UNQ_C6 Number of units in Dense Layer

Course Q&A Natural Language Processing NLP with Attention Models
1

Hi, I have a problem of finding out what number of units to use in tl.Dense()

According to parameters I should use d_ff because (d_ff (int): depth of feed-forward layer ). However if I use d_ff in both tl.Dense(d_ff) it is causing 8 Test cases to pass and 2 Test Cases to fail.

But when I use d_ff in first tl.Dense(d_ff) and d_model in second tl.Dense(d_model) it passes all Test cases.

So my problem is that it is not given that d_model to be used as number of units for second tl.Dense() then why using it passes all test cases while using d_ff for both Dense layers results in 8 passed and 2 failed test cases.

2

Hi @mohit,

Can you tell me which exercise # is this ?

Thanks,
Mubsi

3

Hi @Mubsi It is from Exercise 03 Week 2

UNQ_C6

GRADED FUNCTION: DecoderBlock

def DecoderBlock(d_model, d_ff, n_heads,
dropout, mode, ff_activation):

4

Hi mohit,

The output of the decoder block needs to have the correct dimension so that blocks can be chained. This is why d_ff can be used inside the block, but the last Dense layer of the block requires d_model units.