LLMTabBench: Evaluating LLMs on Binary Tabular Classification From Zero to Few Shots

Supervised classification for tabular data remains a core machine learning task, yet its reliance on large labeled datasets limits applicability in data-scarce domains. For such few-shot scenarios, specialized methods like TabPFN - a state-of-the-art Prior-Data Fitted Network - have set a high standard by leveraging large-scale synthetic pretraining, though they still require a context of labeled examples to function. In contrast, Large Language Models (LLMs) could offer a more flexible alternative via zero- and few-shot in-context learning directly from task descriptions, but their performance on tabular data remains inconsistent and poorly understood. We introduce LLMTabBench, a benchmark designed to systematically evaluate LLMs for tabular classification under data-scarce conditions. LLMTabBench explicitly probes (i) how LLM prior knowledge interacts with in-context information (task descriptions and few-shot examples), and (ii) how model performance scales with increasing data complexity, using both real-world and controlled synthetic datasets. Our findings include: (1) LLMs are highly competitive in zero-shot settings and can outperform alternative models, even when those models have access to few-shot examples; (2) incorporating additional few-shot examples can conflict with LLM prior knowledge, limiting or even degrading performance; and (3) there is a data complexity threshold beyond which LLMs' performance declines and few-shot examples become less effective. Together, these findings reveal fundamental constraints of in-context learning for tabular data and provide practical guidance for deploying LLMs in low-data regimes.

Lightweight error mitigation strategies for post-training N:M activation sparsity in LLMs

The demand for efficient large language model (LLM) inference has intensified the focus on sparsification techniques. While semi-structured (N:M) pruning is well-established for weights, its application to activation pruning remains underexplored despite its potential for dynamic, input-adaptive compression and reductions in I/O overhead. This work presents a comprehensive analysis of methods for post-training N:M activation pruning in LLMs. Across multiple LLMs, we demonstrate that pruning activations enables superior preservation of generative capabilities compared to weight pruning at equivalent sparsity levels. We evaluate lightweight, plug-and-play error mitigation techniques and pruning criteria, establishing strong hardware-friendly baselines that require minimal calibration. Furthermore, we explore sparsity patterns beyond NVIDIA's standard 2:4, showing that the 16:32 pattern achieves performance nearly on par with unstructured sparsity. However, considering the trade-off between flexibility and hardware implementation complexity, we focus on the 8:16 pattern as a superior candidate. Our findings provide both effective practical methods for activation pruning and a motivation for future hardware to support more flexible sparsity patterns. Our code is available https://anonymous.4open.science/r/Structured-Sparse-Activations-Inference-EC3C/README.md .

Tsururu: A Python-based Time Series Forecasting Strategies Library

While current time series research focuses on developing new models, crucial questions of selecting an optimal approach for training such models are underexplored. Tsururu, a Python library introduced in this paper, bridges SoTA research and industry by enabling flexible combinations of global and multivariate approaches and multi-step-ahead forecasting strategies. It also enables seamless integration with various forecasting models. Available at https://github.com/sb-ai-lab/tsururu .

Towards Foundation Time Series Model: To Synthesize Or Not To Synthesize?

The industry is rich in cases when we are required to make forecasting for large amounts of time series at once. However, we might be in a situation where we can not afford to train a separate model for each of them. Such issue in time series modeling remains without due attention. The remedy for this setting is the establishment of a foundation model. Such a model is expected to work in zero-shot and few-shot regimes. However, what should we take as a training dataset for such kind of model? Witnessing the benefits from the enrichment of NLP datasets with artificially-generated data, we might want to adopt their experience for time series. In contrast to natural language, the process of generation of synthetic time series data is even more favorable because it provides full control of series patterns, time horizons, and number of samples. In this work, we consider the essential question if it is advantageous to train a foundation model on synthetic data or it is better to utilize only a limited number of real-life examples. Our experiments are conducted only for regular time series and speak in favor of leveraging solely the real time series. Moreover, the choice of the proper source dataset strongly influences the performance during inference. When provided access even to a limited quantity of short time series data, employing it within a supervised framework yields more favorable results than training on a larger volume of synthetic data. The code for our experiments is publicly available on Github \url{https://github.com/sb-ai-lab/synthesize_or_not}.