v0.3: Fix training hang — precompute tensors, num_workers=0, torch.compile, warmup batch, timing diagnostics

Upload musemorphic/__init__.py

FIX: Use vectorized span masking in notebook (no Python loop over batch)"

PERF FIX: Replace spectral_norm with weight_norm (~50x faster), chunked SSM scan, vectorized masking

Upload requirements.txt

Upload generate.py

Upload MuseMorphic_Training.ipynb

Upload musemorphic/data_pipeline.py

Upload musemorphic/__init__.py

Upload musemorphic/train.py

Upload musemorphic/tokenizer.py

Upload musemorphic/model.py

Upload README.md

initial commit