search

OpenGVLab / DCNv4

[CVPR 2024] Deformable Convolution v4
https://arxiv.org/pdf/2401.06197.pdf
MIT License
482 stars 27 forks source link

RuntimeError: "dcnv4_backward_cuda" not implemented for 'BFloat16' #8

Closed DsaltYfish closed 8 months ago

DsaltYfish commented 8 months ago

I have been using PyTorch's native Automatic Mixed Precision (AMP) feature and specified the data type as bf16 (bfloat16). However, I encountered an error during the backward propagation step. I would like to know if this is due to a lack of framework support or if I made an error in my usage.

YuwenXiong commented 8 months ago

Hi, thanks for raising this up. Now this issue has been fixed with the 1.0.0.post2 release

DsaltYfish commented 8 months ago

After upgrading to version 1.0.0.post2, I encountered an issue. When performing backpropagation, I received an error message saying "expected scalar type Float but found BFloat16".

jiarenyf commented 8 months ago

Meeting the same problem ...

JiaHeng-DLUT commented 2 months ago

+1. same problem.