ValueError: Can not squeeze dim[2], expected a dimension of 1, got 150 for 'QAModel/Squeeze_300' (op: 'Squeeze') with input shapes: [?,300,150].

[feemthan]

I am getting this error when i run this in google collab .

[tbrooks007]

FYI - I don't think this is just a google collab issue. I get this when I train RNET locally as well.
If you are training you can comment this code out, it calculates from RNN probabilities that don't seem to be being used as there is other commented out code that would use these probs.

Once I removed this code on my fork...training proceeds.

[AshayNayak]

I am also getting the same error:
ValueError: Can not squeeze dim[2], expected a dimension of 1, got 150 for 'QAModel/Squeeze_300' (op: 'Squeeze') with input shapes: [?,300,150].

[maakvolunteer]

I am also getting the same error:
ValueError: Can not squeeze dim[2], expected a dimension of 1, got 150 for 'QAModel/Squeeze_300' (op: 'Squeeze') with input shapes: [?,300,150].

[neerajvashistha]

FYI - I don't think this is just a google collab issue. I get this when I train RNET locally as well.
If you are training you can comment this code out, it calculates from RNN probabilities that don't seem to be being used as there is other commented out code that would use these probs.

Once I removed this code on my fork...training proceeds.

@tbrooks007 you have written "this code out" what are you referring to? Can you please help?

Thanks!

[nikitatare]

Can anyone help me with above issue

[libertatis]

I replace tf.squeeze(self.rnet_attention, axis=[2]) with tf.reduce_sum(self.rnet_attention, axis=[2])
and reduce the shape: [batch_size, seq_len, hidden_size] -> shape: [batch_size, seq_len],
So we can use it to compute the RNet Attention Probabilities.

[Dhruvadityamittal]

I replace tf.squeeze(self.rnet_attention, axis=[2]) with tf.reduce_sum(self.rnet_attention, axis=[2])
and reduce the shape: [batch_size, seq_len, hidden_size] -> shape: [batch_size, seq_len],
So we can use it to compute the RNet Attention Probabilities.

Yes its working