Practice
Learn
← Quit
0 / 0 correct
LLMs & AI
Hard
1 of 20
Why is softmax computed as exp(x − max(x)) / Σ exp(x − max(x)) in practice?
A
It makes softmax faster
B
It is required for causal masking
C
It changes the probabilities to favor rare tokens
D
Subtracting the max avoids floating-point overflow without changing the result