From 6cf0b004ca7813cb48db6442438740e1e3dff9b4 Mon Sep 17 00:00:00 2001 From: Piotr Mazurek Date: Wed, 22 Apr 2026 03:32:05 +0200 Subject: [PATCH] [MoE] Add LFM2 MoE tuning support + tuned configs for H100/B200/MI325X (#22791) Co-authored-by: Piotr Mazurek --- .../kernels/fused_moe_triton/common_utils.py | 4 + .../E=32,N=1792,device_name=NVIDIA_B200.json | 154 ++++++++++++++++ ...792,device_name=NVIDIA_H100_80GB_HBM3.json | 154 ++++++++++++++++ .../E=32,N=224,device_name=NVIDIA_B200.json | 154 ++++++++++++++++ ...224,device_name=NVIDIA_H100_80GB_HBM3.json | 154 ++++++++++++++++ .../E=32,N=448,device_name=NVIDIA_B200.json | 154 ++++++++++++++++ ...448,device_name=NVIDIA_H100_80GB_HBM3.json | 154 ++++++++++++++++ .../E=32,N=896,device_name=NVIDIA_B200.json | 154 ++++++++++++++++ ...896,device_name=NVIDIA_H100_80GB_HBM3.json | 154 ++++++++++++++++ .../E=64,N=1536,device_name=NVIDIA_B200.json | 154 ++++++++++++++++ ...536,device_name=NVIDIA_H100_80GB_HBM3.json | 154 ++++++++++++++++ .../E=64,N=192,device_name=NVIDIA_B200.json | 154 ++++++++++++++++ ...192,device_name=NVIDIA_H100_80GB_HBM3.json | 154 ++++++++++++++++ .../E=64,N=384,device_name=NVIDIA_B200.json | 154 ++++++++++++++++ ...384,device_name=NVIDIA_H100_80GB_HBM3.json | 154 ++++++++++++++++ .../E=64,N=768,device_name=NVIDIA_B200.json | 154 ++++++++++++++++ ...768,device_name=NVIDIA_H100_80GB_HBM3.json | 154 ++++++++++++++++ ...=1792,device_name=AMD_Instinct_MI325X.json | 173 ++++++++++++++++++ ...N=224,device_name=AMD_Instinct_MI325X.json | 173 ++++++++++++++++++ ...N=448,device_name=AMD_Instinct_MI325X.json | 173 ++++++++++++++++++ ...N=896,device_name=AMD_Instinct_MI325X.json | 173 ++++++++++++++++++ ...=1536,device_name=AMD_Instinct_MI325X.json | 173 ++++++++++++++++++ ...N=192,device_name=AMD_Instinct_MI325X.json | 173 ++++++++++++++++++ ...N=384,device_name=AMD_Instinct_MI325X.json | 173 ++++++++++++++++++ ...N=768,device_name=AMD_Instinct_MI325X.json | 173 ++++++++++++++++++ 25 files changed, 3852 insertions(+) create mode 100644 python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=1792,device_name=NVIDIA_B200.json create mode 100755 python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=1792,device_name=NVIDIA_H100_80GB_HBM3.json create mode 100644 python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=224,device_name=NVIDIA_B200.json create mode 100755 python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=224,device_name=NVIDIA_H100_80GB_HBM3.json create mode 100644 python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=448,device_name=NVIDIA_B200.json create mode 100755 python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=448,device_name=NVIDIA_H100_80GB_HBM3.json create mode 100644 python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=896,device_name=NVIDIA_B200.json create mode 100755 python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=896,device_name=NVIDIA_H100_80GB_HBM3.json create mode 100644 python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=1536,device_name=NVIDIA_B200.json create mode 100755 python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=1536,device_name=NVIDIA_H100_80GB_HBM3.json create mode 100644 python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=192,device_name=NVIDIA_B200.json create mode 100755 python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=192,device_name=NVIDIA_H100_80GB_HBM3.json create mode 100644 python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=384,device_name=NVIDIA_B200.json create mode 100755 python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=384,device_name=NVIDIA_H100_80GB_HBM3.json create mode 100644 python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=768,device_name=NVIDIA_B200.json create mode 100755 python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=768,device_name=NVIDIA_H100_80GB_HBM3.json create mode 100644 python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_6_0/E=32,N=1792,device_name=AMD_Instinct_MI325X.json create mode 100644 python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_6_0/E=32,N=224,device_name=AMD_Instinct_MI325X.json create mode 100644 python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_6_0/E=32,N=448,device_name=AMD_Instinct_MI325X.json create mode 100644 python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_6_0/E=32,N=896,device_name=AMD_Instinct_MI325X.json create mode 100644 python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_6_0/E=64,N=1536,device_name=AMD_Instinct_MI325X.json create mode 100644 python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_6_0/E=64,N=192,device_name=AMD_Instinct_MI325X.json create mode 100644 python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_6_0/E=64,N=384,device_name=AMD_Instinct_MI325X.json create mode 100644 python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_6_0/E=64,N=768,device_name=AMD_Instinct_MI325X.json diff --git a/benchmark/kernels/fused_moe_triton/common_utils.py b/benchmark/kernels/fused_moe_triton/common_utils.py index 4f4d09b68..ca251cbde 100644 --- a/benchmark/kernels/fused_moe_triton/common_utils.py +++ b/benchmark/kernels/fused_moe_triton/common_utils.py @@ -138,6 +138,10 @@ def get_model_config( E = config.num_experts // ep_size topk = config.top_k_experts intermediate_size = config.moe_intermediate_size + elif architecture == "Lfm2MoeForCausalLM": + E = config.num_experts // ep_size + topk = config.num_experts_per_tok + intermediate_size = config.moe_intermediate_size else: # Default: Mixtral E = config.num_local_experts // ep_size diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=1792,device_name=NVIDIA_B200.json b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=1792,device_name=NVIDIA_B200.json new file mode 100644 index 000000000..332cfa865 --- /dev/null +++ b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=1792,device_name=NVIDIA_B200.json @@ -0,0 +1,154 @@ +{ + "1": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 5 + }, + "2": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 5 + }, + "4": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 4 + }, + "8": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 3 + }, + "16": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 64, + "num_warps": 8, + "num_stages": 4 + }, + "24": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 4 + }, + "32": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 2 + }, + "48": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 2 + }, + "64": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 2 + }, + "96": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 2 + }, + "128": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 2 + }, + "256": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 8, + "num_stages": 4 + }, + "512": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 64, + "num_warps": 4, + "num_stages": 3 + }, + "1024": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 8, + "num_stages": 4 + }, + "1536": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + }, + "2048": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + }, + "3072": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + }, + "4096": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + }, + "8192": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + } +} diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=1792,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=1792,device_name=NVIDIA_H100_80GB_HBM3.json new file mode 100755 index 