Blog/AI Infra

Search

Category

Newest → oldest

Aug 14, 2025

ByteDance Seed Open-Sources VeOmni, Unlocking Any Modality Model Training Reduces engineering development time from weeks to days.

Mar 10, 2025

Comet has been deployed in large-scale clusters, saving millions of GPU hours! MoE communication optimization technology COMET is now open-source!Now open source! Compatible with mainstream large models, accepted with high scores at MLSys 2025.

Feb 12, 2025

Seed Research | New ultra-sparse architecture reduces inference costs by up to 83% compared to MoE!Inference speed is 2-6 times faster than that of MoE. The related paper has been accepted by ICLR 2025.