---
base_model: shinich001/qwen3-4b-lr5e5-ep1-seq2k
datasets:
- u-10bei/dpo-dataset-qwen-cot
language:
- en
license: apache-2.0
library_name: transformers
pipeline_tag: text-generation
tags:
- dpo
- unsloth
- qwen
- alignment
---

# qwen3-4b-seq2k-dpo-merged

This model is a fine-tuned version of **shinich001/qwen3-4b-lr5e5-ep1-seq2k** using **Direct Preference Optimization (DPO)** via the **Unsloth** library.
This repository contains the **full-merged 16-bit weights**. No adapter loading is required.

## Training Configuration
- **Base model**: shinich001/qwen3-4b-lr5e5-ep1-seq2k
- **Method**: DPO 
- **Epochs**: 1
- **Learning rate**: 5e-06
- **Beta**: 0.1
- **Max sequence length**: 2048