Qwen-portuguese-luana-7b
This model was trained with a superset of 250,000 chat in Portuguese. The model comes to help fill the gap in models in Portuguese. Tuned from the Qwen1.5-7B-Chat.
How to use
FULL MODEL : A100
HALF MODEL: L4
8bit or 4bit : T4 or V100
You can use the model in its normal form up to 4-bit quantization. Below we will use both approaches. Remember that verbs are important in your prompt. Tell your model how to act or behave so that you can guide them along the path of their response. Important points like these help models (even smaller models like 7b) to perform much better.
!pip install -q -U transformers
!pip install -q -U accelerate
!pip install -q -U bitsandbytes
from transformers import AutoModelForCausalLM, AutoTokenizer, TextStreamer
from transformers import pipeline
model = AutoModelForCausalLM.from_pretrained("rhaymison/Qwen-portuguese-luana-7b", device_map= {"": 0})
tokenizer = AutoTokenizer.from_pretrained("rhaymison/Qwen-portuguese-luana-7b")
model.eval()
You can use with Pipeline but in this example i will use such as Streaming
prompt = f"""<|im_start|>system
Abaixo estรก uma instruรงรฃo que descreve uma tarefa, juntamente com uma entrada que fornece mais contexto. Escreva uma resposta que complete adequadamente o pedido.<|im_end|>
<|im_start|>user
### instruรงรฃo: Me indique uma programaรงรฃo para fazer no final de semana com minha esposa. <|im_end|>
<|im_start|>"""
pipe = pipeline("text-generation",
model=model,
tokenizer=tokenizer,
do_sample=True,
max_new_tokens=200,
num_beams=2,
temperature=0.3,
top_k=50,
top_p=0.95,
early_stopping=True,
pad_token_id=tokenizer.eos_token_id,
)
pipe(prompt)[0]['generated_text'].split('assistant')[1]
#Claro! Aqui estรก uma sugestรฃo de programaรงรฃo para o final de semana com sua esposa:
#Domingo:
#1. Despertar cedo para um cafรฉ da manhรฃ delicioso juntos.
#2. Faรงa uma caminhada ou uma corrida no parque local para aproveitar o ar fresco e a natureza.
#3. Depois do cafรฉ da manhรฃ, faรงa uma caminhada de compras para escolher algumas roupas ou acessรณrios novos.
#4. Retorne para casa para preparar uma refeiรงรฃo deliciosa juntos.
#5. Depois do almoรงo, vocรช pode assistir a um filme ou jogar jogos de tabuleiro para relaxar
If you are having a memory problem such as "CUDA Out of memory", you should use 4-bit or 8-bit quantization. For the complete model in colab you will need the A100. If you want to use 4bits or 8bits, T4 or L4 will already solve the problem.
4bits example
from transformers import BitsAndBytesConfig
import torch
nb_4bit_config = BitsAndBytesConfig(
load_in_4bit=True,
bnb_4bit_quant_type="nf4",
bnb_4bit_compute_dtype=torch.bfloat16,
bnb_4bit_use_double_quant=True
)
model = AutoModelForCausalLM.from_pretrained(
base_model,
quantization_config=bnb_config,
device_map={"": 0}
)
Open Portuguese LLM Leaderboard Evaluation Results
Detailed results can be found here and on the ๐ Open Portuguese LLM Leaderboard
Metric | Value |
---|---|
Average | 62.49 |
ENEM Challenge (No Images) | 58.36 |
BLUEX (No Images) | 48.12 |
OAB Exams | 42.73 |
Assin2 RTE | 81.05 |
Assin2 STS | 74.25 |
FaQuAD NLI | 57.96 |
HateBR Binary | 70.29 |
PT Hate Speech Binary | 69.92 |
tweetSentBR | 59.69 |
Comments
Any idea, help or report will always be welcome.
email: rhaymisoncristian@gmail.com
- Downloads last month
- 22
Model tree for rhaymison/Qwen-portuguese-luana-7b
Base model
Qwen/Qwen1.5-7BDataset used to train rhaymison/Qwen-portuguese-luana-7b
Space using rhaymison/Qwen-portuguese-luana-7b 1
Evaluation results
- accuracy on ENEM Challenge (No Images)Open Portuguese LLM Leaderboard58.360
- accuracy on BLUEX (No Images)Open Portuguese LLM Leaderboard48.120
- accuracy on OAB ExamsOpen Portuguese LLM Leaderboard42.730
- f1-macro on Assin2 RTEtest set Open Portuguese LLM Leaderboard81.050
- pearson on Assin2 STStest set Open Portuguese LLM Leaderboard74.250
- f1-macro on FaQuAD NLItest set Open Portuguese LLM Leaderboard57.960
- f1-macro on HateBR Binarytest set Open Portuguese LLM Leaderboard70.290
- f1-macro on PT Hate Speech Binarytest set Open Portuguese LLM Leaderboard69.920
- f1-macro on tweetSentBRtest set Open Portuguese LLM Leaderboard59.690