build.toml

[general]
version = "0.0.1"

[torch]
name = "quantization"
src = [
  "core/registration.h",
  "ext-torch/torch_binding.cpp",
  "ext-torch/torch_binding.h"
]
include = [ "." ]
pysrc = [
  "ext-torch/__init__.py"
]

[kernel.cutlass_w8a8]
capabilities = [ "7.5", "8.0", "8.6", "8.7", "8.9", "9.0", "9.0a" ]
src = [
  "cutlass_w8a8/common.hpp",
  "cutlass_w8a8/scaled_mm_c2x.cu",
  "cutlass_w8a8/scaled_mm_c2x.cuh",
  "cutlass_w8a8/scaled_mm_c2x_sm75_dispatch.cuh",
  "cutlass_w8a8/scaled_mm_c2x_sm80_dispatch.cuh",
  "cutlass_w8a8/scaled_mm_c2x_sm89_fp8_dispatch.cuh",
  "cutlass_w8a8/scaled_mm_c2x_sm89_int8_dispatch.cuh",
  "cutlass_w8a8/scaled_mm_entry.cu",
  "cutlass_extensions/epilogue/scaled_mm_epilogues_c2x.hpp",
  "cutlass_extensions/epilogue/broadcast_load_epilogue_c2x.hpp",
]
include = [ "." ]
depends = [ "cutlass", "torch" ]

[kernel.cutlass_w8a8_hopper]
capabilities = [ "9.0", "9.0a" ]
src = [
  "cutlass_w8a8/common.hpp",
  "cutlass_w8a8/scaled_mm_c3x.cu",
  "cutlass_extensions/epilogue/scaled_mm_epilogues_c3x.hpp",
  "cutlass_extensions/epilogue/broadcast_load_epilogue_c3x.hpp",
]
include = [ "." ]
depends = [ "cutlass", "torch" ]

[kernel.fp8_common]
capabilities = [ "7.5", "8.0", "8.6", "8.7", "8.9", "9.0", "9.0a" ]
src = [
  "fp8/common.cu",
  "fp8/common.cuh",
  "dispatch_utils.h"
]
include = [ "." ]
depends = [ "torch" ]

[kernel.fp8_marlin]
capabilities = [ "8.0", "8.6", "8.7", "8.9", "9.0", "9.0a" ]
src = [
  "fp8/fp8_marlin.cu",
  "gptq_marlin/marlin.cuh",
  "gptq_marlin/marlin_dtypes.cuh",
]
#include = [ "." ]
depends = [ "torch" ]

[kernel.int8_common]
capabilities = [ "7.5", "8.0", "8.6", "8.7", "8.9", "9.0", "9.0a" ]
src = [
  "compressed_tensors/int8_quant_kernels.cu",
  "dispatch_utils.h"
]
include = [ "." ]
depends = [ "torch" ]