---
base_model: Qwen/Qwen3-4B-Instruct-2507
datasets:
- takami2022/structured_data_merged_v2v5_0222
language:
- en
license: apache-2.0
library_name: transformers
pipeline_tag: text-generation
tags:
- merged
- sft
- structured-output
---

# qwen3-4b-sft-merged-v2v5ver1

This repository provides a **merged 16-bit model** produced by fine-tuning
**Qwen/Qwen3-4B-Instruct-2507** with QLoRA (4-bit, Unsloth) and then merging the LoRA adapter
into the base model weights.

This model is **fully self-contained** — no adapter loading required.
Intended as the base model for subsequent DPO training.

## Training Configuration

- Base model: Qwen/Qwen3-4B-Instruct-2507
- Dataset: takami2022/structured_data_merged_v2v5_0222
- Method: QLoRA (4-bit) → merged to 16-bit
- Max sequence length: 1024
- Epochs: 3
- Learning rate: 1e-06
- LoRA: r=64, alpha=128
- CoT mask: enabled

## Usage

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

model_id = "takami2022/qwen3-4b-sft-merged-v2v5ver1"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto",
)
```

## Notes

- This merged model is the output of SFT and serves as the starting point for DPO.