From 06e78e172ca757f8eac6f5f60c4a3f56e4ce14f1 Mon Sep 17 00:00:00 2001 From: Jiawen Liu Date: Sat, 18 May 2024 17:36:08 -0700 Subject: [PATCH] Fix f8f8bf16 function signature when cuda disabled (#2605) Summary: Pull Request resolved: https://github.com/pytorch/FBGEMM/pull/2605 Since D57442310 changed f8f8bf16 function signature, changing the function signature back fixes the issue Reviewed By: yuguo68 Differential Revision: D57529592 fbshipit-source-id: afabd1c8de60d4797dd1688863bea0650f14b8fa --- .../gen_ai/src/quantize/cutlass_extensions.cu | 8 ++++++++ 1 file changed, 8 insertions(+) diff --git a/fbgemm_gpu/experimental/gen_ai/src/quantize/cutlass_extensions.cu b/fbgemm_gpu/experimental/gen_ai/src/quantize/cutlass_extensions.cu index d973d28ac..ef44aedb1 100644 --- a/fbgemm_gpu/experimental/gen_ai/src/quantize/cutlass_extensions.cu +++ b/fbgemm_gpu/experimental/gen_ai/src/quantize/cutlass_extensions.cu @@ -1896,6 +1896,14 @@ at::Tensor f8f8bf16_cublas( "CUDA version is older than 12.0"); // requires CUDA>=12 } at::Tensor f8f8bf16( + at::Tensor XQ, // FP8 + at::Tensor WQ, // FP8 + at::Tensor scale, + bool use_fast_accum) { + throw std::runtime_error( + "CUDA version is older than 12.0"); // requires CUDA>=12 +} +at::Tensor f8f8bf16_tensorwise( at::Tensor XQ, // FP8 at::Tensor WQ, // FP8 double scale,