Scholar Feed

KV-cache compression

GPU decompression (CacheGen)

Superseded baseline#149 of 234 most-superseded

Cited as a baseline — critiqued by newer work, not yet beaten on a benchmark here

1 papers critique it · 0 beat it on benchmarks

What papers say

Verbatim critique sentences, each from a paper that cites GPU decompression (CacheGen) as a baseline.

Unfortunately, this introduces a new problem: interference with model computation. We find that concurrently running decompression and LLM serving on the same GPU causes both tasks to slow down, often by ~30%.
— ShadowServe: Interference-Free KV Cache Fetching for Distributed Prefix Caching

What to use instead

Recent methods in the same sub-problem, not yet superseded in the knowledge base — arXiv benchmark leaders, not vetted production recommendations.