huangxinping/HWTCMBench

A benchmark code of hwtcm.

Python

2

4 commits

updated Aug 28, 2024

See the code

README

HWTCMBench

A Comprehensive Benchmark for Evaluating Large Language Models in Traditional Chinese Medicine.

Changelog

  • 2024-08-28: Added 7226 questions.
  • 2024-07-20: Debut.

Dataset

The dataset is available at https://huggingface.co/datasets/Monor/hwtcm

Benchmarking model accuracy

multiple-choice questions(单选题)multiple-answers questions(多选题)True/False questions(判断题)
llama3:8b21.94%17.71%46.56%
phi3:14b-instruct26.93%1.04%38.93%
aya:8b17.85%1.04%34.35%
mistral:7b-instruct21.76%2.08%48.09%
qwen1.5-7b-chat51.35%13.54%46.56%
qwen1.5-14b-chat69.94%78.12%31.30%
huangdi-13b-chat21.73%45.83%0.00%
canggong-14b-chat(SFT)
Ours
55.98%4.17%23.66%
canggong-14b-chat(DPO)
Ours
72.33%2.08%45.80%

canggong-14b-chat is an LLM of traditional Chinese medicine still in training.

benchmark
llm
tcm

Contributors

huangxinping

4 commits

huangxinping/HWTCMBench

A benchmark code of hwtcm.

Python

2

4 commits

updated Aug 28, 2024

See the code

README

HWTCMBench

A Comprehensive Benchmark for Evaluating Large Language Models in Traditional Chinese Medicine.

Changelog

  • 2024-08-28: Added 7226 questions.
  • 2024-07-20: Debut.

Dataset

The dataset is available at https://huggingface.co/datasets/Monor/hwtcm

Benchmarking model accuracy

multiple-choice questions(单选题)multiple-answers questions(多选题)True/False questions(判断题)
llama3:8b21.94%17.71%46.56%
phi3:14b-instruct26.93%1.04%38.93%
aya:8b17.85%1.04%34.35%
mistral:7b-instruct21.76%2.08%48.09%
qwen1.5-7b-chat51.35%13.54%46.56%
qwen1.5-14b-chat69.94%78.12%31.30%
huangdi-13b-chat21.73%45.83%0.00%
canggong-14b-chat(SFT)
Ours
55.98%4.17%23.66%
canggong-14b-chat(DPO)
Ours
72.33%2.08%45.80%

canggong-14b-chat is an LLM of traditional Chinese medicine still in training.

benchmark
llm
tcm

Contributors

huangxinping

4 commits

Languages

Python

100.0%