Skip to main content

GLiClass

📄️ Intro

GLiClass (Generalist and Lightweight Model for Classification) is a zero-shot sequence classification model capable of categorizing text into arbitrary label sets using a bidirectional transformer encoder (BERT-like). Inspired by GLiNER's approach to named entity recognition, GLiClass adapts the single forward pass paradigm for classification tasks, providing a practical alternative to traditional cross-encoder models that require separate inference for each text-label combination, and Large Language Models (LLMs) that, while flexible, are computationally prohibitive for resource-constrained production scenarios requiring fast, scalable text classification.