qwen3-4b-structeval-lora-v2

Qwen3-4B Structured Output LoRA (Score: 0.79019)

This LoRA adapter was developed for the 2025 Final Competition. Final submission score: 0.79019

This repository provides a LoRA adapter fine-tuned from the base model Qwen/Qwen3-4B-Instruct-2507

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: 512
• Epochs: 2
• Learning rate: 2e-05
• LoRA: r=64, alpha=128

Evaluation Result

Final competition score: 0.79019

Configuration used:

• Epochs: 2
• Learning rate: 2e-05
• QLoRA (4-bit, r=64, alpha=128)

Usage

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

base = "Qwen/Qwen3-4B-Instruct-2507"
adapter = "hamini58/qwen3-4b-structeval-lora-v2"

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 of the training dataset (including copyright notice)
• The original license and terms of use of the base model (Apache License 2.0)

This repository distributes LoRA adapter weights only and does not redistribute the base model.