<!-- mobian-agent-page publisher="dailydev" canonical="https://daily.dev/posts/torchtpu-is-now-open-source-run-native-pytorch-on-tpus-tvhj5xd6t" -->

---
title: TorchTPU is now open source: Run native PyTorch on TPUs
description: TorchTPU, Google's native PyTorch stack for TPUs, is now open source on GitHub, co-developed with Meta's PyTorch team. It lets developers train, serve, and...
canonical: https://daily.dev/posts/torchtpu-is-now-open-source-run-native-pytorch-on-tpus-tvhj5xd6t
twitter:card: summary_large_image
twitter:site: @dailydotdev
og:type: website
og:site_name: daily.dev
og:title: TorchTPU is now open source: Run native PyTorch on TPUs | daily.dev
og:description: TorchTPU, Google's native PyTorch stack for TPUs, is now open source on GitHub, co-developed with Meta's PyTorch team. It lets developers train, serve, and...
og:url: https://daily.dev/posts/torchtpu-is-now-open-source-run-native-pytorch-on-tpus-tvhj5xd6t
og:image: https://api.daily.dev/og/posts/tVHJ5xD6t.png
og:image:alt: TorchTPU is now open source: Run native PyTorch on TPUs
og:image:width: 1200
og:image:height: 630
og:locale: en
---

> ## Documentation Index
> Fetch the complete documentation index at: https://daily.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# TorchTPU is now open source: Run native PyTorch on TPUs

**[Google Cloud](https://daily.dev/sources/gcp)** · 10 min read · 2 upvotes · 0 comments

## Summary

TorchTPU, Google's native PyTorch stack for TPUs, is now open source on GitHub, co-developed with Meta's PyTorch team. It lets developers train, serve, and deploy models on TPUs using familiar PyTorch APIs and tools including vLLM, SGLang, llm-d, TorchTitan, Ray, TorchRec, Hugging Face, and LiteRT. Hugging Face Transformers v5 supports TPU deployment via device_map="tpu". Inference integrations include vLLM with Pallas kernels, SGLang with RadixAttention, and llm-d 1.0 (with Red Hat and IBM) for fleet-wide routing. TorchTitan has been validated on up to 8,192 Ironwood TPUs (v7x), reaching 52% Model Flops Utilization on Qwen 3 8B. VeRL gains a TPU backend for reinforcement learning, Ray adds native TPU scheduling, and LiteRT (formerly TensorFlow Lite) joins the PyTorch ecosystem for on-device inference across phones, browsers, and embedded hardware. A TPU agent suite for automated onboarding and performance tuning is also in early preview.

## Full article

daily.dev links to this article rather than hosting it. Read it at the original source: <https://cloud.google.com/blog/topics/ai-infrastructure/torchtpu-is-now-open-source-run-native-pytorch-on-tpus>

## Questions this post answers

### How do I run PyTorch models on Google TPUs now that TorchTPU is open source?

TorchTPU, co-developed with Meta's PyTorch team, lets existing PyTorch code run on TPUs with minimal changes using standard APIs, eager mode for debugging, and torch.compile for production performance. It integrates with vLLM, SGLang, llm-d, TorchTitan, Ray, TorchRec, Hugging Face, and LiteRT, and is available open source on GitHub at google-pytorch/torch_tpu.

_Teams weighing TPUs against GPUs for PyTorch workloads track ecosystem shifts like this on daily.dev._

### How do I deploy a Hugging Face model to a TPU with Transformers v5?

In Transformers v5, deploying a model to a TPU is done by passing device_map="tpu" when loading the model, thanks to native TorchTPU integration built with Hugging Face. The Diffusers library also runs natively on TPUs, supporting eager mode, torch.compile, and tensor parallelism for scaling large generative models across multiple accelerators.

_Developers choosing hardware for Hugging Face deployments follow integrations like this on daily.dev._

### What throughput does TorchTitan achieve training Qwen 3 8B on TPUs?

TorchTitan demonstrated up to 52% Model Flops Utilization (MFU) training Qwen 3 8B on TPUs, with near-linear scaling and stability validated across up to 8,192 Ironwood TPUs (TPU v7x). TorchTitan also provides pre-training recipes for models like Llama 3 70B/405B, Qwen 3 32B/235B MoE, and FLUX.1 12B diffusion transformers.

