Google DeepMind·· 2026-04-22
Decoupled DiLoCo: A new frontier for resilient, distributed AI training
Decoupled DiLoCo: A new frontier for resilient, distributed AI training
AI summary
Google DeepMind published a paper proposing Decoupled DiLoCo, splitting large-scale training into decoupled compute islands that communicate through asynchronous data streams, isolating local hardware failures.
Selection record
Threshold 60Official, first-handFirst 62Second 62
AdmittedSum of both 124 ≥ twice the threshold 120
- Source tier
- Official, first-hand; this tier's threshold is 60
- Pre-filter
- passed：介绍分布式AI训练新架构，实质AI技术
- Why it was chosen
- Measured results for bandwidth, fault tolerance and cross-generation hardware indicate the direction of large-scale pre-training infrastructure.
A model scores each item twice, independently, against one written standard, out of 100. An item is admitted only when the two scores add up to twice the threshold. The threshold is set per source tier.
Source: Google DeepMind · deepmind.google