Instructions to use J-LAB/BRisa-7B-Instruct-v0.2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use J-LAB/BRisa-7B-Instruct-v0.2 with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="J-LAB/BRisa-7B-Instruct-v0.2")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("J-LAB/BRisa-7B-Instruct-v0.2")
model = AutoModelForCausalLM.from_pretrained("J-LAB/BRisa-7B-Instruct-v0.2")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use J-LAB/BRisa-7B-Instruct-v0.2 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "J-LAB/BRisa-7B-Instruct-v0.2"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "J-LAB/BRisa-7B-Instruct-v0.2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/J-LAB/BRisa-7B-Instruct-v0.2

SGLang

How to use J-LAB/BRisa-7B-Instruct-v0.2 with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "J-LAB/BRisa-7B-Instruct-v0.2" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "J-LAB/BRisa-7B-Instruct-v0.2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "J-LAB/BRisa-7B-Instruct-v0.2" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "J-LAB/BRisa-7B-Instruct-v0.2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use J-LAB/BRisa-7B-Instruct-v0.2 with Docker Model Runner:
```
docker model run hf.co/J-LAB/BRisa-7B-Instruct-v0.2
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

BRisa 7B Instruct

This is an instruction model trained for good performance in Portuguese. The initial base is the Mistral 7B v2 Model (source). We utilized the JJhooww/Mistral-7B-v0.2-Base_ptbr version pre-trained on 1 billion tokens in Portuguese (source).

The base model has good performance in Portuguese but faces significant challenges following instructions. We therefore used the version mistralai/Mistral-7B-Instruct-v0.2 and fine-tuned it for responses in Portuguese, then merged it with the base JJhooww/Mistral-7B-v0.2-Base_ptbr (https://huggingface.co/JJhooww/Mistral-7B-v0.2-Base_ptbr).

Developed by: (J-LAB)
Language(s) (NLP): Portuguese
License: APACHE
Finetuned from model: (source)

Model Sources

Demo: (Demonstracao da Versão DPO)

Open Portuguese LLM Leaderboard Evaluation Results

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

Metric	Value
Average	66.19
ENEM Challenge (No Images)	65.08
BLUEX (No Images)	53.69
OAB Exams	43.37
Assin2 RTE	91.50
Assin2 STS	73.61
FaQuAD NLI	68.31
HateBR Binary	74.28
PT Hate Speech Binary	65.12
tweetSentBR	60.77

Downloads last month: 9

Safetensors

Model size

7B params

Tensor type

BF16

Model tree for J-LAB/BRisa-7B-Instruct-v0.2

Quantizations

1 model

Space using J-LAB/BRisa-7B-Instruct-v0.2 1

Evaluation results

accuracy on ENEM Challenge (No Images)
Open Portuguese LLM Leaderboard

65.080
accuracy on BLUEX (No Images)
Open Portuguese LLM Leaderboard

53.690
accuracy on OAB Exams
Open Portuguese LLM Leaderboard

43.370
f1-macro on Assin2 RTE
test set Open Portuguese LLM Leaderboard

91.500
pearson on Assin2 STS
test set Open Portuguese LLM Leaderboard

73.610
f1-macro on FaQuAD NLI
test set Open Portuguese LLM Leaderboard

68.310
f1-macro on HateBR Binary
test set Open Portuguese LLM Leaderboard

74.280
f1-macro on PT Hate Speech Binary
test set Open Portuguese LLM Leaderboard

65.120