Hi-Satoh/adv_sft_dpo_final_4_merged

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:4BQuant:BF16Ctx Length:32kPublished:Feb 28, 2026License:apache-2.0Architecture:Transformer Open Weights Warm

Hi-Satoh/adv_sft_dpo_final_4_merged is a 4 billion parameter causal language model, fine-tuned from Qwen/Qwen3-4B-Instruct-2507 using Direct Preference Optimization (DPO) via Unsloth. This model is specifically optimized to improve reasoning capabilities, particularly Chain-of-Thought, and enhance structured response quality. It is designed for use cases requiring high-quality, aligned outputs based on preference datasets.

Loading preview...

Model Overview

Hi-Satoh/adv_sft_dpo_final_4_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), leveraging the Unsloth library to align its responses with preferred outputs. This repository provides the full-merged 16-bit weights, eliminating the need for adapter loading.

Key Capabilities

  • Enhanced Reasoning: Optimized to improve Chain-of-Thought reasoning processes.
  • Structured Response Quality: Focuses on generating higher quality and more structured outputs.
  • DPO Alignment: Benefits from Direct Preference Optimization for better alignment with desired response characteristics.

Training Details

The model underwent 1 epoch of DPO training with a learning rate of 7e-08 and a beta value of 0.5. It utilized a maximum sequence length of 4096 tokens. The LoRA configuration (r=8, alpha=16) was merged into the base model during the fine-tuning process.

Good For

  • Applications requiring improved reasoning abilities.
  • Scenarios where structured and high-quality responses are critical.
  • Use cases benefiting from models aligned through preference-based learning.