Consider a Multi$-$head Transformer Algorithm as shown in the figure below. The algorithm consists of one encoder and one decoder $($similar to the one discussed in the class$).$ Assume $8$ attention heads, embedding size of $256.$ Assume $16-$bit floating point numbers for inputs, outputs and all intermediate operations. Also notice that some of the blocks have been skipped.

a$.$ Implement the transformer algorithm. $(15)$

b$.$ Comment on the total number of bits to be communicated between two consecutive computing blocks with increasing embedding size. Provide proper explanations and visuals. $(15)$