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
c3dde367
Unverified
Commit
c3dde367
authored
Jun 27, 2024
by
Divakar Verma
Committed by
GitHub
Jun 27, 2024
Browse files
[Kernel][ROCm][AMD] fused_moe Triton configs v2 for mi300X (#5932)
parent
64e8d2a7
Changes
4
Show whitespace changes
Inline
Side-by-side
Showing
4 changed files
with
500 additions
and
194 deletions
+500
-194
vllm/model_executor/layers/fused_moe/configs/E=8,N=14336,device_name=AMD_Instinct_MI300X.json
.../configs/E=8,N=14336,device_name=AMD_Instinct_MI300X.json
+118
-46
vllm/model_executor/layers/fused_moe/configs/E=8,N=1792,device_name=AMD_Instinct_MI300X.json
...e/configs/E=8,N=1792,device_name=AMD_Instinct_MI300X.json
+136
-46
vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=AMD_Instinct_MI300X.json
...e/configs/E=8,N=3584,device_name=AMD_Instinct_MI300X.json
+122
-50
vllm/model_executor/layers/fused_moe/configs/E=8,N=7168,device_name=AMD_Instinct_MI300X.json
...e/configs/E=8,N=7168,device_name=AMD_Instinct_MI300X.json
+124
-52
No files found.
vllm/model_executor/layers/fused_moe/configs/E=8,N=14336,device_name=AMD_Instinct_MI300X.json
View file @
c3dde367
{
{
"1"
:
{
"1"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_K"
:
128
,
"BLOCK_SIZE_K"
:
256
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
,
"num_stages"
:
0
"num_warps"
:
2
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
1
},
},
"2"
:
{
"2"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
6
4
,
"BLOCK_SIZE_N"
:
1
6
,
"BLOCK_SIZE_K"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
,
"num_stages"
:
0
"num_warps"
:
2
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
2
},
},
"4"
:
{
"4"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_K"
:
256
,
"BLOCK_SIZE_K"
:
256
,
"GROUP_SIZE_M"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_stages"
:
1
"num_warps"
:
2
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
2
},
},
"8"
:
{
"8"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
6
4
,
"BLOCK_SIZE_N"
:
1
6
,
"BLOCK_SIZE_K"
:
256
,
"BLOCK_SIZE_K"
:
256
,
"GROUP_SIZE_M"
:
32
,
"GROUP_SIZE_M"
:
1
,
"num_stages"
:
1
"num_warps"
:
1
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
2
},
},
"16"
:
{
"16"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
6
4
,
"BLOCK_SIZE_N"
:
1
6
,
"BLOCK_SIZE_K"
:
256
,
"BLOCK_SIZE_K"
:
256
,
"GROUP_SIZE_M"
:
8
,
"GROUP_SIZE_M"
:
1
,
"num_stages"
:
1
"num_warps"
:
4
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
2
},
},
"24"
:
{
"24"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_K"
:
256
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
64
,
"GROUP_SIZE_M"
:
1
,
"num_stages"
:
1
"num_warps"
:
1
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
2
},
},
"32"
:
{
"32"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
256
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
8
,
"GROUP_SIZE_M"
:
4
,
"num_stages"
:
1
"num_warps"
:
2
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
1
},
},
"48"
:
{
"48"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
6
4
,
"BLOCK_SIZE_N"
:
1
6
,
"BLOCK_SIZE_K"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
8
,
"GROUP_SIZE_M"
:
4
,
"num_stages"
:
0
"num_warps"
:
2
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
2
},
},
"64"
:
{
"64"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_M"
:
32
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
8
,
"GROUP_SIZE_M"
:
4
,
"num_stages"
:
0
"num_warps"
:
8
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
2
},
},
"96"
:
{
"96"
:
{
"BLOCK_SIZE_M"
:
32
,
"BLOCK_SIZE_M"
:
32
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_K"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
16
,
"GROUP_SIZE_M"
:
4
,
"num_stages"
:
0
"num_warps"
:
4
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
2
},
},
"128"
:
{
"128"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
8
,
"GROUP_SIZE_M"
:
4
,
"num_stages"
:
0
"num_warps"
:
8
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
2
},
},
"256"
:
{
"256"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
8
,
"GROUP_SIZE_M"
:
4
,
"num_stages"
:
0
"num_warps"
:
8
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
1
},
},
"512"
:
{
"512"
:
{
"BLOCK_SIZE_M"
:
256
,
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
8
,
"GROUP_SIZE_M"
:
4
,
"num_stages"
:
0
"num_warps"
:
8
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
2
},
},
"1024"
:
{
