I'm reading your works of QTIP, and had one question here.
Why does the size of codebook claim to be (2^(kd) )* d in your paper?
To my perspective, in vector quantization, codebook size is usually (2^k) * d? Am I missing some important information here?
Hope to receive from you guys! Best wishes!
I'm reading your works of QTIP, and had one question here.
Why does the size of codebook claim to be (2^(kd) )* d in your paper?
To my perspective, in vector quantization, codebook size is usually (2^k) * d? Am I missing some important information here?
Hope to receive from you guys! Best wishes!