A lock log says “the lower gate jammed after the upper gate opened.” The model must let “jammed” look at “lower gate,” not just nearby filler words. Which mechanism scores those relationships?
Select an answer to reveal the explanation.
Short Explanation
Jammed must look at lower gate, not just nearby filler. That is self-attention: it scores which tokens in the same sequence matter. A tree split, a learning rate, and NCCL are not that mechanism.
Full Explanation
Self-attention scores how much each position in the same sequence should matter to the current token. That is the associate purpose, not a formula dump. Tree splits, learning rates, and NCCL are not that mechanism.