qwen3-4b-structured-output-lora-v1
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.
Data preprocessing: Markdown fences (```json, ```yaml, etc.) and
text preambles ("Here's the converted ...", etc.) are automatically stripped
from assistant responses before training, ensuring the model learns to produce
clean structured output without formatting artifacts.
Training Configuration
- Base model: Qwen/Qwen3-4B-Instruct-2507
- Method: QLoRA (4-bit)
- Max sequence length: 1024
- Epochs: 2
- Learning rate: 2e-05
- LoRA: r=64, alpha=128
- CoT masking: enabled
- Data preprocessing: markdown fence stripping (enabled)
Usage
1from transformers import AutoModelForCausalLM, AutoTokenizer
2from peft import PeftModel
3import torch
4
5base = "Qwen/Qwen3-4B-Instruct-2507"
6adapter = "a-kuratani/qwen3-4b-structured-output-lora-v5"
7
8tokenizer = AutoTokenizer.from_pretrained(base)
9model = AutoModelForCausalLM.from_pretrained(
10 base,
11 torch_dtype=torch.float16,
12 device_map="auto",
13)
14model = PeftModel.from_pretrained(model, adapter)
Sources & Terms (IMPORTANT)
Training data: u-10bei/structured_data_with_cot_dataset_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.
