Model Overview

This model, nakotsuko13/qwen3-4b-nako13-dpo-qwen-cot-merged, is a specialized variant of the Qwen/Qwen3-4B-Instruct-2507 base model, developed by nakotsuko13. It has undergone a two-stage fine-tuning process to achieve high accuracy and strict adherence to structured output formats.

Key Capabilities

• Structured Data Generation: Highly optimized for generating precise JSON, XML, CSV, and YAML structures.
• Two-Stage Fine-Tuning: Utilizes both Supervised Fine-Tuning (SFT) and Direct Preference Optimization (DPO) for enhanced performance.
  • SFT Stage: Trained on over 16,500 samples to master various structured formats using the nakotsuko13/qwen3-4b-nako13-structured-output-lora adapter.
  • DPO Stage: Further refined using the u-10bei/dpo-dataset-qwen-cot dataset to eliminate conversational filler and ensure direct, raw structured outputs.
• Full-Merged 16-bit Model: Ready for direct use with standard transformers or vLLM libraries.

Ideal Use Cases

• API Response Generation: Creating structured JSON or XML responses for applications.
• Data Extraction: Extracting information into structured formats from unstructured text.
• Configuration File Generation: Producing YAML or other structured configuration files.
• Automated Data Processing: Any task requiring strict, non-conversational structured output.