TL;DR
TAILOR is presented, a token-aware inference framework that jointly optimizes offline partitioning and online routing for autoregressive transformer serving that reduces end-to-end latency and lowers OOM-induced fallback compared with static partitioning baselines, demonstrating robust inference under long-tailed token workloads.
This paper employs five lightweight machine learning models to predict query-specific latency, accuracy, resource usage, and response size for each model-size-quantization-device combination, and design a lightweight heuristic that approximates the BILP solution.
LLM serving is increasingly dominated by long and dynamic decode workloads from agents, reasoning models, and extended conversations. When bursty long-context demand exceeds deployed capacity, existing serving systems typically scale out by launching additional serving instances with model replicas. This instance-level elasticity increases KV capacity only by provisioning another full copy of the model, inheriting startup latency, memory overhead, and batch fragmentation. We present KernelFlume, a decode-centric architecture that disaggregates the stable projection/FFN path from core-attention computation: weight nodes execute dense projection/FFN kernels, while weightless attention nodes store token-range KV partitions and scale with request-state demand. To make this separation elastic, KernelFlume maintains a routing table that maps token ranges to attention-node endpoints. It updates routes at token boundaries and uses host-visible graph signals to drive pre-registered UCX endpoint communication outside the captured CUDA Graph. To preserve low per-token latency after disaggregation, KernelFlume combines query-first core-attention dispatch with inter-layer kernel pipelining, overlapping remote attention and communication with local projection/FFN work. On real GPU testbeds (intra-node A6000 and cross-node H100), under a dynamic long-context agentic workload serving Llama-3.1-8B, KernelFlume sustains flat p99 TPOTs of ~74 ms on A6000 and ~34 ms on H100, while lowering cost per million output tokens by up to 32% and 61%, respectively, relative to full-instance elastic scaling with ServerlessLLM, a state-of-the-art instance-startup method. Replaying the same trace at larger model scale in simulation projects a 56--66% cost reduction over ServerlessLLM, widening to 80--85% with cheaper heterogeneous attention-node hardware and persisting into the million-token context range.
CELLServe formalizes SLO-constrained joint resource provisioning as an optimization problem with a dedicated algorithm, and introduces an opportunistic instance merging strategy for decode phase functions to reclaim fragmented resources.
Cloud-edge speculative decoding places a lightweight draft model at an edge gateway and a higher-quality target model in the cloud, but inserts communication into every speculative block. Under a constrained uplink, candidate messages may queue while the verifier is idle. Stop-and-wait scheduling leaves edge compute underutilized; optimistic same-request runahead can waste work when a rejection or an unexpected bonus token invalidates dependent drafts. We present AsymSpec, which addresses uplink-gated verification and invalid dependent work with two corresponding mechanisms. Its asymmetric verification protocol keeps the common-path acceptance upload compact and moves richer, rejection-only correction information to the downlink. A total-variation (TV) certificate for the residual distribution determines whether a small target top-K response suffices; if not, the protocol progressively escalates through proposal-based exact recovery before falling back to the full distribution. Its confirmed-prefix pipeline exposes only independent, valid requests to the edge scheduler and lets the cloud re-batch arrived blocks, hiding verification waits when another confirmed-prefix request is ready without using same-request runahead. Across three draft-target pairs, two workloads, and three asymmetric network profiles, our end-to-end evaluation shows that AsymSpec delivers 2.82-28.03$\times$ the output-token throughput of the strongest baseline.
Latency-sensitive edge AI services must balance strict deadlines, output quality, and limited compute and energy budgets. However, static CPU provisioning wastes resources because inference cost varies substantially across inputs, model variants, and runtime conditions. We present PRISM, a prediction-guided runtime framework that jointly selects model variants and CPU allocations for containerized edge microservices. Using container-level energy monitoring and lightweight regression models, PRISM adapts each pipeline stage in place and minimizes predicted CPU-package energy under deadline, resource, and offline model-level Quality of Result (QoR) constraints. We evaluate PRISM on more than 52,000 requests in an Automatic License Plate Recognition (ALPR) pipeline with detection and recognition stages. For detection, PRISM reduces energy consumption by 36 % compared to the strongest static configuration while preserving a comparable success rate and using less than half of the average CPU allocation. For recognition, it reaches near-static-best performance with lower average CPU allocation. These results show that predictive in-place adaptation is a practical mechanism for making time-sensitive AI microservice pipelines more energy-efficient at the edge.