qwen3-4b-structured-output-lora-v7

qwen3-4b-structured-output-lora-v7

This repository provides a LoRA adapter (v7) 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: v7 — Data Quality Improvement

This is v7 of the SFT training, focusing on data quality improvement with TOML 3x upsampling, XML escaping, and Deep Structure YAML/JSON. v5.2 achieved 0.74648. v7 aims to improve accuracy by addressing data balance and quality issues.

Changes from v5.2

v7 Key Improvements

1. TOML 3x Upsampling: 611 → 1,833 samples (address TOML underrepresentation)
2. XML Escaping: Proper escaping of &, <, >, ", ' characters
3. Deep Structure: Added complex nested YAML/JSON data
4. Same hyperparameters: Maintained v5.2's proven settings (LR=5e-6, Epoch=1)

Score History

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: 1e-06
• Warmup ratio: 10%
• 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-v7"

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: 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.