Llama 3.3 70 B Instruct abliterated

Introduction

HUIHUI-AI's LLAMA-3.3-70B-INSTRUCT-ABLITERATED is an uncensored, modified version of the Meta Llama 3.3 model, designed for text generation and conversational tasks. This model has been adapted to remove refusals using a process called "abliteration." It supports multiple languages and is optimized for use with the Transformers library.

Architecture

The architecture is based on the Meta Llama 3.3 model, which is a large language model built using the PyTorch framework. It supports eight languages and is compatible with the Transformers library, utilizing safetensors for efficient data handling.

Training

The model was initially trained on a vast corpus of multilingual data using Meta's proprietary techniques. The uncensored version was created as a proof-of-concept to demonstrate how refusals can be removed from language models.

Guide: Running Locally

Basic Steps

1. Install Required Libraries: Ensure you have Python installed and update the Transformers library with pip install --upgrade transformers.

2. Set Up the Model: Use the code snippet below to set up the model with Transformers:

   import transformers
   import torch
   
   model_id = "huihui-ai/Llama-3.3-70B-Instruct-abliterated"
   
   pipeline = transformers.pipeline(
       "text-generation",
       model=model_id,
       model_kwargs={"torch_dtype": torch.bfloat16},
       device_map="auto",
   )
   
   messages = [
       {"role": "system", "content": "You are a pirate chatbot who always responds in pirate speak!"},
       {"role": "user", "content": "Who are you?"},
   ]
   
   outputs = pipeline(
       messages,
       max_new_tokens=256,
   )
   print(outputs[0]["generated_text"][-1])
   

3. Run Inference: Execute the script to generate responses based on the input messages.

Cloud GPUs

To optimize performance, consider using cloud-based GPU services such as AWS, Google Cloud, or Azure.

License

The model is licensed under the LLAMA 3.3 Community License. It provides a non-exclusive, worldwide, non-transferable, and royalty-free license to use, reproduce, and modify the Llama Materials. Redistribution requires adherence to specific conditions, including display of attribution and adherence to Meta's Acceptable Use Policy. For products or services exceeding 700 million monthly active users, additional licensing from Meta is required. The license includes disclaimers and limitations of liability typical of open-source agreements.