Vqrae Representation Quantization

Unify multimodal understanding, generation, and reconstruction using high-dimensional codebooks for semantic information. VQRAE achieves 100% codebook utilization at 1536 dimensions—ideal when you need a single tokenizer for vision-language tasks.

adu2021 Updated 7 repo stars

File contents

adu2021/skillxiv commit e96b17a76d

Frequently asked questions

npx get-skill add adu2021/vqrae-representation-quantization