Add download-verified inference example

Patch tokenizer_class -> PreTrainedTokenizerFast for AutoTokenizer compat

Add tokenizer.json at repo root (was only in checkpoint-300000/)

Remove redundant Training section (merged into Source Code)

Add MolCrawl GitHub source code link to model card

Remove private MolCrawl-HFuploader URL from dataset references

Add model card

Upload ChemBERTa-2 large (300k steps, compounds SMILES)

initial commit