Qwen2 Boundless

Introduction

Qwen2-Boundless is a fine-tuned model derived from Qwen2-1.5B-Instruct, designed to generate responses to a variety of questions, including those involving ethical, illegal, pornographic, and violent content. It is trained on a Chinese dataset to handle complex scenarios effectively and performs optimally in Chinese.

Architecture

Qwen2-Boundless leverages a causal language model architecture, fine-tuned for text-to-text generation. It incorporates features for continuous conversation and streaming responses, making it suitable for applications requiring interactive text generation.

Training

The model was trained using a dataset titled bad_data.json, which contains diverse text content related to ethics, law, pornography, and violence. Additional data from cybersecurity contexts was also used, sourced from the SecGPT project. The training dataset is entirely in Chinese, enhancing the model's performance in that language.

Guide: Running Locally

1. Setup Environment: Ensure you have Python and PyTorch installed. Install the transformers library.
2. Load the Model: Use the provided Python script to load the model and tokenizer.
3. Device Configuration: Set the device to cuda if using a GPU, or to cpu for CPU-only systems.
4. Run the Model: Execute the script to interact with the model in a continuous conversation or streaming mode.
5. Cloud GPUs: For optimal performance, consider using cloud GPU services like AWS EC2 or Google Cloud.

License

The Qwen2-Boundless model and its dataset are released under the Apache 2.0 License, allowing for extensive use with few restrictions.