Optional[str]:
|
|
119
119
|
"""Convert a multiple choice completion to an answer."""
|
|
120
120
|
assert question.question_type == "multiple_choice"
|
|
@@ -131,7 +131,7 @@ class ModelCritiqueClient(CritiqueClient):
|
|
|
131
131
|
return None
|
|
132
132
|
|
|
133
133
|
def _checkbox_completion_to_answer(
|
|
134
|
-
self, question: CritiqueQuestionTemplate, completion:
|
|
134
|
+
self, question: CritiqueQuestionTemplate, completion: GeneratedOutput
|
|
135
135
|
) -> Optional[List[str]]:
|
|
136
136
|
"""Convert a checkbox completion to an answer."""
|
|
137
137
|
assert question.question_type == "checkbox"
|
|
@@ -147,7 +147,9 @@ class ModelCritiqueClient(CritiqueClient):
|
|
|
147
147
|
hlog(f"Error parsing answer: {e}. Skipping question (and so the respondent entirely)")
|
|
148
148
|
return None
|
|
149
149
|
|
|
150
|
-
def _free_response_completion_to_answer(
|
|
150
|
+
def _free_response_completion_to_answer(
|
|
151
|
+
self, question: CritiqueQuestionTemplate, completion: GeneratedOutput
|
|
152
|
+
) -> str:
|
|
151
153
|
"""Convert a free response completion to an answer."""
|
|
152
154
|
assert question.question_type == "free_response"
|
|
153
155
|
return completion.text
|
helm/proxy/example_queries.py
CHANGED
|
@@ -21,6 +21,8 @@ example_queries = [
|
|
|
21
21
|
"""
|
|
22
22
|
temperature: 0.5 # Medium amount of randomness
|
|
23
23
|
stop_sequences: [.] # Stop when you hit a period
|
|
24
|
+
model: openai/gpt-3.5-turbo-0613
|
|
25
|
+
model_deployment: openai/gpt-3.5-turbo-0613
|
|
24
26
|
"""
|
|
25
27
|
),
|
|
26
28
|
environments="",
|
|
@@ -31,7 +33,9 @@ example_queries = [
|
|
|
31
33
|
"""
|
|
32
34
|
temperature: 0.5 # Medium amount of randomness
|
|
33
35
|
stop_sequences: [\\n] # Stop when you hit a newline
|
|
34
|
-
num_completions:
|
|
36
|
+
num_completions: 5 # Generate many samples
|
|
37
|
+
model: openai/gpt-3.5-turbo-0613
|
|
38
|
+
model_deployment: openai/gpt-3.5-turbo-0613
|
|
35
39
|
"""
|
|
36
40
|
),
|
|
37
41
|
environments="",
|
|
@@ -42,7 +46,9 @@ example_queries = [
|
|
|
42
46
|
"""
|
|
43
47
|
echo_prompt: true # Analyze the prompt
|
|
44
48
|
max_tokens: 0 # Don't generate any more
|
|
45
|
-
top_k_per_token:
|
|
49
|
+
top_k_per_token: 5 # Show alternatives for each position
|
|
50
|
+
model: openai/davinci-002
|
|
51
|
+
model_deployment: openai/davinci-002
|
|
46
52
|
"""
|
|
47
53
|
),
|
|
48
54
|
environments=dedent(""),
|
|
@@ -53,6 +59,8 @@ example_queries = [
|
|
|
53
59
|
"""
|
|
54
60
|
temperature: 0 # Deterministic
|
|
55
61
|
max_tokens: 50
|
|
62
|
+
model: openai/gpt-3.5-turbo-0613
|
|
63
|
+
model_deployment: openai/gpt-3.5-turbo-0613
|
|
56
64
|
"""
|
|
57
65
|
),
|
|
58
66
|
environments="",
|
|
@@ -63,13 +71,15 @@ example_queries = [
|
|
|
63
71
|
"""
|
|
64
72
|
temperature: 0
|
|
65
73
|
stop_sequences: [.]
