Upload full run contents (scripts, checkpoints, and results) (part 5)

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_formal_logic.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_formal_logic
+tag: global_mmlu_full_ar_humanities_tasks
+task: global_mmlu_full_ar_formal_logic

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_global_facts.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_global_facts
+tag: global_mmlu_full_ar_other_tasks
+task: global_mmlu_full_ar_global_facts

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_high_school_biology.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_high_school_biology
+tag: global_mmlu_full_ar_stem_tasks
+task: global_mmlu_full_ar_high_school_biology

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_high_school_chemistry.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_high_school_chemistry
+tag: global_mmlu_full_ar_stem_tasks
+task: global_mmlu_full_ar_high_school_chemistry

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_high_school_computer_science.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_high_school_computer_science
+tag: global_mmlu_full_ar_stem_tasks
+task: global_mmlu_full_ar_high_school_computer_science

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_high_school_european_history.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_high_school_european_history
+tag: global_mmlu_full_ar_humanities_tasks
+task: global_mmlu_full_ar_high_school_european_history

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_high_school_geography.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_high_school_geography
+tag: global_mmlu_full_ar_social_sciences_tasks
+task: global_mmlu_full_ar_high_school_geography

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_high_school_government_and_politics.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_high_school_government_and_politics
+tag: global_mmlu_full_ar_social_sciences_tasks
+task: global_mmlu_full_ar_high_school_government_and_politics

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_high_school_macroeconomics.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_high_school_macroeconomics
+tag: global_mmlu_full_ar_social_sciences_tasks
+task: global_mmlu_full_ar_high_school_macroeconomics

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_high_school_mathematics.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_high_school_mathematics
+tag: global_mmlu_full_ar_stem_tasks
+task: global_mmlu_full_ar_high_school_mathematics

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_high_school_microeconomics.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_high_school_microeconomics
+tag: global_mmlu_full_ar_social_sciences_tasks
+task: global_mmlu_full_ar_high_school_microeconomics

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_high_school_physics.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_high_school_physics
+tag: global_mmlu_full_ar_stem_tasks
+task: global_mmlu_full_ar_high_school_physics

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_high_school_psychology.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_high_school_psychology
+tag: global_mmlu_full_ar_social_sciences_tasks
+task: global_mmlu_full_ar_high_school_psychology

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_high_school_statistics.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_high_school_statistics
+tag: global_mmlu_full_ar_stem_tasks
+task: global_mmlu_full_ar_high_school_statistics

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_high_school_us_history.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_high_school_us_history
+tag: global_mmlu_full_ar_humanities_tasks
+task: global_mmlu_full_ar_high_school_us_history

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_high_school_world_history.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_high_school_world_history
+tag: global_mmlu_full_ar_humanities_tasks
+task: global_mmlu_full_ar_high_school_world_history

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_human_aging.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_human_aging
+tag: global_mmlu_full_ar_other_tasks
+task: global_mmlu_full_ar_human_aging

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_human_sexuality.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_human_sexuality
+tag: global_mmlu_full_ar_social_sciences_tasks
+task: global_mmlu_full_ar_human_sexuality

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_international_law.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_international_law
+tag: global_mmlu_full_ar_humanities_tasks
+task: global_mmlu_full_ar_international_law

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_jurisprudence.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_jurisprudence
+tag: global_mmlu_full_ar_humanities_tasks
+task: global_mmlu_full_ar_jurisprudence

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_logical_fallacies.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_logical_fallacies
+tag: global_mmlu_full_ar_humanities_tasks
+task: global_mmlu_full_ar_logical_fallacies

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_machine_learning.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_machine_learning
+tag: global_mmlu_full_ar_stem_tasks
+task: global_mmlu_full_ar_machine_learning

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_management.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_management
+tag: global_mmlu_full_ar_other_tasks
+task: global_mmlu_full_ar_management

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_marketing.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_marketing
+tag: global_mmlu_full_ar_other_tasks
+task: global_mmlu_full_ar_marketing

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_medical_genetics.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_medical_genetics
+tag: global_mmlu_full_ar_other_tasks
+task: global_mmlu_full_ar_medical_genetics

