-
Notifications
You must be signed in to change notification settings - Fork 384
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
SepConv1d/2d for io_parallel with Latency strategy #1012
base: main
Are you sure you want to change the base?
Conversation
@dgburnette This could be followed up in Catapult. I noticed that you don't use the code-gen im2col implementations of conv1d/2d so I didn't want to mess with it |
elif '2D' in node_class: | ||
self._generate_im2col_2d(node) | ||
node_class = node.class_name | ||
if 'Separable' in node_class: |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
If we just had DepthwiseConv2D (https://keras.io/api/layers/convolution_layers/depthwise_convolution2d/) would this work? We nominally support it.
Description
This PR adds implementation of
SeparableConv1D/2D
forio_parallel
andLatency
strategy. It's based on existing code-generation forConv1D/2D
.Type of change
Tests
Added tests in
test_sepconv1d.py
andtest_sepconv2d.py
. 1D tests had a nonsense test with a kernel size of 1, I changed that to 2 to be in line with the 2D tests.Checklist
pre-commit
on the files I edited or added.