Skip to content
Hot eventWatching

Google Releases Open-Source TPU Codesign Guide

1 reports1 reporting sourcesUpdated 1 d ago

Start here

AI digest

Google has launched the open-source AI-Hypercomputer/tpu-codesign guide, which sets out eight principles for designing models around TPU architecture — from 256-aligned dimensions and static shapes to LatentMoE routing and mapping parallel axes onto physical links. Google says the shape rules apply to anyone designing a new model, while developers adapting an existing model can use the Expressing Parallelism section without touching the checkpoint. Google reports measured Ironwood results: a 2.8x cut in on-device permute time, +14% training throughput from an l=768 LatentMoE on a 30B/128-expert model, and +57% tok/s from halving expert count while doubling d_ff on a 63B LatentMoE. The guide also covers TPU 8t and TPU 8i features such as native FP4, Boardfly topology and the Collectives Acceleration Engine, and recommends validating architecture changes with wall-clock ablations on a tuned stack. Google says the companion repository will grow model- and workload-specific sections over time.

Generated from reports · updated 1 h ago

Report timeline

Follow the reports to see the event from different sides.

November 16
  1. Google Developers BlogPicks
    Google releases TPU model design and optimization guide covering shape rules, LatentMoE and parallelism

    Google launched the open-source AI-Hypercomputer/tpu-codesign guide, which sets out eight principles for designing models around TPU architecture, from 256-aligned dimensions and static shapes to LatentMoE routing and mapping parallel axes onto physical links. Measured Ironwood results include a 2.8x cut in on-device permute time and +14% training throughput from an l=768 LatentMoE on a 30B/128-expert model, and +57% tok/s from halving expert count while doubling d_ff on a 63B LatentMoE. The guide also covers TPU 8t and TPU 8i features such as native FP4, Boardfly topology and the Collectives Acceleration Engine, and recommends validating architecture changes with wall-clock ablations on a tuned stack.

Heat for this event

Not enough continuous observations yet to draw a trend.