smzyuki/dpo-qwen-cot-merged

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:4BQuant:BF16Ctx Length:32kPublished:Feb 27, 2026License:apache-2.0Architecture:Transformer Open Weights Warm

The smzyuki/dpo-qwen-cot-merged model is a 4 billion parameter language model, fine-tuned from Qwen/Qwen3-4B-Instruct-2507 using Direct Preference Optimization (DPO). This model specializes in enhancing reasoning capabilities through Chain-of-Thought (CoT) and improving structured response quality. It is designed for applications requiring precise, aligned outputs, particularly in complex reasoning tasks.

Loading preview...

Model Overview

This model, smzyuki/dpo-qwen-cot-merged, 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, with its 16-bit weights fully merged for direct use without adapter loading.

Key Capabilities

  • Enhanced Reasoning: Optimized specifically to improve Chain-of-Thought (CoT) reasoning, making it suitable for tasks requiring multi-step logical deduction.
  • Structured Response Quality: Fine-tuned to produce more aligned and structured outputs based on preference datasets.
  • Efficient Deployment: As a merged model, it simplifies deployment by eliminating the need for separate adapter loading.

Training Details

The model was trained for 2 epochs with a learning rate of 1e-07 and a beta value of 0.1, using a maximum sequence length of 2048. The training leveraged datasets such as u-10bei/structured_data_with_cot_dataset_512_v2, u-10bei/structured_data_with_cot_dataset_512_v5, and daichira/structured-5k-mix-sft.

Licensing

The model operates under the MIT License, consistent with its training data. Users must also adhere to the original base model's license terms.