Merge pull request #816 from EleutherAI/flan-benchmark

[Refactor] Flan benchmark

lm_eval/tasks/mmlu/flan_n_shot/mmlu_log_medical_genetics.yaml

+"dataset_name": "medical_genetics"
+"description": "The following are multiple choice questions (with answers) about medical genetics.\n\n"
+"include": "_mmlu_flan_loglikelihood_template_yaml"
+"task": "mmlu_flan_n_shot_loglikelihood_medical_genetics"

lm_eval/tasks/mmlu/flan_n_shot/mmlu_log_miscellaneous.yaml

+"dataset_name": "miscellaneous"
+"description": "The following are multiple choice questions (with answers) about miscellaneous.\n\n"
+"include": "_mmlu_flan_loglikelihood_template_yaml"
+"task": "mmlu_flan_n_shot_loglikelihood_miscellaneous"

lm_eval/tasks/mmlu/flan_n_shot/mmlu_log_moral_disputes.yaml

+"dataset_name": "moral_disputes"
+"description": "The following are multiple choice questions (with answers) about moral disputes.\n\n"
+"include": "_mmlu_flan_loglikelihood_template_yaml"
+"task": "mmlu_flan_n_shot_loglikelihood_moral_disputes"

lm_eval/tasks/mmlu/flan_n_shot/mmlu_log_moral_scenarios.yaml

+"dataset_name": "moral_scenarios"
+"description": "The following are multiple choice questions (with answers) about moral scenarios.\n\n"
+"include": "_mmlu_flan_loglikelihood_template_yaml"
+"task": "mmlu_flan_n_shot_loglikelihood_moral_scenarios"

lm_eval/tasks/mmlu/flan_n_shot/mmlu_log_nutrition.yaml

+"dataset_name": "nutrition"
+"description": "The following are multiple choice questions (with answers) about nutrition.\n\n"
+"include": "_mmlu_flan_loglikelihood_template_yaml"
+"task": "mmlu_flan_n_shot_loglikelihood_nutrition"

lm_eval/tasks/mmlu/flan_n_shot/mmlu_log_philosophy.yaml

+"dataset_name": "philosophy"
+"description": "The following are multiple choice questions (with answers) about philosophy.\n\n"
+"include": "_mmlu_flan_loglikelihood_template_yaml"
+"task": "mmlu_flan_n_shot_loglikelihood_philosophy"

lm_eval/tasks/mmlu/flan_n_shot/mmlu_log_prehistory.yaml

+"dataset_name": "prehistory"
+"description": "The following are multiple choice questions (with answers) about prehistory.\n\n"
+"include": "_mmlu_flan_loglikelihood_template_yaml"
+"task": "mmlu_flan_n_shot_loglikelihood_prehistory"

lm_eval/tasks/mmlu/flan_n_shot/mmlu_log_professional_accounting.yaml

+"dataset_name": "professional_accounting"
+"description": "The following are multiple choice questions (with answers) about professional accounting.\n\n"
+"include": "_mmlu_flan_loglikelihood_template_yaml"
+"task": "mmlu_flan_n_shot_loglikelihood_professional_accounting"

lm_eval/tasks/mmlu/flan_n_shot/mmlu_log_professional_law.yaml

+"dataset_name": "professional_law"
+"description": "The following are multiple choice questions (with answers) about professional law.\n\n"
+"include": "_mmlu_flan_loglikelihood_template_yaml"
+"task": "mmlu_flan_n_shot_loglikelihood_professional_law"

lm_eval/tasks/mmlu/flan_n_shot/mmlu_log_professional_medicine.yaml

+"dataset_name": "professional_medicine"
+"description": "The following are multiple choice questions (with answers) about professional medicine.\n\n"
+"include": "_mmlu_flan_loglikelihood_template_yaml"
+"task": "mmlu_flan_n_shot_loglikelihood_professional_medicine"

lm_eval/tasks/mmlu/flan_n_shot/mmlu_log_professional_psychology.yaml

+"dataset_name": "professional_psychology"
+"description": "The following are multiple choice questions (with answers) about professional psychology.\n\n"
+"include": "_mmlu_flan_loglikelihood_template_yaml"
+"task": "mmlu_flan_n_shot_loglikelihood_professional_psychology"

lm_eval/tasks/mmlu/flan_n_shot/mmlu_log_public_relations.yaml

+"dataset_name": "public_relations"
+"description": "The following are multiple choice questions (with answers) about public relations.\n\n"
+"include": "_mmlu_flan_loglikelihood_template_yaml"
+"task": "mmlu_flan_n_shot_loglikelihood_public_relations"

lm_eval/tasks/mmlu/flan_n_shot/mmlu_log_security_studies.yaml

+"dataset_name": "security_studies"
+"description": "The following are multiple choice questions (with answers) about security studies.\n\n"
+"include": "_mmlu_flan_loglikelihood_template_yaml"
+"task": "mmlu_flan_n_shot_loglikelihood_security_studies"

lm_eval/tasks/mmlu/flan_n_shot/mmlu_log_sociology.yaml

+"dataset_name": "sociology"
+"description": "The following are multiple choice questions (with answers) about sociology.\n\n"
+"include": "_mmlu_flan_loglikelihood_template_yaml"
+"task": "mmlu_flan_n_shot_loglikelihood_sociology"

lm_eval/tasks/mmlu/flan_n_shot/mmlu_log_us_foreign_policy.yaml

+"dataset_name": "us_foreign_policy"
+"description": "The following are multiple choice questions (with answers) about us foreign policy.\n\n"
+"include": "_mmlu_flan_loglikelihood_template_yaml"
+"task": "mmlu_flan_n_shot_loglikelihood_us_foreign_policy"

lm_eval/tasks/mmlu/flan_n_shot/mmlu_log_virology.yaml

+"dataset_name": "virology"
+"description": "The following are multiple choice questions (with answers) about virology.\n\n"
+"include": "_mmlu_flan_loglikelihood_template_yaml"
+"task": "mmlu_flan_n_shot_loglikelihood_virology"

lm_eval/tasks/mmlu/flan_n_shot/mmlu_log_world_religions.yaml

+"dataset_name": "world_religions"
+"description": "The following are multiple choice questions (with answers) about world religions.\n\n"
+"include": "_mmlu_flan_loglikelihood_template_yaml"
+"task": "mmlu_flan_n_shot_loglikelihood_world_religions"

lm_eval/tasks/super_glue/boolq/t5-prompt.yaml

@@ -9,6 +9,9 @@ output_type: greedy_until
 doc_to_text: "boolq passage: {{passage}} question: {{question}}"
 doc_to_target: label
 doc_to_choice: ['False', 'True']
+generation_kwargs:
+  until:
+    - "</s>"
 metric_list:
   - metric: exact_match
     aggregation: mean

lm_eval/tasks/super_glue/cb/t5-prompt.yaml

@@ -9,6 +9,9 @@ output_type: greedy_until
 doc_to_text: "cb hypothesis: {{hypothesis}} premise: {{premise}}"
 doc_to_target: label
 doc_to_choice: ['entailment', 'contradiction', 'neutral']
+generation_kwargs:
+  until:
+    - "</s>"
 metric_list:
   - metric: exact_match
     aggregation: mean