Mistral-8x7b-Quantized-portuguese-luanaa

This model was trained with a superset of 300,000 instructions in Portuguese. The model comes to help fill the gap in models in Portuguese. Tuned from the Mistral 8x7b and quantized in 4bit for Portuguese, the model was adjusted mainly for instructional tasks.

How to use

A100 GPU

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
model = AutoModelForCausalLM.from_pretrained("rhaymison/Mistral-8x7b-Quantized-portuguese-luana", device_map= {"": 0})
tokenizer = AutoTokenizer.from_pretrained("rhaymison/Mistral-8x7b-Quantized-portuguese-luana")
model.eval()

You can use with Pipeline but in this example i will use such as Streaming


inputs = tokenizer([f"""<s>[INST] Abaixo estΓ‘ uma instruΓ§Γ£o que descreve uma tarefa, juntamente com uma entrada que fornece mais contexto.
Escreva uma resposta que complete adequadamente o pedido.
### instruΓ§Γ£o: aja como um professor de matemΓ‘tica e me explique porque 2 + 2 = 4.
[/INST]"""], return_tensors="pt")

inputs.to(model.device)

streamer = TextStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
_ = model.generate(**inputs, streamer=streamer, max_new_tokens=200)

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}
)

LangChain

Extracting a langChain conversation with a drug petition document as context.

  • Human: Bom dia. Me informe quais medicamentos sΓ£o solicitados no texto. Informe tambΓ©m a posologia e o perΓ­odo de tratamento.

  • Luana: OlΓ‘! No texto sΓ£o solicitados os medicamentos DULOXETINA 30mg,PREGABALINA 100mg ROSUVASTATINA 10mg. A posologia e o perΓ­odo de tratamento nΓ£o sΓ£o especificados, pois isso deve ser indicado periodicamente pelo mΓ©dico.

  • Human: Perfeito. E qual Γ© o valor da causa informado no documento?

  • Luana: O valor da causa informado no documento Γ© R$ 6.132,00 (seis mil cento e trΓͺs).

  • Human: Foi mencionado no texto alguma lei?

  • Luana: Sim, a Lei Complementar nΒΊ 80/94 Γ© mencionada no texto. Ela se refere Γ s prerrogativas institucionais da Defensoria PΓΊblica, incluindo a intimaΓ§Γ£o pessoal de seus membros, a contagem em dobro dos prazos processuais, a manifestaΓ§Γ£o por cotas nos autos e a dispensa da procuraΓ§Γ£o.

  • Human: Perfeito! Muito obrigado pela ajuda. Por hora Γ© somente isso =]

  • Luana: Foi um prazer ajudar! Se vocΓͺ tiver outras dΓΊvidas ou necessitar de mais informaΓ§Γ΅es, nΓ£o hesite em perguntar. Tenha um bom dia!

Open Portuguese LLM Leaderboard Evaluation Results

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

Metric Value
Average 66.05
ENEM Challenge (No Images) 69.63
BLUEX (No Images) 59.11
OAB Exams 49.61
Assin2 RTE 61.21
Assin2 STS 79.95
FaQuAD NLI 78.60
HateBR Binary 72.42
PT Hate Speech Binary 73.01
tweetSentBR 50.90

Comments

Any idea, help or report will always be welcome.

email: [email protected]

Downloads last month
2
Safetensors
Model size
24.2B params
Tensor type
F32
Β·
FP16
Β·
U8
Β·
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support

Model tree for rhaymison/Mistral-8x7b-Quantized-portuguese-luana

Quantized
(35)
this model

Dataset used to train rhaymison/Mistral-8x7b-Quantized-portuguese-luana

Space using rhaymison/Mistral-8x7b-Quantized-portuguese-luana 1

Evaluation results