PyTorch Community,
The Meta team are happy to make our 2025 H2 roadmaps available. We plan on a half year basis and globally optimize across the things we do for our users here at Meta and across the wider community. We’ve removed a few references to internal systems and teams but are eager to share the bulk of what we are planning for this half to encourage robust technical collaboration across the PyTorch community on these topics.
Checkpointing
Compiler Core
Compiler Deployment
Core Performance
Core Systems
DevInfra
Distributed
Docs
Edge
Full Stack Inference
Post-training
TorchCodec, TorchVision & TorchAudio
TorchRec
Triton
Unified Runtime
We hope sharing these detailed roadmaps will help collaborators across the community plan their work and engage with us on making PyTorch an ever stronger platform for AI Innovation.
9 Likes
@gottbrath
The roadmap for creating docs, Edge, TorchVision etc looks pretty exciting. I as a researcher from IIT Madras, would be happy to contribute to this. Would love to understand, how can I contribute and right POC.
1 Like
I appreciate your interest. Lets chat directly about what areas you are most excited about and I can put you in touch with some of the leads.
Am I getting it right that ExecuTorch is basically the successor of the TorchScript runtime?
Hello! I’d like to engage with the community on one of these projects. Specifically, I’m interested in Theme 1 of the DevInfra topic. I have experience using PyTorch and have worked a bit on the documentation before (I was a top contributor in the first PyTorch Docathon). So, I’d like to get a bit more involved with the community now.
Directionally yes. For more details, stay tuned for PTC.
One question: is there any plan to extend and/or add support for training in FP4? I saw that you were already working on a public API, and that the dtype was already available. If so, could someone contribute to the development?