Luxical: High-Speed Lexical-Dense Text Embeddings
This work addresses the problem of efficient web-scale text organization for AI researchers and practitioners, offering a practical tool that balances speed and quality, though it is incremental as it builds on existing methods like TF-IDF and knowledge distillation.
The paper tackles the trade-off between speed and flexibility in text embedding models by introducing Luxical, a library that combines sparse TF-IDF features with a small neural network to approximate transformer embeddings at a fraction of the cost, achieving speedups of 3x to 100x while matching neural baseline quality in retrieval and data curation tasks.
Frontier language model quality increasingly hinges on our ability to organize web-scale text corpora for training. Today's dominant tools trade off speed and flexibility: lexical classifiers (e.g., FastText) are fast but limited to producing classification output scores, while the vector-valued outputs of transformer text embedding models flexibly support numerous workflows (e.g., clustering, classification, and retrieval) but are computationally expensive to produce. We introduce Luxical, a library for high-speed "lexical-dense" text embeddings that aims to recover the best properties of both approaches for web-scale text organization. Luxical combines sparse TF--IDF features, a small ReLU network, and a knowledge distillation training regimen to approximate large transformer embedding models at a fraction of their operational cost. In this technical report, we describe the Luxical architecture and training objective and evaluate a concrete Luxical model in two disparate applications: a targeted webcrawl document retrieval test and an end-to-end language model data curation task grounded in text classification. In these tasks we demonstrate speedups ranging from 3x to 100x over varying-sized neural baselines, and comparable to FastText model inference during the data curation task. On these evaluations, the tested Luxical model illustrates favorable compute/quality trade-offs for large-scale text organization, matching the quality of neural baselines. Luxical is available as open-source software at https://github.com/datologyai/luxical.