Current version might be broken

[medemi68]

Hey Team,

I just spinned up an instance and noticed that the most recent commit to the master branch produces the following error:

Traceback (most recent call last):
File "/parler-tts/stream-working.py", line 5, in
from parler_tts import ParlerTTSForConditionalGeneration, ParlerTTSStreamer
File "/parler-tts/parler_tts/init.py", line 8, in
from .modeling_parler_tts import (
File "/parler-tts/parler_tts/modeling_parler_tts.py", line 64, in
AutoConfig.register("dac", DACConfig)
File "/usr/local/lib/python3.11/dist-packages/transformers/models/auto/configuration_auto.py", line 1070, in register
CONFIG_MAPPING.register(model_type, config, exist_ok=exist_ok)
File "/usr/local/lib/python3.11/dist-packages/transformers/models/auto/configuration_auto.py", line 769, in register
raise ValueError(f"'{key}' is already used by a Transformers config, pick another name.")
ValueError: 'dac' is already used by a Transformers config, pick another name.

Any ideas? Should I try using an older branch/commit tag?

[medemi68]

For anyone who needs to use Parler right away and can't wait for a response, you can revert back to the previous commit using

pip install git+https://github.com/huggingface/parler-tts.git@5d0aca9753ab74ded179732f5bd797f7a8c6f8ee

That is the previous commit.

[bocklucas]

For anyone who needs to use Parler right away and can't wait for a response, you can revert back to the previous commit using

pip install git+https://github.com/huggingface/parler-tts.git@5d0aca9753ab74ded179732f5bd797f7a8c6f8ee

That is the previous commit.

Can confirm it was broken for me as well and reverting fixed it, thanks @medemi68 !

[ylacombe]

Hey everyone, thanks for opening the issue and helping on identifying where it blocks.

It actually comes from the audio encoder. While I still have to figure out how to actually fix it while being backward compatible for person using it with the previous repo version, you can find fixed weights for the current version by using:
model = ParlerTTSForConditionalGeneration.from_pretrained("parler-tts/parler-tts-large-v1", revision= "refs/pr/9")

[bocklucas]

Hey everyone, thanks for opening the issue and helping on identifying where it blocks.

It actually comes from the audio encoder. While I still have to figure out how to actually fix it while being backward compatible for person using it with the previous repo version, you can find fixed weights for the current version by using:
model = ParlerTTSForConditionalGeneration.from_pretrained("parler-tts/parler-tts-large-v1", revision= "refs/pr/9")

Thanks @ylacombe 😊