RadixArk helps bring SGLang to Google TPUs as Meta runs it in production

Google says it co-developed TorchTPU with Meta's PyTorch team; RadixArk worked with Google and Meta on SGLang-TorchTPU, which Meta says is in production.

By · Published

Primary source: Google Developers Blog

Why it matters

Google's TPU challenge to Nvidia depends on software as much as silicon. Meta's deployment puts Google's PyTorch-native approach into production at a major PyTorch shop.

A Google-branded data center with cooling pipes and a lit compute room depicts Meta’s SGLang running on Google TPU infrastructure.

Meta said on October 20th that it is running SGLang-TorchTPU in production, putting Google's custom AI chips to work at a major PyTorch shop. RadixArk, the startup co-founded by Ying Sheng, worked with Google and Meta on the engine.…

When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.

Reader comments

Conversation for this story loads after sign-in.