search

facebookresearch / sam2

The repository provides code for running inference with the Meta Segment Anything Model 2 (SAM 2), links for downloading the trained model checkpoints, and example notebooks that show how to use the model.
Apache License 2.0
12.7k stars 1.19k forks source link

About the cuDNN version #23

Open zzzweakman opened 4 months ago

zzzweakman commented 4 months ago

Great job! Thank the team! I encountered the following issue when using video inference:

x = F.scaled_dot_product_attention(
RuntimeError: cuDNN Frontend error: s_kv not a multiple of 64 or d not a multiple of 64 is not supported with cudnn version below 8.9.6

This is the cuDNN version on my system, and there are no issues when performing image inference:

>>> print(torch.backends.cudnn.version())
8700

Do I need to upgrade the cuDNN version?

diegoasua commented 4 months ago

You need to upgrade cuDNN and potentially CUDA too. run nvcc --version. Ideally CUDA 12.x and cuDNN 9.x.x

zzzweakman commented 4 months ago

You need to upgrade cuDNN and potentially CUDA too. run nvcc --version. Ideally CUDA 12.x and cuDNN 9.x.x

Thanks : )

Jandown commented 3 days ago

You need to upgrade cuDNN and potentially CUDA too. run nvcc --version. Ideally CUDA 12.x and cuDNN 9.x.x

I upgraded CUDA 12. x and CUDNN 9. x, but still encountered an error with the following message: "Runtime Error: CUDNN Frontend error: s_kv not a multiple of 64 with dropout enabled is not supported with CUDNN version below 9.0.0"