"1024"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
,
"num_stages"
:
0
"num_warps"
:
8
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
32
,
"kpack"
:
2
},
},
"1536"
:
{
"1536"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
,
"num_stages"
:
0
"num_warps"
:
8
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
2
},
},
"2048"
:
{
"2048"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
256
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
,
"num_stages"
:
0
"num_warps"
:
8
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
2
},
},
"3072"
:
{
"3072"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
256
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
,
"num_stages"
:
0
"num_warps"
:
8
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
1
},
},
"4096"
:
{
"4096"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
,
"num_stages"
:
0
"num_warps"
:
8
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
1
}
}
}
}
vllm/model_executor/layers/fused_moe/configs/E=8,N=1792,device_name=AMD_Instinct_MI300X.json
100644 → 100755
View file @
c3dde367
{
{
"1"
:
{
"1"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_K"
:
128
,
"BLOCK_SIZE_K"
:
256
,
"GROUP_SIZE_M"
:
64
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
2
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
2
},
},
"2"
:
{
"2"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
32
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
32
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
1
},
},
"4"
:
{
"4"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
8
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
2
},
},
"8"
:
{
"8"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
6
4
,
"BLOCK_SIZE_N"
:
1
6
,
"BLOCK_SIZE_K"
:
256
,
"BLOCK_SIZE_K"
:
256
,
"GROUP_SIZE_M"
:
1
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
2
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
2
},
},
"16"
:
{
"16"
:
{
"BLOCK_SIZE_M"
:
1
6
,
"BLOCK_SIZE_M"
:
6
4
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
256
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
8
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
1
},
},
"24"
:
{
"24"
:
{
"BLOCK_SIZE_M"
:
32
,
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
1
},
"32"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
8
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
4
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
1
},
"32"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
256
,
"GROUP_SIZE_M"
:
4
,
"num_warps"
:
2
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
2
},
},
"48"
:
{
"48"
:
{
"BLOCK_SIZE_M"
:
1
28
,
"BLOCK_SIZE_M"
:
1
6
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
8
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
8
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
2
},
},
"64"
:
{
"64"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
2
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
1
},
},
"96"
:
{
"96"
:
{
"BLOCK_SIZE_M"
:
32
,
"BLOCK_SIZE_M"
:
32
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_K"
:
128
,
"BLOCK_SIZE_K"
:
256
,
"GROUP_SIZE_M"
:
8
"GROUP_SIZE_M"
:
4
,
"num_warps"
:
4
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
2
},
},
"128"
:
{
"128"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
32
"GROUP_SIZE_M"
:
4
,
"num_warps"
:
8
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
1
},
},
"256"
:
{
"256"
:
{
"BLOCK_SIZE_M"
:
32
,
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
"GROUP_SIZE_M"
:
4
,
"num_warps"
:
8
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
1
},
},
"512"
:
{
"512"
:
{
"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_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
2
},
},
"1024"
:
{
"1024"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
8
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
1
},
},
"1536"
:
{
"1536"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
8
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
1
},
},
"2048"
:
{
"2048"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
8
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
2
},
},
"3072"
:
{
"3072"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
8
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
1
},
},
"4096"
:
{
"4096"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
"GROUP_SIZE_M"
:
1
,
"num_warps"
:
8
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
