Title: ELF: A Family of Encoder-Free ECG-Language Models

Abstract: By adapting recent breakthroughs in Multimodal Large Language Models (MLLMs) to the automated interpretation of electrocardiograms (ECGs), ECG-Language Models (ELMs) have emerged as a promising field. Yet, the majority of current ELMs adopt design paradigms from Vision-Language Models (VLMs), necessitating the use of pretrained ECG encoders. This reliance adds significant complexity to both architecture and training procedures. Drawing inspiration from encoder-free VLMs, we present ELF, a suite comprising three distinct encoder-free ELM architectures. Despite their streamlined structures and simplified training workflows, these models demonstrate performance that rivals, and frequently surpasses, existing state-of-the-art ELMs on two benchmark datasets. The associated code and data are publicly accessible at github.com/ELM-Research/ECG-Language-Models.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC