Paper Overview
Field: NLP Authors: Alireza Nadali, Patrick Cooper, Ashutosh Trivedi, Alvaro Velasquez Published: 2026-05-12 arXiv: 2605.12471
Abstract
We introduce KV-Fold, a simple, training-free long-context inference protocol that treats the key-value (KV) cache as the accumulator in a left fold over sequence chunks. At each step, the model processes the next chunk conditioned on the accumulated cache, appends the newly produced keys and values, and passes the enlarged cache forward; the same one-step update is applied repeatedly, analogous to foldl in functional programming. Building on the KV cache concatenation primitive introduced for latent multi-agent communication, we repurpose it as a chunk-to-chunk recurrence for long-context inference. When processing chunk t, the model attends to the KV cache carried from earlier chunks as a prefix, reusing its internal state across segments without modifying or retraining the model.
Despite its simplicity, the induced recurrence is stable: per-step drift rises briefly before saturating into a flat plateau that persists across deep chains. This plateau is insensitive to 10,000x changes in numerical precision, robust across chunk sizes, and consistent across model families. At the task level, KV-Fold preserves exact information over long horizons. On needle-in-a-haystack benchmarks, it achieves 100% exact-match retrieval across 152 trials spanning contexts from 16K to 128K tokens and chain depths up to 511 on Llama-3.1-8B, while remaining within the memory limits of a single 40GB GPU.
--- *Auto-collected on 2026-05-14*