Could not load the custom kernel
environment: Centos 7.6 cuda:11.4. torch:2.0.1 transformers:4.41.0.dev0
When attempting to utilize the demo provided in https://github.com/NielsRogge/Transformers-Tutorials/blob/master/Grounding%20DINO/Inference_with_Grounding_DINO_for_zero_shot_object_detection.ipynb, I encountered a series of errors as follows:
Could not load the custom kernel for multi-scale deformable attention: [Errno 2] No such file or directory: '/home/fconlas/ground_dino/transformers-main/src/transformers/kernels/grounding_dino/vision.cpp'".
Could not load the custom kernel for multi-scale deformable attention: [Errno 2] No such file or directory: '/home/fconlas/ground_dino/transformers-main/src/transformers/kernels/grounding_dino/vision.cpp'".
But this error did not occur when running the demo on a Mac.
same error on the cuda:12.1 torch:2.1.2 transformers: 4.40.1
same
Could not load the custom kernel for multi-scale deformable attention: [Errno 2] No such file or directory: 'D:\\Program Files\\anaconda\\envs\\learning\\lib\\site-packages\\transformers\\kernels\\grounding_dino\\vision.cpp'
@bkfeng @JacksonLopez @colacool could you update the library and try again? It should be fixed with the latest release yesterday
same
@garmanwu which version of transformers are you using?
transformers 4.42.3
It occurs error on windows even if I copied and rename detr instead of missing vision.cpp .
But Macbook is okay.
@garmanwu
you don't need to copy anything I fixed it with a PR a few weeks ago the only thing is that you gotta also have ninja installed to use the kernels. I would recommend trying the latest version of transformers
same error on the cuda 12.1 pytorch 2.4.1, transformers 4.45.2 (pyhd8ed1ab_1 conda-forge), windows
transformers 4.45.0, on linux,Could not load the custom kernel for multi-scale deformable attention: /home/.../.cache/torch_extensions/py39_cu121/MultiScaleDeformableAttention/MultiScaleDeformableAttention.so: cannot open shared object file: No such file or directory
But I can still get the result.
the result is correct.
so actually the above Could not load the custom kernel ... does not matter?
transformers 4.45.0, on linux,
Could not load the custom kernel for multi-scale deformable attention: /home/.../.cache/torch_extensions/py39_cu121/MultiScaleDeformableAttention/MultiScaleDeformableAttention.so: cannot open shared object file: No such file or directory
But I can still get the result.
the result is correct.
so actually the aboveCould not load the custom kernel ...does not matter?
I got the same result with transformers==4.48.3, did you find a solution or did you ignore the warning?