xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-03-31 19:17:05 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Kernel] adding fused moe kernel config for L40S TP4 (#9245)

Browse Source
This commit is contained in:
Burkhard Ringlein 2024-10-11 08:54:22 -07:00 committed by GitHub
parent 7342a7d7f8
commit f710090d8e
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 173 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

173
vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=NVIDIA_L40S.json Normal file
View File

@ -0,0 +1,173 @@
{
"1": {
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 32,
"BLOCK_SIZE_K": 64,
"GROUP_SIZE_M": 1,
"num_warps": 4,
"num_ctas": 1,
"num_stages": 2
},
"2": {
"BLOCK_SIZE_M": 32,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 2,
"num_warps": 4,
"num_ctas": 1,
"num_stages": 7
},
"4": {
"BLOCK_SIZE_M": 32,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 128,
"num_warps": 2,
"num_ctas": 1,
"num_stages": 4
},
"8": {
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 32,
"BLOCK_SIZE_K": 256,
"GROUP_SIZE_M": 1,
"num_warps": 2,
"num_ctas": 1,
"num_stages": 1
},
"16": {
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 32,
"BLOCK_SIZE_K": 256,
"GROUP_SIZE_M": 1,
"num_warps": 2,
"num_ctas": 1,
"num_stages": 1
},
"32": {
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 256,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 2,
"num_warps": 4,
"num_ctas": 1,
"num_stages": 2
},
"48": {
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 256,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 2,
"num_warps": 4,
"num_ctas": 1,
"num_stages": 2
},
"64": {
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 256,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 2,
"num_warps": 4,
"num_ctas": 1,
"num_stages": 2
},
"96": {
"BLOCK_SIZE_M": 32,
"BLOCK_SIZE_N": 256,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 8,
"num_warps": 8,
"num_ctas": 1,
"num_stages": 2
},
"128": {
"BLOCK_SIZE_M": 32,
"BLOCK_SIZE_N": 256,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 8,
"num_warps": 8,
"num_ctas": 1,
"num_stages": 2
},
"192": {
"BLOCK_SIZE_M": 32,
"BLOCK_SIZE_N": 256,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 8,
"num_warps": 8,
"num_ctas": 1,
"num_stages": 2
},
"256": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 256,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1,
"num_warps": 16,
"num_ctas": 1,
"num_stages": 2
},
"512": {
"BLOCK_SIZE_M": 32,
"BLOCK_SIZE_N": 32,
"BLOCK_SIZE_K": 64,
"GROUP_SIZE_M": 128,
"num_warps": 2,
"num_ctas": 1,
"num_stages": 8
},
"1024": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 4,
"num_warps": 8,
"num_ctas": 1,
"num_stages": 3
},
"2048": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 256,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1,
"num_warps": 16,
"num_ctas": 1,
"num_stages": 2
},
"3072": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 256,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1,
"num_warps": 16,
"num_ctas": 1,
"num_stages": 2
},
"4096": {
"BLOCK_SIZE_M": 128,
"BLOCK_SIZE_N": 256,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 64,
"num_warps": 8,
"num_ctas": 1,
"num_stages": 2
},
"6144": {
"BLOCK_SIZE_M": 128,
"BLOCK_SIZE_N": 256,
"BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 64,
"num_warps": 8,
"num_ctas": 1,
"num_stages": 2
},
"8192": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 256,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1,
"num_warps": 16,
"num_ctas": 1,
"num_stages": 2
}
}