OguraHiroyuki/dpo-qwen-cot-merged

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:4BQuant:BF16Ctx Length:32kPublished:Feb 23, 2026License:apache-2.0Architecture:Transformer Open Weights Warm

OguraHiroyuki/dpo-qwen-cot-merged is a 4 billion parameter causal language model, fine-tuned from Qwen/Qwen3-4B-Instruct-2507 using Direct Preference Optimization (DPO). This model is specifically optimized to improve reasoning capabilities through Chain-of-Thought (CoT) and enhance structured response quality. It is designed for tasks requiring aligned and coherent outputs based on preferred data, offering a 32768 token context length.

Loading preview...

Model Overview

This model, OguraHiroyuki/dpo-qwen-cot-merged, is a 4 billion parameter language model derived from Qwen/Qwen3-4B-Instruct-2507. It has been fine-tuned using Direct Preference Optimization (DPO) via the Unsloth library, incorporating a specific preference dataset to guide its learning.

Key Capabilities

  • Enhanced Reasoning: Optimized to improve Chain-of-Thought (CoT) reasoning, enabling more structured and logical problem-solving.
  • Improved Response Quality: Focuses on generating higher-quality, aligned, and structured outputs based on preferred examples.
  • Full-Merged Weights: Provided as full-merged 16-bit weights, eliminating the need for adapter loading and simplifying deployment.

Training Details

The model underwent DPO training for 1 epoch with a learning rate of 1e-07 and a beta value of 0.1. It was trained with a maximum sequence length of 1024, utilizing a LoRA configuration (r=8, alpha=16) that was subsequently merged into the base model. The training data used is sourced from [u-10bei/dpo-dataset-qwen-cot].

Usage

This model can be directly integrated and used with the transformers library for inference, supporting a context length of 32768 tokens. Users should adhere to the MIT License of the training data and the original base model's license terms.