size_t K_strides[3]
Key strides (B, H, L, D = 1)
Definition params.h:30
 
int D
Head Dim.
Definition params.h:15
 
int B
Batch Size.
Definition params.h:13
 
size_t O_strides[3]
Output strides (B, H, L, D = 1)
Definition params.h:32
 
int gqa_factor
Group Query factor.
Definition params.h:20
 
int H
Heads.
Definition params.h:14
 
int NQ
Number of query blocks.
Definition params.h:23
 
int kL
Key Sequence Length.
Definition params.h:18
 
int NQ_aligned
Number of full query blocks.
Definition params.h:26
 
int qL
Query Sequence Length.
Definition params.h:17
 
int NK
Number of key/value blocks.
Definition params.h:24
 
size_t Q_strides[3]
Query strides (B, H, L, D = 1)
Definition params.h:29
 
int NK_aligned
Number of full key/value blocks.
Definition params.h:27
 
size_t V_strides[3]
Value strides (B, H, L, D = 1)
Definition params.h:31
 
float scale
Attention scale.
Definition params.h:21