Eliminate runBatch #1587
Comments
nadavrot
added a commit
to nadavrot/glow
that referenced
this issue
Sep 6, 2018
Add support for the runBatch interface. This commit implements the parts of runBatch that copy parts of the tensors into the placeholder. The implementation is zero-copy because we use tensor views. I hope that this is a temporary solution and that we'll get rid of runBatch soon (see pytorch#1587).
nadavrot
added a commit
that referenced
this issue
Sep 6, 2018
Add support for the runBatch interface. This commit implements the parts of runBatch that copy parts of the tensors into the placeholder. The implementation is zero-copy because we use tensor views. I hope that this is a temporary solution and that we'll get rid of runBatch soon (see #1587).
|
@jfix71 Yes, thank you! |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
The execution engine provides a runBatch method that contains an internal static variable (global variable) that keeps track of the batch size. The batch size is calculated based on the first dimension of the input tensors.
We should remove this method and replace it with some helper utility. The execution engine should not provide this strange interface. The runBatch replacement should rely on the run() interface that the execution engine provides.
The text was updated successfully, but these errors were encountered: