Everyone active in creating or integrating with XLA, including representatives of ML frameworks, hardware platforms, users, and integrators, is eligible to join OpenXLA. Members must request invitations to the SIG Discord and the GitHub organisation before they can participate.
Google Cloud introduced the OpenXLA open source machine learning compiler ecosystem at its Next ’22 conference. A community-led and open source ecosystem of ML compiler and infrastructure projects, OpenXLA was created by Google in collaboration with AMD, Arm, Meta, NVIDIA, AWS, Intel, Apple, and other AI/ML developers. By allowing developers to select the frameworks and hardware they want for their ML projects, OpenXLA attempts to address the problem of framework and hardware incompatibilities, which can hinder the development of machine learning algorithms.
The OpenXLA project charter states that the project’s goal is to enable effective lowering, optimization, and deployment of machine learning models from the majority of major frameworks, including PyTorch and TensorFlow, to any hardware backend (particularly CPUs, GPUs, and ML ASICs), through collaboration with major ML frameworks and hardware vendors.
The XLA compiler, which was created to simplify modelling in TensorFlow by accelerating the training process and lowering overall memory usage, will be the first goal of the new community project, according to Sachin Gupta, Google VP and GM of infrastructure, in a blog post. In order to create StableHLO, a portable ML compute operation set that serves as a portability layer across machine learning frameworks and compilers, OpenXLA will endeavour to isolate the compiler from TensorFlow.
The following objectives are specified for OpenXLA:
- Promote XLA industry collaboration and create a thriving OSS community.
- Exchange ideas and solicit input on OpenXLA’s technical direction to make sure it satisfies the requirements of its key users and contributors.
- Create a new XLA repository or organisation that is independent of hardware and framework, has infrastructure to more easily accept PRs, and has independent build and test.
- Assure that the removal of XLA from TensorFlow causes the fewest possible disruptions for current users and contributors.
- Establish a brand, website, documents, and communication channels for the product.
- Talk about setting up governance outside of TensorFlow.
“At Google, we believe open source software is essential to overcoming the challenges associated with inflexible strategies. And as the leading Cloud Native Computing Foundation contributor, we have over two decades of experience working with the community to turn OSS projects into accessible, transparent catalysts for technological advance. We’re committed to open ecosystems of all kinds, and this commitment extends to AI/ML—we firmly believe no single company should own AI/ML innovation,” said Gupta.