Skip to content

GitLab

Commit f0ac5f23 authored by zhuwenwen's avatar zhuwenwen
Browse files

add deepseek-v2 tp1 fused_moe_config

parent 49ff8ab5
Hide whitespace changes
Inline Side-by-side
Showing with 113 additions and 77 deletions
vllm/model_executor/layers/fused_moe/configs/E=64,N=1408,device_name=BW200.json
View file @ f0ac5f23
{ {
"1": { "1": {
"BLOCK_SIZE_M": 32, "BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 32, "BLOCK_SIZE_N": 32,
"BLOCK_SIZE_K": 128, "BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1, "GROUP_SIZE_M": 1,
"num_warps": 4, "num_warps": 8,
"num_stages": 2 "num_stages": 2,
"waves_per_eu": 0
}, },
"2": { "2": {
"BLOCK_SIZE_M": 32, "BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 32, "BLOCK_SIZE_N": 64,
"BLOCK_SIZE_K": 128, "BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1, "GROUP_SIZE_M": 1,
"num_warps": 4, "num_warps": 4,
"num_stages": 2 "num_stages": 2,
"waves_per_eu": 0
}, },
"4": { "4": {
"BLOCK_SIZE_M": 16, "BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 64, "BLOCK_SIZE_N": 32,
"BLOCK_SIZE_K": 128, "BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1, "GROUP_SIZE_M": 1,
"num_warps": 4, "num_warps": 2,
"num_stages": 2 "num_stages": 2,
"waves_per_eu": 0
}, },
"8": { "8": {
"BLOCK_SIZE_M": 32, "BLOCK_SIZE_M": 32,
"BLOCK_SIZE_N": 32, "BLOCK_SIZE_N": 32,
"BLOCK_SIZE_K": 128, "BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1, "GROUP_SIZE_M": 1,
"num_warps": 4, "num_warps": 2,
"num_stages": 2 "num_stages": 2,
"waves_per_eu": 0
}, },
"16": { "16": {
"BLOCK_SIZE_M": 32, "BLOCK_SIZE_M": 32,
"BLOCK_SIZE_N": 64, "BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128, "BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1, "GROUP_SIZE_M": 1,
"num_warps": 8, "num_warps": 8,
"num_stages": 2 "num_stages": 2,
"waves_per_eu": 0
}, },
"24": { "24": {
"BLOCK_SIZE_M": 32, "BLOCK_SIZE_M": 32,
"BLOCK_SIZE_N": 64, "BLOCK_SIZE_N": 32,
"BLOCK_SIZE_K": 128, "BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1, "GROUP_SIZE_M": 1,
"num_warps": 8, "num_warps": 4,
"num_stages": 2 "num_stages": 2,
"waves_per_eu": 0
}, },
"32": { "32": {
"BLOCK_SIZE_M": 32, "BLOCK_SIZE_M": 32,
...@@ -53,47 +59,53 @@ ...@@ -53,47 +59,53 @@
"BLOCK_SIZE_K": 128, "BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1, "GROUP_SIZE_M": 1,
"num_warps": 4, "num_warps": 4,
"num_stages": 2 "num_stages": 2,
"waves_per_eu": 0
}, },
"48": { "48": {
"BLOCK_SIZE_M": 32, "BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 32, "BLOCK_SIZE_N": 32,
"BLOCK_SIZE_K": 128, "BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1, "GROUP_SIZE_M": 1,
"num_warps": 4, "num_warps": 4,
"num_stages": 2 "num_stages": 2,
"waves_per_eu": 0
}, },
"64": { "64": {
"BLOCK_SIZE_M": 16, "BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 32, "BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128, "BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1, "GROUP_SIZE_M": 4,
"num_warps": 4, "num_warps": 8,
"num_stages": 2 "num_stages": 2,
"waves_per_eu": 0
}, },
"96": { "96": {
"BLOCK_SIZE_M": 16, "BLOCK_SIZE_M": 32,
"BLOCK_SIZE_N": 32, "BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128, "BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1, "GROUP_SIZE_M": 4,
"num_warps": 4, "num_warps": 8,
"num_stages": 2 "num_stages": 2,
