Closed whyhardt closed 3 months ago
Yes, I agree with that. In fact, as part of our code cleanup I was going to ask whether we need any except for TTS. I have only tested TTS and none of the others, so if we still agree that we want to scrape the code down to the bare minimum, then I think we can take all but TTS out for v2.0.
We can keep them on an exploratory branch if we plan on working with them in the future.
I am closing this and leaving it as a task for issue #16
I get an error with the most current pytorch=2.2.2 Does it work with older versions? Cannot downgrade to <2.0 because torchaudio and torchvision are not supported.
Error: RuntimeError: derivative for aten::_scaled_dot_product_efficient_attention_backward is not implemented
Just kick out type=tr and leave only ff and tts?