Metraipolis › archive › dc_728c8bd5f780

Cobalt: Leveraging Expert Co-activation for Efficient Distributed MoE Training — arXiv:2609.36959

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.36959
License (as archived)
CC0-1.0
Archived
2026-10-07T16:59:08.128Z
SHA-256 of the archived text
1f9899141efb716328a38bd5eda0253c3d4390be3ff0f806ed554eabe0eb693f
Stable URL of this page
https://metraipolis.com/documento/dc_728c8bd5f780

How to cite

Cobalt: Leveraging Expert Co-activation for Efficient Distributed MoE Training — arXiv:2609.36959 — arXiv (en), https://arxiv.org/abs/2609.36959 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_728c8bd5f780 (Metraipolis archive dc_728c8bd5f780, sha256 1f9899141efb716328a38bd5eda0253c3d4390be3ff0f806ed554eabe0eb693f)
BibTeX
@misc{dc728c8bd5f780,
  title = {Cobalt: Leveraging Expert Co-activation for Efficient Distributed MoE Training — arXiv:2609.36959},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.36959},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_728c8bd5f780; sha256 1f9899141efb716328a38bd5eda0253c3d4390be3ff0f806ed554eabe0eb693f},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.36959. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Mixture-of-Experts (MoE) has increasingly become a mainstream approach for scaling large language models, as it expands model capacity while keeping computation cost nearly constant. Training large-scale MoE models relies on Expert Parallelism (EP), which distributes expert replicas across GPUs and exchanges tokens through all-to-all communication. The efficiency of EP is often constrained by two system bottlenecks: cross-node token transfers are limited by inter-node bandwidth, while skewed expert workloads lead to imbalanced computation across GPUs. Prior work mitigates these bottlenecks based on per-expert workload statistics, but overlooks the fact that experts could share the communication. In this work, we empirically present the observation that many pairs of experts are frequently co-activated by individual tokens. Motivated by this, we present Cobalt, an efficient MoE training framework that leverages expert co-activation to reduce cross-node traffic and workload imbalance. Cobalt adopts a two-stage expert layout planner that adapts expert layout to the evolving expert co-activation and workload conditions. It periodically co-locates frequently co-activated experts on the same node to reduce the cross-node communication, and performs per-step intra-node adjustment to rebalance the workloads. Subsequently, we develop a communication-aware task assignment method that routes tokens to fewer remote nodes based on the current expert layout. Experiments on 32 B200 GPUs show that Cobalt achieves up to 1.53-2.41 times (1.28-1.89 times on average) of speedup compared to existing MoE training frameworks, while reducing cross-node token traffic by 75.74%-99.26%.

Tema degli esploratori: calcolo e infrastrutture (retro-catalogo arXiv: annunci del 2026-09-30 in cs.DC).
Categorie arXiv: Distributed, Parallel, and Cluster Computing (cs.DC).
Pagina del lavoro: https://arxiv.org/abs/2609.36959 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-07T16:31:32.453Z dalla pagina https://arxiv.org/catchup/cs.DC/2026-09-30?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).