1
}
}
}
}
vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=AMD_Instinct_MI300X.json
View file @
c3dde367
{
{
"1"
:
{
"1"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
8
,
"GROUP_SIZE_M"
:
1
,
"num_stages"
:
0
"num_warps"
:
2
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
1
},
},
"2"
:
{
"2"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
1
28
,
"BLOCK_SIZE_N"
:
1
6
,
"BLOCK_SIZE_K"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
,
"num_stages"
:
0
"num_warps"
:
2
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
2
},
},
"4"
:
{
"4"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_K"
:
32
,
"BLOCK_SIZE_K"
:
256
,
"GROUP_SIZE_M"
:
32
,
"GROUP_SIZE_M"
:
1
,
"num_stages"
:
1
"num_warps"
:
2
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
2
},
},
"8"
:
{
"8"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_K"
:
256
,
"BLOCK_SIZE_K"
:
256
,
"GROUP_SIZE_M"
:
8
,
"GROUP_SIZE_M"
:
1
,
"num_stages"
:
1
"num_warps"
:
2
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
2
},
},
"16"
:
{
"16"
:
{
"BLOCK_SIZE_M"
:
32
,
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_K"
:
128
,
"BLOCK_SIZE_K"
:
256
,
"GROUP_SIZE_M"
:
16
,
"GROUP_SIZE_M"
:
1
,
"num_stages"
:
1
"num_warps"
:
2
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
2
},
},
"24"
:
{
"24"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
256
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
8
,
"GROUP_SIZE_M"
:
1
,
"num_stages"
:
1
"num_warps"
:
4
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
1
},
},
"32"
:
{
"32"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
256
,
"BLOCK_SIZE_N"
:
16
,
"BLOCK_SIZE_K"
:
64
,
"BLOCK_SIZE_K"
:
256
,
"GROUP_SIZE_M"
:
16
,
"GROUP_SIZE_M"
:
4
,
"num_stages"
:
0
"num_warps"
:
2
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
2
},
},
"48"
:
{
"48"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_K"
:
256
,
"BLOCK_SIZE_K"
:
256
,
"GROUP_SIZE_M"
:
16
,
"GROUP_SIZE_M"
:
1
,
"num_stages"
:
1
"num_warps"
:
2
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
2
},
},
"64"
:
{
"64"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_M"
:
32
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_K"
:
64
,
"BLOCK_SIZE_K"
:
256
,
"GROUP_SIZE_M"
:
32
,
"GROUP_SIZE_M"
:
4
,
"num_stages"
:
0
"num_warps"
:
4
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
2
},
},
"96"
:
{
"96"
:
{
"BLOCK_SIZE_M"
:
32
,
"BLOCK_SIZE_M"
:
32
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_K"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
16
,
"GROUP_SIZE_M"
:
4
,
"num_stages"
:
0
"num_warps"
:
4
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
1
},
},
"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"
:
8
,
"GROUP_SIZE_M"
:
4
,
"num_stages"
:
0
"num_warps"
:
8
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
1
},
},
"256"
:
{
"256"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
8
,
"GROUP_SIZE_M"
:
4
,
"num_stages"
:
0
"num_warps"
:
8
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
1
},
},
"512"
:
{
"512"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
,
"num_stages"
:
0
"num_warps"
:
8
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
32
,
"kpack"
:
2
},
},
"1024"
:
{
"1024"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
,
"num_stages"
:
0
"num_warps"
:
8
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
1
},
},
"1536"
:
{
"1536"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
,
"num_stages"
:
0
"num_warps"
:
8
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
2
},
},
"2048"
:
{
"2048"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
,
"num_stages"
:
0
"num_warps"
:
8
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
1
},
},
"3072"
:
{
"3072"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
,
"num_stages"
:
0
"num_warps"
:
8
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
2
},
},
"4096"
:
{
"4096"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
,
"num_stages"
:
0
"num_warps"
:
8
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
1
}
}
}
}
vllm/model_executor/layers/fused_moe/configs/E=8,N=7168,device_name=AMD_Instinct_MI300X.json
View file @
c3dde367
{
{
"1"
:
{
"1"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
6
4
,
"BLOCK_SIZE_N"
:
1
6
,
"BLOCK_SIZE_K"
:
256
,
"BLOCK_SIZE_K"
:
256
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
,
"num_stages"
:
0
"num_warps"
:
2
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
