qwen3-4b-structured-output-lora-v3

This repository provides a LoRA adapter (v3) 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.

Version: v3 — Data Scaling

This is v3 of the SFT training, focusing on data quantity increase. Based on v2's success (score: 0.75074), we doubled the training data.

Changes from v2

Parameter	v2	v3	Rationale
Dataset	1-1_512_v2 (3,933)	Merged (8,541)	+117% data for better pattern learning
MAX_SEQ_LEN	1024	1024	Same as v2
Epochs	1	1	Same as v2
Learning Rate	5e-6	5e-06	Same as v2

Merged Dataset Composition

• 1-1_512_v2: 3,933 samples
• 1-2_512_v4: 4,608 samples
• Total: 8,541 samples

Format distribution: XML(23.4%), JSON(23.3%), YAML(18.2%), TOML(17.8%), CSV(17.3%)

Training Objective

This adapter is trained to improve structured output accuracy (JSON / YAML / XML / TOML / CSV) for the StructEval-T benchmark.

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, Unsloth)
• Max sequence length: 1024
• Epochs: 1
• Learning rate: 5e-06
• Batch size: 2 (effective: 16)
• Gradient accumulation: 8
• LoRA: r=64, alpha=128
• CoT masking: enabled (loss on final output only)

Usage

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

base = "Qwen/Qwen3-4B-Instruct-2507"
adapter = "your_id/qwen3-4b-structured-output-lora-v3"

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: merged dataset (1-1_512_v2 + 1-2_512_v4)

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.