Simulation
Model Sandbox
Tune hyperparameters and watch the synthetic training run respond in real time.
Hyperparameters
live
1.0e-2
Log scale 1e-5 → 1.0 · sweet spot near 1e-3
64
Small batches add gradient noise to the curve
0.08
< 0.02 risks overfit · > 0.3 caps capacity
Final train loss
0.0357
Best val loss
0.0384
Training loss curve
run: sliders@60/64/0.08 · 60 epochs
Training and validation losses track cleanly toward the floor.