Skip to content
Discussion options

You must be logged in to vote

#10585 (comment)
@hawkinsp said:

A few observations:
a) sometimes users do want to hand large model weights to XLA and bake them into their program. The canonical example of this is neural network inference: you might prebake the weights into the program and allow XLA to do things like transpose them into the perfect layout.
b) the excessive memory use should show up in live_executables().
c) we cannot be zero-copy on platforms other than CPU.
There is a mechanism to get what you want: don't lexically capture values, instead pass them as parameters.

Replies: 6 comments 7 replies

Comment options

You must be logged in to vote
5 replies
@PhilipVinc
Comment options

@YouJiacheng
Comment options

@YouJiacheng
Comment options

@C-J-Cundy
Comment options

@YouJiacheng
Comment options

Comment options

You must be logged in to vote
0 replies
Comment options

You must be logged in to vote
0 replies
Comment options

You must be logged in to vote
2 replies
@soraros
Comment options

@C-J-Cundy
Comment options

Comment options

You must be logged in to vote
0 replies
Comment options

You must be logged in to vote
0 replies
Answer selected by C-J-Cundy
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Category
Q&A
Labels
None yet
4 participants