CollectiveSFT

This model is a fine-tuned version of internlm/internlm2_5-7b on some medical datasets.

Model description

CollectiveSFT: Scaling Large Language Models for Chinese Medical Benchmark with Collective Instructions in Healthcare.

Official Code Repo:https://github.com/CAS-SIAT-XinHai/CollectiveSFT

Intended uses & limitations

The model may have limitations in chat functionality.

Training and evaluation data

Language: English

Dataset Name	Style	Size
PubMedQA	QA	273,518
MedMCQA	MCQA	182,822
HeadQA	QA	2,657
Total		458,997

Language: Chinese

Dataset Name	Style	Size
cMedQA2	QA	100,000
cMedDialogu	Dialogue	792,099
webMedQA	QA	252,850
MedicalDialog	Dialogue	2,725,989
CMID	NER	12,254
NLPEC	MCQA	18,703
CMB	MCQA	269,359
MLEC-QA	MCQA	108,988
DISCMe	Dialogue	464,898
Total		4,745,140

For detailed dataset specifications and access instructions, please refer to our paper.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 16
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 2
• total_train_batch_size: 256
• total_eval_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.05
• num_epochs: 3.0

Framework versions

• Transformers 4.42.4
• Pytorch 2.3.1+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1