When will cuda12.1 be supported for GroundingDINO

[yumianhuli]

RT
thanks！

[Ren Tianhe]

Hello, when you're using cuda12.1, what type of error did you meet, to our knowledge, grounding-dino can be successfully worked under cuda12.1 with Torch 2.1.2 @yumianhuli1

[yumianhuli]

Hello, when you're using cuda12.1, what type of error did you meet, to our knowledge, grounding-dino can be successfully worked under cuda12.1 with Torch 2.1.2 @yumianhuli1

Hi！
I have compiled Successfully installed groundingdino-0.1.0
But when I run the demo

1. it shows：The Failed to load custom C++ ops. Running on CPU mode Only!
2. Load the local model and say PermissionError: [Errno 13] Permission denied: ”model path“

I use cuda12.1 and pytorch2.1.1 gpu version

import torch
print(torch.version)
2.1.1+cu121

[Ren Tianhe]

Hello, when you're using cuda12.1, what type of error did you meet, to our knowledge, grounding-dino can be successfully worked under cuda12.1 with Torch 2.1.2 @yumianhuli1

Hi！ I have compiled Successfully installed groundingdino-0.1.0 But when I run the demo

1. it shows：The Failed to load custom C++ ops. Running on CPU mode Only!
2. Load the local model and say PermissionError: [Errno 13] Permission denied: ”model path“

I use cuda12.1 and pytorch2.1.1 gpu version

import torch print(torch.version) 2.1.1+cu121

Hello, have u set the CUDA_HOME environment param to your cuda-12.1 Runtime, you should double-check this, in my env, after setting CUDA_HOME=/path/to/cuda-12.1 the custom Deformable-Attention operator can be successfully installed.

For the second question, could u give us more info? I'm not sure which part of the model can not be loaded well, in loading BERT weight or loading G-DINO weight. And this might be a system permission issue, it should not be an environmental problem, I guess

[yumianhuli]

Hello, when you're using cuda12.1, what type of error did you meet, to our knowledge, grounding-dino can be successfully worked under cuda12.1 with Torch 2.1.2 @yumianhuli1

Hi！ I have compiled Successfully installed groundingdino-0.1.0 But when I run the demo

1. it shows：The Failed to load custom C++ ops. Running on CPU mode Only!
2. Load the local model and say PermissionError: [Errno 13] Permission denied: ”model path“

I use cuda12.1 and pytorch2.1.1 gpu version
import torch print(torch.version) 2.1.1+cu121

Hello, have u set the CUDA_HOME environment param to your cuda-12.1 Runtime, you should double-check this, in my env, after setting CUDA_HOME=/path/to/cuda-12.1 the custom Deformable-Attention operator can be successfully installed.

For the second question, could u give us more info? I'm not sure which part of the model can not be loaded well, in loading BERT weight or loading G-DINO weight. And this might be a system permission issue, it should not be an environmental problem, I guess

Hello, I am using the win11 environment, and set system environment var of CUDA_HOME-- C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.1\bin ，But it didn't work.

[Johan Huynh]

UDA_HOME-- C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.1 only, dont put bin in there