You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
When I ran whisper-timestamped with --threads N or --threads 0 parameter on my CPU-only system the specified number of threads gets created but not used afterwards (only one core is loaded as per htop). Interestingly this behaviour is not observed when I run the "original" whisper, i.e. it uses (and loads) all the cores properly. That leads whisper-timestamped to work literally forever using one core opposed to some reasonable time with raw whisper.
OS: Arch Linux
CPU: AMD EPYC 7B12 x 2
Python version: 3.12.6
Whisper-timestamped version: 1.15.4
Whisper version: 20240927
Test whisper-timestamped commandline (only 1 core is used): whisper_timestamped --model large-v3 --device cpu --task transcribe --language English --vad True --verbose True --accurate /tmp/test.mp3
Test whisper commandline (all non-logical cores are used, as expected): whisper --model large-v3 --device cpu --task transcribe --language English --verbose True --word_timestamps True --highlight_words True /tmp/test.mp3
I can attach the screenshots of htop if needed.
The text was updated successfully, but these errors were encountered:
When I ran whisper-timestamped with
--threads N
or--threads 0
parameter on my CPU-only system the specified number of threads gets created but not used afterwards (only one core is loaded as per htop). Interestingly this behaviour is not observed when I run the "original" whisper, i.e. it uses (and loads) all the cores properly. That leads whisper-timestamped to work literally forever using one core opposed to some reasonable time with raw whisper.OS: Arch Linux
CPU: AMD EPYC 7B12 x 2
Python version: 3.12.6
Whisper-timestamped version: 1.15.4
Whisper version: 20240927
Test whisper-timestamped commandline (only 1 core is used):
whisper_timestamped --model large-v3 --device cpu --task transcribe --language English --vad True --verbose True --accurate /tmp/test.mp3
Test whisper commandline (all non-logical cores are used, as expected):
whisper --model large-v3 --device cpu --task transcribe --language English --verbose True --word_timestamps True --highlight_words True /tmp/test.mp3
I can attach the screenshots of htop if needed.
The text was updated successfully, but these errors were encountered: