qwen3-4b-structured-sft-lora-v05

qwen3-4b-structured-sft-lora-v05

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).

CoT removed in preprocessing; loss applied to full assistant output.

Training Configuration

• Base model: Qwen/Qwen3-4B-Instruct-2507
• Method: QLoRA (4-bit)
• Max sequence length: 1024
• Epochs: 2
• Learning rate: 2e-6
• LoRA: r=64, alpha=128
• Dataset: 9 datasets merged & cleaned (1500 samples)

Usage

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

base = "Qwen/Qwen3-4B-Instruct-2507"
adapter = "deepkick/qwen3-4b-structured-sft-lora-v05"

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: 9 datasets from u-10bei and daichira (see metadata above).

Dataset License: MIT License. Compliance: Users must comply with the MIT license and the base model's original terms of use.