Fix ForceChannelLastForConvPass for nhwc depthwise convolutions#17460
Fix ForceChannelLastForConvPass for nhwc depthwise convolutions#17460mcremon-meta wants to merge 2 commits intomainfrom
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/17460
Note: Links to docs will display an error until the docs builds have been completed. ❌ 2 New FailuresAs of commit a6d7b4a with merge base 6e31609 ( NEW FAILURES - The following jobs have failed:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
meta-cla
bot
added
the
CLA Signed
|
@mcremon-meta has exported this pull request. If you are a Meta employee, you can view the originating Diff in D93188974. |
This PR needs a
|
#17460) Summary: As titled. The current pass doesn't handle the depthwise weights with the right shape, leading to numerical errors. Add a test to catch that as well. Reviewed By: hsharma35, DrJessop Differential Revision: D93188974
facebook-github-bot
force-pushed
the
export-D93188974
branch
from
66c9ada to
70cc273
Compare
Summary: As titled. Much easier to debug that way. Retains the initial tests, adds a few more for 1D cases (will migrate to true 1D kernels soon). We also update the nchw ops to handle `uint8_t` (that got lost in translation when splitting the ops probably). Reviewed By: DrJessop Differential Revision: D93112638
#17460) Summary: As titled. The current pass doesn't handle the depthwise weights with the right shape, leading to numerical errors. Add a test to catch that as well. Reviewed By: hsharma35, DrJessop Differential Revision: D93188974
facebook-github-bot
force-pushed
the
export-D93188974
branch
from
70cc273 to
a6d7b4a
Compare
2 participants
Summary: As titled. The current pass doesn't handle the depthwise weights with the right shape, leading to numerical errors. Add a test to catch that as well.
Reviewed By: DrJessop
Differential Revision: D93188974