Awesome

MED-LLM-BR: Medical Large Language Models for Brazilian Portuguese

MED-LLM-BR is a collaborative project between HAILab and Comsentimento, which aims to develop multiple medical LLMs for Portuguese language, including base models and task-specific models, with different sizes.

Contributions:

Developing Resource-Efficient Clinical LLMs for Brazilian Portuguese

This project leverages LLama and Mistral as base models, adapting them through fine-tuning techniques to enhance their performance in clinical text generation tasks.

To optimize resource utilization during the fine-tuning process, we employed Low-Rank Adaptation (LoRA). This approach enables effective model adaptation with significantly reduced computational and memory requirements, making the fine-tuning process more efficient without compromising the quality of the generated clinical text.

Model Description

LLama: LLama is a state-of-the-art language model known for its scalability and efficiency in handling diverse natural language processing tasks. In this project, LLama serves as one of the base models for fine-tuning, aimed at adapting it to the specific requirements of clinical text generation in Portuguese.

Mistral: Mistral is another advanced language model designed to enhance performance in various text generation applications. By incorporating Mistral into the fine-tuning pipeline, MED-LLM-BR seeks to combine the strengths of multiple models to achieve superior results in generating accurate and contextually relevant clinical notes.

How to use the models with HuggingFace

Link model 1: Clinical-BR-LlaMA-2-7B

Link model 2: Clinical-BR-Mistral-7B-v0.2

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

from huggingface_hub import login
login()

model_id = "pucpr-br/Clinical-BR-LlaMA-2-7B" or "pucpr-br/Clinical-BR-Mistral-7B-v0.2"
tokenizer = AutoTokenizer.from_pretrained(model_id)

model     = AutoModelForCausalLM.from_pretrained(model_id, trust_remote_code=True)
prompt = "Paciente admitido com angina instável, progredindo para infarto agudo do miocárdio (IAM) inferior no primeiro dia de internação; encaminhado para unidade de hemodinâmica, onde foi feita angioplastia com implante de stent na ponte d "	
inputs = tokenizer(prompt, return_tensors="pt", return_token_type_ids=False)
outputs = model.generate(**inputs, max_new_tokens=90)
print(tokenizer.batch_decode(outputs, skip_special_tokens=True)[0])

How to use the models with ollama

Model Clinical-BR-LlaMA-2-7B:

$ ollama run cabelo/clinical-br-llama-2-7b

Model Clinical-BR-Mistral-7B-v0.2:

$ ollama run cabelo/clinical-br-mistral-7b-0.2

How to use the OpenVINO models with HuggingFace

Alessandro de Oliveira Faria (Intel Innovator) is responsible for maintaining the quantization (fp16 and int8) of the MED-LLM-BR model for Intel OpenVINO technology, ensuring optimization and efficiency in production environments.

Link model FP16 : Clinical-BR-LlaMA-2-7B-fp16-ov

Link model Int8 : Clinical-BR-LlaMA-2-7B-int8-ov

Source and more information in MED-LLM-BR-openvino

Provisional Citation:

@inproceedings{pinto2024clinicalLLMs,
  title        = {Developing Resource-Efficient Clinical LLMs for Brazilian Portuguese},
  author       = {João Gabriel de Souza Pinto and Andrey Rodrigues de Freitas and Anderson Carlos Gomes Martins and Caroline Midori Rozza Sawazaki and Caroline Vidal and Lucas Emanuel Silva e Oliveira},
  booktitle    = {Proceedings of the 34th Brazilian Conference on Intelligent Systems (BRACIS)},
  year         = {2024},
  note         = {In press},
}