Google launches open-source TPU model design guide
Google launched the open-source AI-Hypercomputer/tpu-codesign guide to help developers design and optimize models for TPUs. It covers hardware-aligned shapes, architectural choices such as LatentMoE, parallelism, and training and inference for Ironwood (TPU v7x) and later TPU generations. The guide combines published sources with measurements from Google’s own Ironwood profiling program, including results for open 20B–30B MoE models and dense control models.
The guide provides concrete, hardware-grounded rules for building or adapting models for TPU training and serving. It is intended for developers working with dense LLMs, MoE models, multimodal models, world models, and other large-matrix-multiplication workloads.
TPU Model Design Considerations and Optimization Guide - Google Developers Blog
TPU Model Design Considerations and Optimization Guide
NOV. 16, 2026
Gagik Amirkhanyan Staff Software Engineer
Vaibhav Singh Software Engineer
#Google#AI-Hypercomputer/tpu-codesign#TPU#LatentMoE#Ironwood#TPU 8t#TPU 8i#TPU 8tx#TPU 8i Max