Skip to content

GitLab

Unverified Commit fe69f331 authored by Robert Shaw's avatar Robert Shaw Committed by GitHub
Browse files

[Kernels] Improve H200 Fused MoE Config (#28992) 


Signed-off-by: default avatarRobert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com>
parent 3319a493
Hide whitespace changes
Inline Side-by-side
Showing with 36 additions and 36 deletions
vllm/model_executor/layers/fused_moe/configs/E=256,N=256,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128,128].json
View file @ fe69f331
{
"1": {
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1,
"GROUP_SIZE_M": 16,
"num_warps": 4,
"num_stages": 5
"num_stages": 4
},
"2": {
"BLOCK_SIZE_M": 16,
......@@ -13,82 +13,82 @@
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 16,
"num_warps": 4,
"num_stages": 3
"num_stages": 4
},
"4": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 64,
"num_warps": 4,
"num_stages": 4
},
"8": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 64,
"GROUP_SIZE_M": 16,
"num_warps": 4,
"num_stages": 3
"num_stages": 4
},
"16": {
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 256,
"BLOCK_SIZE_K": 64,
"GROUP_SIZE_M": 32,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 16,
"num_warps": 4,
"num_stages": 3
},
"24": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 16,
"num_warps": 4,
"num_stages": 3
"num_stages": 4
},
"32": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1,
"GROUP_SIZE_M": 16,
"num_warps": 4,
"num_stages": 3
"num_stages": 5
},
"48": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 32,
"GROUP_SIZE_M": 64,
"num_warps": 4,
"num_stages": 3
},
"64": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 32,
"GROUP_SIZE_M": 64,
"num_warps": 4,
"num_stages": 3
},
"96": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 32,
"GROUP_SIZE_M": 16,
"num_warps": 4,
"num_stages": 3
},
"128": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 16,
"GROUP_SIZE_M": 32,
"num_warps": 4,
"num_stages": 3
},
"256": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 32,
......@@ -96,10 +96,10 @@
"num_stages": 3
},
"512": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 32,
"GROUP_SIZE_M": 16,
"num_warps": 4,
"num_stages": 3
},
......@@ -109,7 +109,7 @@
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 32,
"num_warps": 4,
"num_stages": 3
"num_stages": 4
},
"1536": {
"BLOCK_SIZE_M": 64,
......@@ -117,21 +117,21 @@
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 32,
"num_warps": 4,
"num_stages": 3
"num_stages": 4
},
"2048": {
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 64,
"GROUP_SIZE_M": 32,
"num_warps": 4,
"num_stages": 3
"num_stages": 4
},
"3072": {
"BLOCK_SIZE_M": 128,
"BLOCK_SIZE_N": 64,
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 16,
"GROUP_SIZE_M": 32,
"num_warps": 4,
"num_stages": 4
},
......@@ -139,7 +139,7 @@
"BLOCK_SIZE_M": 64,
"BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 64,
"GROUP_SIZE_M": 16,
"num_warps": 4,
"num_stages": 3
}
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment