VitisAI

TIMM Model Optimization (Quantization)

This folder contains examples of TIMM (PyTorch Image Models) optimization using Olive workflows, focusing on ONNX quantization with QuarkQuantization pass.

Optimization Workflow

This example optimizes timm/mobilenetv3_small_100.lamb_in1k for CPU or NPU execution by:

• Converting PyTorch model to ONNX

• Applying ONNX quantization

• Model: timm/mobilenetv3_small_100.lamb_in1k

• Dataset: timm/mini-imagenet

---

Running the Optimization

Running with Config File

The provided config.json configuration performs ONNX conversion and quantization.

Install Required Dependencies

pip install -r requirements.txt
olive run --config config.json --setup

Run Model Optimization

olive run --config config.json

After running the above command, the model candidates and corresponding config will be saved in the output directory.