UT-ACA: Uncertainty-Triggered Adaptive Context Allocation for Long-Context Inference
This addresses the problem of inefficient context allocation in long-context inference for AI practitioners, offering an incremental improvement over fixed-budget methods.
The paper tackled the challenge of long-context inference in large language models by proposing UT-ACA, a framework that dynamically adjusts context allocation based on token-wise uncertainty, reducing average context usage while maintaining generation quality.
Long-context inference remains challenging for large language models due to attention dilution and out-of-distribution degradation. Context selection mitigates this limitation by attending to a subset of key-value cache entries, yet most methods allocate a fixed context budget throughout decoding despite highly non-uniform token-level contextual demands. To address this issue, we propose Uncertainty-Triggered Adaptive Context Allocation (UT-ACA), an inference-time framework that dynamically adjusts the context window based on token-wise uncertainty. UT-ACA learns an uncertainty detector that combines semantic embeddings with logit-based confidence while accounting for uncertainty accumulation across decoding steps. When insufficient evidence is indicated, UT-ACA selectively rolls back, expands the context window, and regenerates the token with additional support. Experiments show that UT-ACA substantially reduces average context usage while preserving generation quality in long-context settings.