141M param safety model (not much better than v1, but a great learning)

Study project with experiments in Text MultiClassification Task Field (:

General purpose JUnit categories for better test classification

End-to-end NLP text classification pipeline on AG News, a custom LLaMA-inspired transformer with RoPE/RMSNorm/SwiGLU, Optuna + MLflow hyperparameter tuning, uncertainty-aware evaluation, bundle promotion, FastAPI serving, and Streamlit dashboard. Deployable via Docker & HF Spaces.

Classify text as safe or unsafe using a 141M parameter DeBERTa-v3 model with detailed multi-label categories for content moderation.