See attention add up
Inspect queries, keys, values, and the actual weights behind a tiny attention matrix.
TRY TO FIND OUT
What changes when future positions are masked?
WHAT YOU ARE LOOKING AT
A small numerical attention calculation, not a window into reasoning. The result is an educational calculation, not a claim about a commercial model’s internal state.
CONNECT THE IDEA
Attention: a computed weighted mixture
Queries and keys determine weights; those weights combine values from permitted positions.