QJL: 1-Bit Quantized JL Transform for KV Cache Quantization with Zero Overhead.
Amir ZandiehMajid DaliriInsu HanPublished in: CoRR (2024)
Keyphrases
- transform coefficients
- transform domain
- uniform quantization
- subband
- memory hierarchy
- reconstructed image
- entropy coding
- coding scheme
- transform coding
- resource consumption
- spatial domain
- prefetching
- memory access
- adaptive quantization
- image reconstruction
- image coder
- hash table
- hit rate
- discrete cosine transform
- compression scheme
- bitstream
- vocabulary tree
- filter bank
- image compression
- query processing
- random access memory
- multiscale
- significant bit
- virtual memory
- quantization error
- transmission line
- data access
- wavelet transform
- multiresolution
- computational complexity