diff --git a/vllm/model_executor/layers/fused_moe/configs/E=8,N=14336,device_name=AMD_Instinct_MI300X.json b/vllm/model_executor/layers/fused_moe/configs/E=8,N=14336,device_name=AMD_Instinct_MI300X.json index 66f9106bd1be3..4bf775347eccb 100644 --- a/vllm/model_executor/layers/fused_moe/configs/E=8,N=14336,device_name=AMD_Instinct_MI300X.json +++ b/vllm/model_executor/layers/fused_moe/configs/E=8,N=14336,device_name=AMD_Instinct_MI300X.json @@ -45,8 +45,8 @@ }, "16": { "BLOCK_SIZE_M": 16, - "BLOCK_SIZE_N": 16, - "BLOCK_SIZE_K": 256, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 64, "GROUP_SIZE_M": 1, "num_warps": 2, "num_stages": 2, diff --git a/vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=AMD_Instinct_MI300X.json b/vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=AMD_Instinct_MI300X.json index ed5b655d89937..5a3f415d54142 100644 --- a/vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=AMD_Instinct_MI300X.json +++ b/vllm/model_executor/layers/fused_moe/configs/E=8,N=3584,device_name=AMD_Instinct_MI300X.json @@ -45,8 +45,8 @@ }, "16": { "BLOCK_SIZE_M": 16, - "BLOCK_SIZE_N": 32, - "BLOCK_SIZE_K": 256, + "BLOCK_SIZE_N": 64, + "BLOCK_SIZE_K": 128, "GROUP_SIZE_M": 1, "num_warps": 2, "num_stages": 2, diff --git a/vllm/model_executor/layers/fused_moe/configs/E=8,N=7168,device_name=AMD_Instinct_MI300X.json b/vllm/model_executor/layers/fused_moe/configs/E=8,N=7168,device_name=AMD_Instinct_MI300X.json index 822f04e33e879..8d7b780271854 100644 --- a/vllm/model_executor/layers/fused_moe/configs/E=8,N=7168,device_name=AMD_Instinct_MI300X.json +++ b/vllm/model_executor/layers/fused_moe/configs/E=8,N=7168,device_name=AMD_Instinct_MI300X.json @@ -128,7 +128,7 @@ "num_warps": 8, "num_stages": 2, "waves_per_eu": 0, - "matrix_instr_nonkdim": 32, + "matrix_instr_nonkdim": 16, "kpack": 2 }, "512": {