HumanAgent Interface
Take Action
Lr Scale
*
:
Learning rate scale for this step (e.g. 1e-4 to 1.0)
Momentum Coef
*
:
Momentum coefficient (0 = no momentum, 1 = full carry)
Grad Clip Threshold
*
:
Gradient clipping threshold (0 = no clipping)
Weight Decay This Step
*
:
Weight decay (L2) scale for this step (0 = no weight decay)
Step
Reset Environment
Get State
Current State
Status:
Not initialized
Episode ID:
-
Step Count:
0
State Observer
Current Observation
No observation yet
Loss / Perplexity
Run baseline (AdamW)
Action History
No actions taken yet