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_miscellaneous.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_miscellaneous
+tag: global_mmlu_full_ar_other_tasks
+task: global_mmlu_full_ar_miscellaneous

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_moral_disputes.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_moral_disputes
+tag: global_mmlu_full_ar_humanities_tasks
+task: global_mmlu_full_ar_moral_disputes

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_moral_scenarios.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_moral_scenarios
+tag: global_mmlu_full_ar_humanities_tasks
+task: global_mmlu_full_ar_moral_scenarios

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_nutrition.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_nutrition
+tag: global_mmlu_full_ar_other_tasks
+task: global_mmlu_full_ar_nutrition

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_philosophy.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_philosophy
+tag: global_mmlu_full_ar_humanities_tasks
+task: global_mmlu_full_ar_philosophy

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_prehistory.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_prehistory
+tag: global_mmlu_full_ar_humanities_tasks
+task: global_mmlu_full_ar_prehistory

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_professional_accounting.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_professional_accounting
+tag: global_mmlu_full_ar_other_tasks
+task: global_mmlu_full_ar_professional_accounting

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_professional_law.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_professional_law
+tag: global_mmlu_full_ar_humanities_tasks
+task: global_mmlu_full_ar_professional_law

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_professional_medicine.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_professional_medicine
+tag: global_mmlu_full_ar_other_tasks
+task: global_mmlu_full_ar_professional_medicine

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_professional_psychology.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_professional_psychology
+tag: global_mmlu_full_ar_social_sciences_tasks
+task: global_mmlu_full_ar_professional_psychology

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_public_relations.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_public_relations
+tag: global_mmlu_full_ar_social_sciences_tasks
+task: global_mmlu_full_ar_public_relations

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_security_studies.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_security_studies
+tag: global_mmlu_full_ar_social_sciences_tasks
+task: global_mmlu_full_ar_security_studies

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_sociology.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_sociology
+tag: global_mmlu_full_ar_social_sciences_tasks
+task: global_mmlu_full_ar_sociology

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_us_foreign_policy.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_us_foreign_policy
+tag: global_mmlu_full_ar_social_sciences_tasks
+task: global_mmlu_full_ar_us_foreign_policy

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_virology.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_virology
+tag: global_mmlu_full_ar_other_tasks
+task: global_mmlu_full_ar_virology

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/global_mmlu_full_ar_world_religions.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _ar_template_yaml
+process_docs: !function utils.process_world_religions
+tag: global_mmlu_full_ar_humanities_tasks
+task: global_mmlu_full_ar_world_religions

babylm_eval_repo/multilingual/tasks/global_mmlu/full/ar/utils.py

@@ -0,0 +1,73 @@
+from functools import partial
+
+
+SUBJECTS = [
+    "abstract_algebra",
+    "anatomy",
+    "astronomy",
+    "business_ethics",
+    "clinical_knowledge",
+    "college_biology",
+    "college_chemistry",
+    "college_computer_science",
+    "college_mathematics",
+    "college_medicine",
+    "college_physics",
+    "computer_security",
+    "conceptual_physics",
+    "econometrics",
+    "electrical_engineering",
+    "elementary_mathematics",
+    "formal_logic",
+    "global_facts",
+    "high_school_biology",
+    "high_school_chemistry",
+    "high_school_computer_science",
+    "high_school_european_history",
+    "high_school_geography",
+    "high_school_government_and_politics",
+    "high_school_macroeconomics",
+    "high_school_mathematics",
+    "high_school_microeconomics",
+    "high_school_physics",
+    "high_school_psychology",
+    "high_school_statistics",
+    "high_school_us_history",
+    "high_school_world_history",
+    "human_aging",
+    "human_sexuality",
+    "international_law",
+    "jurisprudence",
+    "logical_fallacies",
+    "machine_learning",
+    "management",
+    "marketing",
+    "medical_genetics",
+    "miscellaneous",
+    "moral_disputes",
+    "moral_scenarios",
+    "nutrition",
+    "philosophy",
+    "prehistory",
+    "professional_accounting",
+    "professional_law",
+    "professional_medicine",
+    "professional_psychology",
+    "public_relations",
+    "security_studies",
+    "sociology",
+    "us_foreign_policy",
+    "virology",
+    "world_religions",
+]
+
+
+def process_docs(dataset, subject):
+    return dataset.filter(lambda x: x["subject"] == subject)
+
+
+process_functions = {
+    f"process_{subject}": partial(process_docs, subject=subject) for subject in SUBJECTS
+}
+
+globals().update(process_functions)

