Maintained implementation availablepytorch

Tutel: Adaptive Mixture-of-Experts at Scale

June 1, 2022arXiv: 2206.03382

3 repos980 stars~a few days to reproduce

Abstract

Task	Dataset	Metric	Value
Adaptive Mixture-of-experts At Scale	ImageNet-22K	Top-1 Accuracy	32
Adaptive Mixture-of-experts At Scale	COCO	Accuracy	85.5
Classification	ImageNet	Accuracy	85.5

Expect multi-day setup/compute for meaningful reproduction based on current guidance.

Tutel MoE: Optimized Mixture-of-Experts Library, Support GptOss/DeepSeek/Kimi-K2/Qwen3 using FP8/NVFP4/MXFP4

980 107 Apr 2026 MIT

License ✓

CI –

Deps –

Docker –

Selected microsoft/tutel as the strongest maintained implementation for new work.
Repository activity is within the last 24 months.
Official repository is preserved separately as historical context.

1
Start with microsoft/tutel and validate setup instructions in README.
2
Reproduce the baseline result with the provided defaults before modifying hyperparameters.
3
Log exact dependency versions and runtime environment for reproducibility.

Time to first repro: a few daysNo CI workflows detectedDependency manifest is missing

No additional official repositories detected.

microsoft/TutelConfidence: low
Tutel MoE: Optimized Mixture-of-Experts Library, Support GptOss/DeepSeek/Kimi-K2/Qwen3 using FP8/NVFP4/MXFP4
Stars: 980Forks: 107Last push: Apr 2026License: MIT

No direct paper-linked artifacts were found. Showing strongest curated related artifacts.

Curated Related