Docker: Gradio Inpainting WebPage shows Connection errored out
prrw opened this issue · comments
Hello @rentainhe @panxinmiao,
I have build the docker image as described. I had to install the package litellm and downgrade the gradio to gradio==3.50.2. After this the server runs. However it shows Error:Connection errored out when i click on Run.
Moreover, the log information after executing python gradio.py. Seems to be a warning from GroundingDINO. I dont understand why it is running in CPU mode. Anyways i dont know if both these things are connected with each other.
I would appreciate any help.
UPDATE:
torch is able to find all the GPUs
However, the system does not have a graphical interface. echo $DISPLAY returns nothing. Is it a problem ?
UPDATE:
The system is running under a proxy server. Should this be a problem ?
Oh, the problem is solved. The proxy server was not allowing the browser to interace with Gradio properly. I added the IP as exception.
Could anyone tell me if it is normal for GroundingDINO to run on CPU ? or there is a solution ?
@rentainhe UPDATE: nvidia-smi is working on the host system (CUDA Version: 12.4,NVIDIA RTX 6000 Ada Generation, Compute Capability: 8.9). Will this architecture be compatible here ? I dont see 8.9 in the TORCH_CUDA_ARCH_LIST.
Also i cannot run nvcc --version because i have not installed Cuda Toolkit in the Host system. So, the Makefile builds docker for CPU and not GPU. But That should not matter or ? Important is to install the correct cuda toolkit in the image.
There is a bug in the Makefile. Since it is mounting the repository from host as volume, the files built by GroundingDINO are not being used. Hence it is running on the CPU.
May be you should update the Makefile to not mount the repository and copy the models during build.