Fetching the paper…

Scalable Training for Vector-Quantized Networks with 100% Codebook Utilization · Around