README.md

# LexGPT-Beta

## Description
LexGPT-Beta is a sophisticated Large Language Model (LLM) that has been fine-tuned using Mistral-7B-Instruct-v0.2 as its base. It is specifically designed to excel in mathematical instruction and reasoning, showcasing a wide array of mathematical and logical capabilities. The model is still in its beta stage, with continuous development and testing ongoing.

## Model Information
- **Base Model**: Mistral-7B-Instruct-v0.2, a versatile LLM renowned for its proficiency in understanding and adhering to intricate instructions. This model surpasses its predecessor, Mistral-7B-Instruct-v0.1, and is widely recognized for its applicability across various AI domains.
- **Fine-tuning Datasets**:
  - [TIGER-Lab/MathInstruct](https://huggingface.co/datasets/TIGER-Lab/MathInstruct)
  - [meta-math/MetaMathQA](https://huggingface.co/datasets/meta-math/MetaMathQA)
  - [LDJnr/Capybara](https://huggingface.co/datasets/LDJnr/Capybara)
- **Fine-tuning Time**: Approximately 90 hours.
- **Parameters**: 7.24 billion.
- **Uncensored**: The model does not include inherent moderation tools.

## Application
LexGPT-Beta can be employed for a multitude of tasks, particularly shining in those involving mathematical reasoning and instruction comprehension. To fully exploit its potential, it's advisable to formulate prompts with concise and explicit instructions.

## Comparison to ChatGPT
Preliminary assessments suggest that LexGPT-Beta's performance is on par with ChatGPT, particularly in mathematical scenarios. However, no Humaneval has been executed to definitively compare the two.

## Development Standing
LexGPT-Beta is presently in the beta phase. Input and collaboration are encouraged to expand the model's functionalities and applications.

## Caution
As LexGPT-Beta is uncensored and in its beta phase, users are urged to exercise caution when utilizing the model. The model's outputs should be scrutinized, especially in delicate contexts.

## Credits
LexGPT-Beta is derived from Mistral-7B-Instruct-v0.2, created by the Mistral AI Team. The fine-tuning process leveraged datasets accessible on Hugging Face, significantly contributing to the model's specialized features.