Zyphra(@ZyphraAI):In Mixture of Expert (MoE) models the cost of moving tokens to their experts can dominate total runtime. At Zyphra research, we use patterns in how tokens are routed to experts to make that communication faster by up to 2.63x on @AMD MI300X GPUs, with the model unchanged.

Zyphra

@ZyphraAI

Full stack open superintelligence

가입 March 2021

40 팔로잉 중 14.9K 팬

Zyphra@ZyphraAI

2026.10.08 15:35

In Mixture of Expert (MoE) models the cost of moving tokens to their experts can dominate total runtime. At Zyphra research, we use patterns in how tokens are routed to experts to make that communication faster by up to 2.63x on @AMD MI300X GPUs, with the model unchanged.