#Does GPU can increase Piper/Whisper (Assist) overall performances of Voice PE ?
1 messages ยท Page 1 of 1 (latest)
whisper set up with gpu acceleration for makes a big difference to STT speed.
Is there a guide on how to set up whisper with GPU ?
i have had some luck with the linuxserver container running on my server that has the gpu in it
heres a cat of my docker-compose i use for it
version: '3'
services:
faster-whisper:
image: lscr.io/linuxserver/faster-whisper:gpu
container_name: faster-whisper
environment:
- PUID=1000
- PGID=1000
- TZ=Etc/UTC
#- WHISPER_MODEL=tiny-int8
#- WHISPER_MODEL=base.en
#- WHISPER_MODEL=medium-int8
- WHISPER_MODEL=cstr/whisper-large-v3-turbo-int8_float32
- WHISPER_BEAM=5 #optional
- WHISPER_LANG=en #optional
volumes:
- ./config/:/config
ports:
- 10300:10300
restart: unless-stopped
devices:
- /dev/dri:/dev/dri
deploy:
resources:
reservations:
devices:
- driver: nvidia
count: all
capabilities: [gpu]
can adjust as needed but its a starting point
I am running HA in a VM running HAOS.
I can easily allocate the GPU to the VM but wondering if there are any particular parameters that I should be aware of for whisper to use it... or does it work "out of the box" once the device is made available ?
Note that version is deprecated and devices should not be needed in your case.
yeah, some of my container stuff is old. i realise i should update it but the whole...
"if it works don't touch it" applys in this case ๐
i dont think the haos addon for whisper supports gpu acceleration (I am not sure though) so I think you would need to run the whisper container on your host OS or another VM.
Sometimes but this can have unintended consequences and docker compose complains about the deprecation on every invocation.
that system may or may not still be using docker-compose and whilst I am not entirely happy about it... it probably wont get "fixed" until I fix a lot of other stuff and rework that server entirely ๐ whilst i keep alot of stuff up to date theres a few things that i havent really touched...
had quite the same setup but i recently moved to Incus for the awesome API, VM and OCI capabilities ๐
And decided to give a chance at HAOS at the same time since it seems to be the prefered way and the most supported way of doing things HA side
Speaking of containes, can someone point me the path where the Docker Containers configuration file are ?
How is it manage ?
What files? You can check the supervisor source code. Are you sure HAOS is the right choice?