SpanMarker is a framework for training Named Entity Recognition models based on the PL-Marker paper. It is built on top of Hugging Face Transformers, leveraging pre-trained encoder models such as BERT and RoBERTa for entity span detection. The library supports common annotation schemes (IOB, IO2, BIOES, BILOU), integrates with the Hugging Face Hub for model sharing and inference, and provides features like hyperparameter optimization, mixed precision training, and 8-bit inference.