arXiv subjects

Kangqi Zhang

Publications and source records attributed to Kangqi Zhang.

4 recordsLinked to original sources

arXiv2026

Fengshui: Demystifying Chiplet Ecosystem and Bespoke Neural Network Accelerator Codesign

Modern ML workloads, with stringent latency and energy constraints, are increasingly hard to run efficiently on homogeneous commodity hardware. We argue that operator-level disaggregation--tailoring microarchitecture, batching, and memory hierarchy to each operator--is essential to overcome these limitations, though the resulting highly bespoke accelerators incur prohibitive Non-Recurring Engineering (NRE) costs. Chiplet-based integration amortizes NRE across applications, but choosing which chiplets to build and how to compose them into accelerators is circularly dependent--a chiplet pool's value depends on the constructed accelerators, while accelerator quality is constrained by available chiplets. This paper introduces Fengshui, a chiplet ecosystem and accelerator co-design framework that jointly optimizes chiplet pool composition and bespoke application-specific integrated circuit (BASIC) design. Fengshui constructs BASICs through operator-level disaggregation, co-exploring chiplet and memory heterogeneity, tensor fusion, and pipeline/tensor/expert parallelism with place-and-route validation for physical implementability. With just 8 strategically selected chiplets, encompassing network switches, processing-in-memory units, and accelerators with diverse microarchitectures, Fengshui-generated BASICs achieve 48.5%, 88.1%, 93.0%, and 97.8% reductions in energy, energy-cost product (EC), energy-delay product (EDP), and energy-delay-cost product (EDPC) over homogeneous accelerators, while scoring within 4.1% of unconstrained heterogeneous designs across diverse neural networks. For datacenter MoE and dense LLM serving, Fengshui reduces prefill energy and EC by up to 16.8% and 28.7%, respectively; for edge autonomous vehicle perception, it achieves 12.0% energy and 23.6% EC reductions under real-time latency constraints.

cs.AR↗

arXiv2026

Agents' Last Exam

Recent AI systems have achieved strong results on a wide range of benchmarks, yet these gains have not translated into economically meaningful deployment across many professional domains. We argue that this gap is largely an evaluation problem: widely used benchmarks lack sustained performance measurement on real and economically valuable workflows. This paper introduces Agents' Last Exam (ALE), a benchmark designed to evaluate AI agents on long horizon, economically valuable, real world tasks with verifiable outcomes. Developed in collaboration with 250+ industry experts, ALE covers non-physical industries defined with reference to O*NET / SOC 2018 (the U.S. federal occupational taxonomy). It is organized around a task taxonomy with 55 sub fields grouped into 13 industry clusters covering 1K+ tasks. Current results show that the hardest tier remains far from saturated: across mainstream harness and backbone configurations, the average full pass rate is below 1%. ALE is designed as a living benchmark: its task pool grows continuously as new workflows and industries are onboarded. More broadly, ALE is intended not merely as another leaderboard, but as an instrument for closing the gap between benchmark success and GDP relevant impact.

cs.AI↗

arXiv2026

Double-P: Hierarchical Top-P Sparse Attention for Long-Context LLMs

As long-context inference becomes central to large language models (LLMs), attention over growing key-value caches emerges as a dominant decoding bottleneck, motivating sparse attention for scalable inference. Fixed-budget top-k sparse attention cannot adapt to heterogeneous attention distributions across heads and layers, whereas top-p sparse attention directly preserves attention mass and provides stronger accuracy guarantees. Existing top-p methods, however, fail to jointly optimize top-p accuracy, selection overhead, and sparse attention cost, which limits their overall efficiency. We present Double-P, a hierarchical sparse attention framework that optimizes all three stages. Double-P first performs coarse-grained top-p estimation at the cluster level using size-weighted centroids, then adaptively refines computation through a second top-p stage that allocates token-level attention only when needed. Across long-context benchmarks, Double-P consistently achieves near-zero accuracy drop, reducing attention computation overhead by up to 1.8x and delivers up to 1.3x end-to-end decoding speedup over state-of-the-art fixed-budget sparse attention methods.

cs.LG↗

arXiv2025

Mozart: A Chiplet Ecosystem-Accelerator Codesign Framework for Composable Bespoke Application Specific Integrated Circuits

Modern AI acceleration faces a fundamental challenge: conventional assumptions about memory requirements, batching effectiveness, and latency-throughput tradeoffs are systemwide generalizations that ignore the heterogeneous computational patterns of individual neural network operators. However, going towards network-level customization and operator-level heterogeneity incur substantial Non-Recurring Engineering (NRE) costs. While chiplet-based approaches have been proposed to amortize NRE costs, reuse opportunities remain limited without carefully identifying which chiplets are truly necessary. This paper introduces Mozart, a chiplet ecosystem and accelerator codesign framework that systematically constructs low cost bespoke application-specific integrated circuits (BASICs). BASICs leverage operator-level disaggregation to explore chiplet and memory heterogeneity, tensor fusion, and tensor parallelism, with place-and-route validation ensuring physical implementability. The framework also enables constraint-aware system-level optimization across deployment contexts ranging from datacenter inference serving to edge computing in autonomous vehicles. The evaluation confirms that with just 8 strategically selected chiplets, Mozart-generated composite BASICs achieve 43.5%, 25.4%, 67.7%, and 78.8% reductions in energy, energy-cost product, energy-delay product (EDP), and energy-delay-cost product compared to traditional homogeneous accelerators. For datacenter LLM serving, Mozart achieves 15-19% energy reduction and 35-39% energy-cost improvement. In speculative decoding, Mozart delivers throughput improvements of 24.6-58.6% while reducing energy consumption by 38.6-45.6%. For autonomous vehicle perception, Mozart reduces energy-cost by 25.54% and energy by 10.53% under real-time constraints.

cs.AR↗