specific method · filed under model architecture
Block-sparse grouped-query attention
A grouped-query attention variant that uses block sparsity in the attention stack.
Also called block-sparse GQA.
- source
- 1
- model
- 1
- lab adopt it
- 1
- strongest
- core
How sources treat it
One count per evidence span, weakest treatment to strongest.
core 1
Documented in
Evidence
1 span quoted from the sources, strongest treatment first.
The full attention stack gets replaced with block-sparse GQA.
coremodel architecturein MiniMax-M3MiniMax
Filed alongside
Other methods under model architecture :: token mixer :: softmax attention :: grouped-query attention.