mamba icon indicating copy to clipboard operation
mamba copied to clipboard

Mamba-S4

Open turian opened this issue 2 years ago • 6 comments

Could you share code for Mamba-S4, the audio model for longform signals?

turian avatar Dec 20 '23 21:12 turian

We don't have that code in a releasable form. It's essentially the S4 model put inside the Mamba block, so it's pretty easy to implement.

albertfgu avatar Dec 21 '23 06:12 albertfgu

@albertfgu one question to bother you.

Is there the minor module of S4 in mamba or s4 repos? Thanks.

ftgreat avatar Jan 10 '24 08:01 ftgreat

We aren't using submodules. You'll need to incorporate the code from the S4 repo separately.

albertfgu avatar Jan 10 '24 17:01 albertfgu

@albertfgu Thank you for your work on mamba. I was wondering if you could elaborate on the architecture of the S4+mamba block a bit more? I am trying to reproduce mamba+s4 with a unet as described in the paper, but I so far only have working forward pass for a sashimi-like model with just mamba blocks. Any additional detail you can provide would be much appreciated!

btrude avatar Jan 22 '24 15:01 btrude

Take the Mamba block and replace the S6 layer with S4.

albertfgu avatar Jan 22 '24 22:01 albertfgu

@albertfgu To my understanding, Mamba-S4 is this block with the S4Block as mixer. Is that correct?

hungdche avatar Apr 28 '24 01:04 hungdche