Open AgentScrubbles opened 6 months ago
This appears to work when I run your direct container from Dockerhub, daswer123/xtts-api-server
, but not when I build the container myself. Maybe it's just a documentation thing that I'm missing?
Hi docker was not created by me, I was helped, it was quite a long time ago, I unfortunately do not have the opportunity to test the docker. After the holidays I will try to fix this problem with the help of a friend's computer.
Thanks, I took a quick stab at it but I need to understand the underlying code more. I think it just needs a clearer mapping of the model directory to a specific folder. It may be a good idea to extract that to an environment variable if it isn't already, then always reference that environment variable for saving/loading models
I upgraded my pc and now I can fully work with docker. I've redone the docker file and it should work now
Thanks I'll try it out!
Update, sorry for the lateness, so I've tried it and it's running, but not picking up my model either. I'm still poking at it, I want to double check to see if it's my specific image or if there's something else wrong with it, will have an update soon.
Ive got the docker running on URAID... the docker runs and I can connect to XTTS in ST, but there are no voices. I put the speakers folder in the app folder and voices still not found. Im thinking that its not looking at the correct location for the files it needs.
CMD ["bash", "-c", "python3 -m xtts_api_server -hs 127.0.0.1 -p 8020 -sf 'xtts-server/speakers' -o 'xtts-server/output' -mf 'xtts-server/models' --deepspeed"]
Would be nice if we could use an environment variable and then set our own startup options. For example I would like to use streaming mode.
It looks like the download functions for the models is currently broken. I've tried the default image and have also tried to tinker with it myself.
The models will download, and I can verify they are on the disk, but it appears that it immediately has no idea where the model/configs are. In the code it seems to confuse /app/models and /app/xtts_api_server/models, so maybe there's something there?
Full output from running the docker image: