NvTensorRtRtx

Qwen2.5-7B-Instruct optimization

This folder contains examples of Olive recipes for Qwen2.5-7B-Instruct optimization.

INT4 Model Building

The olive recipe Qwen2.5-7B-Instruct_model_builder_int4.json uses ModelBuilder and MatMulNBitsToQDQ passes to generate the INT4 model for NvTensorRTRTXExecutionProvider (aka NvTensorRtRtx EP).

Setup

1. Install Olive

2. Install onnxruntime-genai package that has support for NvTensorRTRTXExecutionProvider.

Steps to run

Use the following command to export the model using Olive with NvTensorRTRTXExecutionProvider:

olive run --config Qwen2.5-7B-Instruct_model_builder_int4.json