activation probessparse autoencoderspublished Nov 20, 2026 · arXiv:2611.24619
GateWatch: Low-Overhead LLM Monitoring with Boolean Circuits
KV cache offloadinglong-context inferencepublished Nov 6, 2026 · arXiv:2611.07542
FlashOffloading: Low-Latency KV Cache Offloading for Long-Context LLM Serving
LLM inferencedeterministic inferencepublished Nov 6, 2026 · arXiv:2611.06977
BlockLock: Reproducible LLM Inference with Fixed Batch Geometry
mixture-of-expertsLLM servingpublished Nov 5, 2026 · arXiv:2611.06147
SafeStart: Mitigating the Overlooked Safety Cost of Active Expert Reduction in MoE LLM Serving
adaptive speculative decodingmodel-based draftingpublished Nov 5, 2026 · arXiv:2611.05611
Hybrid Spark: Adaptive Speculative Decoding That Lets the Content Choose the Drafter
coding agentsmulti-agent systemspublished Oct 28, 2026 · arXiv:2610.36848
Evolving Agent Teams