000000000..f77a16300 --- /dev/null +++ b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=1792,device_name=NVIDIA_H100_80GB_HBM3.json @@ -0,0 +1,154 @@ +{ + "1": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 5 + }, + "2": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 3 + }, + "4": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 3 + }, + "8": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 256, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 3 + }, + "16": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 4 + }, + "24": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 64, + "num_warps": 4, + "num_stages": 2 + }, + "32": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 5 + }, + "48": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2 + }, + "64": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2 + }, + "96": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 2 + }, + "128": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 3 + }, + "256": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 3 + }, + "512": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + }, + "1024": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 4 + }, + "1536": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 4 + }, + "2048": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 4 + }, + "3072": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 4 + }, + "4096": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 4 + }, + "8192": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 8, + "num_stages": 4 + } +} diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=224,device_name=NVIDIA_B200.json b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=224,device_name=NVIDIA_B200.json new file mode 100644 index 000000000..e5bcc0e57 --- /dev/null +++ b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=224,device_name=NVIDIA_B200.json @@ -0,0 +1,154 @@ +{ + "1": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 5 + }, + "2": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 64, + "num_warps": 4, + "num_stages": 5 + }, + "4": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 4 + }, + "8": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 4 + }, + "16": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 64, + "num_warps": 4, + "num_stages": 4 + }, + "24": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 4 + }, + "32": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 4 + }, + "48": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 4 + }, + "64": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 4 + }, + "96": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 4 + }, + "128": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 4 + }, + "256": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 4 + }, + "512": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 4 + }, + "1024": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 64, + "num_warps": 8, + "num_stages": 4 + }, + "1536": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 64, + "num_warps": 8, + "num_stages": 4 + }, + "2048": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 64, + "num_warps": 8, + "num_stages": 3 + }, + "3072": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 8, + "num_stages": 3 + }, + "4096": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2 + }, + "8192": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + } +} diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=224,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=224,device_name=NVIDIA_H100_80GB_HBM3.json new file mode 100755 index 000000000..1661e7784 --- /dev/null +++ b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=224,device_name=NVIDIA_H100_80GB_HBM3.json @@ -0,0 +1,154 @@ +{ + "1": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 5 + }, + "2": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 64, + "num_warps": 4, + "num_stages": 3 + }, + "4": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 4 + }, + "8": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 64, + "num_warps": 4, + "num_stages": 4 + }, + "16": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 5 + }, + "24": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 3 + }, + "32": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 3 + }, + "48": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 3 + }, + "64": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 64, + "num_warps": 4, + "num_stages": 3 + }, + "96": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 3 + }, + "128": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 5 + }, + "256": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 4 + }, + "512": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 8, + "num_stages": 4 + }, + "1024": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 4 + }, + "1536": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 4 + }, + "2048": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 8, + "num_stages": 4 + }, + "3072": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 8, + "num_stages": 4 + }, + "4096": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 4 + }, + "8192": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 8, + "num_stages": 4 + } +} diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=448,device_name=NVIDIA_B200.json b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=448,device_name=NVIDIA_B200.json new file mode 100644 index 000000000..eee2b5714 --- /dev/null +++ b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=448,device_name=NVIDIA_B200.json @@ -0,0 +1,154 @@ +{ + "1": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 5 + }, + "2": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 5 + }, + "4": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 64, + "num_warps": 4, + "num_stages": 5 + }, + "8": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 5 + }, + "16": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 4 + }, + "24": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 4 + }, + "32": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 4 + }, + "48": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 4 + }, + "64": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 4 + }, + "96": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 4 + }, + "128": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 4 + }, + "256": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 8, + "num_stages": 4 + }, + "512": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 3 + }, + "1024": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 8, + "num_stages": 3 + }, + "1536": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + }, + "2048": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + }, + "3072": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 8, + "num_stages": 3 + }, + "4096": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 4 + }, + "8192": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + } +} diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=448,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=448,device_name=NVIDIA_H100_80GB_HBM3.json new file mode 100755 index 000000000..f69b70704 --- /dev/null +++ b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=448,device_name=NVIDIA_H100_80GB_HBM3.json @@ -0,0 +1,154 @@ +{ + "1": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 5 + }, + "2": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 4 + }, + "4": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 