tmmluplus / README.md
ikala-avery's picture
Update README.md
c0e8ae9 verified
metadata
license: mit
license_name: mit
task_categories:
  - question-answering
language:
  - zh
tags:
  - traditional chinese
  - finance
  - medical
  - taiwan
  - benchmark
  - zh-tw
  - zh-hant
pretty_name: tmmlu++
size_categories:
  - 100K<n<1M
configs:
  - config_name: engineering_math
    data_files:
      - split: train
        path: data/engineering_math_dev.csv
      - split: validation
        path: data/engineering_math_val.csv
      - split: test
        path: data/engineering_math_test.csv
  - config_name: dentistry
    data_files:
      - split: train
        path: data/dentistry_dev.csv
      - split: validation
        path: data/dentistry_val.csv
      - split: test
        path: data/dentistry_test.csv
  - config_name: traditional_chinese_medicine_clinical_medicine
    data_files:
      - split: train
        path: data/traditional_chinese_medicine_clinical_medicine_dev.csv
      - split: validation
        path: data/traditional_chinese_medicine_clinical_medicine_val.csv
      - split: test
        path: data/traditional_chinese_medicine_clinical_medicine_test.csv
  - config_name: clinical_psychology
    data_files:
      - split: train
        path: data/clinical_psychology_dev.csv
      - split: validation
        path: data/clinical_psychology_val.csv
      - split: test
        path: data/clinical_psychology_test.csv
  - config_name: technical
    data_files:
      - split: train
        path: data/technical_dev.csv
      - split: validation
        path: data/technical_val.csv
      - split: test
        path: data/technical_test.csv
  - config_name: culinary_skills
    data_files:
      - split: train
        path: data/culinary_skills_dev.csv
      - split: validation
        path: data/culinary_skills_val.csv
      - split: test
        path: data/culinary_skills_test.csv
  - config_name: mechanical
    data_files:
      - split: train
        path: data/mechanical_dev.csv
      - split: validation
        path: data/mechanical_val.csv
      - split: test
        path: data/mechanical_test.csv
  - config_name: logic_reasoning
    data_files:
      - split: train
        path: data/logic_reasoning_dev.csv
      - split: validation
        path: data/logic_reasoning_val.csv
      - split: test
        path: data/logic_reasoning_test.csv
  - config_name: real_estate
    data_files:
      - split: train
        path: data/real_estate_dev.csv
      - split: validation
        path: data/real_estate_val.csv
      - split: test
        path: data/real_estate_test.csv
  - config_name: general_principles_of_law
    data_files:
      - split: train
        path: data/general_principles_of_law_dev.csv
      - split: validation
        path: data/general_principles_of_law_val.csv
      - split: test
        path: data/general_principles_of_law_test.csv
  - config_name: finance_banking
    data_files:
      - split: train
        path: data/finance_banking_dev.csv
      - split: validation
        path: data/finance_banking_val.csv
      - split: test
        path: data/finance_banking_test.csv
  - config_name: anti_money_laundering
    data_files:
      - split: train
        path: data/anti_money_laundering_dev.csv
      - split: validation
        path: data/anti_money_laundering_val.csv
      - split: test
        path: data/anti_money_laundering_test.csv
  - config_name: ttqav2
    data_files:
      - split: train
        path: data/ttqav2_dev.csv
      - split: validation
        path: data/ttqav2_val.csv
      - split: test
        path: data/ttqav2_test.csv
  - config_name: marketing_management
    data_files:
      - split: train
        path: data/marketing_management_dev.csv
      - split: validation
        path: data/marketing_management_val.csv
      - split: test
        path: data/marketing_management_test.csv
  - config_name: business_management
    data_files:
      - split: train
        path: data/business_management_dev.csv
      - split: validation
        path: data/business_management_val.csv
      - split: test
        path: data/business_management_test.csv
  - config_name: organic_chemistry
    data_files:
      - split: train
        path: data/organic_chemistry_dev.csv
      - split: validation
        path: data/organic_chemistry_val.csv
      - split: test
        path: data/organic_chemistry_test.csv
  - config_name: advance_chemistry
    data_files:
      - split: train
        path: data/advance_chemistry_dev.csv
      - split: validation
        path: data/advance_chemistry_val.csv
      - split: test
        path: data/advance_chemistry_test.csv
  - config_name: physics
    data_files:
      - split: train
        path: data/physics_dev.csv
      - split: validation
        path: data/physics_val.csv
      - split: test
        path: data/physics_test.csv
  - config_name: secondary_physics
    data_files:
      - split: train
        path: data/secondary_physics_dev.csv
      - split: validation
        path: data/secondary_physics_val.csv
      - split: test
        path: data/secondary_physics_test.csv
  - config_name: human_behavior
    data_files:
      - split: train
        path: data/human_behavior_dev.csv
      - split: validation
        path: data/human_behavior_val.csv
      - split: test
        path: data/human_behavior_test.csv
  - config_name: national_protection
    data_files:
      - split: train
        path: data/national_protection_dev.csv
      - split: validation
        path: data/national_protection_val.csv
      - split: test
        path: data/national_protection_test.csv
  - config_name: jce_humanities
    data_files:
      - split: train
