pallas

#pallas

@googledevs: Scaling frontier Mixture-of-Experts (MoE) models takes more than trial-and-error tuning Explore how Qwen 3.5-397B was o…

X AI KOLs Following ↗ · 20h ago Cached

Google Cloud details how they optimized Qwen 3.5-397B MoE on Ironwood TPUs using a modular, model-agnostic engineering playbook, achieving 3.1× decode and 4.7× prefill performance gains.

0 favorites 0 likes

pallas

@googledevs: Scaling frontier Mixture-of-Experts (MoE) models takes more than trial-and-error tuning Explore how Qwen 3.5-397B was o…

Submit Feedback