This is a LoRA adapter fine-tuned on a custom dataset optimized for structured output tasks (JSON/XML). The model was trained using Unsloth and QLoRA on a Tesla T4 GPU.

Training Details

Base Model: Qwen/Qwen3-4B-Instruct-2507
Training Data: A filtered and sampled combination (1,500 entries) of:
- u-10bei/structured_data_with_cot_dataset_512_v2
- daichira/structured-hard-sft-4k
Sequence Length: 1024
Epochs: 2
Rank: 64
Alpha: 128

Usage

python

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

base_model = "Qwen/Qwen3-4B-Instruct-2507"
# Replace with your username/repo
adapter_name = "NTA2/qwen3-4b-structured-mix"

model = AutoModelForCausalLM.from_pretrained(base_model, load_in_4bit=True, device_map="auto")
model = PeftModel.from_pretrained(model, adapter_name)
tokenizer = AutoTokenizer.from_pretrained(base_model)