There is rapidly growing interest in using Bayesian optimization to tune model and inference hyperparameters for machine learning algorithms that take a long time to run.

With the rapid adoption of Large Language Models (LLMs), LLM-adapters have become increasingly common, providing lightweight specialization of large-scale models. Serving hundreds or thousands of these adapters on a single GPU allows request aggregation, increasing throughput, but may also cause request starvation if GPU memory limits are exceeded. To address this issue, this study focuses on determining the joint configuration of concurrent and parallel adapters that maximizes GPU throughput without inducing starvation, given heterogeneous adapter and traffic properties. We propose a data-driven ML approach leveraging interpretable models to tackle this caching problem and introduce the first Digital Twin capable of reproducing an LLM-adapter serving system, enabling efficient training data generation. Experiments with the vLLM framework and LoRA adapters show that the Digital Twin reproduces throughput within 5.1% of real results, while the ML approach predicts optimal numbers of concurrent and parallel adapters with an error of at most 7.2% under heterogeneous, real-world workloads. The code is publicly available at https://github.com/FerranAgulloLopez/GPULLMAdapterOptimization.

Input data preprocessing is a common bottleneck when concurrently training multimedia machine learning (ML) models in modern systems. To alleviate these bottlenecks and reduce the training time for concurrent jobs, we present Seneca, a data loading system that optimizes cache partitioning and data sampling for the data storage and ingestion (DSI) pipeline. The design of Seneca contains two key techniques. First, Seneca uses a performance model for the data pipeline to optimally partition the cache for three different forms of data (encoded, decoded, and augmented). Second, Seneca opportunistically serves cached data over uncached ones during random batch sampling so that concurrent jobs benefit from each other. We implement Seneca by modifying PyTorch and demonstrate its effectiveness by comparing it against several state-of-the-art caching systems for DNN training. Seneca reduces the makespan by 45.23% compared to PyTorch and increases data processing throughput by up to 3.45x compared to the next best dataloader.

ONERA - DTIS Universit e de Toulouse Toulouse, France Abstract--Accurate aircraft trajectory prediction is critical for air traffic management, airline operations, and environmental assessment. This paper introduces NODE-FDM, a Neural Ordinary Differential Equations-based Flight Dynamics Model trained on Quick Access Recorder (QAR) data. By combining analytical kinematic relations with data-driven components, NODE-FDM achieves a more accurate reproduction of recorded trajectories than state-of-the-art models such as a BADA-based trajectory generation methodology (BADA4 performance model combined with trajectory control routines), particularly in the descent phase of the flight. The analysis demonstrates marked improvements across altitude, speed, and mass dynamics. Despite current limitations, including limited physical constraints and the limited availability of QAR data, the results demonstrate the potential of physics-informed neural ordinary differential equations as a high-fidelity, data-driven approach to aircraft performance modelling. Future work will extend the framework to incorporate a full modelling of the lateral dynamics of the aircraft. Accurate aircraft trajectory prediction is essential to air traffic management (A TM), airline operations, and aviation environmental assessment.

In the hardware design space exploration process, it is critical to optimize both hardware parameters and algorithm-to-hardware mappings. Previous work has largely approached this simultaneous optimization problem by separately exploring the hardware design space and the mapspace - both individually large and highly nonconvex spaces - independently. The resulting combinatorial explosion has created significant difficulties for optimizers. In this paper, we introduce DOSA, which consists of differentiable performance models and a gradient descent-based optimization technique to simultaneously explore both spaces and identify high-performing design points. Experimental results demonstrate that DOSA outperforms random search and Bayesian optimization by 2.80x and 12.59x, respectively, in improving DNN model energy-delay product, given a similar number of samples. We also demonstrate the modularity and flexibility of DOSA by augmenting our analytical model with a learned model, allowing us to optimize buffer sizes and mappings of a real DNN accelerator and attain a 1.82x improvement in energy-delay product.

Large Language Models (LLMs) have revolutionized Artificial Intelligence (AI) with significant advancements such as OpenAI's ChatGPT, Meta's Llama, and Databricks' DBRX. This paper addresses the cost and scalability challenges encountered when constructing private LLM systems for personal or small group services, as aimed by Apple Intelligence. A Mac Studio cluster with Apple's M2 Ultra chips is established as a cost-efficient solution to host and accelerate the pretrained DBRX model with the Mixture-of-Experts (MoE) architecture. Our performance analysis reveal that parallel execution of the model's experts across two to four machine nodes significantly reduces inference time. We find that computation time for the experts is comparable to the communication time for exchanging their outputs, emphasizing the importance of network latency over bandwidth. We also observe significant management overhead due to Apple software stack's memory management logic. Based on these findings, we develop optimization schemes to eliminate the memory management overhead. As a result, the Mac Studio cluster is 1.15 times more cost-efficient than the state-of-the-art AI supercomputer with NVIDIA H100 GPUs. In addition, we construct a performance model to estimate system performance under varying configurations, and the model provides valuable insights for designing private LLM systems.

Mixture of Experts (MoE) LLMs, characterized by their sparse activation patterns, offer a promising approach to scaling language models while avoiding proportionally increasing the inference cost. However, their large parameter sizes present deployment challenges in resource-constrained environments with limited GPU memory capacity, as GPU memory is often insufficient to accommodate the full set of model weights. Consequently, typical deployments rely on CPU-GPU hybrid execution: the GPU handles compute-intensive GEMM operations, while the CPU processes the relatively lightweight attention mechanism. This setup introduces a key challenge: how to effectively optimize resource utilization across CPU and GPU? Prior work has designed system optimizations based on performance models with limited scope. Specifically, such models do not capture the complex interactions between hardware properties and system execution mechanisms. Therefore, previous approaches neither identify nor achieve the hardware limit. This paper presents MoE-Lens, a high-throughput MoE LLM inference system designed through holistic performance modeling for resource-constrained environments. Our performance model thoroughly analyzes various fundamental system components, including CPU memory capacity, GPU compute power, and workload characteristics, to understand the theoretical performance upper bound of MoE inference. Furthermore, it captures the system execution mechanisms to identify the key hardware bottlenecks and accurately predict the achievable throughput. Informed by our performance model, MoE-Lens introduces an inference system approaching hardware limits. Evaluated on diverse MoE models and datasets, MoE-Lens outperforms the state-of-the-art solution by 4.6x on average (up to 25.5x), with our theoretical model predicting performance with an average 94% accuracy.

Speculative decoding (SD) has been shown to reduce the latency of autoregressive decoding (AD) by 2-3x for small batch sizes. However, increasing throughput and therefore reducing the cost per token requires decoding with large batch sizes. Recent work shows that SD can accelerate decoding with large batch sizes too if the context is sufficiently long and the draft model's KV cache is sparse. We introduce SPIRe, a draft model that combines static sparse attention, pruned initialization, and feedback memory to increase the modeled throughput of speculative decoding by over 100% compared to speculation with a much smaller draft model and by over 35% compared to the strong baseline of sparse self-speculation. Our approach is particularly effective when context lengths vary significantly across requests.