Modelpublic

TSUTAYA/qwen3-4b-struct-lora-4

sourceHugging Faceapache-2.0updated 7mo agoView on Hugging Face

0likes7downloads

Model Card

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.

Overview

This adapter is designed to improve structured output reliability across formats such as:

—JSON
—YAML
—XML
—TOML
—CSV

The focus is on increasing formatting stability, bracket correctness, key-value alignment, and structural consistency.

Training Objective

The model is trained via Supervised Fine-Tuning (SFT) with the following design:

—Loss is applied only to the final assistant output
—Intermediate reasoning (Chain-of-Thought) is masked
—Training emphasizes output formatting precision

This setup prioritizes structured output correctness over verbose reasoning.

Training Configuration

—Base model: Qwen/Qwen3-4B-Instruct-2507
—Method: QLoRA (4-bit)
—Max sequence length: 1024
—Epochs: 1
—Learning rate: 2e-05
—LoRA configuration:
—r = 128
—alpha = 256

Usage

python

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

base = "Qwen/Qwen3-4B-Instruct-2507"
adapter = "your_id/your-repo"

tokenizer = AutoTokenizer.from_pretrained(base)

model = AutoModelForCausalLM.from_pretrained(
    base,
    torch_dtype=torch.float16,
    device_map="auto",
)

model = PeftModel.from_pretrained(model, adapter)
model.eval()

Sources & Terms (IMPORTANT)

Training data: daichira/structured-hard-sft-4k

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.