Edit model card

internlm-chatbode-20b

ChatBode Logo

O InternLm-ChatBode é um modelo de linguagem ajustado para o idioma português, desenvolvido a partir do modelo InternLM2. Este modelo foi refinado através do processo de fine-tuning utilizando o dataset UltraAlpaca.

Características Principais

  • Modelo Base: internlm/internlm2-chat-20b
  • Dataset para Fine-tuning: UltraAlpaca
  • Treinamento: O treinamento foi realizado a partir do fine-tuning, usando QLoRA, do internlm2-chat-20b.

Exemplo de uso

A seguir um exemplo de código de como carregar e utilizar o modelo:

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM
tokenizer = AutoTokenizer.from_pretrained("recogna-nlp/internlm-chatbode-20b", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("recogna-nlp/internlm-chatbode-20b", torch_dtype=torch.float16, trust_remote_code=True).cuda()
model = model.eval()
response, history = model.chat(tokenizer, "Olá", history=[])
print(response)
response, history = model.chat(tokenizer, "O que é o Teorema de Pitágoras? Me dê um exemplo", history=history)
print(response)

As respostas podem ser geradas via stream utilizando o método stream_chat:

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_path = "recogna-nlp/internlm-chatbode-20b"
model = AutoModelForCausalLM.from_pretrained(model_path, torch_dtype=torch.float16, trust_remote_code=True).cuda()
tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)

model = model.eval()
length = 0
for response, history in model.stream_chat(tokenizer, "Olá", history=[]):
    print(response[length:], flush=True, end="")
    length = len(response)

Open Portuguese LLM Leaderboard Evaluation Results

Detailed results can be found here and on the 🚀 Open Portuguese LLM Leaderboard

Metric Value
Average 71.68
ENEM Challenge (No Images) 65.78
BLUEX (No Images) 58.69
OAB Exams 43.33
Assin2 RTE 91.53
Assin2 STS 78.95
FaQuAD NLI 81.36
HateBR Binary 81.72
PT Hate Speech Binary 73.66
tweetSentBR 70.11
Downloads last month
8
Safetensors
Model size
19.9B params
Tensor type
FP16
·
Inference Examples
Inference API (serverless) does not yet support model repos that contain custom code.

Space using recogna-nlp/internlm-chatbode-20b 1

Evaluation results