tf.keras Serialization fix #432
Conversation
Signed-off-by: Shah, Karan <[email protected]>
* Remove `layers.py`, declare model/opt/loss in notebook * Update shape blobs to MNIST 2D input shape Signed-off-by: Shah, Karan <[email protected]>
|
CLA Assistant Lite bot All contributors have signed the CLA ✍️ ✅ |
|
Hi @MasterSkepticista, thanks for linking a PR. I had talked to @psfoley regarding the issue and having a tensorflow version check before invoking the hotfix function might be a better way to go about it. This way, we can support tutorials having older tensorflow versions as well. |
|
@mansishr Version check is a possible solution. If there are existing outside projects using lower TF versions, a version check will be required. |
|
@MasterSkepticista that's right. I'll submit a PR with a version check before invoking the hotfix function. Thanks for your help! |
openfl-tutorials/interactive_api/Tensorflow_MNIST/workspace/Tensorflow_MNIST.ipynb
Outdated
Show resolved
Hide resolved
|
@MasterSkepticista A version check would be useful here for the reason that you mentioned; external projects may rely on an earlier TF version. @mansishr if you want to create a separate PR for that, then I think once cell outputs are cleared in the jupyter notebook this is ready to merge |
Signed-off-by: Shah, Karan <[email protected]>
|
@MasterSkepticista could you please sign the CLA? |
|
I have read the CLA Document and I hereby sign the CLA |
|
@psfoley @MasterSkepticista I've created a PR #435 for TF version check. |
This closes #422.
Serialization of
tf.kerasmodels is natively supported as of keras-team/keras#14748.This PR removes hotfix function to allow serialization on TF2.7 and greater.