Closed lonzi closed 1 year ago
@lonzi i was just trying to see if you were paying attenti.. just kidding
that was one of my high school teacher's favorite line :laughing:
thank you for catching this Alon!
are you doing TTS?
Thanks for fixing! No, I work on drums generation and using the mentioned transformer above a pre-trained EnCodec model.
Hi, thanks for the implementation, it is really helpful!
While using the Transformer class, I noticed a significant performance degradation when activating the optional cross attention branch. I suspect this is due to a missing skip connection in the following line of code: https://github.com/lucidrains/audiolm-pytorch/blob/main/audiolm_pytorch/audiolm_pytorch.py#L456
I suggest changing to x = cross_attn(x, context = context, mask = context_mask) + x