I am running one Streamlit application on Huggingface spaces. The application is working partially on CPU, but I want my application on GPU. So I am using the NvidiaA10G large, but the page is not loading and showing connection timeout. As my model is quite large and requires at least NvidiaA10G large I cannot run the whole pipeline on CPU. But, it’s strange that I am able to run my application “partially” on Nvidia T4 Small and stops as the model size is big. Quite Weird. Is there any issue with the specific GPU configuration?
any help in this issue? Is this specific to streamlit application?
Hi @xelpmocAI what is the size of your model ? The space might get restarted because it’s consuming too much disk or getting OOM killed for lack of memory leading to the connection error.
The available memory on a T4 is 15GB.
Hello @chris-rannou , I am using NvidiaA10G container which has 24GB VRAM and 46 GB RAM. It is giving connection timeout at the application startup.