Embeddings - Why does t_embeddings have 3 dimensions instead of 2?

[jdesouza-ai]

Hi,

I have used a script from another StackGAN repo to generate embeddings for sentences. The result for a set of sentences when I load in the t7 file and convert it numpy is a 2D matrix - where there is a 1 dimensional embedding for each sentence - this is to be expected.

However, when I look at the code for this repo is shows the below reference:
t_embeddings[:, i, :]

indicating that t_embeddings is 3D - where does the extra dimension come from?

[mansijoshii]

Facing the same issue. Tried reducing it to 2 dimensions but getting a matmul error later in model.py