add step3-vl k100-ai config

375fdfb0 · zhuwenwen · f54639f9 · 375fdfb0
Commit 375fdfb0 authored Aug 05, 2025 by zhuwenwen
Show whitespace changes
Inline Side-by-side

Showing with 164 additions and 0 deletions

vllm/model_executor/layers/fused_moe/configs/E=48,N=320,device_name=K100_AI_nn.json .../fused_moe/configs/E=48,N=320,device_name=K100_AI_nn.json +164 -0

No files found.
--- a/vllm/model_executor/layers/fused_moe/configs/E=48,N=320,device_name=K100_AI_nn.json
+++ b/vllm/model_executor/layers/fused_moe/configs/E=48,N=320,device_name=K100_AI_nn.json
+{
+    "1": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 32,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 2,
+        "num_ldmatrixes": 1
+    },
+    "2": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 32,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 2,
+        "num_ldmatrixes": 1
+    },
+    "4": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 2,
+        "num_ldmatrixes": 1
+    },
+    "8": {
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 64,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 8,
+        "num_stages": 2,
+        "num_ldmatrixes": 1
+    },
+    "16": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 256,
+        "BLOCK_SIZE_K": 64,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 2,
+        "num_ldmatrixes": 1
+    },
+    "24": {
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 256,
+        "BLOCK_SIZE_K": 32,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 8,
+        "num_stages": 3,
+        "num_ldmatrixes": 1
+    },
+    "32": {
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 64,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 3,
+        "num_ldmatrixes": 1
+    },
+    "48": {
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 64,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 8,
+        "num_stages": 4,
+        "num_ldmatrixes": 1
+    },
+    "64": {
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 64,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 4,
+        "num_ldmatrixes": 1
+    },
+    "96": {
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 64,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 8,
+        "num_stages": 3,
+        "num_ldmatrixes": 1
+    },
+    "128": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 64,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 3,
+        "num_ldmatrixes": 1
+    },
+    "256": {
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 64,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 8,
+        "num_stages": 3,
+        "num_ldmatrixes": 1
+    },
+    "512": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 2,
+        "num_ldmatrixes": 1
+    },
+    "1024": {
+        "BLOCK_SIZE_M": 128,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 32,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 2,
+        "num_ldmatrixes": 1
+    },
+    "1536": {
+        "BLOCK_SIZE_M": 128,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 64,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 3,
+        "num_ldmatrixes": 1
+    },
+    "2048": {
+        "BLOCK_SIZE_M": 256,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 32,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 2,
+        "num_ldmatrixes": 1
+    },
+    "3072": {
+        "BLOCK_SIZE_M": 256,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 32,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 2,
+        "num_ldmatrixes": 1
+    },
+    "4096": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 8,
+        "num_stages": 2,
+        "num_ldmatrixes": 1
+    }
+}