-
Notifications
You must be signed in to change notification settings - Fork 86
Modify RobertaEmbedding forward as custom op method #996
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
base: habana_main
Are you sure you want to change the base?
Conversation
@kzawora-intel As @michalkuligowski is off now, can you please review this PR? Thanks. |
@michalkuligowski @kzawora-intel Can you please advise me how you want to change this PR? Our customer waits for roberta embedding enablement. |
964be54
to
0ddcf15
Compare
/run-gaudi-tests |
From those two failed test logs, I see that they were actually passed but somehow couldn't exit the process normally with this message. "Received notify event: Due to an error on node g3-srv179-c03w-idc a jira ticket https://jira.habana-labs.com/browse/SW-225420 was opened, your resource vllm-fork-996-79cqyb8h7e-tfjob might be effected" I think they are not real issues. And the same PR for v1.21.0-next branch which is #1049, all CI passed on it. |
Same PR as #996. Just for v1.21.0_next branch.
This is custom op change as PR #786 follow-up.
Removed RobertaEmbedding class from model file and implemented it as CustomOp class in new file.
forward_cuda() is the original forward function and forward_hpu() is our specific change.