Equation 13 · How Llama's Architecture Actually Works, Generation by Generation
What does this equation mean?
Read the formula alongside the article passage below. Each part has a deeper page with its role in the equation, the supporting passage and nearby citations.
This mathematical expression combines the displayed quantities; its precise role follows from the surrounding article text. Read the equation part by part below; each part has a contextual explanation and a link to its mathematical background.
Read it piece by piece
Symbol H
H is a part of this expression. Its role is fixed by the surrounding article and by the operations shown in the formula.
How to interpret it
Read this expression with the definitions, units, and assumptions supplied by the article.
What the article says around this equation
Multi-head attention gives every query head its own key and value projections. That is expensive at inference time for an autoregressive model, because every generated token requires reading the cached keys and values for every previous token, once per head, from memory — and memory bandwidth, not arithmetic, is usually the bottleneck in decoding. Multi-query attention collapses all query heads onto a single shared key-value pair, cutting that cache to a fraction of its multi-head size at some cost in quality. Ainslie and colleagues proposed the middle path that all three Llama generations actually use: grouped-query attention, in which H query heads are partitioned into G groups, and every…
Read the full surrounding passage
Multi-head attention gives every query head its own key and value projections. That is expensive at inference time for an autoregressive model, because every generated token requires reading the cached keys and values for every previous token, once per head, from memory — and memory bandwidth, not arithmetic, is usually the bottleneck in decoding. Multi-query attention collapses all query heads onto a single shared key-value pair, cutting that cache to a fraction of its multi-head size at some cost in quality. Ainslie and colleagues proposed the middle path that all three Llama generations actually use: grouped-query attention, in which H query heads are partitioned into G groups, and every query head within a group shares one key-value head pair [ 8 ] . The two earlier schemes are the boundary cases of the same construction:
Sources cited in the surrounding passage
These citations give research context. Read each source to check which claims it supports.
Return to How Llama's Architecture Actually Works, Generation by Generation