Are you training on TPUs?
Yes, I am training on colab TPUs.
You cannot save a model locally. You need to use a GCS Bucket.
model.save("gs://you-bucket-name/you-model-name")
should work given that COlab has been provided access to the Bucket.
You can use the following:
from google.colab import auth
auth.authenticate_user()
Hi Sayak,
Actually I have been saving models to google drive when using other architectures, so I wonder what the difference is. But thanks, I can try using gsbucket as you suggested.
Best,
John
When working with TPUs, I am not if Google Drive would work for saving models. Using a GCS Bucket is the recommended solution (I can’t find the official resource that suggests it at the moment).
Hello Sayak can you please explain why is get_config method necessary to save_weights of model in .h5 format. I’m trying to save a model with custom layers in this notebook but it fails. It seems I also have issue regarding the get_config methods implementation.