5 + }, + "8": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 4 + }, + "16": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + }, + "24": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 3 + }, + "32": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 3 + }, + "48": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 3 + }, + "64": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 8, + "num_stages": 3 + }, + "96": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 3 + }, + "128": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 64, + "num_warps": 4, + "num_stages": 3 + }, + "256": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 3 + }, + "512": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 8, + "num_stages": 3 + }, + "1024": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 4 + }, + "1536": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 4 + }, + "2048": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 8, + "num_stages": 4 + }, + "3072": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + }, + "4096": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + }, + "8192": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + } +} diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=896,device_name=NVIDIA_B200.json b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=896,device_name=NVIDIA_B200.json new file mode 100644 index 000000000..db9010449 --- /dev/null +++ b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=896,device_name=NVIDIA_B200.json @@ -0,0 +1,154 @@ +{ + "1": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 5 + }, + "2": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 5 + }, + "4": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 4 + }, + "8": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 4 + }, + "16": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 3 + }, + "24": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 3 + }, + "32": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 3 + }, + "48": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 2 + }, + "64": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2 + }, + "96": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2 + }, + "128": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 3 + }, + "256": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 4 + }, + "512": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 4 + }, + "1024": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 8, + "num_stages": 4 + }, + "1536": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 8, + "num_stages": 3 + }, + "2048": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 4 + }, + "3072": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 4 + }, + "4096": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + }, + "8192": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + } +} diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=896,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=896,device_name=NVIDIA_H100_80GB_HBM3.json new file mode 100755 index 000000000..6da1eb2c6 --- /dev/null +++ b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=32,N=896,device_name=NVIDIA_H100_80GB_HBM3.json @@ -0,0 +1,154 @@ +{ + "1": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 4 + }, + "2": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 3 + }, + "4": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 64, + "num_warps": 4, + "num_stages": 3 + }, + "8": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 2 + }, + "16": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 8, + "num_stages": 2 + }, + "24": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 2 + }, + "32": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 64, + "num_warps": 4, + "num_stages": 3 + }, + "48": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 64, + "num_warps": 4, + "num_stages": 5 + }, + "64": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 3 + }, + "96": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 3 + }, + "128": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 3 + }, + "256": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 64, + "num_warps": 4, + "num_stages": 3 + }, + "512": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + }, + "1024": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 4 + }, + "1536": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 8, + "num_stages": 4 + }, + "2048": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 4 + }, + "3072": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 4 + }, + "4096": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 4 + }, + "8192": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 4 + } +} diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=1536,device_name=NVIDIA_B200.json b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=1536,device_name=NVIDIA_B200.json new file mode 100644 index 000000000..d047ab956 --- /dev/null +++ b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=1536,device_name=NVIDIA_B200.json @@ -0,0 +1,154 @@ +{ + "1": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 5 + }, + "2": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 5 + }, + "4": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 8, + "num_stages": 4 + }, + "8": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2 + }, + "16": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 3 + }, + "24": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + }, + "32": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 64, + "num_warps": 4, + "num_stages": 4 + }, + "48": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 3 + }, + "64": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 2 + }, + "96": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 64, + "num_warps": 4, + "num_stages": 2 + }, + "128": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 2 + }, + "256": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 2 + }, + "512": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + }, + "1024": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 4 + }, + "1536": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 4 + }, + "2048": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 4 + }, + "3072": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + }, + "4096": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + }, + "8192": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 8, + "num_stages": 3 + } +} diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=1536,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=1536,device_name=NVIDIA_H100_80GB_HBM3.json new file mode 100755 index 000000000..d1a664e1d --- /dev/null +++ b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=1536,device_name=NVIDIA_H100_80GB_HBM3.json @@ -0,0 +1,154 @@ +{ + "1": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 4 + }, + "2": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 5 + }, + "4": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 64, + "num_warps": 4, + "num_stages": 2 + }, + "8": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 256, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 3 + }, + "16": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 5 + }, + "24": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 256, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 3 + }, + "32": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 256, + "GROUP_SIZE_M": 64, + "num_warps": 4, + "num_stages": 