Overview

This model, sakuraimo/your-lora-repo-dpo, is a 4 billion parameter language model derived from Qwen/Qwen3-4B-Instruct-2507. It has undergone Direct Preference Optimization (DPO) using the Unsloth library, specifically targeting improvements in reasoning and structured response generation. The repository provides the full-merged 16-bit weights, eliminating the need for adapter loading.

Key Capabilities

• Enhanced Reasoning: Optimized for Chain-of-Thought reasoning, aiming for more logical and coherent multi-step problem-solving.
• Structured Output Quality: Improved ability to generate well-formatted and structured responses based on preference data.
• Direct Use: As a merged model, it can be directly loaded and used with the transformers library without additional LoRA configuration.

Training Details

The model was fine-tuned for 1 epoch with a learning rate of 5e-06 and a beta value of 0.05. It utilized a maximum sequence length of 4096 tokens during DPO training. The training data used was [u-10bei/dpo-dataset-qwen-cot].

Good For

• Applications requiring improved logical reasoning.
• Scenarios where structured and high-quality outputs are critical.
• Developers looking for a readily deployable DPO-tuned Qwen3-4B variant.