Autocompletion is slow in huggingface/transformers #1258
Labels
fixed in next version (main)
A fix has been implemented and will appear in an upcoming version
Comments
|
Ok I think the |
|
Is this code available somewhere? Lines like: Are concerning; 6 seconds to check a file is pretty large, and it'd be helpful to be able to see if there's something to optimize here. |
savannahostrowski
added
the
waiting for user response
|
I have reinstalled the transformers package from hugging face (https://huggingface.co/transformers/index.html) using conda which installs it with an egg file. As for now pylance is not slow anymore since it can't parse this library. I don't know why it was so slow though. |
|
Hi again, LogsLogs Pylance
torm.py file
from transformers import AutoTokenizer
AutoTokenizer. from Do you know if there's a solution or if I could remove transformers library from Pylance check? |
|
Yes, transformers on conda was unfortunately one of the main reasons to add support (so I was waiting for you to re-report this, sorry). Can you create trace logs again so we can pin down which files may be problematic in transformers? |
|
How can I do that? I mean, the logs I have linked are Python language server's log when autocompletion is slow on my torm.py file. I did not see any reference to what blocks in transformers library |
|
I was comparing it to the logs that you had pasted in your original issue; the new logs are much, much shorter and don't show any of the analysis work. You're definitely using the "Trace" log level, right? |
|
Oh sorry, I was logged onto another server and forgot to switch to Trace log level Logs
|
|
Thanks, I can reproduce this. |
jakebailey
added
needs investigation
jakebailey
changed the title
|
since we can repro, with this hidden option we probably can find which type is taking a long time in type evaluation and fix it by either adding partial type stub or improve type evaluation code around it. |
heejaechang
added
fixed in next version (main)
|
fixed in 2021.8.0 release. |
|
@heejaechang Hi, I'm using pyright and facing this issue. Can I get partial stubs somewhere? |
|
you can generate your own stub. add a strict comment to a file #pyright: strict
from transformers import AutoTokenizerthen a code action should appear on i'll add them here tmr https://github.com/microsoft/python-type-stubs or you can install pylance and manually look for the bundled stub folder |
|
i'll add them here tmr https://github.com/microsoft/python-type-stubs |
|
updated |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Environment data
Actual behaviour
It has been a week that Pylance seems to be very slow. It takes around 10 seconds to have the autocompletions. However it depends on which way you import the library. For instance
The auto completion is instantaneous. However with this
The autocompletion will take 6-7 seconds
Logs
Logs on https://pastebin.com/9cpHJ8SS (The logs was too long to put it there)
The text was updated successfully, but these errors were encountered: