Under you can see the last version of the app but restarting the app or even restarting the space does not use the latest version. Do I really always need to run a factory rebuild?
1 Like
Maybe. Ongoing issue.
Incidentally, it seems that we can report bugs on the Hugging Face website using the following github.
If there is a dedicated github repository for each library, etc., we should of course use that for reporting, but there probably isn’t one for Spaces.
I have a lot of trouble trying to deploy my app on a gradio space. It runs locally. I can deploy it to the 72h environment by using launch(share=True) and it works. But it seams to be very difficult to run it in a space with zeroGPU (as a pro user). When calling the inferenceAPI it starts to allocate the GPU and give no response, no log entry, no exception. When restarting the app the port 7860 is still in use and this is the only one the app tries to get - no range - no flexibility. With restarting the space the port gets free at least. But the source code is taken from the cache until you run a factory rebuild.
1 Like
Zero GPU Spaces are convenient, but the way they’re used is quite unique…
I think it’s the most unique of all the Hugging Face Spaces because it has a lot of shared resources.
The functions with spaces.GPU decorators would be in a separate process, they’re only for Gradio Spaces, they have limited versions of PyTorch, and sometimes the behavior via the API is different…
Well, if you encounter any bugs with Zero GPU, I think it’s a good idea to check the Zero GPU community first.
Spaces other than the Zero GPU space are relatively easy to set up, apart from the YAML part of the README.md and the dependency settings.