replicate/flash-mla

1[general]2name = "flash_mla"3 4[torch]5src = ["torch-ext/torch_binding.cpp", "torch-ext/torch_binding.h"]6 7 8[kernel.activation]9cuda-capabilities = [10  # "7.0", "7.2", "7.5", "8.0", "8.6", "8.7", "8.9", 11 12  # Only available on H100 and H200 13  "9.0", # (Hopper)14]15src = [16  "flash_mla/flash_mla_api.cu",17  "flash_mla/flash_fwd_mla_bf16_sm90.cu",18  "flash_mla/flash_fwd_mla_fp16_sm90.cu",19  "flash_mla/flash_fwd_mla_kernel.h",20  "flash_mla/flash_fwd_mla_metadata.cu",21  "flash_mla/flash_mla.h",22  "flash_mla/named_barrier.h",23  "flash_mla/softmax.h",24  "flash_mla/static_switch.h",25  "flash_mla/utils.h",26]27depends = ["torch", "cutlass_3_6"]28