forked from Karylab-cklius/vllm
[ROCm] Add tuned selective_state_update config for AMD MI350 (#48159)
Signed-off-by: Giuseppe Grossi <ggrossi@amd.com>
This commit is contained in:
+51
@@ -0,0 +1,51 @@
|
||||
{
|
||||
"triton_version": "3.6.0",
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"num_warps": 4
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"num_warps": 4
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"num_warps": 2
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 8,
|
||||
"num_warps": 1
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 8,
|
||||
"num_warps": 1
|
||||
},
|
||||
"8192": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"num_warps": 2
|
||||
},
|
||||
"16384": {
|
||||
"BLOCK_SIZE_M": 8,
|
||||
"num_warps": 1
|
||||
},
|
||||
"32768": {
|
||||
"BLOCK_SIZE_M": 8,
|
||||
"num_warps": 1
|
||||
},
|
||||
"65536": {
|
||||
"BLOCK_SIZE_M": 8,
|
||||
"num_warps": 1
|
||||
},
|
||||
"131072": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"num_warps": 1
|
||||
},
|
||||
"196608": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"num_warps": 1
|
||||
},
|
||||
"262144": {
|
||||
"BLOCK_SIZE_M": 8,
|
||||
"num_warps": 1
|
||||
}
|
||||
}
|
||||
+51
@@ -0,0 +1,51 @@
|
||||
{
|
||||
"triton_version": "3.6.0",
|
||||
"128": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"num_warps": 4
|
||||
},
|
||||
"256": {
|
||||
"BLOCK_SIZE_M": 16,
|
||||
"num_warps": 4
|
||||
},
|
||||
"1024": {
|
||||
"BLOCK_SIZE_M": 8,
|
||||
"num_warps": 1
|
||||
},
|
||||
"2048": {
|
||||
"BLOCK_SIZE_M": 8,
|
||||
"num_warps": 1
|
||||
},
|
||||
"4096": {
|
||||
"BLOCK_SIZE_M": 4,
|
||||
"num_warps": 1
|
||||
},
|
||||
"8192": {
|
||||
"BLOCK_SIZE_M": 32,
|
||||
"num_warps": 2
|
||||
},
|
||||
"16384": {
|
||||
"BLOCK_SIZE_M": 4,
|
||||
"num_warps": 1
|
||||
},
|
||||
"32768": {
|
||||
"BLOCK_SIZE_M": 4,
|
||||
"num_warps": 1
|
||||
},
|
||||
"65536": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"num_warps": 8
|
||||
},
|
||||
"131072": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"num_warps": 8
|
||||
},
|
||||
"196608": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"num_warps": 2
|
||||
},
|
||||
"262144": {
|
||||
"BLOCK_SIZE_M": 64,
|
||||
"num_warps": 8
|
||||
}
|
||||
}
|
||||
Reference in New Issue
Block a user