_Engineers benchmarking large-scale training efficiency follow results like this on daily.dev._

## Similar posts on daily.dev

- [TorchTPU: Running PyTorch Natively on TPUs at Google Scale](https://daily.dev/posts/torchtpu-running-pytorch-natively-on-tpus-at-google-scale-wnrth1jyb) · Google Developers · 0 upvotes · 0 comments
- [TPU 8t and TPU 8i technical deep dive](https://daily.dev/posts/tpu-8t-and-tpu-8i-technical-deep-dive-u5h1xk1ga) · Google Cloud · 0 upvotes · 0 comments
- [Google splits its TPU line in two for the agentic era](https://daily.dev/posts/google-splits-its-tpu-line-in-two-for-the-agentic-era-zbjc2yyc7) · The New Stack · 0 upvotes · 0 comments
- [Our eighth generation TPUs: two chips for the agentic era](https://daily.dev/posts/our-eighth-generation-tpus-two-chips-for-the-agentic-era-0d0ehgeeq) · Hacker News · 0 upvotes · 0 comments

---

Tags: [#machine-learning](https://daily.dev/tags/machine-learning), [#gcp](https://daily.dev/tags/gcp), [#pytorch](https://daily.dev/tags/pytorch), [#vllm](https://daily.dev/tags/vllm)

[View this post on daily.dev](https://daily.dev/posts/torchtpu-is-now-open-source-run-native-pytorch-on-tpus-tvhj5xd6t)

```json
{"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]}
{"@context":"https://schema.org","@type":"TechArticle","headline":"TorchTPU is now open source: Run native PyTorch on TPUs","url":"https://daily.dev/posts/torchtpu-is-now-open-source-run-native-pytorch-on-tpus-tvhj5xd6t","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/torchtpu-is-now-open-source-run-native-pytorch-on-tpus-tvhj5xd6t"},"datePublished":"2026-10-20T11:40:46.534Z","dateModified":"2026-10-21T18:50:33.555Z","description":"TorchTPU, Google's native PyTorch stack for TPUs, is now open source on GitHub, co-developed with Meta's PyTorch team. It lets developers train, serve, and...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/fae798aa3baebc72680bb6fa8c180d93?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/fae798aa3baebc72680bb6fa8c180d93?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Google Cloud","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Google Cloud","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/gcp","url":"https://daily.dev/sources/gcp"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/torchtpu-is-now-open-source-run-native-pytorch-on-tpus-tvhj5xd6t","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":2},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"machine-learning,gcp,pytorch,vllm","timeRequired":"PT10M"}
{"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Google Cloud","item":"https://daily.dev/sources/gcp"},{"@type":"ListItem","position":3,"name":"TorchTPU is now open source: Run native PyTorch on TPUs"}]}
{"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/torchtpu-is-now-open-source-run-native-pytorch-on-tpus-tvhj5xd6t#faq","mainEntity":[{"@type":"Question","name":"How do I run PyTorch models on Google TPUs now that TorchTPU is open source?","acceptedAnswer":{"@type":"Answer","text":"TorchTPU, co-developed with Meta's PyTorch team, lets existing PyTorch code run on TPUs with minimal changes using standard APIs, eager mode for debugging, and torch.compile for production performance. It integrates with vLLM, SGLang, llm-d, TorchTitan, Ray, TorchRec, Hugging Face, and LiteRT, and is available open source on GitHub at google-pytorch/torch_tpu. Teams weighing TPUs against GPUs for PyTorch workloads track ecosystem shifts like this on daily.dev."}},{"@type":"Question","name":"How do I deploy a Hugging Face model to a TPU with Transformers v5?","acceptedAnswer":{"@type":"Answer","text":"In Transformers v5, deploying a model to a TPU is done by passing device_map=\"tpu\" when loading the model, thanks to native TorchTPU integration built with Hugging Face. The Diffusers library also runs natively on TPUs, supporting eager mode, torch.compile, and tensor parallelism for scaling large generative models across multiple accelerators. Developers choosing hardware for Hugging Face deployments follow integrations like this on daily.dev."}},{"@type":"Question","name":"What throughput does TorchTitan achieve training Qwen 3 8B on TPUs?","acceptedAnswer":{"@type":"Answer","text":"TorchTitan demonstrated up to 52% Model Flops Utilization (MFU) training Qwen 3 8B on TPUs, with near-linear scaling and stability validated across up to 8,192 Ironwood TPUs (TPU v7x). TorchTitan also provides pre-training recipes for models like Llama 3 70B/405B, Qwen 3 32B/235B MoE, and FLUX.1 12B diffusion transformers. Engineers benchmarking large-scale training efficiency follow results like this on daily.dev."}}]}
```