NVIDIA /

TransformerEngine

A library for accelerating Transformer models on NVIDIA GPUs, including using 8-bit and 4-bit floating point (FP8 and FP4) precision on Hopper, Ada and Blackwell GPUs, to provide better performance with lower memory utilization in both training and inference.

Solana × Meteora

Launch a repository token.

Create a community token linked to this repository using Meteora’s bonding curve on Solana.

Start a token launchExplore Solana launches →

A community launch does not establish ownership or endorsement of this repository.

Behind the repository

Default branch: main

License: Apache-2.0

Open issues / PRs: 366

Created: Sep 20, 2022

cudadeep-learningfp4fp8gpujaxmachine-learningpythonpytorch

Top contributors

ksivamantimmoon10phu0ngngptrendxcyanguwapggPL

Contributors are shown as project information, not fee recipients.