search

KaihuaTang / Scene-Graph-Benchmark.pytorch

A new codebase for popular Scene Graph Generation methods (2020). Visualization & Scene Graph Extraction on custom images/datasets are provided. It's also a PyTorch implementation of paper “Unbiased Scene Graph Generation from Biased Training CVPR 2020”
MIT License
1.03k stars 228 forks source link

Cuda Mismatch Issue #93

Open kyteinsky opened 3 years ago

kyteinsky commented 3 years ago

❓ Questions and Help

Do I need to recompile PyTorch 1.4 with Cuda 11.0 in order to build apex? It raises an error saying that it was compiled using Cuda 10.1 and may result in another issue afterwards.

Is there any option to not use GPU at all for just inference from pre-trained model like from the demo folder?

Any help here?

Thanks!

yanan1989 commented 3 years ago

I encountered a same problem even though I compile pytorch 1.4 with cuda 11.0 during inference time

yanan1989 commented 3 years ago

I solved this issue by using python3.7 instead of python 3.8. pytorch1.4 with cuda10.1 is no problem. Please follow the installation order that kaihui showed us.

kyteinsky commented 3 years ago

@wangyanan326 That is the problem, I am using fedora and it doesn't allow degrading the CUDA version (fedora WS being a rolling release). I am currently not much involved in this project, once I get back to it, I will definitely try compiling pytorch with the latest cuda and just hope no further errors come up.