Skip to content
GitLab
Menu
Projects
Groups
Snippets
Loading...
Help
Help
Support
Community forum
Keyboard shortcuts
?
Submit feedback
Contribute to GitLab
Sign in
Toggle navigation
Menu
Open sidebar
OpenDAS
vllm_cscc
Commits
18ec9eaa
"vscode:/vscode.git/clone" did not exist on "ecc3dd66cceebcd9fc48216ae3852a0e994fbdc1"
Commit
18ec9eaa
authored
Mar 27, 2025
by
yangql
Browse files
更新bw以及k500sm_ai的支持
parent
d01a8fa8
Changes
4
Hide whitespace changes
Inline
Side-by-side
Showing
4 changed files
with
580 additions
and
43 deletions
+580
-43
vllm/model_executor/layers/fused_moe/configs/E=256,N=128,device_name=BW200,dtype=int4_w4a16.json
...nfigs/E=256,N=128,device_name=BW200,dtype=int4_w4a16.json
+52
-43
vllm/model_executor/layers/fused_moe/configs/E=256,N=128,device_name=K500SM_AI,dtype=int4_w4a16_120.json
...256,N=128,device_name=K500SM_AI,dtype=int4_w4a16_120.json
+182
-0
vllm/model_executor/layers/fused_moe/configs/E=256,N=64,device_name=BW200,dtype=int4_w4a16.json
...onfigs/E=256,N=64,device_name=BW200,dtype=int4_w4a16.json
+173
-0
vllm/model_executor/layers/fused_moe/configs/E=256,N=64,device_name=K500SM_AI,dtype=int4_w4a16_120.json
...=256,N=64,device_name=K500SM_AI,dtype=int4_w4a16_120.json
+173
-0
No files found.
vllm/model_executor/layers/fused_moe/configs/E=256,N=128,device_name=BW200,dtype=int4_w4a16.json
View file @
18ec9eaa
{
{
"1"
:
{
"1"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_M"
:
32
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_K"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
4
,
"num_stages"
:
2
,
"num_ldmatrixes"
:
0
"num_ldmatrixes"
:
0
},
},
"2"
:
{
"2"
:
{
...
@@ -20,145 +20,154 @@
...
@@ -20,145 +20,154 @@
"4"
:
{
"4"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
32
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_stages"
:
2
,
"num_ldmatrixes"
:
0
"num_ldmatrixes"
:
0
},
},
"8"
:
{
"8"
:
{
"BLOCK_SIZE_M"
:
32
,
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
32
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
8
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_stages"
:
1
,
"num_ldmatrixes"
:
0
"num_ldmatrixes"
:
0
},
},
"16"
:
{
"16"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
32
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_stages"
:
1
,
"num_ldmatrixes"
:
0
"num_ldmatrixes"
:
0
},
},
"24"
:
{
"24"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
32
,
"BLOCK_SIZE_K"
:
32
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_stages"
:
1
,
"num_ldmatrixes"
:
0
"num_ldmatrixes"
:
0
},
},
"32"
:
{
"32"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
32
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_stages"
:
1
,
"num_ldmatrixes"
:
0
"num_ldmatrixes"
:
0
},
},
"48"
:
{
"48"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
32
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_stages"
:
1
,
"num_ldmatrixes"
:
0
"num_ldmatrixes"
:
0
},
},
"64"
:
{
"64"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
32
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_stages"
:
1
,
"num_ldmatrixes"
:
0
"num_ldmatrixes"
:
0
},
},
"96"
:
{
"96"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
32
,
"BLOCK_SIZE_K"
:
32
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
2
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_stages"
:
1
,
"num_ldmatrixes"
:
0
"num_ldmatrixes"
:
0
},
},
"128"
:
{
"128"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
32
,
"BLOCK_SIZE_K"
:
32
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_stages"
:
1
,
"num_ldmatrixes"
:
0
"num_ldmatrixes"
:
0
},
},
"256"
:
{
"256"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
32
,
"BLOCK_SIZE_K"
:
32
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_stages"
:
1
,
"num_ldmatrixes"
:
0
"num_ldmatrixes"
:
0
},
},
"512"
:
{
"512"
:
{
"BLOCK_SIZE_M"
:
32
,
"BLOCK_SIZE_M"
:
32
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
32
,
"BLOCK_SIZE_K"
:
32
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_stages"
:
1
,
"num_ldmatrixes"
:
0
"num_ldmatrixes"
:
0
},
},
"1024"
:
{
"1024"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
32
,
"BLOCK_SIZE_K"
:
32
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_stages"
:
1
,
"num_ldmatrixes"
:
0
"num_ldmatrixes"
:
0
},
},
"1536"
:
{
"1536"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
32
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_stages"
:
1
,
"num_ldmatrixes"
:
0
"num_ldmatrixes"
:
0
},
},
"2048"
:
{
"2048"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
32
,
"BLOCK_SIZE_K"
:
32
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_stages"
:
1
,
"num_ldmatrixes"
:
0
"num_ldmatrixes"
:
0
},
},
"3072"
:
{
"3072"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
32
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_stages"
:
1
,
"num_ldmatrixes"
:
0
"num_ldmatrixes"
:
0
},
},
"4096"
:
{
"4096"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
32
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
8
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_stages"
:
1
,
"num_ldmatrixes"
:
0
},
"8192"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
1
,
"num_ldmatrixes"
:
0
"num_ldmatrixes"
:
0
}
}
}
}
vllm/model_executor/layers/fused_moe/configs/E=256,N=128,device_name=K500SM_AI,dtype=int4_w4a16_120.json
0 → 100644
View file @
18ec9eaa
{
"1"
:
{
"BLOCK_SIZE_M"
:
32
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_ldmatrixes"
:
0
},
"2"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_ldmatrixes"
:
0
},
"4"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_ldmatrixes"
:
0
},
"8"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_ldmatrixes"
:
0
},
