taming-transformers
taming-transformers copied to clipboard
About VectorQuantizer2
Hi. I have seen this sentence in doc-string of class VectorQuantizer2
(link):
Mostly avoids costly matrix multiplications and allows for post-hoc remapping of indices.
But I do not quite understand how it "avoids costly matrix multiplications" compared with VectorQuantizer
? It'll be appreciated if you can share some advice.
Looking forward to your reply!