Source · Tier A · Peer-reviewedSource · Shortcut-connected Expert Parallelism for Accelerating Mixture of Experts
Shortcut-connected Expert Parallelism for Accelerating Mixture of Experts
W. Cai, J. Jiang, L. Qin, J. Cui, S. Kim, J. Huang. 2025. ICML 2025, Proceedings of Machine Learning Research 267.
| Original | https://proceedings.mlr.press/v267/cai25c.html |
|---|---|
| Accessed | 2026-09-25 |
| Note | The paper reports all-to-all expert-parallel communication as a bottleneck in both MoE training and inference. |