A head is reading "the black cat slept". The token "slept" has already done its scoring and its softmax, so its attention weights are settled:
| token | attention weight |
|---|---|
| the | |
| black | |
| cat | |
| slept |
Every token in the sentence projected a key and a value. Both tables are given in full:
| token | key | value |
|---|---|---|
| the | ||
| black | ||
| cat | ||
| slept |
The blending stage produces one context vector for "slept".
What is the third component of that context vector?
Round your answer to 2 decimal places.