2 + }, + "48": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 5 + }, + "64": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 2 + }, + "96": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 2 + }, + "128": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 2 + }, + "256": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2 + }, + "512": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 3 + }, + "1024": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + }, + "1536": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + }, + "2048": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 4 + }, + "3072": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 4 + }, + "4096": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 4 + }, + "8192": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 4 + } +} diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=192,device_name=NVIDIA_B200.json b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=192,device_name=NVIDIA_B200.json new file mode 100644 index 000000000..cf9f9b4c8 --- /dev/null +++ b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=192,device_name=NVIDIA_B200.json @@ -0,0 +1,154 @@ +{ + "1": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 256, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 3 + }, + "2": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 64, + "num_warps": 4, + "num_stages": 5 + }, + "4": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 5 + }, + "8": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 5 + }, + "16": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 5 + }, + "24": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 64, + "num_warps": 4, + "num_stages": 5 + }, + "32": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 5 + }, + "48": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 4 + }, + "64": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 4 + }, + "96": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 64, + "num_warps": 4, + "num_stages": 5 + }, + "128": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 5 + }, + "256": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 4 + }, + "512": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 8, + "num_stages": 3 + }, + "1024": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + }, + "1536": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 8, + "num_stages": 4 + }, + "2048": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + }, + "3072": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 8, + "num_stages": 3 + }, + "4096": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + }, + "8192": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 8, + "num_stages": 3 + } +} diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=192,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=192,device_name=NVIDIA_H100_80GB_HBM3.json new file mode 100755 index 000000000..c6527a560 --- /dev/null +++ b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=192,device_name=NVIDIA_H100_80GB_HBM3.json @@ -0,0 +1,154 @@ +{ + "1": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 5 + }, + "2": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 5 + }, + "4": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 5 + }, + "8": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 5 + }, + "16": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 4 + }, + "24": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 64, + "num_warps": 4, + "num_stages": 4 + }, + "32": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 4 + }, + "48": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 4 + }, + "64": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 64, + "num_warps": 4, + "num_stages": 5 + }, + "96": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 4 + }, + "128": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 3 + }, + "256": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 3 + }, + "512": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 3 + }, + "1024": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 8, + "num_stages": 3 + }, + "1536": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + }, + "2048": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 2 + }, + "3072": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 8, + "num_stages": 4 + }, + "4096": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 8, + "num_stages": 3 + }, + "8192": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 8, + "num_stages": 3 + } +} diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=384,device_name=NVIDIA_B200.json b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=384,device_name=NVIDIA_B200.json new file mode 100644 index 000000000..e90c622e9 --- /dev/null +++ b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=384,device_name=NVIDIA_B200.json @@ -0,0 +1,154 @@ +{ + "1": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 5 + }, + "2": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 5 + }, + "4": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 5 + }, + "8": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 4 + }, + "16": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 4 + }, + "24": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 4 + }, + "32": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 3 + }, + "48": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 3 + }, + "64": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 3 + }, + "96": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 3 + }, + "128": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 3 + }, + "256": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 3 + }, + "512": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 64, + "num_warps": 8, + "num_stages": 3 + }, + "1024": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 8, + "num_stages": 4 + }, + "1536": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 8, + "num_stages": 3 + }, + "2048": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 8, + "num_stages": 4 + }, + "3072": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 8, + "num_stages": 3 + }, + "4096": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 8, + "num_stages": 3 + }, + "8192": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 8, + "num_stages": 3 + } +} diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=384,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=384,device_name=NVIDIA_H100_80GB_HBM3.json new file mode 100755 index 000000000..f55b5d695 --- /dev/null +++ b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=384,device_name=NVIDIA_H100_80GB_HBM3.json @@ -0,0 +1,154 @@ +{ + "1": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 4 + }, + "2": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 5 + }, + "4": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 5 + }, + "8": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 64, + "num_warps": 4, + "num_stages": 4 + }, + "16": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + }, + "24": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 64, + "num_warps": 4, + "num_stages": 2 + }, + "32": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2 + }, + "48": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 64, + "num_warps": 4, + "num_stages": 3 + }, + "64": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 3 + }, + "96": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + }, + "128": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 