Update app.py

app.py

@@ -47,12 +47,13 @@ training_args = GRPOConfig(
     max_prompt_length=512,
     max_completion_length=96,
     num_generations=8,
-    optim="adamw_8bit",
     num_train_epochs=1,
-    bf16=True,
     report_to=["wandb"],
     remove_unused_columns=False,
     logging_steps=1,
+    bf16=False,
+    fp16=True,                     # если есть GPU
+    optim="adamw_torch_fused",     # НЕ "adamw_8bit"
 )
 
 trainer = GRPOTrainer(