TTL-based KV-cache pinning that uses predicted tool-call durations and queueing-delay costs cuts agent job completion time by up to 8x.
Mamba: Linear-time sequence modeling with selective state spaces
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.OS 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Continuum: Efficient and Robust Multi-Turn LLM Agent Scheduling with KV Cache Time-to-Live
TTL-based KV-cache pinning that uses predicted tool-call durations and queueing-delay costs cuts agent job completion time by up to 8x.