Periscope Extends Frozen LMs to Million-Token Contexts Without Training

KAUST's Periscope is a training-free inference method that lets frozen language models answer questions over documents far beyond their context window, enabling a 27B model on a single 80GB GPU to process about 4.5 million tokens.

2026-10-06 ~ 2026-10-06 · 2 related posts

• Periscope: training-free method reads 1M-token texts with a frozen 9k-token window — _reachsumit · 2026-10-06
• KAUST's training-free Periscope lets a 27B model read 4.5M-token contexts on one 80GB GPU — KAUST · 2026-10-06