config: production epochs 100→200, min_epochs_before_stopping 50→100
H100 has the compute budget — 200 epochs with cosine decay gives the model full room to learn through C51 transition and converge. Early stopping can't fire before epoch 100. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
@@ -7,7 +7,7 @@
|
||||
# - Cosine LR decay: 1e-4 → 1e-5 over 100 epochs
|
||||
|
||||
[training]
|
||||
epochs = 100
|
||||
epochs = 200
|
||||
batch_size = 1024
|
||||
learning_rate = 0.0001
|
||||
gamma = 0.99
|
||||
@@ -42,7 +42,7 @@ per_beta_start = 0.4
|
||||
[early_stopping]
|
||||
enabled = true
|
||||
patience = 20
|
||||
min_epochs_before_stopping = 50
|
||||
min_epochs_before_stopping = 100
|
||||
|
||||
[experience]
|
||||
initial_capital = 100000.0
|
||||
|
||||
Reference in New Issue
Block a user