I have been trying to deploy my app using different methods but am running into various related errors:
Deploying using environment.yml file gives me the following error: Installer returned a non-zero exit code; Error during processing dependencies; Streamlit server consistently failed status checks
When previously I was deploying with a requirements.txt file, I was able to deploy but was getting strange errors when trying to use the functionality. For instance, the keras models would not load and I would get the following error: You may be trying to load on a different device from the computational device. Consider setting the
***experimental _io_device*** option in
tf.saved_model.LoadOptions to the io device such as ‘/job:localhost’.
Similarly, some of my other models would not load with errors such as ‘GradientBoostingClassifier’ object has no attribute ‘_loss’ etc.
The closest I got to deployment was with the simple requirements.txt file where some of my models would work and some would give out error codes.
The local version of the app works without any glitches.
Thanks for your help.
Can you share a link to your public github repo?
I suspect that the ML models you have locally may have been created with a different (e.g. much older or incompatible) version of the respective ML library.
Hi Franky - there you go. Interesting hypothesis - very possible as I wrote some of the code a couple of months ago so any updates in that time would certainly not be reflected. I saved the models a few days ago however after rerunning the Python script so I can only think that the environment.yml file in my repo has some unsupported libraries. Thanks
I did the same, and i had the same issues you described, see my pull request.
I think the models were trained with older versions of the ML frameworks.
Thanks Franky - what would you recommend? Is it better to retrain the models using the new frameworks and try to deploy again?
To update the frameworks - do I need to run any command line code in terminal i.e. update the packages in the local environment that I am running?
I got some models working by downgrading some of the ML libraries, see my pull request.
I am not an ML expert and this ecosystem is changing so fast, but if you have the ability to retrain your models, i would do that with the newer ML libraries.
I would setup a clean local dev environment, either with docker or with a python virtual environment. Otherwise you will end up with a mess
Thanks Franky - that makes sense. Appreciate your help, I will retrain the models in a clean local dev environment and attempt to redeploy
Hello, Have you been able to fix the error and deploy successfully? I’m having the same error and I would like to know how you went about it.
Since this is a very unspecific error message, please share your public github repo.
Hey Franky - I was able to retrain in a virtual environment and deploy successfully. Thanks very much for the help.