Paper

PReM: Learning What to Preserve and When to Refresh for Context Compression

arXiv:2607.14327v1 Announce Type: cross Abstract: Efficient long-context inference is not only about reducing memory cost, but also about keeping useful contextual evidence accessible as generation proceeds. However, existing compression-oriented approaches, such as key-value (KV) cache compression and context compression, often either make an early decision about which contextual information to keep or rely on an external compressor. Such designs make it difficult to adapt the compressed context to the evidence needed by later reasoning steps. This paper introduces PReM (Preserve and Refresh…

Authors:

Related discussions