Spaces:

Ajsaxena
/

deceit1

Paused

Jayant-Kernel commited on 13 days ago

Commit

42f691c

unverified ·

1 Parent(s): eed0745

fix: batch size 4 to match num_generations 4

Files changed (1) hide show

train.py CHANGED Viewed

@@ -159,7 +159,7 @@ trainer = GRPOTrainer(
     args=GRPOConfig(
         output_dir="./deceit-1.5b",
         max_steps=150,
-        per_device_train_batch_size=2,
         num_generations=4,
         learning_rate=5e-6,
         warmup_steps=5,

     args=GRPOConfig(
         output_dir="./deceit-1.5b",
         max_steps=150,
+        per_device_train_batch_size=4,
         num_generations=4,
         learning_rate=5e-6,
         warmup_steps=5,