281M and 235M param models too.
https://github.com/facebookresearch/seamless_communication/b...
I don't really know how the metrics they list compare to whisper, I'm very curious if these are fast enough for realtime speech2text? I think whisper technically could but it was difficult to do or something like that?