MiniGPT-v1.0-10M-BrazilianPortuguese

Hello! My name is Arthur Samuel, I am 13 years old, and I am the creator of this project.

This repository contains MiniGPT, a small transformer-based language model with 10 million parameters. The model is designed to learn reasoning and conversational skills using a custom dataset in Brazilian Portuguese(loboGOAT/Reasoning-MiniGPT-brazilian-portuguese).

About the Model

• Architecture: Decoder-only Transformer (GPT-style)
• Parameters: 10M
• Purpose: To generate text, reason logically, and have casual conversations in Brazilian Portuguese.
• Special Tokens Used: <|PROMPT|>, <|THINK|>, <|MINIGPT|>
• Training Dataset: Brazilian Portuguese dataset with dialogues, reasoning steps, daily-life conversations, jokes, school math (8th grade+), and tips for studying, all formatted using special tokens.

Example Usage

from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("loboGOAT/MiniGPT-v1.0-10M-BrazilianPortuguese")
model = AutoModelForCausalLM.from_pretrained("loboGOAT/MiniGPT-v1.0-10M-BrazilianPortuguese")

prompt = "<|PROMPT|> Olá, MiniGPT! Como você está hoje?\n<|THINK|>"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_length=50)
print(tokenizer.decode(outputs[0]))
License
This project is released under the Apache-2.0 License.

⚠️ REMEMBER: This model is still in training. This is a commercial beta test.

Future plans: An English version will be released, trained on English datasets and optimized for reasoning in English as well.

Created by Arthur Samuel, age 13. MiniGPT is a small but capable transformer trained specifically for reasoning and conversation in Brazilian Portuguese. 🚀