[Kernel] Retune Mixtral 8x22b configs for FP8 on H100 (#5294)
This commit is contained in:
parent
4efff036f0
commit
abe855d637
@ -1,113 +1,113 @@
|
|||||||
{
|
{
|
||||||
"1": {
|
"1": {
|
||||||
"BLOCK_SIZE_M": 16,
|
"BLOCK_SIZE_M": 64,
|
||||||
"BLOCK_SIZE_N": 32,
|
"BLOCK_SIZE_N": 128,
|
||||||
"BLOCK_SIZE_K": 64,
|
"BLOCK_SIZE_K": 128,
|
||||||
"GROUP_SIZE_M": 1,
|
"GROUP_SIZE_M": 16,
|
||||||
"num_warps": 8,
|
"num_warps": 4,
|
||||||
"num_stages": 3
|
"num_stages": 5
|
||||||
},
|
},
|
||||||
"2": {
|
"2": {
|
||||||
"BLOCK_SIZE_M": 16,
|
"BLOCK_SIZE_M": 64,
|
||||||
"BLOCK_SIZE_N": 64,
|
"BLOCK_SIZE_N": 256,
|
||||||
"BLOCK_SIZE_K": 64,
|
"BLOCK_SIZE_K": 128,
|
||||||
"GROUP_SIZE_M": 16,
|
"GROUP_SIZE_M": 16,
|
||||||
"num_warps": 4,
|
"num_warps": 4,
|
||||||
"num_stages": 5
|
"num_stages": 5
|
||||||
},
|
},
|
||||||
"4": {
|
"4": {
|
||||||
"BLOCK_SIZE_M": 64,
|
"BLOCK_SIZE_M": 64,
|
||||||
"BLOCK_SIZE_N": 32,
|
"BLOCK_SIZE_N": 256,
|
||||||
"BLOCK_SIZE_K": 256,
|
"BLOCK_SIZE_K": 128,
|
||||||
"GROUP_SIZE_M": 64,
|
"GROUP_SIZE_M": 32,
|
||||||
"num_warps": 8,
|
"num_warps": 4,
|
||||||
"num_stages": 5
|
"num_stages": 4
|
||||||
},
|
},
|
||||||
"8": {
|
"8": {
|
||||||
"BLOCK_SIZE_M": 16,
|
"BLOCK_SIZE_M": 64,
|
||||||
"BLOCK_SIZE_N": 64,
|
"BLOCK_SIZE_N": 256,
|
||||||
"BLOCK_SIZE_K": 256,
|
"BLOCK_SIZE_K": 128,
|
||||||
"GROUP_SIZE_M": 1,
|
"GROUP_SIZE_M": 1,
|
||||||
"num_warps": 4,
|
"num_warps": 4,
|
||||||
"num_stages": 3
|
"num_stages": 5
|
||||||
},
|
},
|
||||||
"16": {
|
"16": {
|
||||||
"BLOCK_SIZE_M": 128,
|
"BLOCK_SIZE_M": 64,
|
||||||
"BLOCK_SIZE_N": 64,
|
"BLOCK_SIZE_N": 64,
|
||||||
"BLOCK_SIZE_K": 64,
|
"BLOCK_SIZE_K": 256,
|
||||||
"GROUP_SIZE_M": 1,
|
"GROUP_SIZE_M": 32,
|
||||||
"num_warps": 8,
|
"num_warps": 4,
|
||||||
"num_stages": 3
|
"num_stages": 3
|
||||||
},
|
},
|
||||||
"24": {
|
"24": {
|
||||||
"BLOCK_SIZE_M": 64,
|
"BLOCK_SIZE_M": 64,
|
||||||
"BLOCK_SIZE_N": 256,
|
"BLOCK_SIZE_N": 64,
|
||||||
"BLOCK_SIZE_K": 128,
|
"BLOCK_SIZE_K": 128,
|
||||||
"GROUP_SIZE_M": 64,
|
"GROUP_SIZE_M": 1,
|
||||||
"num_warps": 8,
|
"num_warps": 4,
|
||||||
"num_stages": 2
|
"num_stages": 3
|
||||||
},
|
},
|
||||||
"32": {
|
"32": {
|
||||||
"BLOCK_SIZE_M": 64,
|
"BLOCK_SIZE_M": 64,
|
||||||
"BLOCK_SIZE_N": 256,
|
"BLOCK_SIZE_N": 64,
|
||||||
"BLOCK_SIZE_K": 64,
|
"BLOCK_SIZE_K": 128,
|
||||||
"GROUP_SIZE_M": 1,
|
"GROUP_SIZE_M": 16,
|
||||||
"num_warps": 8,
|
"num_warps": 4,
