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.
Speed up the training of AI models on NVIDIA graphics cards while using less memory.
Claim its page: indexed whatever its rank, translated into six languages, and enriched with what you write yourself.
Get an email alert on its next release or when it starts trending — never miss the moment.
Free · no card · unsubscribe anytimeA 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.
TransformerEngine has 3.4k stars on GitHub. It has been forked 774 times. TransformerEngine is written mainly in Python. It has been in active development since 2022. TransformerEngine is available under the Apache-2.0 license. Its main topics are cuda, deep-learning, fp4, fp8.
What is 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.
Is TransformerEngine open source?
TransformerEngine is an open-source project. It is released under the Apache-2.0 license.
Is TransformerEngine free?
Yes. TransformerEngine is free and open source — you can use, modify and self-host it.
What license does TransformerEngine use?
TransformerEngine is available under the Apache-2.0 license.
What language is TransformerEngine written in?
TransformerEngine is written mainly in Python.
Add this live badge to your README — your GitHub stars and directory rank, refreshed daily.
[](https://olud.ai/project/nvidia-transformerengine.html)
Measured from GitHub topics shared by both projects, weighted by how rare each topic is.