|
--- |
|
language: |
|
- pt |
|
license: llama3 |
|
library_name: transformers |
|
tags: |
|
- portuguese |
|
- llama |
|
- cabra |
|
- llama-3 |
|
datasets: |
|
- botbot-ai/Cabra3k |
|
model-index: |
|
- name: CabraLlama3-70b |
|
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: 82.02 |
|
name: accuracy |
|
source: |
|
url: https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=botbot-ai/CabraLlama3-70b |
|
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: 70.1 |
|
name: accuracy |
|
source: |
|
url: https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=botbot-ai/CabraLlama3-70b |
|
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: 68.52 |
|
name: accuracy |
|
source: |
|
url: https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=botbot-ai/CabraLlama3-70b |
|
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: 93.21 |
|
name: f1-macro |
|
source: |
|
url: https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=botbot-ai/CabraLlama3-70b |
|
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: 83.32 |
|
name: pearson |
|
source: |
|
url: https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=botbot-ai/CabraLlama3-70b |
|
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: 80.6 |
|
name: f1-macro |
|
source: |
|
url: https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=botbot-ai/CabraLlama3-70b |
|
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: 81.62 |
|
name: f1-macro |
|
source: |
|
url: https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=botbot-ai/CabraLlama3-70b |
|
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: 72.72 |
|
name: f1-macro |
|
source: |
|
url: https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=botbot-ai/CabraLlama3-70b |
|
name: Open Portuguese LLM Leaderboard |
|
- task: |
|
type: text-generation |
|
name: Text Generation |
|
dataset: |
|
name: tweetSentBR |
|
type: eduagarcia/tweetsentbr_fewshot |
|
split: test |
|
args: |
|
num_few_shot: 25 |
|
metrics: |
|
- type: f1_macro |
|
value: 73.85 |
|
name: f1-macro |
|
source: |
|
url: https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard?query=botbot-ai/CabraLlama3-70b |
|
name: Open Portuguese LLM Leaderboard |
|
--- |
|
# Cabra Llama-3 70B |
|
|
|
O Cabra Llama-3 70B é uma versão aprimorada do Meta Llama 3 70B Instruct, refinado com o uso do dataset Cabra 30k. Este modelo foi especialmente otimizado para compreender e responder em português (pt-br). |
|
|
|
**Conheça os nossos outros [modelos e datasets](https://huggingface.co/collections/botbot-ai/models-6604c2069ceef04f834ba99b), e o [Cabra Llama 3 8b](https://huggingface.co/botbot-ai/CabraLlama3-8b).** |
|
|
|
## Detalhes do modelo base |
|
|
|
### Modelo: Meta-Llama-3-70B-Instruct |
|
|
|
A Meta desenvolveu e lançou a família de modelos Llama 3, uma coleção de modelos de texto generativos pré-treinados e ajustados por instruções nos tamanhos de 8B e 70B. Os modelos Llama 3 ajustados por instruções são otimizados para casos de uso em diálogos e superam muitos dos modelos de chat de código aberto disponíveis em benchmarks comuns da indústria. Além disso, ao desenvolver esses modelos, tomamos grande cuidado para otimizar a utilidade e a segurança. |
|
|
|
Arquitetura do Modelo: Llama 3 é um modelo de linguagem auto-regressivo que usa uma arquitetura de transformador otimizada. As versões ajustadas utilizam o aprimoramento supervisionado (SFT) e aprendizado por reforço com feedback humano (RLHF) para se alinhar às preferências humanas quanto à utilidade e segurança. |
|
|
|
### Dataset: Cabra 30k |
|
|
|
Dataset interno para fine-tuning. Vamos lançar em breve. |
|
|
|
### Quantização / GGUF |
|
|
|
Colocamos diversas versões (GGUF) quantanizadas no branch "quantanization". |
|
|
|
# Open Portuguese LLM Leaderboard Evaluation Results |
|
|
|
Detailed results can be found [here](https://huggingface.co/datasets/eduagarcia-temp/llm_pt_leaderboard_raw_results/tree/main/botbot-ai/CabraLlama3-70b) and on the [🚀 Open Portuguese LLM Leaderboard](https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard) |
|
|
|
| Metric | Value | |
|
|--------------------------|---------| |
|
|Average |**78.44**| |
|
|ENEM Challenge (No Images)| 82.02| |
|
|BLUEX (No Images) | 70.10| |
|
|OAB Exams | 68.52| |
|
|Assin2 RTE | 93.21| |
|
|Assin2 STS | 83.32| |
|
|FaQuAD NLI | 80.60| |
|
|HateBR Binary | 81.62| |
|
|PT Hate Speech Binary | 72.72| |
|
|tweetSentBR | 73.85| |
|
|
|
|