Skip to content

README.md

Latest commit

 

History

History
21 lines (12 loc) · 644 Bytes

README.md

File metadata and controls

21 lines (12 loc) · 644 Bytes
Raw

Llama-3.1-8B-Instruct optimization

This folder contains examples of Olive recipes for Llama-3.1-8B-Instruct optimization.

INT4 Model Building

The olive recipe Llama-3.1-8B-Instruct_model_builder_int4.json uses ModelBuilder pass 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 Llama-3.1-8B-Instruct_model_builder_int4.json