search

yuhongtian17 / Spatial-Transform-Decoupling

MIT License
89 stars 7 forks source link

Dependencies compatibility #7

Closed theoh-io closed 5 months ago

theoh-io commented 6 months ago

Hello,

Thanks for sharing the code and weights for this model.

I am currently trying to setup the project on my personal laptop I previously tried with my regular setup: WSL Ubuntu20.04, python3.10, cuda12.1 and latest pytorch version gut got an error.

Following your recommendation I started a new WSL instance on Ubuntu 18 with python3 and made sure to install cuda toolkit 10.2 and pytorch 1.7, I get torch.cuda is available however I still get RuntimeError: CUDA error: no kernel image is available for execution on the device.

From what I understood my GPU(RTX4060) needs latest (12.x) CUDA version.

Do you have any idea on how I could solve this ? Do you plan on releasing compatible updates with cuda 12.x ?

yuhongtian17 commented 6 months ago

NVIDIA do not recommend using CUDA 10.2 with RTX 40 series GPU. Our RTX 4090 GPU with {cuda toolkit 11.6, python 3.8, pytorch 1.12.1, mmcv==1.6.1, mmdet==2.25.1, mmrotate==0.3.4, numpy==1.21.5, yapf==0.40.1} can run the code.