        path: data/jce_humanities_dev.csv
      - split: validation
        path: data/jce_humanities_val.csv
      - split: test
        path: data/jce_humanities_test.csv
  - config_name: politic_science
    data_files:
      - split: train
        path: data/politic_science_dev.csv
      - split: validation
        path: data/politic_science_val.csv
      - split: test
        path: data/politic_science_test.csv
  - config_name: agriculture
    data_files:
      - split: train
        path: data/agriculture_dev.csv
      - split: validation
        path: data/agriculture_val.csv
      - split: test
        path: data/agriculture_test.csv
  - config_name: official_document_management
    data_files:
      - split: train
        path: data/official_document_management_dev.csv
      - split: validation
        path: data/official_document_management_val.csv
      - split: test
        path: data/official_document_management_test.csv
  - config_name: financial_analysis
    data_files:
      - split: train
        path: data/financial_analysis_dev.csv
      - split: validation
        path: data/financial_analysis_val.csv
      - split: test
        path: data/financial_analysis_test.csv
  - config_name: pharmacy
    data_files:
      - split: train
        path: data/pharmacy_dev.csv
      - split: validation
        path: data/pharmacy_val.csv
      - split: test
        path: data/pharmacy_test.csv
  - config_name: educational_psychology
    data_files:
      - split: train
        path: data/educational_psychology_dev.csv
      - split: validation
        path: data/educational_psychology_val.csv
      - split: test
        path: data/educational_psychology_test.csv
  - config_name: statistics_and_machine_learning
    data_files:
      - split: train
        path: data/statistics_and_machine_learning_dev.csv
      - split: validation
        path: data/statistics_and_machine_learning_val.csv
      - split: test
        path: data/statistics_and_machine_learning_test.csv
  - config_name: management_accounting
    data_files:
      - split: train
        path: data/management_accounting_dev.csv
      - split: validation
        path: data/management_accounting_val.csv
      - split: test
        path: data/management_accounting_test.csv
  - config_name: introduction_to_law
    data_files:
      - split: train
        path: data/introduction_to_law_dev.csv
      - split: validation
        path: data/introduction_to_law_val.csv
      - split: test
        path: data/introduction_to_law_test.csv
  - config_name: computer_science
    data_files:
      - split: train
        path: data/computer_science_dev.csv
      - split: validation
        path: data/computer_science_val.csv
      - split: test
        path: data/computer_science_test.csv
  - config_name: veterinary_pathology
    data_files:
      - split: train
        path: data/veterinary_pathology_dev.csv
      - split: validation
        path: data/veterinary_pathology_val.csv
      - split: test
        path: data/veterinary_pathology_test.csv
  - config_name: accounting
    data_files:
      - split: train
        path: data/accounting_dev.csv
      - split: validation
        path: data/accounting_val.csv
      - split: test
        path: data/accounting_test.csv
  - config_name: fire_science
    data_files:
      - split: train
        path: data/fire_science_dev.csv
      - split: validation
        path: data/fire_science_val.csv
      - split: test
        path: data/fire_science_test.csv
  - config_name: optometry
    data_files:
      - split: train
        path: data/optometry_dev.csv
      - split: validation
        path: data/optometry_val.csv
      - split: test
        path: data/optometry_test.csv
  - config_name: insurance_studies
    data_files:
      - split: train
        path: data/insurance_studies_dev.csv
      - split: validation
        path: data/insurance_studies_val.csv
      - split: test
        path: data/insurance_studies_test.csv
  - config_name: pharmacology
    data_files:
      - split: train
        path: data/pharmacology_dev.csv
      - split: validation
        path: data/pharmacology_val.csv
      - split: test
        path: data/pharmacology_test.csv
  - config_name: taxation
    data_files:
      - split: train
        path: data/taxation_dev.csv
      - split: validation
        path: data/taxation_val.csv
      - split: test
        path: data/taxation_test.csv
  - config_name: trust_practice
    data_files:
      - split: train
        path: data/trust_practice_dev.csv
      - split: validation
        path: data/trust_practice_val.csv
      - split: test
        path: data/trust_practice_test.csv
  - config_name: geography_of_taiwan
    data_files:
      - split: train
        path: data/geography_of_taiwan_dev.csv
      - split: validation
        path: data/geography_of_taiwan_val.csv
      - split: test
        path: data/geography_of_taiwan_test.csv
  - config_name: physical_education
    data_files:
      - split: train
        path: data/physical_education_dev.csv
      - split: validation
        path: data/physical_education_val.csv
      - split: test
        path: data/physical_education_test.csv
  - config_name: auditing
    data_files:
      - split: train
        path: data/auditing_dev.csv
      - split: validation
        path: data/auditing_val.csv
      - split: test
        path: data/auditing_test.csv
  - config_name: administrative_law
    data_files:
      - split: train
        path: data/administrative_law_dev.csv
      - split: validation
        path: data/administrative_law_val.csv
      - split: test
        path: data/administrative_law_test.csv
