Medium training run

Fixed compiling issue, added legnth penalty, and atttempting freezing encoder layers 0-5 to lower parameters and preserve T5's langauge understanding.

Clean up codebase and fix training bugs

Add demo data for HF Space (books + news samples)

Expand datasets: GoEmotions (28 emotions), Yahoo Answers (10 topics), BookSum

Fix: Regenerate tokenizer without add_prefix_space to fix CI

Fix tokenizer files, remove LFS dependency for .model files

feat: Add FLAN-T5 compatibility with relative position bias

Made necessary .json files available for hugging face hosting