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.
Star History
Momentum
+89
STARS · LAST 30 DAYS
3
PER DAY
#201
MOST-STARRED Python
| Window | 7 days | 30 days | 90 days |
|---|---|---|---|
| Stars gained | +21 | +89 | +270 |
| Per day | 3 | 3 | 3 |
| Forks gained | +4 | +17 | +42 |
TransformerEngine gained 89 stars in the last 30 days, about 3 a day, and now has 3.6K. It is about 4 years old and has averaged roughly 889 stars a year. It ranks #201 among Python repositories and #1,350 across all languages on GitHubRepo.
Trending Record
5
DAYS ON TRENDING
#640
BEST RANK
Sep 24, 2026
FIRST APPEARANCE
Active
STATUS TODAY
TransformerEngine has maintained a continuous presence across global trending indexes, peaking at #640. Below is the 30-day activity profile:
💡 Overview
TransformerEngine is an open-source project written in Python: 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.
Engineered for speed, consistency, and developer ease, it solves common hurdles in cuda, deep-learning, fp4. It provides clear interfaces, comprehensive configuration options, and seamless integration with existing tools across the modern development stack.
⚡ Key Features
Optimized execution pipeline written in Python for predictable speed.
Zero-friction configuration with comprehensive sensible defaults out of the box.
Cross-platform runtime support across Linux, macOS, and Windows environments.
Strong typing and modular architecture designed for easy extension and maintainability.
Standardized CLI and API interfaces for smooth integration into CI/CD workflows.
Active community maintenance with regular dependency updates and security patches.
📥 Installation
$ pip install TransformerEngine⚙ System Requirements
Platforms
- • macOS
- • Linux
- • Windows
Runtime & Dependencies
Python >= 3.9, pip, virtualenv
Architecture
x86_64, ARM64 (Apple Silicon & Graviton)
🧠 How It Works
TransformerEngine coordinates its core functionality through a modular Python pipeline. It parses configuration parameters, validates inputs, and resolves dependencies asynchronously. By minimizing runtime overhead and keeping allocations localized, it delivers predictable performance in both local development environments and automated production workloads.
🎯 Production Use Cases
Production System Integration
Embed TransformerEngine into Python backend services to handle core application logic.
CI/CD Automated Pipelines
Run automated validation, builds, and integration suites during deployments.
Developer Tooling & Workflows
Accelerate developer onboarding with pre-configured project utilities.
Open Source Extension
Fork and customize internal modules under the repository's open Apache-2.0 license.
🚀 Getting Started
Install TransformerEngine using your package manager: `pip install TransformerEngine`
Initialize your project workspace or configuration file for TransformerEngine.
Import TransformerEngine into your codebase or invoke it directly from your terminal.
Execute your test suite or run `TransformerEngine --help` to verify successful setup.
👍 Strengths
⚠️ Considerations
⇄ Alternatives & Direct Competitors
👥 Who Should Use This
Developers and engineering teams building with Python, seeking reliable, tested, and actively maintained tooling for production workloads.
🏆 Nearby in the Rankings
NVIDIA/TransformerEngine is currently ranked #1,350 by stars across every repository tracked on GitHubRepo. These are adjacent projects:
| Rank | Repository | Language | Stars | Action |
|---|---|---|---|---|
| #1,345 | symfony/framework-bundle | PHP | ★ 3.6K | Compare ↗ |
| #1,345 | denoland/std | TypeScript | ★ 3.6K | Compare ↗ |
| #1,347 | huxingyi/dust3d | C++ | ★ 3.6K | Compare ↗ |
| #1,348 | jordan-gibbs/hyperresearch | Python | ★ 3.6K | Compare ↗ |
| #1,349 | grafana/alloy | Go | ★ 3.6K | Compare ↗ |
| #1,350 | NVIDIA/TransformerEngine This Project | Python | ★ 3.6K | |
| #1,351 | okteto/okteto | Go | ★ 3.5K | Compare ↗ |
| #1,352 | gmh5225/awesome-game-security | Python | ★ 3.5K | Compare ↗ |
| #1,353 | unitycatalog/unitycatalog | Java | ★ 3.5K | Compare ↗ |
| #1,354 | neo-project/neo | C# | ★ 3.5K | Compare ↗ |
| #1,355 | fasiondog/hikyuu | C++ | ★ 3.5K | Compare ↗ |
Frequently Asked Questions
What does TransformerEngine do? +
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.
What language is TransformerEngine written in? +
The primary language is Python. Topics include: cuda, deep-learning, fp4, fp8, gpu.
Is TransformerEngine actively maintained? +
Yes, the last recorded push was on Sep 25, 2026 with 357 open issues being tracked.
How many stars does TransformerEngine have? +
TransformerEngine has 3,556 stars and 839 forks on GitHub.
How does TransformerEngine rank among GitHub repositories? +
With 3,556 stars, NVIDIA/TransformerEngine is ranked #1,350 globally across all repositories tracked on GitHubRepo and #201 among Python projects.
What license is TransformerEngine distributed under? +
The repository reports a Apache-2.0 license. Always verify the repository LICENSE file for legal terms.
