arXiv subjects

Ata Fatahi Baarzi

Publications and source records attributed to Ata Fatahi Baarzi.

3 recordsLinked to original sources

arXiv2026

LLM Query Scheduling with Prefix Reuse and Latency Constraints

The efficient deployment of large language models (LLMs) in online settings requires optimizing inference performance under stringent latency constraints, particularly the time-to-first-token (TTFT) and time-per-output-token (TPOT). This paper focuses on the query scheduling problem for LLM inference with prefix reuse, a technique that leverages shared prefixes across queries to reduce computational overhead. Our work reveals previously unknown limitations of the existing first-come-first-serve (FCFS) and longest-prefix-match (LPM) scheduling strategies with respect to satisfying latency constraints. We present a formal theoretical framework for LLM query scheduling under RadixAttention, a prefix reuse mechanism that stores and reuses intermediate representations in a radix tree structure. Our analysis establishes the NP-hardness of the scheduling problem with prefix reuse under TTFT constraints and proposes a novel scheduling algorithm, $k$-LPM, which generalizes existing methods by balancing prefix reuse and fairness in query processing. Theoretical guarantees demonstrate that $k$-LPM achieves improved TTFT performance under realistic traffic patterns captured by a data generative model. Empirical evaluations in a realistic serving setting validates our findings, showing significant reductions in P99 TTFT compared to baseline methods.

cs.DS↗

arXiv2025

Scaling Up Efficient Small Language Models Serving and Deployment for Semantic Job Search

Large Language Models (LLMs) have demonstrated impressive quality when applied to predictive tasks such as relevance ranking and semantic search. However, deployment of such LLMs remains prohibitively expensive for industry applications with strict latency and throughput requirements. In this work, we present lessons and efficiency insights from developing a purely text-based decoder-only Small Language Model (SLM) for a semantic search application at LinkedIn. Particularly, we discuss model compression techniques such as pruning that allow us to reduce the model size by up to $40\%$ while maintaining the accuracy. Additionally, we present context compression techniques that allow us to reduce the input context length by up to $10$x with minimal loss of accuracy. Finally, we present practical lessons from optimizing the serving infrastructure for deploying such a system on GPUs at scale, serving millions of requests per second. Taken together, this allows us to increase our system's throughput by $10$x in a real-world deployment, while meeting our quality bar.

cs.IR↗

arXiv2023

Cluster Resource Management for Dynamic Workloads by Online Optimization

Over the past ten years, many different approaches have been proposed for different aspects of the problem of resources management for long running, dynamic and diverse workloads such as processing query streams or distributed deep learning. Particularly for applications consisting of containerized microservices, researchers have attempted to address problems of dynamic selection of, for example: types and quantities of virtualized services (e.g., IaaS/VMs), vertical and horizontal scaling of different microservices, assigning microservices to VMs, task scheduling, or some combination thereof. In this context, we argue that frameworks like simulated annealing are highly suitable for online navigation of trade-offs between performance (SLO) and cost, particularly when the complex workloads and cloud-service offerings vary over time. Based on a macroscopic objective that combines both performance and cost terms, annealing facilitates light-weight and coherent policies of exploration and exploitation. In this paper, we first give some background on simulated annealing and then experimentally demonstrate its usefulness for different case studies, including service selection for both a single type of workload (e.g., distributed deep learning) and a mixture of workload types (exploring a partially categorical set of options), and container sizing for microservice benchmarks. We conclude with a discussion of how the basic annealing platform can be applied to other resource-management problems, hybridized with other methods, and accommodate user-specified rules of thumb.

cs.PF↗