Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Pin 2024 03 27 #15

Merged
merged 31 commits into from
Mar 27, 2024
Merged

Pin 2024 03 27 #15

merged 31 commits into from
Mar 27, 2024

Conversation

mars1248
Copy link

Pin 2024 03 27
TORCHXLA_PIN_COMMIT=27ae6dc955734b223e2b93e1c5bbcd681b0537fd

mbzomowski and others added 30 commits January 8, 2024 10:21
…rch#6268)

This commit reenables test/test_zero1.py for GPU. It is passing with latest torch/xla (commit e1c94df).

pytorch#6260
This is useful for scoping up the single graph, and seeing how compiler is able to optimise it.
This is is optional i.e not added by default. Once we start setting up another benchmark suite with profiling info etc. this should be run additionally, but separately so CUPTI interface won't interact with pure CUDA events.

We do not measure compilation time for this option yet.
With the pure wall time infrastructure came nice
ability to do microbenchmarks easily. I added
the sample microbenchmark for matmul.
This reverts commit 0c81365.

It causes OOMs for some training workloads on GPUs, and slows down others.
This PR introduces, and integrates a verification module under the --verify flag. The verification module gets model output as an input, model/benchmark reconstruction args, then runs the eager and finally calculates the relative error and compares it against the provided threshold to return the appropriate error code.
@zjjott zjjott merged commit 7b8d407 into master Mar 27, 2024
0 of 2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.