Help Please-Pulling My Hair Out!

[sidsidney]

Rookie here trying to get StyleTTS2 running in Docker on Fedora 41 with RTX 5090 plus 64G Ram. Gong back and forth between Grok and Claude for days now and they have got me most of the way there but just can't close the deal. UI loads at http://127.0.0.1:7860/ but anytime I input text I get a 'Model Loading Failed' error. Any assistance that you can offer would be greatly appreciated as I am close to bailing on this project.

api_server.py
app.py

logs8.txt
requirements.txt
start.sh

config.yml.txt

config_ft.yml.txt
config_libritts.yml.txt

[martinambrus]

These lines would suggest that your model is either trained and saved in a different form than expected (for example storing keys under a "model." prefix/namespace, which happens when you train it in DDP using accelerate) OR something that's loading the model is wrapping it (or your libraries, such as the mentioned Munch library) in some kind of an object wrapper.

styletts2-1 | ERROR:app:Error loading models: decoder
styletts2-1 | Traceback (most recent call last):
styletts2-1 | File "/opt/conda/lib/python3.10/site-packages/munch/init.py", line 116, in getattr
styletts2-1 | return object.getattribute(self, k)
styletts2-1 | AttributeError: 'Munch' object has no attribute 'decoder'

Another issue could be that your config (if you're using config_ft.yml.txt) is incorrect. The decoder in this config is outside of the "model" key and there are also other wrong entries in there. Compare it with https://github.com/yl4579/StyleTTS2/blob/main/Configs/config_ft.yml

[sidsidney]

Thanks Martin, I'll give it a look