Bert Maher
Bert Maher
This is great, thank you @davidberard98 ! (And sorry I'm slow to closing this issue haha)
Awesome, I will rebase onto those in the morning (US East Coast Time :-) )
Let's see where we are! I haven't solved the Triton wheel build failures yet but I think we've knocked out all the known test failures?
> Note [triton-lang/triton#5089](https://github.com/triton-lang/triton/pull/5089) is known to resolve a pretty big regression on AMD backend, will be nice to move pin after this at some point, or cherry pick. @jataylo Do...
@chuanqi129 @Stonepia @etaf The Triton XPU wheels are failing to build, because the upstream Triton version # is bumping to 3.2. How would you like to proceed?
> @bertmaher Yes, we also want to do it by PR #137886. Seems we need to land this PR with xpu triton build failure to bump up the triton version...
Pushed a rebase this morning. There are a few expected failures left: - the `test_convolution_as_mm` failure, which has been intermittent, but is now confirmed to be real. @davidberard98, @embg, and...
@pytorchbot merge -f "Remaining failures are either expected (xpu, and Triton wheel version) or due to self-hosted runner timeouts"
@jataylo Thanks for the heads up. There are a ton of cuda jobs timing/losing connection on this PR too, and it's not obvious to me if it's just infra in...
@chuanqi129 @etaf This version bump has landed, if you'd like to go ahead with bumping the xpu triton version, please proceed!