text stringlengths 0 1.14k |
|---|
--- CONTRIBUTING.md --- |
# Contributing to TPU Inference |
Thank you for your interest in contributing to TPU Inference! Our community is open to everyone and welcomes all kinds of contributions, no matter how small or large. There are several ways you can contribute to the project: |
* Identify and report any issues or bugs. |
* Request or add support for a new model. |
* Suggest or implement new features. |
* Improve documentation or contribute a how-to guide. |
We also believe in the power of community support; thus, answering queries, offering PR reviews, and assisting others are also highly regarded and beneficial contributions. |
Finally, one of the most impactful ways to support us is by raising awareness about TPU Inference. Talk about it in your blog posts and highlight how it's driving your incredible projects. Express your support on social media if you're using TPU Inference, or simply offer your appreciation by starring our repository! |
## Getting Started |
We recommend filtering on the “Good First Issue” tag in the [Issues](https://github.com/vllm-project/tpu-inference/issues) section of Github if it's your first time contributing! |
## Issues |
If you encounter a bug or have a feature request, please search [existing issues](https://github.com/vllm-project/tpu-inference/issues) first to see if it has already been reported. If not, please [file a new issue](https://github.com/vllm-project/tpu-inference/issues/new/choose), providing as much relevant information... |
## Directory Structure |
We choose to follow a similar directory structure as vLLM: |
* `tpu_inference/layers/`: |
* `common` contains layers that are common to both vLLM and JAX |
* `jax` contains layers that are only used by JAX models |
* `vllm` contains layers that are only used by vLLM models |
* `tpu_inference/models/` |
* `common` contains model implementations/functionalities that are used by both vLLM and JAX |
* `jax` contains model implementations/functionalities that are only used by JAX models |
* `vllm` contains model implementations/functionalities that are only used by vLLM models |
### Torchax v/s JAX-Native implementation |
We prefer enabling Pytorch based vLLM models via Torchax first and then following up with integrating JAX-Native/FLAX implementation of the models. |
## Testing |
When checking in a new feature, we expect that you you add relevant unit tests as well as CI tests. You can read more about the latter [here](https://github.com/vllm-project/tpu-inference/tree/main/.buildkite#adding-a-new-feature-to-ci). |
## Setting up linting, formatting, and static type checking |
``` |
pip install pre-commit |
# Linting, formatting and static type checking |
pre-commit install --hook-type pre-commit --hook-type commit-msg |
# You can manually run pre-commit with |
pre-commit run --all-files |
``` |
## Thank You! |
We wanted to thank you for taking the time to read these guidelines and for your interest in contributing to TPU Inference. All of your contributions help make TPU Infernece a great tool and community for everyone! |
--- README.md --- |
<p align="center"> |
<!-- This image will ONLY show up in GitHub's dark mode --> |
<img src="docs/assets/tpu_inference_dark_mode_short.png#gh-dark-mode-only" alt="vLLM TPU" style="width: 86%;"> |
<!-- This image will ONLY show up in GitHub's light mode (and on other platforms) --> |
<img src="docs/assets/tpu_inference_light_mode_short.png#gh-light-mode-only" alt="vLLM TPU" style="width: 86%;"> |
</p> |
<p align="center"> |
| <a href="https://docs.vllm.ai/projects/tpu/en/latest/"><b>Documentation</b></a> | <a href="https://blog.vllm.ai/"><b>Blog</b></a> | <a href="https://discuss.vllm.ai/c/hardware-support/google-tpu-support/27"><b>User Forum</b></a> | <a href="https://slack.vllm.ai"><b>Developer Slack</b></a> (#sig-tpu) | |
</p> |
--- |
<p> |
<b>🤝 Contribute to the Project</b><br> |
<sub><i>Looking to help? Click a badge below to find issues that need your attention.</i></sub> |
</p> |
<!-- START: issue_badges --> |
[](https://github.com/vllm-project/tpu-inference/issues?q=is%3Aissue+is%3Aopen+label%3A%22bug%22) [](https://github.com/vllm-project/tpu-inference/issues?q... |
<!-- END: issue_badges --> |
## Latest News |
- [Announcing Gemma 4 on vLLM](https://vllm.ai/blog/gemma4) Byte for byte, the most capable open models - available on TPUs on Day 0! |
<details markdown="1"> |
<summary><i>Previous News</i> 🔥</summary> |
- [Pytorch Conference](https://pytorchconference.sched.com/event/27QCh/sponsored-session-everything-everywhere-all-at-once-vllm-hardware-optionality-with-spotify-and-google-brittany-rockwell-google-shireen-kheradpey-spotify) Learn how Spotify uses vLLM with both GPUs and TPUs to drive down costs and improve user experi... |
- [Ray Summit, November 3-5](https://www.anyscale.com/ray-summit/2025) in San Francisco! |
- [JAX DevLab on November 18th](https://rsvp.withgoogle.com/events/devlab-fall-2025) in Sunnyvale! |
- [2025/10] [vLLM TPU: A New Unified Backend Supporting PyTorch and JAX on TPU](https://blog.vllm.ai/2025/10/16/vllm-tpu.html) |
</details> |
<br> |
## About |
vLLM TPU is now powered by `tpu-inference`, an expressive and powerful new hardware plugin unifying JAX and PyTorch under a single lowering path within the vLLM project. The new backend now provides a framework for developers to: |
- Push the limits of TPU hardware performance in open source. |
- Provide more flexibility to JAX and PyTorch users by running PyTorch model definitions performantly on TPU without any additional code changes, while also extending native support to JAX. |
End of preview. Expand in Data Studio
No dataset card yet
- Downloads last month
- 48