BehaviorBox: Automated Discovery of Fine-Grained Performance Differences Between Language Models
This work addresses the problem of brittle and vague language model evaluation for researchers and practitioners by providing an automated tool to discover interpretable performance differences, though it is incremental as it builds on existing embedding and comparison techniques.
The authors tackled the challenge of automatically finding fine-grained performance differences between language models by proposing BehaviorBox, a method that uses performance-aware contextual embeddings to identify specific text features where one model outperforms another, such as conditional phrases or punctuation contexts, and applied it to compare models of varying sizes and families.
Language model evaluation is a daunting task: prompts are brittle, corpus-level perplexities are vague, and the choice of benchmarks are endless. Finding examples that show meaningful, generalizable differences between two LMs is crucial to understanding where one model succeeds and another fails. Can this process be done automatically? In this work, we propose methodology for automated comparison of language models that uses performance-aware contextual embeddings to find fine-grained features of text where one LM outperforms another. Our method, which we name BehaviorBox, extracts coherent features that demonstrate differences with respect to the ease of generation between two LMs. Specifically, BehaviorBox finds features that describe groups of words in fine-grained contexts, such as "conditional 'were' in the phrase 'if you were'" and "exclamation marks after emotional statements", where one model outperforms another within a particular datatset. We apply BehaviorBox to compare models that vary in size, model family, and post-training, and enumerate insights into specific contexts that illustrate meaningful differences in performance which cannot be found by measures such as corpus-level perplexity alone.