Llama 3.1 8 B Lexi Uncensored V2 G G U F

Introduction

Llama-3.1-8B-Lexi-Uncensored-V2 is a text generation model developed by Orenguteng, based on Llama-3.1-8b-Instruct. It is designed to provide uncensored responses, allowing for higher compliance in various applications. Users are advised to implement an alignment layer before deploying the model as a service.

Architecture

The model, with a size of 8B parameters, supports text generation tasks. It is built to be compliant with requests, but users should be aware of its potential to generate unethical content without proper alignment.

Training

The model has been evaluated on several datasets with the following results:

• IFEval (0-Shot): 77.92% strict accuracy
• BBH (3-Shot): 29.69% normalized accuracy
• MATH Lvl 5 (4-Shot): 16.92% exact match
• GPQA (0-Shot): 4.36% normalized accuracy
• MuSR (0-Shot): 7.77% normalized accuracy
• MMLU-PRO (5-Shot): 30.9% accuracy

These results are sourced from the Open LLM Leaderboard.

Guide: Running Locally

1. System Requirements: Ensure you have the necessary dependencies installed for running the model.
2. Model Download: Obtain the model from Hugging Face's model hub.
3. Setup Environment: Configure your environment to accommodate the model's requirements.
4. Inference: Use the same template as the official Llama 3.1 8B instruct. System tokens must be present during inference.
5. Cloud GPUs: For optimal performance, consider utilizing cloud GPUs such as those offered by AWS, Google Cloud, or Azure.

License

This model is licensed under the META LLAMA 3.1 COMMUNITY LICENSE AGREEMENT. It permits any use, including commercial applications, in accordance with Meta's Llama-3.1 license. Users are responsible for the content generated by the model.