2
},
},
"2"
:
{
"2"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
2
56
,
"BLOCK_SIZE_K"
:
3
2
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
,
"num_stages"
:
1
"num_warps"
:
4
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
1
},
},
"4"
:
{
"4"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_K"
:
256
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
32
,
"GROUP_SIZE_M"
:
1
,
"num_stages"
:
1
"num_warps"
:
4
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
1
},
},
"8"
:
{
"8"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_K"
:
256
,
"BLOCK_SIZE_K"
:
256
,
"GROUP_SIZE_M"
:
8
,
"GROUP_SIZE_M"
:
1
,
"num_stages"
:
1
"num_warps"
:
2
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
1
},
},
"16"
:
{
"16"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
6
4
,
"BLOCK_SIZE_N"
:
1
6
,
"BLOCK_SIZE_K"
:
256
,
"BLOCK_SIZE_K"
:
256
,
"GROUP_SIZE_M"
:
8
,
"GROUP_SIZE_M"
:
1
,
"num_stages"
:
1
"num_warps"
:
4
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
2
},
},
"24"
:
{
"24"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_M"
:
32
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_K"
:
256
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
8
,
"GROUP_SIZE_M"
:
1
,
"num_stages"
:
1
"num_warps"
:
8
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
1
},
},
"32"
:
{
"32"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_K"
:
256
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
16
,
"GROUP_SIZE_M"
:
4
,
"num_stages"
:
0
"num_warps"
:
2
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
2
},
},
"48"
:
{
"48"
:
{
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_M"
:
16
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_K"
:
256
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
16
,
"GROUP_SIZE_M"
:
4
,
"num_stages"
:
0
"num_warps"
:
2
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
1
},
},
"64"
:
{
"64"
:
{
"BLOCK_SIZE_M"
:
32
,
"BLOCK_SIZE_M"
:
32
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_K"
:
256
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
8
,
"GROUP_SIZE_M"
:
4
,
"num_stages"
:
1
"num_warps"
:
4
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
2
},
},
"96"
:
{
"96"
:
{
"BLOCK_SIZE_M"
:
32
,
"BLOCK_SIZE_M"
:
32
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_N"
:
32
,
"BLOCK_SIZE_K"
:
128
,
"BLOCK_SIZE_K"
:
128
,
"GROUP_SIZE_M"
:
8
,
"GROUP_SIZE_M"
:
4
,
"num_stages"
:
0
"num_warps"
:
4
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
2
},
},
"128"
:
{
"128"
:
{
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_M"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_K"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
8
,
"GROUP_SIZE_M"
:
4
,
"num_stages"
:
0
"num_warps"
:
8
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
1
},
},
"256"
:
{
"256"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
64
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
8
,
"GROUP_SIZE_M"
:
4
,
"num_stages"
:
0
"num_warps"
:
8
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
32
,
"kpack"
:
2
},
},
"512"
:
{
"512"
:
{
"BLOCK_SIZE_M"
:
256
,
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
8
,
"GROUP_SIZE_M"
:
1
,
"num_stages"
:
0
"num_warps"
:
8
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
1
},
},
"1024"
:
{
"1024"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
,
"num_stages"
:
0
"num_warps"
:
8
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
1
},
},
"1536"
:
{
"1536"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
,
"num_stages"
:
0
"num_warps"
:
8
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
2
},
},
"2048"
:
{
"2048"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
,
"num_stages"
:
0
"num_warps"
:
8
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
1
},
},
"3072"
:
{
"3072"
:
{
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
64
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
,
"num_stages"
:
0
"num_warps"
:
8
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
2
},
},
"4096"
:
{
"4096"
:
{
"BLOCK_SIZE_M"
:
256
,
"BLOCK_SIZE_M"
:
128
,
"BLOCK_SIZE_N"
:
256
,
"BLOCK_SIZE_N"
:
128
,
"BLOCK_SIZE_K"
:
32
,
"BLOCK_SIZE_K"
:
64
,
"GROUP_SIZE_M"
:
1
,
"GROUP_SIZE_M"
:
1
,
"num_stages"
:
0
"num_warps"
:
8
,
"num_stages"
:
0
,
"waves_per_eu"
:
0
,
"matrix_instr_nonkdim"
:
16
,
"kpack"
:
1
}
}
}
}
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