Spaces:
Running
on
T4
This space runs on XTTS_V1, not XTTS_V2
Hi,
It's written that the linked model is coqui/XTTS-v2 but it's not the case. It's written so because the coqui/XTTS-v2 model is explicitely mentionned in README.md. If you remove this mention, the linked model becomes coqui/XTTS-v1. But I don't know how to use coqui/XTTS-v2...
Regards
You sure? The mentioned model within code is _v2 here:
https://huggingface.co/spaces/coqui/xtts/blob/main/app.py#L50-L72
To check it, duplicate the space, remove the mention in the README.md and you will see that the actual model is v1, even if the parameter passes v2. HuggingFace will show that the space downloads the model v1.
I did. I get:
Downloading if not downloaded Coqui XTTS V2
> Downloading model to /home/user/.local/share/tts/tts_models--multilingual--multi-dataset--xtts_v2
> Model's license - CPML
> Check https://coqui.ai/cpml.txt for more info.
XTTS downloaded
While it says "to <path>", it is also the model name that it downloads as defined in the following repo:
https://github.com/coqui-ai/TTS/blob/dev/TTS/utils/manage.py#L371
Within the HF README.md, you are conflating models parameter with preload_from_hub, models only links this space to the HF model. See:
https://huggingface.co/docs/hub/spaces-config-reference
XTTS-v1 is even given as an example there. Removing models from README.md makes HF search for the model name within code... at least... I don't think it searches within binary files.
[edit] Also for me even without models within the README.md, it still lists -v2
Ah OK, maybe I'm wrong.