[onert/train] support Concat for training - Githubissues

Samsung / ONE

On-device Neural Engine

Other

434 stars 157 forks source link

[onert/train] support Concat for training #14216

Open icodo98 opened 2 weeks ago

icodo98 commented 2 weeks ago

what

Let's support Concat operation for training.

why

The InceptionV3 model from TensorFlow applications has Concanation operation. However, onert does not support training this operation.

related

from : #13829

icodo98 commented 2 weeks ago

Error message is : Concat operation is not trainable yet

zetwhite commented 2 weeks ago

Thanks for your help! I added GoodFirstIssue to get the newcomer's help.

icodo98 commented 1 week ago

For newcomer who interested in this issue.

How to reproduce

follow this docs to install one_rt and train
- https://github.com/Samsung/ONE/blob/master/docs/howto/how-to-build-runtime.md
- https://github.com/Samsung/ONE/blob/master/docs/runtime/training_cnn_on_mnist.md
Use InceptionV3 model for training.

Tips

Unlike #13829 , Concat layer is not implemented yet. you may implement it first.
Some PRs you may refer
- MaxPool https://github.com/Samsung/ONE/issues/11427
- Mean https://github.com/Samsung/ONE/issues/12357