README: drop _80M filename note; checkpoints renamed to _120M for clarity

Remove misleading _80M filename for final_model_v2_80M.pt

Rename final_model_v2_80M.pt to final_model_v2_120M.pt (correct param count: 119.6M)

Remove misleading _80M filename for best_model_v2_80M.pt

Rename best_model_v2_80M.pt to best_model_v2_120M.pt (correct param count: 119.6M)

Remove legacy filename

Add PRAXIS-named tokenizer

Add PRAXIS-AUSCULT model card

Upload vast_train.py with huggingface_hub

Upload zovonotes_tokenizer_v2.model with huggingface_hub

Upload best_model_v2_80M.pt with huggingface_hub

Upload final_model_v2_80M.pt with huggingface_hub

initial commit