NvTensorRtRtx

DeepSeek-R1-Distill-Qwen-1.5B optimization

This folder contains examples of Olive recipes for DeepSeek-R1-Distill-Qwen-1.5B optimization.

FP16 Model Building

The olive recipe DeepSeek-R1-Distill-Qwen-1.5B_model_builder_fp16.json uses ModelBuilder pass to generate the FP16 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 DeepSeek-R1-Distill-Qwen-1.5B_model_builder_fp16.json