Open lmmx opened 3 years ago
Currently this is blocked by CUDA out of memory because it's apparently not released following transcription ?!
(Also please turn off the annoying huggingface parallelism warning triggered by the quill standup
and remote_push_manifest
calls, which make the quill output unreadable)
Now that it's possible to obtain the MP4 source URLs automatically (without manually providing the M4S segment URLs or even the episode ID) and therefore obtain a transcript without any requirements whatsoever other than the series ID already being stored in the library, it's conceivable to have a stream that not only transcribes upon creation but then serves the exported transcript(s) if/when successful.
Add this as another flag (
quill_export=None
) totap.scrape.grab.load_stream
Such that
stream = tap.scrape.load_stream(transcribe=True, queck_export="poll")
would carry out the