logarithmic scale of hyperparams

new experiments + faster models + optuna + somehow faster local training...

adding optuna

new experiments + labels + examples

deleted experiments + stable app.py

obtained logs