Can't find version information

[AaronWatson2975]

Description

When I tried to run this code both on my home server and Kaggle I ran into the following error.

Could not build a TypeSpec for <KerasTensor: shape=(None, 512) dtype=float32 (created by layer 'tf.math.multiply')> with type KerasTensor

After some googling I think it might be related to my Tensorflow version. Since this repo hasn't had any commits in a year or two that makes sense.

Possible Solution

It would be nice if there was a requirements.txt file in the repo so people can ensure they are running the same versions that were used in the original experiment.

Thanks

I've gotta say, this is a really cool project. It's crazy, the fact that you could train it in a day on a 1080 is mind blowing, incredible results.

[someone2639]

Had this same issue, seemed to be related to the Lambda function, and converting it to a classical lambda let it through with now-solvable errors

Look for this line style[-1] = Lambda(lambda x: x * trunc)(style[-1]) and replace with yy = lambda x: x * trunc; style[-1] = yy(style[-1])

(I'm not entirely sure if this works so i will be updating this whenever the model is finished training)