A guide to building long-term compounding knowledge infrastructure. See details on GitHub .

Transformer Engine

Transformer Engine provides high-performance kernels and mixed-precision support for Transformer models.

Overview

Transformer Engine provides optimized kernels and FP8/mixed-precision support to accelerate Transformer training and inference on NVIDIA hardware.

Key features

  • FP8 convergence recipes and optimized kernels.
  • Integrations with PyTorch and other frameworks.

Use cases

  • High-performance Transformer training and inference.

Technical highlights

  • Low-level kernel optimizations targeting NVIDIA accelerators.

Comments

Transformer Engine
Resource Info
Author NVIDIA
Added Date 2025-10-02
Open Source Since 2022-09-20
Tags
ML Platform Open Source