babylm_eval_repo/multilingual/tasks/global_mmlu/full/bn/_bn_template_yaml

@@ -0,0 +1,16 @@
+dataset_path: CohereLabs/Global-MMLU
+dataset_name: bn
+test_split: test
+fewshot_split: dev
+fewshot_config:
+  sampler: first_n
+output_type: multiple_choice
+doc_to_text: "{{question.strip()}}\nA. {{option_a}}\nB. {{option_b}}\nC. {{option_c}}\nD. {{option_d}}\nAnswer:"
+doc_to_choice: ["A", "B", "C", "D"]
+doc_to_target: answer
+metric_list:
+  - metric: acc
+    aggregation: mean
+    higher_is_better: true
+metadata:
+  version: 0.0

babylm_eval_repo/multilingual/tasks/global_mmlu/full/bn/_global_mmlu_full_bn.yaml

@@ -0,0 +1,11 @@
+group: global_mmlu_full_bn
+task:
+  - global_mmlu_full_bn_stem
+  - global_mmlu_full_bn_other
+  - global_mmlu_full_bn_social_sciences
+  - global_mmlu_full_bn_humanities
+aggregate_metric_list:
+  - metric: acc
+    weight_by_size: True
+metadata:
+  version: 0.0

babylm_eval_repo/multilingual/tasks/global_mmlu/full/bn/_global_mmlu_full_bn_humanities.yaml

@@ -0,0 +1,8 @@
+group: global_mmlu_full_bn_humanities
+task:
+  - global_mmlu_full_bn_humanities_tasks
+aggregate_metric_list:
+  - metric: acc
+    weight_by_size: true
+metadata:
+  version: 0.0

babylm_eval_repo/multilingual/tasks/global_mmlu/full/bn/_global_mmlu_full_bn_other.yaml

@@ -0,0 +1,8 @@
+group: global_mmlu_full_bn_other
+task:
+  - global_mmlu_full_bn_other_tasks
+aggregate_metric_list:
+  - metric: acc
+    weight_by_size: true
+metadata:
+  version: 0.0

babylm_eval_repo/multilingual/tasks/global_mmlu/full/bn/_global_mmlu_full_bn_social_sciences.yaml

@@ -0,0 +1,8 @@
+group: global_mmlu_full_bn_social_sciences
+task:
+  - global_mmlu_full_bn_social_sciences_tasks
+aggregate_metric_list:
+  - metric: acc
+    weight_by_size: true
+metadata:
+  version: 0.0

babylm_eval_repo/multilingual/tasks/global_mmlu/full/bn/_global_mmlu_full_bn_stem.yaml

@@ -0,0 +1,8 @@
+group: global_mmlu_full_bn_stem
+task:
+  - global_mmlu_full_bn_stem_tasks
+aggregate_metric_list:
+  - metric: acc
+    weight_by_size: true
+metadata:
+  version: 0.0

babylm_eval_repo/multilingual/tasks/global_mmlu/full/bn/global_mmlu_full_bn_abstract_algebra.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _bn_template_yaml
+process_docs: !function utils.process_abstract_algebra
+tag: global_mmlu_full_bn_stem_tasks
+task: global_mmlu_full_bn_abstract_algebra

babylm_eval_repo/multilingual/tasks/global_mmlu/full/bn/global_mmlu_full_bn_anatomy.yaml

@@ -0,0 +1,5 @@
+# Generated by _generate_configs.py
+include: _bn_template_yaml
+process_docs: !function utils.process_anatomy
+tag: global_mmlu_full_bn_stem_tasks
+task: global_mmlu_full_bn_anatomy