Full training results & evaluation with BERTScore

Fix regression: restore balanced training settings, add technical manual filter

Major improvements: real titles, anti-overfitting, better demo UX

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

Improve emotion task architecture and fix visualizations

Clean up codebase and fix training bugs

Update LexiMind: improved training, model architecture, and evaluation

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

Fix Pylance type errors, add inductor compilation support

Full training run, code cleanup, mypy/ruff fixes

feat: Add FLAN-T5 compatibility with relative position bias

Refactor: Consolidate dependencies, improve testing, and add CI/CD

Added pretrained BART wights for the encoder/decoder while keeping custom architectiure and task heads

chore: snapshot current refinements

Reformatted Project Structure