Each MuseNet encoding token is one of roughly 4500 values. That takes about 12 bits to represent. They are currently represented as 5 ascii characters (including a space) = 40 bits. We should be able to reduce the file size by at least 70%.
This probably isn't worth doing, but I've been thinking about it so I opened an issue to keep track of my thoughts.
Each MuseNet encoding token is one of roughly 4500 values. That takes about 12 bits to represent. They are currently represented as 5 ascii characters (including a space) = 40 bits. We should be able to reduce the file size by at least 70%.
This probably isn't worth doing, but I've been thinking about it so I opened an issue to keep track of my thoughts.