Implement custom model with docker

[maglat]

Hey hey,

sorry I am not so familiar with it. I would like to use a custom model https://huggingface.co/stlohrey/chatterbox_de/tree/main
inside the docker. how I can point to this huggingface or my local download of it?
Inside the docker-compose config, model location is pointed to /cache which is internal. How could I point to my local folder where I downloaded the custom model?

[camfirem]

Hey,

I maybe found a solution, cause I was having the same issue. I guess you wanna have a German Model.
It isn't the most beautiful way, but it works for me.
I edited the "tts_model.py" in the running container, located with the path:

/app/app/core/tts_model.py

There, in function initialize_model(), under the "_model"-variable (should be line 84).
Add following lines:
checkpoint_path = hf_hub_dowload(repo_id="SebastianBodza/Kartoffelbox-v0.1", filename="t3_kartoffelbox.safetensors")
t3_state = load_file(checkpoint_path, device="cpu")
_model.t3.load_state_dict(t3_state)

This gave me the best results.

Another way I found was changing the "REPO_ID"-variable in /app/.venv/lib/python3.11/site-packages/chatterbox/tts.py (inside the Docker as well) ,
from:
REPO_ID = "ResembleAI/chatterbox"
to:
REPO_ID = "stlohrey/chatterbox_de"

This responded unfortunately with very bad results.

I hope I could help with it.