|
||||||
"num_stages": 3
|
"num_stages": 3
|
||||||
},
|
},
|
||||||
"48": {
|
"48": {
|
||||||
"BLOCK_SIZE_M": 64,
|
"BLOCK_SIZE_M": 64,
|
||||||
"BLOCK_SIZE_N": 256,
|
"BLOCK_SIZE_N": 64,
|
||||||
"BLOCK_SIZE_K": 128,
|
"BLOCK_SIZE_K": 128,
|
||||||
"GROUP_SIZE_M": 64,
|
"GROUP_SIZE_M": 1,
|
||||||
"num_warps": 8,
|
"num_warps": 4,
|
||||||
"num_stages": 4
|
"num_stages": 3
|
||||||
},
|
},
|
||||||
"64": {
|
"64": {
|
||||||
"BLOCK_SIZE_M": 128,
|
|
||||||
"BLOCK_SIZE_N": 256,
|
|
||||||
"BLOCK_SIZE_K": 128,
|
|
||||||
"GROUP_SIZE_M": 64,
|
|
||||||
"num_warps": 8,
|
|
||||||
"num_stages": 2
|
|
||||||
},
|
|
||||||
"96": {
|
|
||||||
"BLOCK_SIZE_M": 64,
|
"BLOCK_SIZE_M": 64,
|
||||||
"BLOCK_SIZE_N": 256,
|
"BLOCK_SIZE_N": 64,
|
||||||
"BLOCK_SIZE_K": 128,
|
"BLOCK_SIZE_K": 128,
|
||||||
"GROUP_SIZE_M": 16,
|
"GROUP_SIZE_M": 16,
|
||||||
"num_warps": 4,
|
"num_warps": 4,
|
||||||
"num_stages": 4
|
"num_stages": 3
|
||||||
|
},
|
||||||
|
"96": {
|
||||||
|
"BLOCK_SIZE_M": 64,
|
||||||
|
"BLOCK_SIZE_N": 64,
|
||||||
|
"BLOCK_SIZE_K": 128,
|
||||||
|
"GROUP_SIZE_M": 16,
|
||||||
|
"num_warps": 4,
|
||||||
|
"num_stages": 3
|
||||||
},
|
},
|
||||||
"128": {
|
"128": {
|
||||||
"BLOCK_SIZE_M": 64,
|
"BLOCK_SIZE_M": 64,
|
||||||
"BLOCK_SIZE_N": 256,
|
"BLOCK_SIZE_N": 64,
|
||||||
"BLOCK_SIZE_K": 64,
|
"BLOCK_SIZE_K": 128,
|
||||||
"GROUP_SIZE_M": 32,
|
"GROUP_SIZE_M": 16,
|
||||||
"num_warps": 8,
|
"num_warps": 4,
|
||||||
"num_stages": 3
|
"num_stages": 3
|
||||||
},
|
},
|
||||||
"256": {
|
"256": {
|
||||||
"BLOCK_SIZE_M": 128,
|
|
||||||
"BLOCK_SIZE_N": 64,
|
|
||||||
"BLOCK_SIZE_K": 128,
|
|
||||||
"GROUP_SIZE_M": 64,
|
|
||||||
"num_warps": 8,
|
|
||||||
"num_stages": 4
|
|
||||||
},
|
|
||||||
"512": {
|
|
||||||
"BLOCK_SIZE_M": 64,
|
"BLOCK_SIZE_M": 64,
|
||||||
"BLOCK_SIZE_N": 256,
|
"BLOCK_SIZE_N": 128,
|
||||||
"BLOCK_SIZE_K": 128,
|
"BLOCK_SIZE_K": 128,
|
||||||
"GROUP_SIZE_M": 64,
|
"GROUP_SIZE_M": 64,
|
||||||
"num_warps": 4,
|
"num_warps": 4,
|
||||||
"num_stages": 3
|
"num_stages": 3
|
||||||
},
|
},
|
||||||
|
"512": {
|
||||||
|
"BLOCK_SIZE_M": 128,
|
||||||
|
"BLOCK_SIZE_N": 128,
|
||||||
|
"BLOCK_SIZE_K": 128,
|
||||||
|
"GROUP_SIZE_M": 64,
|
||||||
|
"num_warps": 8,
|
||||||
|
"num_stages": 5
|
||||||
|
},
|
||||||
"1024": {
|
"1024": {
|
||||||
"BLOCK_SIZE_M": 128,
|
"BLOCK_SIZE_M": 128,
|
||||||
"BLOCK_SIZE_N": 256,
|
"BLOCK_SIZE_N": 256,
|
||||||
"BLOCK_SIZE_K": 128,
