fix batchnorm dimension issue

by prathikr - opened Jun 24

base: refs/heads/main

←

from: refs/pr/4

Discussion Files changed

-2

This PR is in draft mode

prathikr

Jun 24

•

edited Jun 24

According to the ONNX Spec for BatchNormalization, there is no explicitly defined behavior for input of rank 2. There is precedent for input of rank 1 where we can assume a 2nd dimension of size 1. Extending that precedent, this change adds an unsqueeze node before all BatchNormalization operators to add a 3rd dimension of size 1 and then squeezes away that dimension for downstream operators.

➡️

add unsqueeze/squeezeca7c27ee

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

Publish this branch

This branch is in draft mode, publish it to be able to merge.

· Sign up or log in to comment