  - config_name: education_(profession_level)
    data_files:
      - split: train
        path: data/education_(profession_level)_dev.csv
      - split: validation
        path: data/education_(profession_level)_val.csv
      - split: test
        path: data/education_(profession_level)_test.csv
  - config_name: economics
    data_files:
      - split: train
        path: data/economics_dev.csv
      - split: validation
        path: data/economics_val.csv
      - split: test
        path: data/economics_test.csv
  - config_name: veterinary_pharmacology
    data_files:
      - split: train
        path: data/veterinary_pharmacology_dev.csv
      - split: validation
        path: data/veterinary_pharmacology_val.csv
      - split: test
        path: data/veterinary_pharmacology_test.csv
  - config_name: nautical_science
    data_files:
      - split: train
        path: data/nautical_science_dev.csv
      - split: validation
        path: data/nautical_science_val.csv
      - split: test
        path: data/nautical_science_test.csv
  - config_name: occupational_therapy_for_psychological_disorders
    data_files:
      - split: train
        path: data/occupational_therapy_for_psychological_disorders_dev.csv
      - split: validation
        path: data/occupational_therapy_for_psychological_disorders_val.csv
      - split: test
        path: data/occupational_therapy_for_psychological_disorders_test.csv
  - config_name: basic_medical_science
    data_files:
      - split: train
        path: data/basic_medical_science_dev.csv
      - split: validation
        path: data/basic_medical_science_val.csv
      - split: test
        path: data/basic_medical_science_test.csv
  - config_name: macroeconomics
    data_files:
      - split: train
        path: data/macroeconomics_dev.csv
      - split: validation
        path: data/macroeconomics_val.csv
      - split: test
        path: data/macroeconomics_test.csv
  - config_name: trade
    data_files:
      - split: train
        path: data/trade_dev.csv
      - split: validation
        path: data/trade_val.csv
      - split: test
        path: data/trade_test.csv
  - config_name: chinese_language_and_literature
    data_files:
      - split: train
        path: data/chinese_language_and_literature_dev.csv
      - split: validation
        path: data/chinese_language_and_literature_val.csv
      - split: test
        path: data/chinese_language_and_literature_test.csv
  - config_name: tve_design
    data_files:
      - split: train
        path: data/tve_design_dev.csv
      - split: validation
        path: data/tve_design_val.csv
      - split: test
        path: data/tve_design_test.csv
  - config_name: junior_science_exam
    data_files:
      - split: train
        path: data/junior_science_exam_dev.csv
      - split: validation
        path: data/junior_science_exam_val.csv
      - split: test
        path: data/junior_science_exam_test.csv
  - config_name: junior_math_exam
    data_files:
      - split: train
        path: data/junior_math_exam_dev.csv
      - split: validation
        path: data/junior_math_exam_val.csv
      - split: test
        path: data/junior_math_exam_test.csv
  - config_name: junior_chinese_exam
    data_files:
      - split: train
        path: data/junior_chinese_exam_dev.csv
      - split: validation
        path: data/junior_chinese_exam_val.csv
      - split: test
        path: data/junior_chinese_exam_test.csv
  - config_name: junior_social_studies
    data_files:
      - split: train
        path: data/junior_social_studies_dev.csv
      - split: validation
        path: data/junior_social_studies_val.csv
      - split: test
        path: data/junior_social_studies_test.csv
  - config_name: tve_mathematics
    data_files:
      - split: train
        path: data/tve_mathematics_dev.csv
      - split: validation
        path: data/tve_mathematics_val.csv
      - split: test
        path: data/tve_mathematics_test.csv
  - config_name: tve_chinese_language
    data_files:
      - split: train
        path: data/tve_chinese_language_dev.csv
      - split: validation
        path: data/tve_chinese_language_val.csv
      - split: test
        path: data/tve_chinese_language_test.csv
  - config_name: tve_natural_sciences
    data_files:
      - split: train
        path: data/tve_natural_sciences_dev.csv
      - split: validation
        path: data/tve_natural_sciences_val.csv
      - split: test
        path: data/tve_natural_sciences_test.csv
  - config_name: junior_chemistry
    data_files:
      - split: train
        path: data/junior_chemistry_dev.csv
      - split: validation
        path: data/junior_chemistry_val.csv
      - split: test
        path: data/junior_chemistry_test.csv
  - config_name: music
    data_files:
      - split: train
        path: data/music_dev.csv
      - split: validation
        path: data/music_val.csv
      - split: test
        path: data/music_test.csv
  - config_name: education
    data_files:
      - split: train
        path: data/education_dev.csv
      - split: validation
        path: data/education_val.csv
      - split: test
        path: data/education_test.csv
  - config_name: three_principles_of_people
    data_files:
      - split: train
        path: data/three_principles_of_people_dev.csv
      - split: validation
        path: data/three_principles_of_people_val.csv
      - split: test
        path: data/three_principles_of_people_test.csv
  - config_name: taiwanese_hokkien
    data_files:
      - split: train
        path: data/taiwanese_hokkien_dev.csv
      - split: validation
        path: data/taiwanese_hokkien_val.csv
      - split: test
        path: data/taiwanese_hokkien_test.csv