"waves_per_eu": 0
}, },
"128": { "128": {
"BLOCK_SIZE_M": 32, "BLOCK_SIZE_M": 32,
"BLOCK_SIZE_N": 32, "BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128, "BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1, "GROUP_SIZE_M": 4,
"num_warps": 4, "num_warps": 8,
"num_stages": 2 "num_stages": 2,
"waves_per_eu": 0
}, },
"256": { "256": {
"BLOCK_SIZE_M": 32, "BLOCK_SIZE_M": 32,
"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_warps": 4,
"num_stages": 2 "num_stages": 2,
"waves_per_eu": 0
}, },
"512": { "512": {
"BLOCK_SIZE_M": 64, "BLOCK_SIZE_M": 64,
...@@ -101,15 +113,17 @@ ...@@ -101,15 +113,17 @@
"BLOCK_SIZE_K": 128, "BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1, "GROUP_SIZE_M": 1,
"num_warps": 4, "num_warps": 4,
"num_stages": 2 "num_stages": 2,
"waves_per_eu": 0
}, },
"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_warps": 8, "num_warps": 4,
"num_stages": 2 "num_stages": 2,
"waves_per_eu": 0
}, },
"1536": { "1536": {
"BLOCK_SIZE_M": 128, "BLOCK_SIZE_M": 128,
...@@ -117,30 +131,34 @@ ...@@ -117,30 +131,34 @@
"BLOCK_SIZE_K": 64, "BLOCK_SIZE_K": 64,
"GROUP_SIZE_M": 1, "GROUP_SIZE_M": 1,
"num_warps": 4, "num_warps": 4,
"num_stages": 2 "num_stages": 2,
"waves_per_eu": 0
}, },
"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": 4,
"num_warps": 4, "num_warps": 4,
"num_stages": 2 "num_stages": 2,
"waves_per_eu": 0
}, },
"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": 8,
"num_warps": 4, "num_warps": 4,
"num_stages": 2 "num_stages": 2,
"waves_per_eu": 0
}, },
"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": 4,
"num_warps": 4, "num_warps": 4,
"num_stages": 2 "num_stages": 2,
"waves_per_eu": 0
} }
} }
vllm/model_executor/layers/fused_moe/configs/E=64,N=1408,device_name=BW200_nn.json
View file @ f0ac5f23
{ {
"1": { "1": {
"BLOCK_SIZE_M": 16, "BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 128, "BLOCK_SIZE_N": 64,
"BLOCK_SIZE_K": 64, "BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 16, "GROUP_SIZE_M": 1,
"num_warps": 4, "num_warps": 8,
"num_stages": 2, "num_stages": 2,
"num_ldmatrixes": 1 "num_ldmatrixes": 1,
"waves_per_eu": 0
}, },
"2": { "2": {
"BLOCK_SIZE_M": 16, "BLOCK_SIZE_M": 16,
"BLOCK_SIZE_N": 128, "BLOCK_SIZE_N": 256,
"BLOCK_SIZE_K": 128, "BLOCK_SIZE_K": 32,
"GROUP_SIZE_M": 1, "GROUP_SIZE_M": 1,
"num_warps": 4, "num_warps": 4,
"num_stages": 2, "num_stages": 2,
"num_ldmatrixes": 1 "num_ldmatrixes": 1,
"waves_per_eu": 0
}, },
"4": { "4": {
"BLOCK_SIZE_M": 16, "BLOCK_SIZE_M": 16,
...@@ -24,16 +26,18 @@ ...@@ -24,16 +26,18 @@
"GROUP_SIZE_M": 1, "GROUP_SIZE_M": 1,
"num_warps": 4, "num_warps": 4,
"num_stages": 2, "num_stages": 2,
"num_ldmatrixes": 1 "num_ldmatrixes": 1,
"waves_per_eu": 0
}, },
"8": { "8": {
"BLOCK_SIZE_M": 16, "BLOCK_SIZE_M": 32,
"BLOCK_SIZE_N": 128, "BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 128, "BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1, "GROUP_SIZE_M": 1,
"num_warps": 4, "num_warps": 8,
"num_stages": 2, "num_stages": 2,
"num_ldmatrixes": 1 "num_ldmatrixes": 1,
"waves_per_eu": 0
}, },
"16": { "16": {
"BLOCK_SIZE_M": 32, "BLOCK_SIZE_M": 32,
...@@ -42,7 +46,8 @@ ...@@ -42,7 +46,8 @@
"GROUP_SIZE_M": 1, "GROUP_SIZE_M": 1,
"num_warps": 8, "num_warps": 8,
"num_stages": 2, "num_stages": 2,
"num_ldmatrixes": 1 "num_ldmatrixes": 1,
"waves_per_eu": 0
}, },
"24": { "24": {
