search

meronym / speaker-transcription

Transcription with speaker diarization pipeline
MIT License
83 stars 17 forks source link

ModelError: Prediction interrupted; please retry (code: PA) #5

Open gkamradt opened 1 year ago

gkamradt commented 1 year ago

Unsure if this is a replicate issue not.

I run the model on in a 8 minute mp3 via the api which ran fine in the expected amount of time.

However I'm trying a 32 minute mp3 and the prediction is stays in "starting" stage and doesn't move.

Comes back with an error after a while: ModelError: Prediction interrupted; please retry (code: PA)

The same 32 minute file runs fine on the web interface, but can't get it going via the API.

Thoughts?

ShinnosukeUesaka commented 1 year ago

same error

ShinnosukeUesaka commented 1 year ago

It worked for me when I uploaded a file to somewhere else and sent a url to the replicate endpoint rather than the actual file.

gkamradt commented 1 year ago

@ShinnosukeUesaka Mind sharing that code so it's documented?

Then I'll close this one out.

gkamradt commented 1 year ago

@ShinnosukeUesaka down for that?

EticalPy commented 1 year ago

How can I send the url to the replicate endpoint?

jasoncox commented 1 year ago

I'm not working with this model specifically, but ran into a similar problem on Replicate. Digging a little deeper into the cog docs:

https://github.com/replicate/cog/blob/main/docs/http.md

All Path and File inputs for the HTTP API are passed as URLs, so it looks like you have the option to pass in a data url (https://developer.mozilla.org/en-US/docs/Web/HTTP/Basics_of_HTTP/Data_URLs) or an external url directly.