IT之家 · 智能时代T2·78 pts
- Targets trillion-parameter MoE pretraining and RL on 10k-GPU clusters
- Supports blockwise FP8 training to boost computational efficiency
- Integrates DeepEP and ACE async communication for EP-computation overlap
HOT RADAR
computeHEAT 11.7°
AI CURATED HOT STORY · 10/28/2026
Moore Threads Open-Sources Megatron-MUSA-Titan for Trillion-Parameter MoE Training
1 reports archived1 independent sourcesupdated 10/28/2026, 18:16:40
Synthesis & Latest Updates
1 Sources Cross-ValidatedMoore Threads has open-sourced Megatron-MUSA-Titan, an infrastructure project optimized for training trillion-parameter MoE models on 10k-GPU clusters. It integrates blockwise FP8, DeepEP, and async communication to enhance Model FLOPs Utilization (MFU) and stability across pretraining and RL scenarios.
LATEST/Moore Threads has open-sourced Megatron-MUSA-Titan, an infrastructure project optimized for training trillion-parameter MoE models on 10k-GPU clusters. It integrates blockwise FP8, DeepEP, and async communication to enhance Model FLOPs Utilization (MFU) and stability across pretraining and RL scenarios.
Hourly heat curve
2 fully observed hoursFewer than 3 fully observed hours; trend not plotted yet.