Mixture-of-experts (MoE) models activate few experts per token, yet batched decoding can access nearly the entire expert pool, making expert-weight traffic a major bottleneck.
← Back to all developments · Signals are a development reported by more than one source. See the methodology.