84basi/lora-10-2

qwen3-4b-structured-output-lora

This repository provides a full fine-tuned model based on Qwen/Qwen3-4B-Instruct-2507 using BF16 full fine-tuning + NEFTune.

This repository contains the complete model weights.

Training Objective

This model is trained to improve structured output accuracy (JSON / YAML / XML / TOML / CSV).

CoT (Chain-of-Thought) is removed from training data during preprocessing.

Training Configuration

• Base model: Qwen/Qwen3-4B-Instruct-2507
• Method: Full fine-tuning (BF16)
• Max sequence length: 2048
• Epochs: 1
• Learning rate: 2e-05
• NEFTune noise alpha: 5.0

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "84basi/lora-10-2"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto",
)

Sources & Terms (IMPORTANT)

Training data: u-10bei/structured_data_with_cot_dataset_512_v5

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.