Update supernova/train.py

Update supernova/train.py

Update supernova/model.py

fixed gradient norm error

\Resolve merge conflicts in supernova/train.py\n\n- Keep improved parameter formatting for TokenChunkDataset\n- Use standard torch.cuda.amp.GradScaler initialization \n- Implement proper validation with wikitext-2 validation split\n- Maintain consistent code style and comments"

fixed training datasets

Update supernova/train.py

Update supernova/train.py

Update supernova/data.py

Update supernova/reasoning_engine.py

Update supernova/data.py

Update supernova/train.py

Update supernova/__init__.py

Update supernova/train.py

Upload folder using huggingface_hub