Meta PyTorch Team 2025 H1 Roadmaps

10

In the Triton Roadmap, it’s mentioned that:

  • a intra-kernel tracing tool for Triton that gives kernel writers a Kineto-like interface into kernel execution was developed
  • multiple chasing the speed-of-light analyses to see why Triton perf keeps falling off a cliff were completed.

Are these publicly available?

Thanks!