TMMLU+ : Large scale traditional chinese massive multitask language understanding

A close-up image of a neat paper note with a white background. The text 'TMMLU+' is written horizontally across the center of the note in bold, black. Join us to work in multimodal LLM : https://ikala.ai/recruit/

We present TMMLU+, a traditional Chinese massive multitask language understanding dataset. TMMLU+ is a multiple-choice question-answering dataset featuring 66 subjects, ranging from elementary to professional level.

The TMMLU+ dataset is six times larger and contains more balanced subjects compared to its predecessor, TMMLU. We have included benchmark results in TMMLU+ from closed-source models and 20 open-weight Chinese large language models, with parameters ranging from 1.8B to 72B. The benchmark results show that Traditional Chinese variants still lag behind those trained on major Simplified Chinese models.

from datasets import load_dataset
task_list = [
             'engineering_math', 'dentistry', 'traditional_chinese_medicine_clinical_medicine', 'clinical_psychology', 'technical', 'culinary_skills', 'mechanical', 'logic_reasoning', 'real_estate',
             'general_principles_of_law', 'finance_banking', 'anti_money_laundering', 'ttqav2', 'marketing_management', 'business_management', 'organic_chemistry', 'advance_chemistry',
             'physics', 'secondary_physics', 'human_behavior', 'national_protection', 'jce_humanities', 'politic_science', 'agriculture', 'official_document_management',
             'financial_analysis', 'pharmacy', 'educational_psychology', 'statistics_and_machine_learning', 'management_accounting', 'introduction_to_law', 'computer_science', 'veterinary_pathology',
             'accounting', 'fire_science', 'optometry', 'insurance_studies', 'pharmacology', 'taxation', 'trust_practice', 'geography_of_taiwan', 'physical_education', 'auditing', 'administrative_law',
             'education_(profession_level)', 'economics', 'veterinary_pharmacology', 'nautical_science', 'occupational_therapy_for_psychological_disorders',
             'basic_medical_science', 'macroeconomics', 'trade', 'chinese_language_and_literature', 'tve_design', 'junior_science_exam', 'junior_math_exam', 'junior_chinese_exam',
             'junior_social_studies', 'tve_mathematics', 'tve_chinese_language', 'tve_natural_sciences', 'junior_chemistry', 'music', 'education', 'three_principles_of_people',
             'taiwanese_hokkien'
            ]
for task in task_list:
  val = load_dataset('ikala/tmmluplus', task)['validation']
  dev = load_dataset('ikala/tmmluplus', task)['train']
  test = load_dataset('ikala/tmmluplus', task)['test']

For each dataset split

for row in test:
  print(row)
  break
>> Dataset({
    features: ['question', 'A', 'B', 'C', 'D', 'answer'],
    num_rows: 11
})