"16"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_ldmatrixes"
:
0
},
"24"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_ldmatrixes"
:
0
},
"32"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
1
,
"num_ldmatrixes"
:
0
},
"48"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_ldmatrixes"
:
0
},
"64"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_ldmatrixes"
:
0
},
"96"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_ldmatrixes"
:
0
},
"128"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_ldmatrixes"
:
0
},
"256"
:
{
"BLOCK_SIZE_M"
:
32
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
1
,
"num_ldmatrixes"
:
0
},
"512"
:
{
"BLOCK_SIZE_M"
:
32
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_ldmatrixes"
:
0
},
"1024"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_ldmatrixes"
:
0
},
"1536"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
1
,
"num_ldmatrixes"
:
0
},
"2048"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
1
,
"num_ldmatrixes"
:
0
},
"3072"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
1
,
"num_ldmatrixes"
:
0
},
"4096"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
32
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
4
,
"num_ldmatrixes"
:
0
},
"6144"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
1
,
"num_ldmatrixes"
:
0
},
"8192"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
32
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
8
,
"num_stages"
:
1
,
"num_ldmatrixes"
:
0
}
}
vllm/model_executor/layers/fused_moe/configs/E=256,N=64,device_name=BW200,dtype=int4_w4a16.json
0 → 100644
View file @
18ec9eaa
{
"1"
:
{
"BLOCK_SIZE_M"
:
32
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_ldmatrixes"
:
0
},
"2"
:
{
"BLOCK_SIZE_M"
:
32
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_ldmatrixes"
:
0
},
"4"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_ldmatrixes"
:
0
},
"8"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_ldmatrixes"
:
0
},
"16"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_ldmatrixes"
:
0
},
"24"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
1
,
"num_ldmatrixes"
:
0
},
"32"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
1
,
"num_ldmatrixes"
:
0
},
"48"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
32
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_ldmatrixes"
:
0
},
"64"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
32
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_ldmatrixes"
:
0
},
"96"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
32
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_ldmatrixes"
:
0
},
"128"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
32
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
1
,
"num_ldmatrixes"
:
0
},
"256"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
32
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
1
,
"num_ldmatrixes"
:
0
},
"512"
:
{
"BLOCK_SIZE_M"
:
32
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
32
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
1
,
"num_ldmatrixes"
:
0
},
"1024"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
32
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
1
,
"num_ldmatrixes"
:
0
},
"1536"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
1
,
"num_ldmatrixes"
:
0
},
"2048"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
32
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
1
,
"num_ldmatrixes"
:
0
},
"3072"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
1
,
"num_ldmatrixes"
:
0
},
"4096"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
1
,
"num_ldmatrixes"
:
0
},
"8192"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
1
,
"num_ldmatrixes"
:
0
}
}
vllm/model_executor/layers/fused_moe/configs/E=256,N=64,device_name=K500SM_AI,dtype=int4_w4a16_120.json
0 → 100644
View file @
18ec9eaa
{
"1"
:
{
"BLOCK_SIZE_M"
:
32
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_ldmatrixes"
:
0
},
"2"
:
{
"BLOCK_SIZE_M"
:
32
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_ldmatrixes"
:
0
},
"4"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_ldmatrixes"
:
0
},
"8"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_ldmatrixes"
:
0
},
"16"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_ldmatrixes"
:
0
},
"24"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_ldmatrixes"
:
0
},
"32"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
32
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
1
,
"num_ldmatrixes"
:
0
},
"48"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_ldmatrixes"
:
0
},
"64"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_ldmatrixes"
:
0
},
"96"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_ldmatrixes"
:
0
},
"128"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_ldmatrixes"
:
0
},
"256"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_ldmatrixes"
:
0
},
"512"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_ldmatrixes"
:
0
},
"1024"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_ldmatrixes"
:
0
},
"1536"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_ldmatrixes"
:
0
},
"2048"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
1
,
"num_ldmatrixes"
:
0
},
"3072"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
1
,
"num_ldmatrixes"
:
0
},
"4096"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
2
,
"num_ldmatrixes"
:
0
},
"8192"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
1
,
"num_ldmatrixes"
:
0
}
}
Write
Preview
Markdown
is supported
0%
Try again
or
attach a new file
.
Attach a file
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Cancel
Please
register
or
sign in
to comment
