Baidu Intelligent Cloud Tech Hub

Jun 29, 2026 · Artificial Intelligence

How to Maximize Cosmos‑3 Training Throughput Without NVLink Using AI Infra Optimizations

By applying systematic AI Infra engineering—optimizing data loading, I/O pipelines, activation checkpointing, torch.compile, and multi‑node scaling—the Cosmos‑3‑Nano‑Policy‑DROID model achieved an 89× faster startup, 99.3% higher single‑node throughput, 0.42 MFU, and 98.3% scaling efficiency across 12 nodes, all without NVLink.