New LowRAM method compresses KV caches with low-rank keys and amortised values
A new arXiv paper introduces LowRAM, a test-time KV cache compression method using low-rank keys and amortised values to cut memory use during autoregressive decoding.
Read the full story at arXiv cs.LG ↗
Timeline · 1 report
- 2026-11-06 05:00 · arXiv cs.LG
LowRAM: Low-Rank Keys and Amortised Values for Composable KV Cache Compression at Test-Time