This repository contains the merged model that merged base model (Qwen/Qwen3-4B-Instruct-2507) and LoRA adapter weights (TakaTaka3/qwen3-4b-lora-adapter_V4) .

Training Objective

This adapter is trained to improve structured output accuracy (JSON / YAML / XML / TOML / CSV).

Loss is applied only to the final assistant output, while intermediate reasoning (Chain-of-Thought) is masked.

Training Configuration

Base model: Qwen/Qwen3-4B-Instruct-2507
Method: QLoRA (4-bit)
Max sequence length: 2048
Epochs: 1
Learning rate: 2e-06
LoRA: r=64, alpha=128

Usage

python

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "TakaTaka3/Qwen3-4B-Instruct-2507-sft-merged_V2"

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 & Terms (IMPORTANT)

Training data: u-10bei/structured_data_with_cot_dataset_512_v2

Dataset License: MIT License. This dataset is used and distributed under the terms of the MIT License. Compliance: Users must comply with the MIT license (including copyright notice) and the base model's original terms of use.