From 0f52fb55ecd992f856892dc9fdec281e467e1ca9 Mon Sep 17 00:00:00 2001 From: Mick Date: Mon, 17 Mar 2025 14:51:58 +0800 Subject: [PATCH] config: Update fused moe config (#4493) --- ...dtype=fp8_w8a8,block_shape=[128, 128].json | 146 ++++++++++++++++++ ...dtype=fp8_w8a8,block_shape=[128, 128].json | 22 +-- 2 files changed, 157 insertions(+), 11 deletions(-) create mode 100644 python/sglang/srt/layers/moe/fused_moe_triton/configs/E=64,N=1024,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/E=64,N=1024,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/fused_moe_triton/configs/E=64,N=1024,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json new file mode 100644 index 000000000..e9c5b9251 --- /dev/null +++ b/python/sglang/srt/layers/moe/fused_moe_triton/configs/E=64,N=1024,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json @@ -0,0 +1,146 @@ +{ + "1": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 3 + }, + "2": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 4 + }, + "4": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 3 + }, + "8": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 3 + }, + "16": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 64, + "num_warps": 4, + "num_stages": 3 + }, + "24": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 3 + }, + "32": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 3 + }, + "48": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 3 + }, + "64": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 3 + }, + "96": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 64, + "num_warps": 4, + "num_stages": 3 + }, + "128": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 64, + "num_warps": 4, + "num_stages": 3 + }, + "256": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 3 + }, + "512": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 3 + }, + "1024": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 3 + }, + "1536": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 4 + }, + "2048": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 3 + }, + "3072": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 3 + }, + "4096": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 3 + } +} diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/E=64,N=512,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json b/python/sglang/srt/layers/moe/fused_moe_triton/configs/E=64,N=512,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json index 566707e78..3483f05a2 100644 --- a/python/sglang/srt/layers/moe/fused_moe_triton/configs/E=64,N=512,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json +++ b/python/sglang/srt/layers/moe/fused_moe_triton/configs/E=64,N=512,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json @@ -1,11 +1,11 @@ { "1": { "BLOCK_SIZE_M": 64, - "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_N": 32, "BLOCK_SIZE_K": 128, - "GROUP_SIZE_M": 1, + "GROUP_SIZE_M": 16, "num_warps": 4, - "num_stages": 5 + "num_stages": 3 }, "2": { "BLOCK_SIZE_M": 64, @@ -35,7 +35,7 @@ "BLOCK_SIZE_M": 64, "BLOCK_SIZE_N": 64, "BLOCK_SIZE_K": 128, - "GROUP_SIZE_M": 16, + "GROUP_SIZE_M": 32, "num_warps": 4, "num_stages": 3 }, @@ -43,7 +43,7 @@ "BLOCK_SIZE_M": 64, "BLOCK_SIZE_N": 64, "BLOCK_SIZE_K": 128, - "GROUP_SIZE_M": 64, + "GROUP_SIZE_M": 32, "num_warps": 4, "num_stages": 3 }, @@ -51,7 +51,7 @@ "BLOCK_SIZE_M": 64, "BLOCK_SIZE_N": 128, "BLOCK_SIZE_K": 128, - "GROUP_SIZE_M": 32, + "GROUP_SIZE_M": 64, "num_warps": 4, "num_stages": 3 }, @@ -59,7 +59,7 @@ "BLOCK_SIZE_M": 64, "BLOCK_SIZE_N": 128, "BLOCK_SIZE_K": 128, - "GROUP_SIZE_M": 32, + "GROUP_SIZE_M": 64, "num_warps": 4, "num_stages": 3 }, @@ -67,7 +67,7 @@ "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 }, @@ -91,7 +91,7 @@ "BLOCK_SIZE_M": 64, "BLOCK_SIZE_N": 128, "BLOCK_SIZE_K": 128, - "GROUP_SIZE_M": 32, + "GROUP_SIZE_M": 64, "num_warps": 4, "num_stages": 3 }, @@ -99,7 +99,7 @@ "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": 3 }, @@ -139,7 +139,7 @@ "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": 3 }