Flair: AttributeError: module 'tensorflow_core.keras.activations' has no attribute 'swish'

[ishan42d]

I am very new to this flair package but belive that it can help me resolve the current challenges I am facing right now. I am just trying to get some basics right and while using the following code:

from flair.data import Sentence
sentence = Sentence('The grass is green.')
print(sentence)

I got this error:
AttributeError: module 'tensorflow_core.keras.activations' has no attribute 'swish'

Can someone explain this to me why is this happening? TF is already there on my system and I am not sure why I am getting this error

[whoisjones]

Hi @ishan42d, you're code is working just fine. Here's my print out:
Sentence: "The grass is green ." [− Tokens: 5]
Flair is built up on PyTorch, not on TensorFlow - so i guess it is rather another error on your side and nothing related to flair. You might wanna look at this issue here which seems more related to your problem: Issue

[ishan42d]

@whoisjones : Thank you so much! Let me look into this.

[ishan42d]

Reinstalling Tensorflow helped actually! PyTorch was running fine and was not the issue, TF was. Closing this issue

[rafelps]

Hi, I had the same exact problem, which I can reproduce by just doing:

from flair.data import Sentence
flair_sentence = Sentence('hello')

However, looking at the repo it seems there is nothing about TF nor Keras. Do you have any insight of why this is happening? Maybe some cross-imports?

By the way, as mentioned, the issue can be resolved by changing TF version. The question is: Why a TF error?

Thanks for your work and support!

[whoisjones]

As mentioned in this issue, your problem arises from using transformers (which we are using in flair) and this is not compatible with TF versions >= 2.1

[rafelps]

As mentioned in this issue, your problem arises from using transformers (which we are using in flair) and this is not compatible with TF versions >= 2.1

It makes a lot of sense now! Flair uses PyTorch but it has a "far" requirement with TF through transformers.

Thank you very much!