Attention--Feed-Forward Network (FFN) Disaggregation (AFD) is emerging as a promising architecture for serving Mixture-of-Experts (MoE) language models. While existing AFD systems improve the efficiency of disaggregated execution, they leave a deployment question unanswered: under the same model, workload, time-per-output-token (TPOT) service-level objective (SLO), hardware budget, hardware catalog, and runtime capabilities, does AFD provide higher throughput than the best collocated deployment?
arXiv:2608. 04502v1 Announce Type: cross Abstract: Attention--Feed-Forward Network (FFN) Disaggregation (AFD) is emerging as a promising architecture for serving Mixture-of-Experts (MoE) language models.
arXiv:2601. 21351v4 Announce Type: replace-cross Abstract: Attentio-FFN disaggregation (AFD) is an emerging architecture for LLM decoding that separates state-heavy, KV-cache-dominated Attention computation from stateless, compute-intensive FFN computation, connected by per-step communication.
arXiv:2609.14237v1 Announce Type: cross Abstract: LLM serving systems increasingly disaggregate inference into finer-grained stages, with recent approaches separating attention from FFN or MoE execut...
The paper argues that AI deployment performance depends on interactions among compression, compiler transformations, and serving policies rather than just model architecture. It introduces a three‑layer taxonomy—model‑level techniques, compiler transformations, and system policies—and frames deployment as a constrained multi‑objective optimization problem over accuracy, latency, throughput, memory footprint, and energy. The authors propose an evidence protocol for comparable benchmarking and synthesize data from edge and data‑center platforms to show that cross‑layer interactions drive deployment outcomes, concluding with a constraint‑aware selection procedure and open research problems.
HoliBench is a modular benchmarking and deployment toolkit that jointly measures accuracy, latency, and energy for foundation models across a wide range of devices, from single-board computers to GPU servers. It provides a platform abstraction layer that calibrates cross-device measurements and supports multiple model modalities, inference engines, and quantization levels. Using HoliBench, the authors evaluated 20 models on 7 device types, revealing tradeoffs such as limited latency gains from quantization on low‑precision hardware and diminishing accuracy returns relative to energy consumption, while also showing that single-model profiles can predict multi-model pipeline performance within a few percent.
arXiv:2606. 15004v1 Announce Type: cross Abstract: Deploying neural networks on low-power microcontrollers (MCUs) requires selecting model architectures under tight memory, latency, and energy constraints.
arXiv:2605. 21312v2 Announce Type: replace-cross Abstract: Modern LLM serving is no longer homogeneous or monolithic.
arXiv:2607. 05876v1 Announce Type: cross Abstract: LLM serving optimization typically benchmarks many configurations and reaches for heavy profilers when latency targets are missed.
arXiv:2607. 24434v1 Announce Type: cross Abstract: Large Mixture-of-Experts (MoE) language models are attractive for end-device deployment because only a small subset of experts is active per token, but their routed expert weights often exceed accelerator memory.
arXiv:2608. 11361v1 Announce Type: new Abstract: Tokenizer vocabulary size is a foundational design choice in large language model (LLM) infrastructure, yet it is typically fixed at training time based on convention rather than deployment analysis.
Atlas is a framework that optimizes the deployment of compound AI workflows on heterogeneous clusters by selecting execution plans that satisfy service level objectives (SLOs). It introduces MAP, a Markovian Accuracy Predictor, which estimates configuration accuracy using local conditional accuracy transitions between adjacent workflow stages, avoiding exhaustive end‑to‑end profiling. Atlas formulates plan selection as a mixed‑integer linear program, achieving near‑oracle accuracy while reducing deployment cost by up to 42% and profiling cost by up to 2.6×.
arXiv:2608. 26418v1 Announce Type: cross Abstract: Modern AI workloads and the hardware that runs them evolve on different timescales: architectural definition precedes volume silicon by years, while target workloads shift in months.