64, + "num_warps": 4, + "num_stages": 3 + }, + "256": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 2 + }, + "512": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 8, + "num_stages": 3 + }, + "1024": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 3 + }, + "1536": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + }, + "2048": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + }, + "3072": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 4 + }, + "4096": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 8, + "num_stages": 4 + }, + "8192": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 4 + } +} diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=768,device_name=NVIDIA_B200.json b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=768,device_name=NVIDIA_B200.json new file mode 100644 index 000000000..a6bafd3cf --- /dev/null +++ b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=768,device_name=NVIDIA_B200.json @@ -0,0 +1,154 @@ +{ + "1": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 5 + }, + "2": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 64, + "num_warps": 4, + "num_stages": 5 + }, + "4": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 5 + }, + "8": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 3 + }, + "16": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 3 + }, + "24": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 3 + }, + "32": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 3 + }, + "48": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 64, + "num_warps": 8, + "num_stages": 3 + }, + "64": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 64, + "num_warps": 8, + "num_stages": 3 + }, + "96": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + }, + "128": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + }, + "256": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2 + }, + "512": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + }, + "1024": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + }, + "1536": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 8, + "num_stages": 4 + }, + "2048": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 8, + "num_stages": 4 + }, + "3072": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + }, + "4096": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + }, + "8192": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + } +} diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=768,device_name=NVIDIA_H100_80GB_HBM3.json b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=768,device_name=NVIDIA_H100_80GB_HBM3.json new file mode 100755 index 000000000..b21f14c6c --- /dev/null +++ b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_5_1/E=64,N=768,device_name=NVIDIA_H100_80GB_HBM3.json @@ -0,0 +1,154 @@ +{ + "1": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 64, + "num_warps": 4, + "num_stages": 4 + }, + "2": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 5 + }, + "4": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 3 + }, + "8": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 64, + "num_warps": 4, + "num_stages": 2 + }, + "16": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 3 + }, + "24": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 64, + "num_warps": 4, + "num_stages": 5 + }, + "32": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 16, + "num_warps": 8, + "num_stages": 3 + }, + "48": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 64, + "num_warps": 4, + "num_stages": 5 + }, + "64": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 64, + "num_warps": 4, + "num_stages": 2 + }, + "96": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 2 + }, + "128": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 2 + }, + "256": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2 + }, + "512": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 3 + }, + "1024": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + }, + "1536": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 3 + }, + "2048": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 4 + }, + "3072": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 4 + }, + "4096": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 4 + }, + "8192": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 4 + } +} diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_6_0/E=32,N=1792,device_name=AMD_Instinct_MI325X.json b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_6_0/E=32,N=1792,device_name=AMD_Instinct_MI325X.json new file mode 100644 index 000000000..f7639b0c5 --- /dev/null +++ b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_6_0/E=32,N=1792,device_name=AMD_Instinct_MI325X.json @@ -0,0 +1,173 @@ +{ + "1": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 4, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "2": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "4": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "8": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 256, + "GROUP_SIZE_M": 4, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "16": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "24": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "32": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "48": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "64": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "96": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "128": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "256": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "512": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "1024": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "1536": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "2048": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "3072": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "4096": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 4, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "8192": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 4, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + } +} diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_6_0/E=32,N=224,device_name=AMD_Instinct_MI325X.json b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_6_0/E=32,N=224,device_name=AMD_Instinct_MI325X.json new file mode 100644 index 000000000..117782fb8 --- /dev/null +++ b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_6_0/E=32,N=224,device_name=AMD_Instinct_MI325X.json @@ -0,0 +1,173 @@ +{ + "1": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 256, + "GROUP_SIZE_M": 32, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "2": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 256, + "GROUP_SIZE_M": 8, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "4": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 4, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "8": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 8, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "16": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 4, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "24": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "32": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 8, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "48": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 16, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "64": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "96": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "128": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "256": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "512": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 4, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "1024": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 4, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "1536": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "2048": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "3072": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 8, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "4096": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 8, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "8192": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 4, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + } +} diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_6_0/E=32,N=448,device_name=AMD_Instinct_MI325X.json b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_6_0/E=32,N=448,device_name=AMD_Instinct_MI325X.json new file mode 100644 index 000000000..08441dee7 --- /dev/null +++ b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_6_0/E=32,N=448,device_name=AMD_Instinct_MI325X.json @@ -0,0 +1,173 @@ +{ + "1": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 256, + "GROUP_SIZE_M": 16, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "2": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 8, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "4": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 4, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "8": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "16": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "24": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "32": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "48": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 8, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "64": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 8, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "96": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "128": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 8, + "num_warps": 1, + "num_stages": 2, + "waves_per_eu": 0 + }, + "256": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "512": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "1024": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 4, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "1536": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 4, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "2048": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 4, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "3072": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "4096": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "8192": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 8, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + } +} diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_6_0/E=32,N=896,device_name=AMD_Instinct_MI325X.json b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_6_0/E=32,N=896,device_name=AMD_Instinct_MI325X.json new file mode 100644 index 000000000..ed8e457f6 --- /dev/null +++ b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_6_0/E=32,N=896,device_name=AMD_Instinct_MI325X.json @@ -0,0 +1,173 @@ +{ + "1": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 256, + "GROUP_SIZE_M": 8, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "2": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "4": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "8": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "16": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "24": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "32": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 8, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "48": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "64": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 8, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "96": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 256, + "GROUP_SIZE_M": 4, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "128": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 8, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "256": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "512": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "1024": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 8, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "1536": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 4, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "2048": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 4, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "3072": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 4, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "4096": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 4, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "8192": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 8, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + } +} diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_6_0/E=64,N=1536,device_name=AMD_Instinct_MI325X.json b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_6_0/E=64,N=1536,device_name=AMD_Instinct_MI325X.json new file mode 100644 index 000000000..2c1e90e0f --- /dev/null +++ b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_6_0/E=64,N=1536,device_name=AMD_Instinct_MI325X.json @@ -0,0 +1,173 @@ +{ + "1": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "2": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "4": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 1, + "num_stages": 2, + "waves_per_eu": 0 + }, + "8": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 1, + "num_stages": 2, + "waves_per_eu": 0 + }, + "16": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 256, + "GROUP_SIZE_M": 4, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "24": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 256, + "GROUP_SIZE_M": 4, + "num_warps": 1, + "num_stages": 2, + "waves_per_eu": 0 + }, + "32": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 256, + "GROUP_SIZE_M": 1, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "48": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "64": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "96": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "128": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "256": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 32, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "512": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 256, + "GROUP_SIZE_M": 32, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "1024": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "1536": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "2048": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "3072": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "4096": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "8192": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 4, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + } +} diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_6_0/E=64,N=192,device_name=AMD_Instinct_MI325X.json b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_6_0/E=64,N=192,device_name=AMD_Instinct_MI325X.json new file mode 100644 index 000000000..809218dd5 --- /dev/null +++ b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_6_0/E=64,N=192,device_name=AMD_Instinct_MI325X.json @@ -0,0 +1,173 @@ +{ + "1": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 256, + "GROUP_SIZE_M": 8, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "2": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 256, + "GROUP_SIZE_M": 4, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "4": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "8": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "16": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 8, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "24": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 8, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "32": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 8, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "48": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "64": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "96": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 8, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "128": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "256": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "512": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "1024": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 4, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "1536": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 4, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "2048": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 4, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "3072": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 8, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "4096": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 8, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "8192": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 8, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + } +} diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_6_0/E=64,N=384,device_name=AMD_Instinct_MI325X.json b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_6_0/E=64,N=384,device_name=AMD_Instinct_MI325X.json new file mode 100644 index 000000000..a360c5d33 --- /dev/null +++ b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_6_0/E=64,N=384,device_name=AMD_Instinct_MI325X.json @@ -0,0 +1,173 @@ +{ + "1": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 256, + "GROUP_SIZE_M": 8, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "2": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 4, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "4": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 4, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "8": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 8, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "16": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 8, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "24": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 1, + "num_stages": 2, + "waves_per_eu": 0 + }, + "32": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 8, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "48": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 8, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "64": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "96": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "128": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "256": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 4, + "num_warps": 1, + "num_stages": 2, + "waves_per_eu": 0 + }, + "512": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 1, + "num_stages": 2, + "waves_per_eu": 0 + }, + "1024": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 4, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "1536": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 32, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "2048": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 4, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "3072": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 8, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "4096": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 4, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "8192": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 8, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + } +} diff --git a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_6_0/E=64,N=768,device_name=AMD_Instinct_MI325X.json b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_6_0/E=64,N=768,device_name=AMD_Instinct_MI325X.json new file mode 100644 index 000000000..b796b745e --- /dev/null +++ b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_6_0/E=64,N=768,device_name=AMD_Instinct_MI325X.json @@ -0,0 +1,173 @@ +{ + "1": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 256, + "GROUP_SIZE_M": 8, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "2": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 16, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 16, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "4": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 16, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "8": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 1, + "num_warps": 1, + "num_stages": 2, + "waves_per_eu": 0 + }, + "16": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 256, + "GROUP_SIZE_M": 4, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "24": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 256, + "GROUP_SIZE_M": 4, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "32": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 256, + "GROUP_SIZE_M": 4, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "48": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 256, + "GROUP_SIZE_M": 4, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "64": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 256, + "GROUP_SIZE_M": 4, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "96": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 256, + "GROUP_SIZE_M": 4, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "128": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 256, + "GROUP_SIZE_M": 4, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "256": { + "BLOCK_SIZE_M": 32, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 256, + "GROUP_SIZE_M": 8, + "num_warps": 2, + "num_stages": 2, + "waves_per_eu": 0 + }, + "512": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 256, + "GROUP_SIZE_M": 4, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "1024": { + "BLOCK_SIZE_M": 64, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 1, + "num_warps": 4, + "num_stages": 2, + "waves_per_eu": 0 + }, + "1536": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 8, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "2048": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 128, + "GROUP_SIZE_M": 4, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "3072": { + "BLOCK_SIZE_M": 128, + "BLOCK_SIZE_N": 128, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 8, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "4096": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 4, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + }, + "8192": { + "BLOCK_SIZE_M": 256, + "BLOCK_SIZE_N": 256, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 4, + "num_warps": 8, + "num_stages": 2, + "waves_per_eu": 0 + } +}