diff --git a/moe_kernels/triton_3_3_1/E=128,N=192,device_name=NVIDIA_GeForce_RTX_3090.json b/moe_kernels/triton_3_3_1/E=128,N=192,device_name=NVIDIA_GeForce_RTX_3090.json new file mode 100644 index 0000000..79dc228 --- /dev/null +++ b/moe_kernels/triton_3_3_1/E=128,N=192,device_name=NVIDIA_GeForce_RTX_3090.json @@ -0,0 +1,10 @@ +{ + "64": { + "BLOCK_SIZE_M": 16, + "BLOCK_SIZE_N": 32, + "BLOCK_SIZE_K": 64, + "GROUP_SIZE_M": 64, + "num_warps": 4, + "num_stages": 3 + } +} diff --git a/moe_kernels/triton_3_3_1/E=128,N=384,device_name=NVIDIA_GeForce_RTX_3090.json b/moe_kernels/triton_3_3_1/E=128,N=384,device_name=NVIDIA_GeForce_RTX_3090.json deleted file mode 100644 index 661af2d..0000000 --- a/moe_kernels/triton_3_3_1/E=128,N=384,device_name=NVIDIA_GeForce_RTX_3090.json +++ /dev/null @@ -1,10 +0,0 @@ -{ - "64": { - "BLOCK_SIZE_M": 16, - "BLOCK_SIZE_N": 128, - "BLOCK_SIZE_K": 128, - "GROUP_SIZE_M": 32, - "num_warps": 8, - "num_stages": 2 - } -}