---
base_model: Qwen/Qwen3-4B-Instruct-2507
datasets:
- dpo_train_brushed_v4_balanced.json
language:
- en
license: apache-2.0
library_name: transformers
pipeline_tag: text-generation
tags:
- dpo
- unsloth
- qwen
- alignment
---

# 東京大学 松尾・岩澤研究室　大規模言語モデル　応用講座2025-2026

## Author and Acknowledgments

- **Author:** Toshiki Demizu (出水 利樹) — GitHub/Hugging Face ID: [@demimomi](https://huggingface.co/demimomi)  
- **Affiliation:** ソフトバンク株式会社、MONET Technologies株式会社
- **Course:** Large Language Model Development Lecture Advanced (Winter 2025－2026)  
- **Participants:** 3800名参加 

## メインコンペ(2026年2月2日～3月2日)

- **状況:**

  2026年2月8日現在 293位（現時点で497人が提出） 0.70044点

  2026年2月11日現在 261位（現時点で646人が提出）0.73407点
  https://huggingface.co/demimomi/demimomi44taomax-qwen3-4b-structured-output-lora

  T4（TPU）だと日次Limitにすぐ達するため、A100（GPU）にて学習/推論コードを実施。

- **ルール:**

基準点：0.7　※コード脳死で回すだけでは超えられない

　１Google Colabで実行可能なモデル・実装であること 

　２評価は StructEval（Text）のみを使用

　３提出物は推論結果JSONとHugging Face上のモデルURL

　４運営指定モデル・データのみ使用可

　５Omnicampusに提出すると自動採点・順位付け

# ＜demimomi-max44-qwen3-4b-dpo-qwen-cot-merged（ 0.70044点版）＞

## model

This model is a fine-tuned version of **Qwen/Qwen3-4B-Instruct-2507** using **Direct Preference Optimization (DPO)** via the **Unsloth** library.

This repository contains the **full-merged 16-bit weights**. No adapter loading is required.

## Training Objective
This model has been optimized using DPO to align its responses with preferred outputs, focusing on improving reasoning (Chain-of-Thought) and structured response quality based on the provided preference dataset.

## Training Configuration
- **Base model**: Qwen/Qwen3-4B-Instruct-2507
- **Method**: DPO (Direct Preference Optimization)
- **Epochs**: 2
- **Learning rate**: 1e-06
- **Beta**: 0.05
- **Max sequence length**: 1536
- **LoRA Config**: r=8, alpha=16 (merged into base)

## Usage
Since this is a merged model, you can use it directly with `transformers`.

```python
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "demimomi/dpo-qwen-cot-merged"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.float16,
    device_map="auto"
)

# Test inference
prompt = "Your question here"
inputs = tokenizer.apply_chat_template([{"role": "user", "content": prompt}], tokenize=True, add_generation_prompt=True, return_tensors="pt").to("cuda")
outputs = model.generate(**inputs, max_new_tokens=512)
print(tokenizer.decode(outputs[0]))

```

## Sources & License (IMPORTANT)

* **Training Data**: [dpo_train_brushed_v4_balanced.json]
* **License**: MIT License. (As per dataset terms).
* **Compliance**: Users must follow the original base model's license terms.