Llama 3 8 B Lexi Uncensored

Introduction

Llama-3-8B-Lexi-Uncensored is an advanced text generation model based on Llama-3-8b-Instruct. This model is designed to provide uncensored, open conversational capabilities while allowing flexibility for compliance through user-implemented alignment layers.

Architecture

The model architecture is based on the Llama-3 framework, which supports a variety of tasks in text generation. It leverages advanced transformer technology to generate human-like text responses across diverse contexts.

Training

The model has been evaluated on several datasets, achieving the following performance metrics:

• AI2 Reasoning Challenge (25-Shot): 59.56% normalized accuracy
• HellaSwag (10-Shot): 77.88% normalized accuracy
• MMLU (5-Shot): 67.68% accuracy
• TruthfulQA (0-shot): 47.72% multiple-choice accuracy
• Winogrande (5-shot): 75.85% accuracy
• GSM8k (5-shot): 68.39% accuracy

These evaluations have been documented on the Open LLM Leaderboard.

Guide: Running Locally

1. Environment Setup:

   • Ensure you have Python installed, along with transformers and safetensors libraries.

2. Model Download:

   • Clone the model repository from Hugging Face: git clone https://huggingface.co/Orenguteng/Llama-3-8B-Lexi-Uncensored.

3. Load the Model:

   • Use the transformers library to load the model.

   from transformers import AutoModelForCausalLM, AutoTokenizer
   
   tokenizer = AutoTokenizer.from_pretrained("Orenguteng/Llama-3-8B-Lexi-Uncensored")
   model = AutoModelForCausalLM.from_pretrained("Orenguteng/Llama-3-8B-Lexi-Uncensored")
   

4. Inference:

   • Use the model to generate text:

   inputs = tokenizer("Your input text here", return_tensors="pt")
   outputs = model.generate(**inputs)
   print(tokenizer.decode(outputs[0]))
   

5. Cloud GPUs:

   • For optimal performance, utilize cloud GPU services such as AWS EC2, Google Cloud, or Azure to handle the computational requirements of the model.

License

Llama-3-8B-Lexi-Uncensored is licensed under the META LLAMA 3 COMMUNITY LICENSE AGREEMENT. Users have the flexibility to use the model, including for commercial purposes, as long as they comply with the terms outlined in Meta's Llama-3 license. The model is uncensored and requires users to implement their own compliance measures.