rajkumar4466/nj-housing-qlora-onnx

QLoRA Fine-tuned Qwen2.5-0.5B (ONNX) — NJ Housing Price Prediction

Qwen2.5-0.5B fine-tuned with QLoRA (4-bit NF4) on NJ housing data, then merged and exported to ONNX format for CPU inference.

Metrics (held-out test set)

Metric	QLoRA (this model)	XGBoost Baseline
MAE	$140,141	$128,013
RMSE	$190,172	$168,135
R²	0.6359	0.7154
MAPE	23.0%	22.7%

Model Details

• Base model: Qwen/Qwen2.5-0.5B
• Fine-tuning: QLoRA (rank=16, alpha=32, 4-bit NF4 quantization)
• Export: Merged to fp32, exported to ONNX
• Inference: ONNX Runtime (CPU), no GPU required

Prompt Format

Property: Single Family in zip 07650. 3 bedrooms, 2.0 bathrooms, 1500 sqft living area, 0.25 acre lot, built in 1990. Predicted price: $

The model generates the price value after the $ prefix.

Usage

import onnxruntime as ort
from transformers import AutoTokenizer
from huggingface_hub import snapshot_download

path = snapshot_download("rajkumar4466/nj-housing-qlora-onnx")
session = ort.InferenceSession(f"{path}/model.onnx")
tokenizer = AutoTokenizer.from_pretrained(path)

Dataset

Trained on rajkumar4466/nj-housing-prices

Comparison

See rajkumar4466/nj-housing-xgboost-baseline for the XGBoost baseline comparison.