CatalogWantedMirrorsLedgerOperating StatementDocsStudio statusCard system
alt.freeweights.tools / Transformer Engine
From: NVIDIA <github.com/NVIDIA>Newsgroups: alt.freeweights.toolsSubject:

Transformer Engine

Release: v2.21.0 (2026-11-10)Commit: 7457e47License: Apache-2.0

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.

trainingv2.21.0releaseTransformer EnginePythonApache-2.0
FactsLicenseApache-2.0LanguagePythonKindtrainingRelease2026-11-10PlatformsLinux (NVIDIA)Links to the projectpermissive
$ pip install --no-build-isolation "transformer_engine[pytorch]"
OverviewInstallUse with FreeweightsLicenseProvenance

Overview

FP8 and mixed-precision building blocks for transformer training on NVIDIA GPUs.

Kind
training
License
Apache-2.0
Language
Python
Platforms
Linux (NVIDIA)
Latest release
v2.21.0 (2026-11-10)
Repository
NVIDIA/TransformerEngine
Topicscudadeep-learningfp4fp8gpujaxmachine-learningpythonpytorch

Freeweights links to the project and does not host tool binaries.

Install

$ pip install --no-build-isolation "transformer_engine[pytorch]"
$ pip install --no-build-isolation "transformer_engine[pytorch]"

Checked against the project's README on 2026-11-15. The project's own instructions win when they differ: install docs

Freeweights links to the project and does not host tool binaries.

Use with Freeweights

Fine-tune a catalog model

Download the weights from Freeweights first (HF_ENDPOINT or fw.py), then point the tool at the folder.

Catalog models it can fine-tune (3137)

135M-lc-100k-rope-12Ba2-rl-e2egit_large-50-32BAesCode-32BAesCode-8BAFM-4.5BAFM-4.5B-BaseAgentCPM-ReportAgents-A1-4BAI21-Jamba-Reasoning-3BAI21-Jamba2-MiniAlchemistCoder-L-7BAlexandria-Gemma-4-12B-it-Ornith9-Summaries-LoRA-r128Alexandria-Gemma-4-12B-it-Ornith9-Summaries-LoRA-r64Alexandria-Gemma-4-12B-it-Qwen27-Knowledge-Units-LoRA-r128Alexandria-Gemma-4-12B-it-Qwen27-Summaries-LoRA-r128Alexandria-Gemma-4-12B-it-Qwen27-Summaries-LoRA-r64Alexandria-Gemma-4-E4B-it-Qwen27-Knowledge-Units-LoRA-r128ALLaM-7B-Instruct-previewAM-Thinking-v1AMD-OLMo-1BAMD-OLMo-1B-SFTAMD-OLMo-1B-SFT-DPOApertus-70B-2509Apertus-70B-Instruct-2509Apertus-8B-2509Apertus-8B-Instruct-2509Apertus-8B-Instruct-2509-unsloth-bnb-4bitApertus-v1.1-0.5BApertus-v1.1-0.5B-InstructApertus-v1.1-1.5BApertus-v1.1-1.5B-InstructApertus-v1.1-4BApertus-v1.1-4B-InstructApex-35B-V1.0Apodex-1.1-miniApriel-1.5-15b-ThinkerApriel-1.6-15b-ThinkerApriel-Nemotron-15b-ThinkerAquila-135M-IntermediateAquila-VL-2B-IntermediateAquilaDense-16BAquilaMoEAquilaMoE-SFTArcee-Maestro-7B-PreviewArete-27BAREX-2AREX-2-6bitAREX-BaseAREX-TurboAstaBrief_8BAstaBrief_8B_SFTAurora-Code-1BaguettotronBaichuan-M2-32BBAR-2x7B-BaseBAR-2x7B-CodeBAR-2x7B-Code-SFTBAR-2x7B-MathBAR-2x7B-Math-SFTBAR-2x7B-Safety

and 3077 more.

License

License
Apache-2.0
Class
Permissive

Read the license

Freeweights links to the project and does not host tool binaries.

Provenance

From: NVIDIA <github.com/NVIDIA>Newsgroups: alt.freeweights.toolsSubject: Transformer EngineRelease: v2.21.0 (2026-11-10)Commit: 7457e47License: Apache-2.0
Repository
NVIDIA/TransformerEngine
Pinned release
v2.21.0, 2026-11-10
Release commit
7457e47d26d9d02dfb783007335dfbe75e1b90a8
Default branch
main 4f5c08e48b3e190d27c731c1f9515f021dec255a
Fetched
2026-11-15
Takedowns
None.

Freeweights links to the project and does not host tool binaries.

Get it
InstallProject repository
Works with

3137 catalog models. See the list