AI TrackerFrontier desk · AEST
Back to news
normalaitech

ACCT: Asymmetric Cross-LLM Context Transfer

arXiv cs.AI · 31 Oct 2026, 03:50 AEST

ACCT (Asymmetric Cross-LLM Context Transfer) proposes a method for efficient context reuse across different large language models. This approach reduces computational costs by transforming KV cache from a source model into a format usable by a target model.

Key facts

Read aloud

ACCT Asymmetric Cross-L L M Context Transfer proposes a method for efficient context reuse across different large language models. This approach reduces computational costs by transforming KV cache from a source model into a format usable by a target model. In real-world deployments, large language model L L M systems are increasingly evolving from single-model invocation toward heterogeneous deployments in which different models are dynamically invoked for inference. ACCT transforms the KV cache produced by a source model into sparse KV states that can be directly consumed by a target model, thereby enabling cross-model reuse of context computation and decoupling context computation from downstream inference. Experiments show that ACCT matches or exceeds the target model's original inference performance even when reducing the KV cache size by seventy five percent. At a context length of approximately 19K tokens, ACCT's KV conversion achieves a six point five-twenty one point two times speedup over target context prefill.

Sources in this cluster