diff --git a/dsv4/kernels/cuda/kv_quantize.cu b/dsv4/kernels/cuda/kv_quantize.cu
index d29246ee..689c5d04 100644
--- a/dsv4/kernels/cuda/kv_quantize.cu
+++ b/dsv4/kernels/cuda/kv_quantize.cu
@@ -293,7 +293,7 @@ std::tuple<torch::Tensor, torch::Tensor> quantize_nvfp4_from_fp32_cuda(
     return {out_fp4.view(torch::kFloat4_e2m1fn_x2), out_sf.view(torch::kFloat8_e4m3fn)};
 }
 
-std::tuple<torch::Tensor, torch::Tensor, torch::Tensor> quantize_fp8_e4m3_from_fp32_cuda(
+std::tuple<torch::Tensor, torch::Tensor> quantize_fp8_e4m3_from_fp32_cuda(
     torch::Tensor input
 ) {
     int M = input.size(0);