-
Notifications
You must be signed in to change notification settings - Fork 425
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
[Question] What does device / embedding_specs.compute_device parameter in ctor of TBE mean? #2395
Comments
Hi @JacoCheung
There is a list of constraints being checked at runtime TBE construction, including:
If the constraints are not met, an error with detailed messages will be thrown, which will help guide you to create TBE with the correct combination of parameters. We will update our docs to explain this more in details. Let us know if you have other questions. cc @sryap |
@q10 Thanks for your reply! What if the |
@JacoCheung If |
Thanks @q10 . I have another question (but may be beyond this issue's scope). How are the pyhsical tables allocated (Assume EmbeddingLocation are all device)? Do multiple tables share single memory chunk? For example, I have 2 embedding tables with different embedding dimensions. Will there be 2 separate memory buffers or single? And how many lookup kernels will be launched while doing the forward? |
Hi team, I am confused by the following parameters related to device context ( in TBE ctor. )
What combinations of those parameters are legal? Could anyone provide some hints?
Thanks!
The text was updated successfully, but these errors were encountered: