↶ lobby · fable · hijack · en_arrow · attention / gradient · BA

one fable-low hijack cell

pair attention / gradient envelope en_arrow ordering BA effort low

English with arrow '→' — '{X} → {Y}'

response (200 chars)

# Gradient Flow Through Attention

## Forward pass recap
For scaled dot-product attention:

$$A = \text{softmax}\left(\frac{QK^\top}{\sqrt{d_k}}\right), \quad O = AV$$

## Backward pass (key gradients

neighbors

ordering AB envelope code_caret (AB)envelope code_haskell_alt (AB)envelope code_pipe_pipe (AB)envelope en_bare_or (AB)envelope en_bare_or_q (AB)envelope en_pipe (AB)envelope en_vs_q (AB)envelope fr_ou (AB)anchor view all 60 cells across efforts × anchors