Sounds good, but my main question is: is this a fork, or a new backend they're building in (like MPS)?
They write that they use PrivateUse1, so it’s a custom out-of-tree backend
TorchTPU: Running PyTorch Natively on TPUs at Google Scale
11–20 of 22 posts
Re: TorchTPU: Running PyTorch Natively on TPUs at Google Scale
#12pitch basically boils down to 'just change one line and it works' which sounds too good to be true, but if they actually pull it off at 100k-chip scale, that's genuinely a big deal
Re: TorchTPU: Running PyTorch Natively on TPUs at Google Scale
#13Adding a support for new hardware to PyTorch is actually quite convenient. I did that with WebGPU using the same PrivateUse1 mechanism TorchTPU used. Every hardware has its own slot and identifier, and when you want to add a support for a new one without merging it into PyTorch, PrivateUse1 works essentially like plug-in slot
Re: TorchTPU: Running PyTorch Natively on TPUs at Google Scale
#14Is it just me, or does it feel like everyone now uses AI to write any kind of blog?
These parts here somehow trigger me:
- Enter TorchTPU. As an engineering team, our mandate was to build a stack that leads with usability, portability, and excellent performance.
- Engineering the TorchTPU Stack: The Technical Reality
- Eager First: Flexibility Without Compromise
- The breakthrough, however, is our fused eager mode.
- The Road Ahead: 2026 and Beyond
I have mixed feelings about this. On one hand, we all seem to be using the same tools and converging to the same style. On the other hand, if we all use the same models with the same system prompts, we might lose a lot of creativity and diversity in online content.
Re: TorchTPU: Running PyTorch Natively on TPUs at Google Scale
#15Is it just me, or does it feel like everyone now uses AI to write any kind of blog? These parts here somehow trigger me: - Enter TorchTPU. As an engineering team, our mandate was to build a stack that leads with usability, portability, and excellent performance. - Engineering the TorchTPU Stack: The Technical Reality - Eager First: Flexibility Without Compromise - The breakthrough, however, is our fused eager mode. -…
it's just sad
Re: TorchTPU: Running PyTorch Natively on TPUs at Google Scale
#16So Google is basically admitting PyTorch/XLA on TPUs didn't work — TorchTPU looks like them rebuilding what should have worked day one. Its hard to run production ML on a toolchain engineers can't trust, no matter how fast the silicon is.
Re: TorchTPU: Running PyTorch Natively on TPUs at Google Scale
#17Is it just me, or does it feel like everyone now uses AI to write any kind of blog? These parts here somehow trigger me: - Enter TorchTPU. As an engineering team, our mandate was to build a stack that leads with usability, portability, and excellent performance. - Engineering the TorchTPU Stack: The Technical Reality - Eager First: Flexibility Without Compromise - The breakthrough, however, is our fused eager mode. -…
More like, everyone now has KPIs and OKRs attached to AI for their quartal appraisals.
If you don't, I envy you.
Re: TorchTPU: Running PyTorch Natively on TPUs at Google Scale
#18Is it just me, or does it feel like everyone now uses AI to write any kind of blog? These parts here somehow trigger me: - Enter TorchTPU. As an engineering team, our mandate was to build a stack that leads with usability, portability, and excellent performance. - Engineering the TorchTPU Stack: The Technical Reality - Eager First: Flexibility Without Compromise - The breakthrough, however, is our fused eager mode. -…
More like, everyone now has KPIs and OKRs attached to AI for their quartal appraisals. If you don't, I envy you.
This right here is exactly right. If it hasn't come for you yet, it's coming.
Re: TorchTPU: Running PyTorch Natively on TPUs at Google Scale
#19Is it just me, or does it feel like everyone now uses AI to write any kind of blog? These parts here somehow trigger me: - Enter TorchTPU. As an engineering team, our mandate was to build a stack that leads with usability, portability, and excellent performance. - Engineering the TorchTPU Stack: The Technical Reality - Eager First: Flexibility Without Compromise - The breakthrough, however, is our fused eager mode. -…
Yeah but everyone also uses AI to summarize blogs for them.
Re: TorchTPU: Running PyTorch Natively on TPUs at Google Scale
#20So Google is basically admitting PyTorch/XLA on TPUs didn't work — TorchTPU looks like them rebuilding what should have worked day one. Its hard to run production ML on a toolchain engineers can't trust, no matter how fast the silicon is.
This does still look like it uses XLA, fwiw.