joaocarloscruz/Qwen3-4B-China-Uncensored-DPO

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:4BQuant:BF16Ctx Length:32kPublished:Jan 8, 2026Architecture:Transformer0.0K Warm

The joaocarloscruz/Qwen3-4B-China-Uncensored-DPO is a 4 billion parameter Qwen3-based language model, fine-tuned using DPO and converted to GGUF format. This model is specifically designed for uncensored Chinese language tasks, offering a distinct alternative to standard Qwen3 models. Its GGUF conversion, facilitated by Unsloth, makes it suitable for efficient local deployment and inference.

Loading preview...

Model Overview

The joaocarloscruz/Qwen3-4B-China-Uncensored-DPO is a 4 billion parameter language model based on the Qwen3 architecture. It has been fine-tuned using Direct Preference Optimization (DPO) to provide an "uncensored" experience, particularly for Chinese language content. The model was converted to the GGUF format using Unsloth, which also facilitated a 2x faster training process.

Key Characteristics

  • Architecture: Based on the Qwen3 model family.
  • Parameter Count: 4 billion parameters, offering a balance between performance and computational efficiency.
  • Fine-tuning: Utilizes DPO for specific content generation characteristics, focusing on uncensored Chinese output.
  • Format: Available in GGUF format (Qwen3-4B-Instruct-China-Uncensored.Q4_K_M.gguf), optimized for local inference with tools like llama.cpp.
  • Training Efficiency: Benefited from Unsloth's optimizations for faster training.

Usage and Deployment

This model is designed for deployment with llama.cpp and similar tools that support the GGUF format. Example command-line usage is provided for both text-only and multimodal llama.cpp clients, leveraging the --jinja flag for prompt templating. Its GGUF format makes it highly accessible for local machine inference.