Streamlit is stopping

[nemesis]

Ask the community or our support engineers for answers to questions.

I am writing an app to take a sentence, perform some named entity recognition on it with a model and return the output. It works perfectly in the python app but when I run the streamlit main file, it is stopping. Can someone help me out?

I can provide link to the source on github.

[andfanilo]

Hello @nemesis, welcome to the community

This does look like odd behavior, but it’s going to be hard for us to debug without a bit of source code. If possible for you, are you able to provide us with a smaller reproducible code snippet with the problem, or at the very least point the line of code where it seems to stop inside your source code on Github + a link to the project ?

Fanilo

[nemesis]

Here is a link to the main file on my github:

Soumi7/St_webapp/blob/master/main.py

Unfortunately the model file is too big to upload to github, I can attach a drive link if required.

[Gal_Benshlomo]

Hey @nemesis , did you happen to find a solution to this one? I’m having similar issues (also using simpletransformers, oddly enough). the only difference in my case was that I cached the function that loads the model, so it won’t load each time we want to make a prediction

[Tomasz_Kulakowski]

EDIT2: GOT IT
After reading the source code of the predict function I decided to check if multiprocess is being used and what will happen if I will omit it. And it worked.

model = T5Model(“t5-small”, use_cuda=False)
print(model.args.use_multiprocessed_decoding)
model.args.use_multiprocessed_decoding = False

This made a lot of sense because there were new processes on the “ps aux | grep streamlit” list

And now the previous message

Having the same problem right now. I’ve also used the cache for caching model (simpletransformers.t5) and it’s stopping after decoding outputs. Two different PC’s , ubuntu on both.

EDIT: I did found out 1 of problems on mi side with try/catch code block:

try:
#code
except Exception as e:
print(getattr(e, ‘message’, repr(e)))
print(getattr(e, ‘message’, str(e)))

Linux has opened file limit ( ulimit set to 1024 by default ) and I had it low. After changing it I still have “Stopping…” issue, but maybe you guys will have more luck.

[Stark]

Hey @nemesis !!

I am also encountering the same issue. Did you find any fixes yet ? It would be great if you can add something.

Thanks

[ziad1988]

Hello,

I don’t know if this issue has been solved but I encountered the same problem in my code and I fixed it but actually changing the model args of my bert model to :
model.args.use_multiprocessing_for_evaluation=False
I am guessing when the predict is launched, streamlit doesn’t like the multiprocess part

[acadavid79]

Hi. I’m having the same problem. I built the app on my M1 mac, used multiprocessing perfectly. Now I have deployed to an EC2 aws instance (only 2 cores) but expected to work at least. The Stopping … is everything that happens in the backend.