|
|
66
|
-
|
|
74
|
+
# Try out multiple models
|
|
75
|
+
model: ${model}
|
|
76
|
+
model_deployment: ${model}
|
|
67
77
|
"""
|
|
68
78
|
),
|
|
69
79
|
environments=dedent(
|
|
70
80
|
"""
|
|
71
81
|
occupation: [mathematician, lawyer, doctor]
|
|
72
|
-
|
|
82
|
+
model: [openai/gpt-3.5-turbo-0613, openai/gpt-3.5-turbo-1106]
|
|
73
83
|
"""
|
|
74
84
|
),
|
|
75
85
|
),
|
|
@@ -88,12 +98,14 @@ example_queries = [
|
|
|
88
98
|
temperature: 0.5
|
|
89
99
|
stop_sequences: [\\n]
|
|
90
100
|
num_completions: 5
|
|
91
|
-
|
|
101
|
+
# Try out multiple models
|
|
102
|
+
model: ${model}
|
|
103
|
+
model_deployment: ${model}
|
|
92
104
|
"""
|
|
93
105
|
),
|
|
94
106
|
environments=dedent(
|
|
95
107
|
"""
|
|
96
|
-
|
|
108
|
+
model: [openai/gpt-3.5-turbo-0613, openai/gpt-3.5-turbo-1106]
|
|
97
109
|
"""
|
|
98
110
|
),
|
|
99
111
|
),
|
|
@@ -122,20 +134,23 @@ example_queries = [
|
|
|
122
134
|
temperature: 0
|
|
123
135
|
max_tokens: 1
|
|
124
136
|
top_k_per_token: 4
|
|
125
|
-
|
|
137
|
+
# Try out multiple models
|
|
138
|
+
model: ${model}
|
|
139
|
+
model_deployment: ${model}
|
|
126
140
|
"""
|
|
127
141
|
),
|
|
128
142
|
environments=dedent(
|
|
129
143
|
"""
|
|
130
|
-
|
|
144
|
+
model: [openai/gpt-3.5-turbo-0613, openai/gpt-3.5-turbo-1106]
|
|
131
145
|
"""
|
|
132
146
|
),
|
|
133
147
|
),
|
|
134
148
|
Query(
|
|
135
|
-
prompt="
|
|
149
|
+
prompt="Write a Python function that takes two vectors a and b and returns their Euclidean distance.",
|
|
136
150
|
settings=dedent(
|
|
137
151
|
"""
|
|
138
|
-
|
|
152
|
+
model: openai/gpt-3.5-turbo-0613
|
|
153
|
+
model_deployment: openai/gpt-3.5-turbo-0613
|
|
139
154
|
"""
|
|
140
155
|
),
|
|
141
156
|
environments="",
|
|
@@ -144,19 +159,16 @@ example_queries = [
|
|
|
144
159
|
prompt="The quick brown fox",
|
|
145
160
|
settings=dedent(
|
|
146
161
|
"""
|
|
147
|
-
model_deployment: ${model_deployment}
|
|
148
162
|
temperature: 0.3
|
|
149
163
|
stop_sequences: [\\n]
|
|
164
|
+
# Try out multiple models
|
|
165
|
+
model: ${model}
|
|
166
|
+
model_deployment: ${model}
|
|
150
167
|
"""
|
|
151
168
|
),
|
|
152
169
|
environments=dedent(
|
|
153
170
|
"""
|
|
154
|
-
|
|
155
|
-
"openai/davinci", "openai/text-davinci-002",
|
|
156
|
-
"openai/text-davinci-003", "ai21/j1-grande-v2-beta",
|
|
157
|
-
"together/gpt-j-6b", "together/gpt-jt-6b-v1",
|
|
158
|
-
"together/bloom", "together/opt-175b"
|
|
159
|
-
]
|
|
171
|
+
model: [openai/gpt-3.5-turbo-0613, openai/gpt-3.5-turbo-1106]
|
|
160
172
|
"""
|
|
161
173
|
),
|
|
162
174
|
),
|