David Braun

@claude please work on this.

@claude Please try again.

@claude I think the assumptions about the preset format are incorrect. Work on the code and so that all tests pass locally before commiting.

Thanks for taking a look. I've been plotting histograms and I've observed that I can get the same behavior between PyTorch and JAX with this procedure: 1. Use the same...

@pseeth @seethlord can you take a look?

Sorry for replying so late. 1) This might happen if you're playing MIDI notes several octaves above the center note (set to 60). The sampler [test](https://github.com/DBraun/DawDreamer/blob/main/tests/test_sampler.py) is working well for...

1. I think the EnCodec-MMD is still on this branch: https://github.com/jmlemercier/audiocraft/blob/encodec-mmd/docs/MMD.md 2. Look at this stack trace: * https://github.com/facebookresearch/audiocraft/blob/adf0b04a4452f171970028fcf80f101dd5e26e19/audiocraft/models/lm.py#L257 * https://github.com/facebookresearch/audiocraft/blob/adf0b04a4452f171970028fcf80f101dd5e26e19/audiocraft/modules/transformer.py#L708 * https://github.com/facebookresearch/audiocraft/blob/adf0b04a4452f171970028fcf80f101dd5e26e19/audiocraft/modules/transformer.py#L565 The "Linear" in the diagram probably refers...

I see something similar. I modified `torchinfo` in this way: https://github.com/TylerYep/torchinfo/issues/254 Then I did ```python from audiocraft.models import MusicGen import torchinfo model = MusicGen.get_pretrained('facebook/musicgen-small') model.set_generation_params(duration=2) with torch.autocast("cuda"): model.lm.eval() descriptions =...

There's an explanation for `1/sqrt(3)`. It's because the variance of a uniform distribution between -1 and 1 is 1/3, so the standard deviation is `1/sqrt(3)`. I hope that's a clue...

I would suggest using a replicator where the main base is something with an Audio Play CHOP. You could have 100 replicants. Then use TOP processing to figure out how...