Beyond the Vote: When Do LLMs Add Value to Retrieval-Based Tabular Classification?

Large language models (LLMs) perform table-centric prediction through in-context learning, making demonstration selection critical to performance. Existing retrieval methods prioritize similarity to the query, but similar demonstrations often reinforce the model's likely prediction rather than reveal the distinctions needed for difficult decisions. We propose EdgeLM, a retrieval framework that instead selects edge evidence, demonstrations that are both relevant to the query and informative about the decision boundary. EdgeLM retrieves two complementary forms of edge evidence by selecting data edges, nearby examples with different ground-truth labels, and model edges, similar examples previously misclassified by the deployed model. EdgeLM requires neither model retraining nor task-specific engineering. Across five data wrangling tasks, fifteen datasets, and five open-weight and proprietary LLMs, EdgeLM consistently achieves the best or near-best performance in every setting, while ablations show that the two forms of edge evidence provide complementary benefits. Our code and datasets are publicly available at https://github.com/soroushomidvar/EdgeLM.

Foundational models for tabular data have made significant progress in recent years, with TabICLv2 reporting state-of-the-art performance on several tabular classification tasks. However, full-context tabular ICL still suffers from attention cost that grows with the training-context size, which limits its ability to handle large datasets efficiently. Localized TabICLv2 introduces a method that reduces the inference cost of TabICLv2 by retrieving only the k nearest training neighbours for each test point, measured by similarity in the model's Stage 2 row-representation space, rather than using the full training context. This requires no architectural changes, and we show that accuracy retention can be improved through additional Stage 2 and Stage 3 fine-tuning. On TabArena classification tasks, the fine-tuned localized model retains 98.64% of Full TabICLv2 accuracy and it achieves a median 2.18$\times$ speedup in batch inference, and reaches approximately 249$\times$ median speedup in the single-query serving setting.

Every tabular classification task requires sufficient labeled data. When labels are scarce, practitioners can collect more examples, encode domain expertise into features, or accept lower performance. Large language models offer an alternative by leveraging knowledge acquired during pretraining. This raises three questions: (i) can zero-shot inference reduce the need for labeled data; (ii) can a richer prompt reduce the need for more examples; and (iii) can encoded domain knowledge compensate for missing data? To answer each under explicit, testable conditions, we introduce TabLens, a benchmark of 91 real-world and synthetic datasets across eight domains, plus eight industrial datasets with free-text columns. First, below a complexity threshold estimable from a dataset's intrinsic dimensionality before running an LLM, zero-shot inference matches the accuracy of up to about 16 labeled examples. This holds whether the model generates an answer or, more cheaply, scores candidate labels by log-probabilities, which loses almost no accuracy except for the smallest models. Above the threshold, the advantage disappears. On datasets with free-text columns, however, zero-shot LLMs outperform 16-shot TabPFN-3 Plus without manual feature engineering. Second, a richer prompt reduces the number of required demonstrations only when it includes both real column names and a plain-language task description; providing only one can be worse than neither, particularly at higher shot counts. Third, on LLM-synthetic datasets, explicit if--then expert rules help only when the prompt provides enough task context to interpret them, and even then, they substitute for at most about 16 real examples. Together, these results establish testable conditions under which LLMs can reduce data-collection and feature-engineering requirements for tabular classification.