|
"BLOCK_SIZE_K": 128,
|
||||||
"GROUP_SIZE_M": 32,
|
"GROUP_SIZE_M": 64,
|
||||||
"num_warps": 8,
|
"num_warps": 8,
|
||||||
"num_stages": 4
|
"num_stages": 4
|
||||||
},
|
},
|
||||||
@ -139,7 +139,7 @@
|
|||||||
"BLOCK_SIZE_M": 128,
|
"BLOCK_SIZE_M": 128,
|
||||||
"BLOCK_SIZE_N": 256,
|
"BLOCK_SIZE_N": 256,
|
||||||
"BLOCK_SIZE_K": 128,
|
"BLOCK_SIZE_K": 128,
|
||||||
"GROUP_SIZE_M": 32,
|
"GROUP_SIZE_M": 16,
|
||||||
"num_warps": 8,
|
"num_warps": 8,
|
||||||
"num_stages": 4
|
"num_stages": 4
|
||||||
}
|
}
|
||||||
|
|||||||
@ -1,17 +1,17 @@
|
|||||||
{
|
{
|
||||||
"1": {
|
"1": {
|
||||||
"BLOCK_SIZE_M": 16,
|
"BLOCK_SIZE_M": 64,
|
||||||
"BLOCK_SIZE_N": 32,
|
"BLOCK_SIZE_N": 128,
|
||||||
"BLOCK_SIZE_K": 64,
|
"BLOCK_SIZE_K": 128,
|
||||||
"GROUP_SIZE_M": 1,
|
"GROUP_SIZE_M": 1,
|
||||||
"num_warps": 4,
|
"num_warps": 4,
|
||||||
"num_stages": 5
|
"num_stages": 5
|
||||||
},
|
},
|
||||||
"2": {
|
"2": {
|
||||||
"BLOCK_SIZE_M": 64,
|
"BLOCK_SIZE_M": 64,
|
||||||
"BLOCK_SIZE_N": 64,
|
"BLOCK_SIZE_N": 128,
|
||||||
"BLOCK_SIZE_K": 64,
|
"BLOCK_SIZE_K": 128,
|
||||||
"GROUP_SIZE_M": 16,
|
"GROUP_SIZE_M": 64,
|
||||||
"num_warps": 4,
|
"num_warps": 4,
|
||||||
"num_stages": 4
|
"num_stages": 4
|
||||||
},
|
},
|
||||||
@ -20,59 +20,59 @@
|
|||||||
"BLOCK_SIZE_N": 128,
|
"BLOCK_SIZE_N": 128,
|
||||||
"BLOCK_SIZE_K": 128,
|
"BLOCK_SIZE_K": 128,
|
||||||
"GROUP_SIZE_M": 64,
|
"GROUP_SIZE_M": 64,
|
||||||
"num_warps": 8,
|
"num_warps": 4,
|
||||||
"num_stages": 2
|
"num_stages": 3
|
||||||
},
|
},
|
||||||
"8": {
|
"8": {
|
||||||
"BLOCK_SIZE_M": 64,
|
|
||||||
"BLOCK_SIZE_N": 256,
|
|
||||||
"BLOCK_SIZE_K": 128,
|
|
||||||
"GROUP_SIZE_M": 32,
|
|
||||||
"num_warps": 4,
|
|
||||||
"num_stages": 3
|
|
||||||
},
|
|
||||||
"16": {
|
|
||||||
"BLOCK_SIZE_M": 64,
|
|
||||||
"BLOCK_SIZE_N": 128,
|
|
||||||
"BLOCK_SIZE_K": 256,
|
|
||||||
"GROUP_SIZE_M": 1,
|
|
||||||
"num_warps": 8,
|
|
||||||
"num_stages": 2
|
|
||||||
},
|
|
||||||
"24": {
|
|
||||||
"BLOCK_SIZE_M": 64,
|
|
||||||
"BLOCK_SIZE_N": 128,
|
|
||||||
"BLOCK_SIZE_K": 256,
|
|
||||||
"GROUP_SIZE_M": 1,
|
|
||||||
"num_warps": 4,
|
|
||||||
"num_stages": 3
|
|
||||||
},
|
|
||||||
"32": {
|
|
||||||
"BLOCK_SIZE_M": 64,
|
"BLOCK_SIZE_M": 64,
|
||||||
"BLOCK_SIZE_N": 64,
|
"BLOCK_SIZE_N": 64,
|
||||||
"BLOCK_SIZE_K": 128,
|
"BLOCK_SIZE_K": 128,
|
||||||
"GROUP_SIZE_M": 1,
|
"GROUP_SIZE_M": 1,
|
||||||
"num_warps": 8,
|
"num_warps": 4,
|
