NVIDIA® Transformer Engine is a library for accelerating Transformer models on NVIDIA GPUs, including using 8-bit floating point (FP8) precision on Hopper and Ada GPUs, to provide better performance with lower memory utilization in both training and inference.
These pages contain documentation for Transformer Engine release 1.0.0 and earlier releases.
The following documents are provided:
User Guide : Demonstrates how to install and use Transformer Engine release 1.0.0.
Release Notes : Describe the key features, software enhancements and improvements, and known issues for Transformer Engine release 1.0.0.
Software License Agreement (SLA): The software license subject to which Transformer Engine is published. This license is identical to the Apache License, version 2.0, an open source license defined and maintained by the Apache Software Foundation . By accepting this agreement, you agree to comply with all the terms and conditions applicable to the specific product(s) included herein.
Documentation Archive: User Guide and Release Notes for all releases of Transformer Engine, from the first release through the current release.
Notices: Trademark and copyright notices and other legal information relating to Transformer Engine.