LSTM of size "d_model"

[lbaiao]

In the lessons, we usually have a LSTM with size “d_model”, where “d_model” is the word embedding size.

What does a LSTM with size “d_model” mean? Does it mean we have a layer with “d_model” LSTM units?

[arvyzukai]

Hi @lbaiao

@lbaiao

What does a LSTM with size “d_model” mean? Does it mean we have a layer with “d_model” LSTM units?

Essentially - yes.
In reality, it means the dimensionality of the LSTM output (how many units, or features, or in other words - numbers (vector of numbers) does each input/sample produce). Often the LSTM input dimensionality (output from the Embedding layer) and LSTM output dimensionality (often to some Dense layer) match (meaning, embedding and lstm sizes match). But that is not a requirement - you can set any LSTM size and you will receive the output of that size.