"BLOCK_SIZE_M": 16, "BLOCK_SIZE_M": 16,
...@@ -51,7 +56,8 @@ ...@@ -51,7 +56,8 @@
"GROUP_SIZE_M": 1, "GROUP_SIZE_M": 1,
"num_warps": 4, "num_warps": 4,
"num_stages": 2, "num_stages": 2,
"num_ldmatrixes": 1 "num_ldmatrixes": 1,
"waves_per_eu": 0
}, },
"32": { "32": {
"BLOCK_SIZE_M": 16, "BLOCK_SIZE_M": 16,
...@@ -60,7 +66,8 @@ ...@@ -60,7 +66,8 @@
"GROUP_SIZE_M": 1, "GROUP_SIZE_M": 1,
"num_warps": 4, "num_warps": 4,
"num_stages": 2, "num_stages": 2,
"num_ldmatrixes": 1 "num_ldmatrixes": 1,
"waves_per_eu": 0
}, },
"48": { "48": {
"BLOCK_SIZE_M": 16, "BLOCK_SIZE_M": 16,
...@@ -69,7 +76,8 @@ ...@@ -69,7 +76,8 @@
"GROUP_SIZE_M": 1, "GROUP_SIZE_M": 1,
"num_warps": 4, "num_warps": 4,
"num_stages": 2, "num_stages": 2,
"num_ldmatrixes": 1 "num_ldmatrixes": 1,
"waves_per_eu": 0
}, },
"64": { "64": {
"BLOCK_SIZE_M": 16, "BLOCK_SIZE_M": 16,
...@@ -78,7 +86,8 @@ ...@@ -78,7 +86,8 @@
"GROUP_SIZE_M": 1, "GROUP_SIZE_M": 1,
"num_warps": 4, "num_warps": 4,
"num_stages": 2, "num_stages": 2,
"num_ldmatrixes": 1 "num_ldmatrixes": 1,
"waves_per_eu": 0
}, },
"96": { "96": {
"BLOCK_SIZE_M": 16, "BLOCK_SIZE_M": 16,
...@@ -87,34 +96,38 @@ ...@@ -87,34 +96,38 @@
"GROUP_SIZE_M": 1, "GROUP_SIZE_M": 1,
"num_warps": 4, "num_warps": 4,
"num_stages": 2, "num_stages": 2,
"num_ldmatrixes": 1 "num_ldmatrixes": 1,
"waves_per_eu": 0
}, },
"128": { "128": {
"BLOCK_SIZE_M": 32, "BLOCK_SIZE_M": 32,
"BLOCK_SIZE_N": 256, "BLOCK_SIZE_N": 128,
"BLOCK_SIZE_K": 64, "BLOCK_SIZE_K": 128,
"GROUP_SIZE_M": 1, "GROUP_SIZE_M": 1,
"num_warps": 8, "num_warps": 8,
"num_stages": 2, "num_stages": 2,
"num_ldmatrixes": 1 "num_ldmatrixes": 1,
"waves_per_eu": 0
}, },
"256": { "256": {
"BLOCK_SIZE_M": 64, "BLOCK_SIZE_M": 64,
"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_warps": 4,
"num_stages": 2, "num_stages": 2,
"num_ldmatrixes": 1 "num_ldmatrixes": 1,
"waves_per_eu": 0
}, },
"512": { "512": {
"BLOCK_SIZE_M": 64, "BLOCK_SIZE_M": 64,
"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_warps": 4,
"num_stages": 2, "num_stages": 2,
"num_ldmatrixes": 1 "num_ldmatrixes": 1,
"waves_per_eu": 0
}, },
"1024": { "1024": {
"BLOCK_SIZE_M": 128, "BLOCK_SIZE_M": 128,
...@@ -123,7 +136,8 @@ ...@@ -123,7 +136,8 @@
"GROUP_SIZE_M": 1, "GROUP_SIZE_M": 1,
"num_warps": 4, "num_warps": 4,
"num_stages": 2, "num_stages": 2,
"num_ldmatrixes": 1 "num_ldmatrixes": 1,
"waves_per_eu": 0
}, },
"1536": { "1536": {
"BLOCK_SIZE_M": 128, "BLOCK_SIZE_M": 128,
...@@ -132,33 +146,37 @@ ...@@ -132,33 +146,37 @@
"GROUP_SIZE_M": 1, "GROUP_SIZE_M": 1,
"num_warps": 4, "num_warps": 4,
"num_stages": 2, "num_stages": 2,
"num_ldmatrixes": 1 "num_ldmatrixes": 1,
"waves_per_eu": 0
}, },
"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": 16, "GROUP_SIZE_M": 4,
"num_warps": 4, "num_warps": 4,
"num_stages": 2, "num_stages": 2,
"num_ldmatrixes": 1 "num_ldmatrixes": 1,
"waves_per_eu": 0
}, },
"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": 16, "GROUP_SIZE_M": 4,
"num_warps": 4, "num_warps": 4,
"num_stages": 2, "num_stages": 2,
"num_ldmatrixes": 1 "num_ldmatrixes": 1,
"waves_per_eu": 0
}, },
"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": 4,
"num_warps": 4, "num_warps": 4,
"num_stages": 2, "num_stages": 2,
"num_ldmatrixes": 1 "num_ldmatrixes": 1,
"waves_per_eu": 0
} }
} }
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