Qw Q 32 B Preview abliterated

Introduction

The QWQ-32B-PREVIEW-ABLITERATED model is an uncensored variant of the Qwen/QwQ-32B-Preview, designed to enhance text generation capabilities by removing refusals. This model employs a technique known as "abliteration" to achieve its uncensored status. It is a proof-of-concept implementation aimed at demonstrating how refusals can be eliminated from a large language model (LLM) without using TransformerLens.

Architecture

• Base Model: Qwen/QwQ-32B-Preview
• Language: English
• Pipeline Tag: Text Generation
• Library: Transformers
• Tags: Chat, Abliterated, Uncensored

Training

The model uses a process termed "abliteration," implemented as per the methodology outlined in the remove-refusals-with-transformers repository. The goal of this process is to modify the model to avoid refusals during inference, allowing for more fluid and interactive conversation generation.

Guide: Running Locally

To run the model locally, you can use the ollama command-line tool. Below are the basic steps:

1. Install Ollama: Ensure that Ollama is installed on your system to facilitate running models.
2. Run the Model: Execute the following command:

   ollama run huihui_ai/qwq-abliterated
   

3. Cloud GPU Suggestion: For optimal performance, consider using cloud-based GPUs such as those provided by AWS, Google Cloud, or Azure, which can handle larger models more efficiently.

License

This model is licensed under the Apache-2.0 License. For details, view the license file at LICENSE.