EasyDeepLearn

Describe a single transformer encoder block in detail.

hard

Answer

  • Input x.
  • Sublayer 1: y = x + Dropout(MultiHeadAttention(LayerNorm(x))).
  • Sublayer 2: z = y + Dropout(MLP(LayerNorm(y))), where MLP = Linear → GELU → Linear with dffn  =  4d_{\mathrm{ffn}}\; = \;4d.
  • Modern variants use RMSNorm instead of LN, GLU-family MLPs, and pre-norm placement.
  • All ops are parallel across sequence positions — no recurrence.
Check yourself — multiple choice
  • MLP + attention with no residuals
  • Attention sublayer + MLP sublayer, each wrapped in pre-norm + residual + dropout; MLP has dffnd_{\mathrm{ffn}} ≈ 4d
  • Uses recurrence
  • No MLP

Transformer block: attn sublayer + MLP sublayer, both with pre-norm + residual + dropout.

#transformers#attention#architectures

Practise Deep Learning

214 interview questions in this topic.

Related questions