portuguese-tom-cat-13b
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 Llama-2-13b
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 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/portuguese-tom-cat-13b", device_map= {"": 0})
tokenizer = AutoTokenizer.from_pretrained("rhaymison/portuguese-tom-cat-13b")
model.eval()
You can use with Pipeline.
from transformers import pipeline
pipe = pipeline("text-generation",
model=model,
tokenizer=tokenizer,
do_sample=True,
max_new_tokens=512,
num_beams=2,
temperature=0.3,
top_k=50,
top_p=0.95,
early_stopping=True,
pad_token_id=tokenizer.eos_token_id,
)
def format_question(input:str)-> str:
base_instruction = """Abaixo está uma instrução que descreve uma tarefa, juntamente com uma entrada que fornece mais contexto. Escreva uma resposta que complete adequadamente o pedido."""
_input = f"""<s>[INST] <<SYS>>\n {base_instruction}
<</SYS>> {input} [/INST]
"""
return _input.strip()
prompt = "Me explique sobre os romanos"
pipe(format_question(prompt))
Os romanos foram um povo que viveu na Itália antiga, entre o século VIII a.C. e o século V d.C.
Eles eram conhecidos por sua habilidade em construir estradas, edifícios e aquedutos, e também por suas conquistas militares.
O Império Romano, que durou de 27 a.C. a 476 d.C., foi o maior império da história, abrangendo uma área que ia da Grécia até a Inglaterra.
Os romanos também desenvolveram um sistema de leis e instituições políticas que influenciaram profundamente a cultura ocidental.
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 🚀 Open Portuguese LLM Leaderboard
Metric | Value |
---|---|
Average | 57.86 |
ENEM Challenge (No Images) | 42.76 |
BLUEX (No Images) | 45.62 |
OAB Exams | 39.09 |
Assin2 RTE | 77.41 |
Assin2 STS | 58.44 |
FaQuAD NLI | 68.14 |
HateBR Binary | 84.13 |
PT Hate Speech Binary | 56.27 |
tweetSentBR | 48.86 |
Comments
Any idea, help or report will always be welcome.
email: rhaymisoncristian@gmail.com
- Downloads last month
- 7
Model tree for rhaymison/portuguese-tom-cat-13b
Base model
meta-llama/Llama-2-13bDataset used to train rhaymison/portuguese-tom-cat-13b
Space using rhaymison/portuguese-tom-cat-13b 1
Evaluation results
- accuracy on ENEM Challenge (No Images)Open Portuguese LLM Leaderboard42.760
- accuracy on BLUEX (No Images)Open Portuguese LLM Leaderboard45.620
- accuracy on OAB ExamsOpen Portuguese LLM Leaderboard39.090
- f1-macro on Assin2 RTEtest set Open Portuguese LLM Leaderboard77.410
- pearson on Assin2 STStest set Open Portuguese LLM Leaderboard58.440
- f1-macro on FaQuAD NLItest set Open Portuguese LLM Leaderboard68.140
- f1-macro on HateBR Binarytest set Open Portuguese LLM Leaderboard84.130
- f1-macro on PT Hate Speech Binarytest set Open Portuguese LLM Leaderboard56.270
- f1-macro on tweetSentBRtest set Open Portuguese LLM Leaderboard48.860