Model Overview

This model, helloworldabc/dpo-qwen-cot-merged, is a 4 billion parameter language model derived from the Qwen/Qwen3-4B-Instruct-2507 base model. It has undergone Direct Preference Optimization (DPO) using the Unsloth library, resulting in a fully merged 16-bit weight model that requires no adapter loading.

Key Capabilities

• Enhanced Reasoning: Optimized to improve Chain-of-Thought (CoT) reasoning, making it suitable for tasks requiring logical deduction and multi-step problem-solving.
• Structured Response Quality: Fine-tuned to produce more coherent and structured outputs, aligning with preferred response formats.
• DPO Alignment: Utilizes DPO to align model responses with human preferences, based on a specific preference dataset (u-10bei/dpo-dataset-qwen-cot).

Training Details

The model was trained for 1 epoch with a learning rate of 1e-07 and a beta value of 0.1. It supports a maximum sequence length of 1024 tokens during training. The LoRA configuration (r=8, alpha=16) was merged directly into the base model.

Usage

As a merged model, it can be directly integrated and used with the transformers library for inference, offering straightforward deployment for various applications.