EfficientMoE/MoE-Infinity GitHub repository

Fields

id

sources/src-moe-infinity-github

URL

https://github.com/EfficientMoE/MoE-Infinity

Publisher

University of Edinburgh / EfficientMoE

Kind

repo

Accessed

2026-10-18

notes

Apache-2.0 PyTorch library. The README describes expert weights offloaded to host memory and SSD, an activation-aware cache of hot experts kept on the GPU, and tracing plus prefetching to hide the transfer, plus a HuggingFace-compatible MoE class and an OpenAI-compatible server. 378 commits, 373 stars and 46 forks at access on 2026-10-18. Cited only for the code URL and the license, not for any performance number.

Notes

Apache-2.0 PyTorch library. The README describes expert weights offloaded to host memory and SSD, an activation-aware cache of hot experts kept on the GPU, and tracing plus prefetching to hide the transfer, plus a HuggingFace-compatible MoE class and an OpenAI-compatible server. 378 commits, 373 stars and 46 forks at access on 2026-10-18. Cited only for the code URL and the license, not for any performance number.

Cross-references

This record references no other records.