Statistic on all four categories : STEM, Social Science, Humanities, Other

Category Test Dev Validation
STEM 3458 70 385
Social Sciences 5958 90 665
Humanities 1763 35 197
Other (Business, Health, Misc.) 8939 135 995
Total 20118 330 2242

Benchmark on direct prompting

model STEM Social Science Humanities Other Average
Gemini-1.5-pro 66.18 70.29 61.84 60.30 64.65
Qwen/Qwen-72B 61.12 71.65 63.00 61.31 64.27
gpt-4-0613 60.36 67.36 56.03 57.62 60.34
Qwen-max 59.92 66.95 57.43 56.48 60.20
Qwen/Qwen-72B-Chat 55.15 66.20 55.65 57.19 58.55
Qwen/Qwen-14B 46.94 56.69 49.43 48.81 50.47
Gemini-pro 45.38 57.29 48.80 48.21 49.92
01-ai/Yi-34B-Chat 40.24 56.77 53.99 47.58 49.64
Gemini-1.5-flash 53.47 53.42 42.99 46.56 49.11
Reka Flash 45.26 52.91 46.31 43.76 47.06
Qwen/Qwen-14B-Chat 43.86 53.29 44.78 45.13 46.77
Qwen/Qwen1.5-14B-Chat 39.65 52.76 43.90 44.95 45.31
01-ai/Yi-6B-Chat 39.62 50.24 44.44 44.26 44.64
Claude-1.3 42.65 49.33 42.16 44.14 44.57
MediaTek-Research/Breeze-7B-Instruct-v0_1 36.46 48.38 45.11 40.75 42.67
gpt-3.5-turbo-0613 41.56 46.72 36.73 42.03 41.76
CausalLM/14B 39.83 44.50 39.61 41.97 41.48
Skywork/Skywork-13B-base 36.93 47.27 41.04 40.10 41.33
Claude-3-opus 42.95 45.49 35.79 40.24 41.12
Qwen/Qwen-7B 37.53 45.48 38.09 38.96 40.01
meta-llama/Llama-3-70b-chat-hf 34.44 47.02 37.50 39.51 39.62
Qwen/Qwen-7B-Chat 33.32 44.64 40.27 39.89 39.53
vivo-ai/BlueLM-7B-Base 33.94 41.52 37.38 38.74 37.90
baichuan-inc/Baichuan2-13B-Chat 29.64 43.73 37.36 39.88 37.65
Qwen/Qwen-1_8B 32.65 38.95 38.34 35.27 36.30
Claude-2 39.65 39.09 28.59 37.47 36.20
THUDM/chatglm3-6b 31.05 39.31 35.64 35.60 35.40
deepseek-ai/deepseek-llm-7b-chat 29.82 42.29 34.24 34.31 35.17
CausalLM/7B 31.03 38.17 35.87 35.39 35.11
Azure99/blossom-v3_1-mistral-7b 32.80 36.91 32.36 34.53 34.15
google/gemma-7b-it 31.89 35.70 34.00 33.79 33.84
Reka Edge 30.02 39.40 31.84 32.36 33.41
microsoft/Orca-2-13b 24.69 39.18 33.60 31.99 32.37
Qwen/Qwen-1_8B-Chat 26.60 36.36 31.81 31.96 31.68
meta-llama/Llama-3-8b-chat-hf 31.52 34.19 28.91 31.79 31.60
TigerResearch/tigerbot-13b-chat-v3 24.73 29.63 25.72 27.22 26.82
hongyin/mistral-7b-80k 24.26 23.76 22.56 24.57 23.79
deepseek-ai/deepseek-llm-67b-chat 19.10 26.06 21.51 21.77 22.11
yentinglin/Taiwan-LLM-13B-v2.0-chat 18.53 27.65 17.77 21.49 21.36
GeneZC/MiniChat-3B 17.66 23.35 22.71 20.34 21.02
LinkSoul/Chinese-Llama-2-7b 16.55 18.39 12.97 16.13 16.01
yentinglin/Taiwan-LLM-7B-v2.1-chat 14.99 16.23 15.00 16.22 15.61
Claude-instant-1 12.52 17.13 15.10 13.57 14.58
FlagAlpha/Atom-7B 5.60 13.57 7.71 11.84 9.68

Results via ievals ( settings : 0-shot direct answering )

Citation

@article{ikala2024improved,
  title={An Improved Traditional Chinese Evaluation Suite for Foundation Model},
  author={Tam, Zhi-Rui and Pai, Ya-Ting and Lee, Yen-Wei and Cheng, Sega and Shuai, Hong-Han},
  journal={arXiv preprint arXiv:2403.01858},
  year={2024}
}