xla
xla copied to clipboard
openxla
A machine learning compiler for GPUs, CPUs, and ML accelerators
Do not crash on autotuner verification errors in int4 tests This is the only test we have in the codebase that considers this an error. I don't think the test...
![copybara-service[bot] avatar](https://avatars.githubusercontent.com/in/44061?v=4 "copybara-service[bot] avatar"){kind=avatar}
Expose getters and setters for marking nodes as selective resource occupiers/releasers.
[XLA:GPU] Collect GEMM cost model stats.
PR #25661: Report an error message when user specify a non-existed pgle profile file Imported from GitHub PR https://github.com/openxla/xla/pull/25661 Copybara import of the project: -- b68998f19251298a1e70978146b495b1ea7d52f9 by Shawn Wang :...
This implements `GetCompiledMemoryStats` for ahead-of-time compiled executables. With this patch, one can estimate memory consumption of a JAX function even without access to a GPU. Unfortunately, the patch duplicates code...
Reverts 13ab3cfb9885529954e3db4bdae082612496c46a
@xla-rotation , would you please take a look?
Decouple HloGumgraph from hlo_diff_summary. This change removes the dependency of the `hlo_diff_summary` on the `HloGumgraph`. Now the summary is constructed based only on HLO modules and diff result.