Francesco Antici

Francesco Antici, Andrea Borghesi, Zeynep Kiziltan

Modern High Performance Computing (HPC) systems are complex machines, with major impacts on economy and society. Along with their computational capability, their energy consumption is also steadily raising, representing a critical issue given the ongoing environmental and energetic crisis. Therefore, developing strategies to optimize HPC system management has paramount importance, both to guarantee top-tier performance and to improve energy efficiency. One strategy is to act at the workload level and highlight the jobs that are most likely to fail, prior to their execution on the system. Jobs failing during their execution unnecessarily occupy resources which could delay other jobs, adversely affecting the system performance and energy consumption. In this paper, we study job failure prediction at submit-time using classical machine learning algorithms. Our novelty lies in (i) the combination of these algorithms with Natural Language Processing (NLP) tools to represent jobs and (ii) the design of the approach to work in an online fashion in a real system. The study is based on a dataset extracted from a production machine hosted at the HPC centre CINECA in Italy. Experimental results show that our approach is promising.

Hiari Pizzini Cavagna, Andrea Proia, Giacomo Madella et al.

Large Language Models (LLMs) inference is central to modern AI applications, dominating worldwide datacenter workloads, making it critical to predict its energy footprint. Existing approaches estimate energy consumption as a simple linear function of input and output sequence. However, by analyzing the autoregressive structure of Transformers, which implies a fundamentally non-linear relationship between input and output sequence lengths and energy consumption, we demonstrate the existence of a generation energy minima. Peak efficiency occurs with short-to-moderate inputs and medium-length outputs, while efficiency drops sharply for long inputs or very short outputs. Consequently, we propose SweetSpot, an analytical model derived from the computational and memory-access complexity of the Transformer architecture, which accurately characterizes the efficiency curve as a function of input and output lengths. To assess accuracy, we measure energy consumption using TensorRT-LLM on NVIDIA H100 GPUs across a diverse set of LLMs ranging from 1B to 9B parameters, including OPT, LLaMA, Gemma, Falcon, Qwen2, and Granite. We test input and output lengths from 64 to 4096 tokens and achieve a mean MAPE of 1.79%. Our results show that aligning sequence lengths with these efficiency "sweet spots" reduce energy usage, up to 33.41x, enabling informed truncation, summarization, and adaptive generation strategies in production systems.

Francesco Antici, Andrea Galassi, Federico Ruggeri et al.

We develop novel annotation guidelines for sentence-level subjectivity detection, which are not limited to language-specific cues. We use our guidelines to collect NewsSD-ENG, a corpus of 638 objective and 411 subjective sentences extracted from English news articles on controversial topics. Our corpus paves the way for subjectivity detection in English and across other languages without relying on language-specific tools, such as lexicons or machine translation. We evaluate state-of-the-art multilingual transformer-based models on the task in mono-, multi-, and cross-language settings. For this purpose, we re-annotate an existing Italian corpus. We observe that models trained in the multilingual setting achieve the best performance on the task.