Model Overview

This model, toshiyuki-kato/dpo-qwen-cot-merged, is a 4 billion parameter language model based on the Qwen3-4B-Instruct-2507 architecture. It has been fine-tuned by toshiyuki-kato using Direct Preference Optimization (DPO) via the Unsloth library, with its 16-bit weights fully merged into the base model.

Key Capabilities & Optimization

The primary objective of this DPO fine-tuning was to align the model's responses with preferred outputs, specifically focusing on:

• Improved Reasoning: Enhancing Chain-of-Thought (CoT) capabilities.
• Structured Response Quality: Delivering more coherent and structured outputs based on a preference dataset.

Training Details

The model underwent 1 epoch of DPO training with a learning rate of 1e-07 and a beta value of 0.1. The maximum sequence length used during training was 1024 tokens. The training utilized a LoRA configuration (r=8, alpha=16) which has since been merged into the base model, meaning no adapter loading is required for inference.

Usage & Licensing

As a merged model, it can be directly used with the transformers library. The training data, u-10bei/dpo-dataset-qwen-cot, is licensed under the MIT License, and users must also comply with the original base model's license terms.