Search papers, labs, and topics across Lattice.
This paper introduces Texo, a formula recognition model with only 20M parameters, designed for efficient deployment. Texo achieves comparable performance to larger state-of-the-art models like UniMERNet-T and PPFormulaNet-S through attentive design, distillation, and transfer learning of the vocabulary and tokenizer. The reduced model size allows for real-time inference on consumer hardware and in-browser applications, demonstrated via a developed web application.
Formula recognition gets a lot more practical: Texo shrinks state-of-the-art performance into a 20M-parameter model deployable on consumer hardware.
In this paper we present Texo, a minimalist yet highperformance formula recognition model that contains only 20 million parameters. By attentive design, distillation and transfer of the vocabulary and the tokenizer, Texo achieves comparable performance to state-of-the-art models such as UniMERNet-T and PPFormulaNet-S, while reducing the model size by 80% and 65%, respectively. This enables real-time inference on consumer-grade hardware and even in-browser deployment. We also developed a web application to demonstrate the model capabilities and facilitate its usage for end users.
