Preprint
Jul 2026
SeDeM: Selective Decompression of Hidden-State Memories for Long-Context Question Answering
SeDeM is proposed, a selective decompression framework that decouples compact memory storage from decoder conditioning and reduces online time-to-first-token and improves autoregressive decoding throughput relative to ICAE.
Maryam Haghifam, Jason Cong, Yizhou Sun
· 0 citations