||||||
"num_stages": 3
|
"num_stages": 5
|
||||||
|
},
|
||||||
|
"16": {
|
||||||
|
"BLOCK_SIZE_M": 64,
|
||||||
|
"BLOCK_SIZE_N": 64,
|
||||||
|
"BLOCK_SIZE_K": 128,
|
||||||
|
"GROUP_SIZE_M": 16,
|
||||||
|
"num_warps": 4,
|
||||||
|
"num_stages": 5
|
||||||
|
},
|
||||||
|
"24": {
|
||||||
|
"BLOCK_SIZE_M": 64,
|
||||||
|
"BLOCK_SIZE_N": 64,
|
||||||
|
"BLOCK_SIZE_K": 128,
|
||||||
|
"GROUP_SIZE_M": 16,
|
||||||
|
"num_warps": 4,
|
||||||
|
"num_stages": 5
|
||||||
|
},
|
||||||
|
"32": {
|
||||||
|
"BLOCK_SIZE_M": 64,
|
||||||
|
"BLOCK_SIZE_N": 128,
|
||||||
|
"BLOCK_SIZE_K": 256,
|
||||||
|
"GROUP_SIZE_M": 1,
|
||||||
|
"num_warps": 4,
|
||||||
|
"num_stages": 4
|
||||||
},
|
},
|
||||||
"48": {
|
"48": {
|
||||||
"BLOCK_SIZE_M": 64,
|
"BLOCK_SIZE_M": 64,
|
||||||
"BLOCK_SIZE_N": 128,
|
"BLOCK_SIZE_N": 64,
|
||||||
"BLOCK_SIZE_K": 128,
|
"BLOCK_SIZE_K": 128,
|
||||||
"GROUP_SIZE_M": 64,
|
"GROUP_SIZE_M": 1,
|
||||||
"num_warps": 4,
|
"num_warps": 4,
|
||||||
"num_stages": 2
|
"num_stages": 3
|
||||||
},
|
},
|
||||||
"64": {
|
"64": {
|
||||||
"BLOCK_SIZE_M": 64,
|
"BLOCK_SIZE_M": 64,
|
||||||
"BLOCK_SIZE_N": 64,
|
"BLOCK_SIZE_N": 64,
|
||||||
"BLOCK_SIZE_K": 256,
|
"BLOCK_SIZE_K": 128,
|
||||||
"GROUP_SIZE_M": 16,
|
"GROUP_SIZE_M": 1,
|
||||||
"num_warps": 4,
|
"num_warps": 4,
|
||||||
"num_stages": 2
|
"num_stages": 3
|
||||||
},
|
},
|
||||||
"96": {
|
"96": {
|
||||||
"BLOCK_SIZE_M": 128,
|
"BLOCK_SIZE_M": 64,
|
||||||
"BLOCK_SIZE_N": 64,
|
"BLOCK_SIZE_N": 64,
|
||||||
"BLOCK_SIZE_K": 128,
|
"BLOCK_SIZE_K": 128,
|
||||||
"GROUP_SIZE_M": 1,
|
"GROUP_SIZE_M": 1,
|
||||||
@ -81,11 +81,11 @@
|
|||||||
},
|
},
|
||||||
"128": {
|
"128": {
|
||||||
"BLOCK_SIZE_M": 64,
|
"BLOCK_SIZE_M": 64,
|
||||||
"BLOCK_SIZE_N": 128,
|
"BLOCK_SIZE_N": 64,
|
||||||
"BLOCK_SIZE_K": 256,
|
"BLOCK_SIZE_K": 128,
|
||||||
"GROUP_SIZE_M": 1,
|
"GROUP_SIZE_M": 1,
|
||||||
"num_warps": 8,
|
"num_warps": 4,
|
||||||
"num_stages": 2
|
"num_stages": 3
|
||||||
},
|
},
|
||||||
"256": {
|
"256": {
|
||||||
"BLOCK_SIZE_M": 128,
|
"BLOCK_SIZE_M": 128,
|
||||||
@ -93,7 +93,7 @@
|
|||||||
"BLOCK_SIZE_K": 128,
|
"BLOCK_SIZE_K": 128,
|
||||||
"GROUP_SIZE_M": 1,
|
"GROUP_SIZE_M": 1,
|
||||||
"num_warps": 8,
|
"num_warps": 8,
|
||||||
"num_stages": 5
|
"num_stages": 3
|
||||||
},
|
},
|
||||||
"512": {
|
"512": {
|
||||||
"BLOCK_SIZE_M": 128,
|
"BLOCK_SIZE_M": 128,
|
||||||
|
|||||||
Loading…
Reference in New Issue
Block a user