From 37e9ef8cfc9b16c51e7c3a9402c468b63334ee20 Mon Sep 17 00:00:00 2001 From: Jiawen Liu Date: Sat, 18 May 2024 13:54:27 -0700 Subject: [PATCH] Fix f8f8bf16 function signature when cuda disabled Summary: Since D57442310 changed f8f8bf16 function signature, changing the function signature back fixes the issue Differential Revision: D57529592 --- .../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,