Practice
Learn
← Quit
0 / 0 correct
LLMs & AI
Hard
1 of 40
What is the 'unembedding' (output) layer, and why is it often tied to the input embedding?
A
It deletes embeddings after use
B
It maps the final hidden vector to logits over the vocabulary; weight tying reuses the embedding matrix to save parameters
C
It converts logits to text
D
It's the tokenizer's lookup table