Kauhiro/qwen3-4b-instruct-2507-structured-output-lora-ver18

qwen3-4b-instruct-2507-structured-output-lora

This repository provides a LoRA adapter fine-tuned from Qwen/Qwen3-4B-Instruct-2507 using QLoRA (4-bit, Unsloth).

This repository contains LoRA adapter weights only. The base model must be loaded separately.

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: 1024
• Epochs: 2
• Learning rate: 1e-6
• Grad Accum: 8
• LoRA: r=64, alpha=128
• Dropout: 0.1
• Upsampling: 1
• Upsampling rules json: "xml_to_yaml": 2.0, "json_to_xml": 1.8, "text_to_yaml": 1.6

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
import torch

base = "Qwen/Qwen3-4B-Instruct-2507"
adapter = "Kauhiro/qwen3-4b-instruct-2507-structured-output-lora-ver18"

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

Sources & Terms (IMPORTANT)

Training data: Kauhiro/structured_data_with_cot_dataset_512_collected_from_v2v4v5.
This data was gathered from version 2, 4, and 5 of u-10bei/structured_data_with_cot_dataset_512,
the type of which was conversion only. Total records: 5530

Original datasets:
https://huggingface.co/datasets/u-10bei/structured_data_with_cot_dataset_512_v2
https://huggingface.co/datasets/u-10bei/structured_data_with_cot_dataset_512_v4
https://huggingface.co/datasets/u-10bei/structured_data_with_cot_dataset_512_v5

Dataset License: The dataset used inherits the license of these original datasets.

Compliance: Users must comply with the MIT license (including copyright notice) and the base model's original terms of use.
Base model license: Apache-2.0 https://www.apache.org/licenses/LICENSE-2.0