arXiv subjects

Pol G. Recasens

Publications and source records attributed to Pol G. Recasens.

6 recordsLinked to original sources

arXiv2026

SLIM: Saturation-Aware Lightweight Performance Modeling for LLM Serving

Large language model (LLM) serving commonly increases batch size to improve throughput, but performance eventually reaches a deployment-dependent plateau beyond which larger batches provide marginal gains while increasing latency and GPU memory consumption. Previous studies have attributed this behavior to HBM/DRAM bandwidth limitations, but the underlying causes have primarily been supported by conceptual arguments or high-level performance observations. As our first contribution, we present a detailed GPU characterization using hardware profiling techniques, demonstrating that throughput saturation originates in the attention kernels during the decode phase. Specifically, we show that their nearly constant arithmetic intensity as active-context lengths increases -not merely larger batch sizes- drives DRAM-bandwidth saturation, while the achieved compute throughput remains far below the hardware limit. Building on this analysis, we present the Batching Configuration Advisor (BCA), which selects the highest-throughput batching configuration satisfying a target latency constraint and identifies up to 55 GB of GPU memory allocation that can be avoided for the evaluated OPT models with minimal throughput loss. To enable these recommendations, we introduce SLIM (Saturation-Aware Lightweight Performance Model), a semi-analytical model that predicts LLM inference throughput and latency from analytical formulations of Transformer computation and memory traffic. Across the evaluated scenarios, SLIM outperforms representative performance-modeling baselines while successfully generalizing to previously unseen operating conditions.

cs.DC↗

arXiv2025

In-Context Bias Propagation in LLM-Based Tabular Data Generation

Large Language Models (LLMs) are increasingly used for synthetic tabular data generation through in-context learning (ICL), offering a practical solution for data augmentation in data scarce scenarios. While prior work has shown the potential of LLMs to improve downstream task performance through augmenting underrepresented groups, these benefits often assume access to a subset of unbiased in-context examples, representative of the real dataset. In real-world settings, however, data is frequently noisy and demographically skewed. In this paper, we systematically study how statistical biases within in-context examples propagate to the distribution of synthetic tabular data, showing that even mild in-context biases lead to global statistical distortions. We further introduce an adversarial scenario where a malicious contributor can inject bias into the synthetic dataset via a subset of in-context examples, ultimately compromising the fairness of downstream classifiers for a targeted and protected subgroup. Finally, we evaluate mitigation strategies based on preprocessing in-context examples, demonstrating that while such interventions can attenuate disparity, the inherent sensitivity of LLMs to adversarial prompts remains a persistent challenge. Our findings highlight a critical new vulnerability in LLM-based data generation pipelines within sensitive domains.

cs.LG↗

arXiv2025

Mind the Memory Gap: Unveiling GPU Bottlenecks in Large-Batch LLM Inference

Large language models have been widely adopted across different tasks, but their auto-regressive generation nature often leads to inefficient resource utilization during inference. While batching is commonly used to increase throughput, performance gains plateau beyond a certain batch size, especially with smaller models, a phenomenon that existing literature typically explains as a shift to the compute-bound regime. In this paper, through an in-depth GPU-level analysis, we reveal that large-batch inference remains memory-bound, with most GPU compute capabilities underutilized due to DRAM bandwidth saturation as the primary bottleneck. To address this, we propose a Batching Configuration Advisor (BCA) that optimizes memory allocation, reducing GPU memory requirements with minimal impact on throughput. The freed memory and underutilized GPU compute capabilities can then be leveraged by concurrent workloads. Specifically, we use model replication to improve serving throughput and GPU utilization. Our findings challenge conventional assumptions about LLM inference, offering new insights and practical strategies for improving resource utilization, particularly for smaller language models. The code is publicly available at https://github.com/FerranAgulloLopez/vLLMBatchingMemoryGap.

cs.DC↗

arXiv2024

FRIDA: Free-Rider Detection using Privacy Attacks

Federated learning is increasingly popular as it enables multiple parties with limited datasets and resources to train a machine learning model collaboratively. However, similar to other collaborative systems, federated learning is vulnerable to free-riders - participants who benefit from the global model without contributing. Free-riders compromise the integrity of the learning process and slow down the convergence of the global model, resulting in increased costs for honest participants. To address this challenge, we propose FRIDA: free-rider detection using privacy attacks. Instead of focusing on implicit effects of free-riding, FRIDA utilizes membership and property inference attacks to directly infer evidence of genuine client training. Our extensive evaluation demonstrates that FRIDA is effective across a wide range of scenarios.

cs.LG↗

arXiv2024

Towards Pareto Optimal Throughput in Small Language Model Serving

Large language models (LLMs) have revolutionized the state-of-the-art of many different natural language processing tasks. Although serving LLMs is computationally and memory demanding, the rise of Small Language Models (SLMs) offers new opportunities for resource-constrained users, who now are able to serve small models with cutting-edge performance. In this paper, we present a set of experiments designed to benchmark SLM inference at performance and energy levels. Our analysis provides a new perspective in serving, highlighting that the small memory footprint of SLMs allows for reaching the Pareto-optimal throughput within the resource capacity of a single accelerator. In this regard, we present an initial set of findings demonstrating how model replication can effectively improve resource utilization for serving SLMs.

cs.CL↗

arXiv2023

On Masked Pre-training and the Marginal Likelihood

Masked pre-training removes random input dimensions and learns a model that can predict the missing values. Empirical results indicate that this intuitive form of self-supervised learning yields models that generalize very well to new domains. A theoretical understanding is, however, lacking. This paper shows that masked pre-training with a suitable cumulative scoring function corresponds to maximizing the model's marginal likelihood, which is de facto the Bayesian model selection measure of generalization. Beyond shedding light on the success of masked pre-training, this insight also suggests that Bayesian models can be trained with appropriately designed self-supervision. Empirically, we confirm the developed theory and explore the main learning principles of masked pre-training in large language models.

stat.ML↗