language:
- pt
license: apache-2.0
tags:
- text-generation-inference
- transformers
- unsloth
- mistral
- trl
base_model: unsloth/mistral-7b-bnb-4bit
datasets:
- cnmoro/GPT4-500k-Augmented-PTBR-Clean
widget:
- text: Me conte a história do Boto
model-index:
- name: boto-7B
results:
- task:
type: text-generation
name: Text Generation
dataset:
name: ENEM Challenge (No Images)
type: eduagarcia/enem_challenge
split: train
args:
num_few_shot: 3
metrics:
- type: acc
value: 59.97
name: accuracy
source:
url: >-
https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=lucianosb/boto-7B
name: Open Portuguese LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: BLUEX (No Images)
type: eduagarcia-temp/BLUEX_without_images
split: train
args:
num_few_shot: 3
metrics:
- type: acc
value: 48.82
name: accuracy
source:
url: >-
https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=lucianosb/boto-7B
name: Open Portuguese LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: OAB Exams
type: eduagarcia/oab_exams
split: train
args:
num_few_shot: 3
metrics:
- type: acc
value: 43.37
name: accuracy
source:
url: >-
https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=lucianosb/boto-7B
name: Open Portuguese LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: Assin2 RTE
type: assin2
split: test
args:
num_few_shot: 15
metrics:
- type: f1_macro
value: 89.58
name: f1-macro
source:
url: >-
https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=lucianosb/boto-7B
name: Open Portuguese LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: Assin2 STS
type: eduagarcia/portuguese_benchmark
split: test
args:
num_few_shot: 15
metrics:
- type: pearson
value: 69.87
name: pearson
source:
url: >-
https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=lucianosb/boto-7B
name: Open Portuguese LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: FaQuAD NLI
type: ruanchaves/faquad-nli
split: test
args:
num_few_shot: 15
metrics:
- type: f1_macro
value: 55.57
name: f1-macro
source:
url: >-
https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=lucianosb/boto-7B
name: Open Portuguese LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: HateBR Binary
type: ruanchaves/hatebr
split: test
args:
num_few_shot: 25
metrics:
- type: f1_macro
value: 77.04
name: f1-macro
source:
url: >-
https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=lucianosb/boto-7B
name: Open Portuguese LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: PT Hate Speech Binary
type: hate_speech_portuguese
split: test
args:
num_few_shot: 25
metrics:
- type: f1_macro
value: 58.84
name: f1-macro
source:
url: >-
https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=lucianosb/boto-7B
name: Open Portuguese LLM Leaderboard
- task:
type: text-generation
name: Text Generation
dataset:
name: tweetSentBR
type: eduagarcia-temp/tweetsentbr
split: test
args:
num_few_shot: 25
metrics:
- type: f1_macro
value: 57.2
name: f1-macro
source:
url: >-
https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=lucianosb/boto-7B
name: Open Portuguese LLM Leaderboard
Boto 7B
Boto é um fine-tuning do Mistral 7B para língua portuguesa. O Boto é bem "falante", as respostas tendem a ser longas e nem sempre objetivas por padrão.
Acesse a demonstração online disponível. E cante junto:
Boto é um nome dado a vários tipos de golfinhos e botos nativos do Amazonas e dos afluentes do rio Orinoco. Alguns botos existem exclusivamente em água doce, e estes são frequentemente considerados golfinhos primitivos.
O “boto” das regiões do rio Amazonas no norte do Brasil é descrito de acordo com o folclore local como assumindo a forma de um humano, também conhecido como Boto cor-de-rosa, e com o hábito de seduzir mulheres humanas e engravidá-las.
Métricas de avaliação em andamento...
English description
Boto is a fine-tuning of Mistral 7B for portuguese language. Responses tend to be verbose.
Try the demo.
Boto is a Portuguese name given to several types of dolphins and river dolphins native to the Amazon and the Orinoco River tributaries. A few botos exist exclusively in fresh water, and these are often considered primitive dolphins.
The "boto" of the Amazon River regions of northern Brazil are described according to local lore as taking the form of a human or merman, also known as Boto cor-de-rosa ("Pink Boto" in Portuguese) and with the habit of seducing human women and impregnating them.
How to Run on Colab T4
from transformers import AutoTokenizer, pipeline
import torch
model_id = "lucianosb/boto-7B"
tokenizer = AutoTokenizer.from_pretrained(model_id)
pipe = pipeline(
"text-generation",
model=model_id,
torch_dtype=torch.float16,
device_map="cuda:0"
)
def make_prompt(question):
return f"""Abaixo está uma instrução que descreve uma tarefa, combinada com uma entrada que fornece contexto adicional.
Escreva uma resposta que complete adequadamente a solicitação.
### Instruction:
{question}
### Response:
"""
question = "Conte a história do boto"
prompt = make_prompt(question)
sequences = pipe(
prompt,
do_sample=True,
num_return_sequences=1,
eos_token_id=tokenizer.eos_token_id,
max_length=2048,
temperature=0.9,
top_p=0.6,
repetition_penalty=1.15
)
print(sequences[0]["generated_text"])
Métricas
| Tasks | Version | Filter | n-shot | Metric | Value | Stderr | |
|---|---|---|---|---|---|---|---|
| bluex | 1.1 | all | 3 | acc | 0.0083 | ± | 0.0020 |
| enem | 1.1 | all | 3 | acc | 0.0014 | ± | 0.0006 |
| oab_exams | 1.5 | all | 3 | acc | 0.0096 | ± | 0.0012 |
| assin2_rte | 1.1 | all | 15 | f1_macro | 0.9032 | ± | 0.0042 |
| all | 15 | acc | 0.9032 | ± | 0.0042 | ||
| assin2_sts | 1.1 | all | 15 | pearson | 0.4912 | ± | 0.0141 |
| all | 15 | mse | 1.3185 | ± | N/A | ||
| faquad_nli | 1.1 | all | 15 | f1_macro | 0.6104 | ± | 0.0137 |
| all | 15 | acc | 0.6292 | ± | 0.0134 | ||
| hatebr_offensive_binary | 1 | all | 25 | f1_macro | 0.7888 | ± | 0.0078 |
| all | 25 | acc | 0.7936 | ± | 0.0077 | ||
| portuguese_hate_speech_binary | 1 | all | 25 | f1_macro | 0.5503 | ± | 0.0121 |
| all | 25 | acc | 0.5523 | ± | 0.0121 |
Uploaded model
- Developed by: lucianosb
- License: apache-2.0
- Finetuned from model : unsloth/mistral-7b-bnb-4bit
This mistral model was trained 2x faster with Unsloth and Huggingface's TRL library.
Open Portuguese LLM Leaderboard Evaluation Results
Detailed results can be found here
| Metric | Value |
|---|---|
| Average | 62.25 |
| ENEM Challenge (No Images) | 59.97 |
| BLUEX (No Images) | 48.82 |
| OAB Exams | 43.37 |
| Assin2 RTE | 89.58 |
| Assin2 STS | 69.87 |
| FaQuAD NLI | 55.57 |
| HateBR Binary | 77.04 |
| PT Hate Speech Binary | 58.84 |
| tweetSentBR | 57.20 |