Add `"add_prefix_space": true,`; this allows for much stronger token-level performance (e.g. NER, ColBERT)

Update tokenizer: Set lstrip=True for [MASK]

Upload ModernBERT-base model