fix: remove misplaced import inside GRPOConfig args

train.py

@@ -175,10 +175,9 @@ trainer = GRPOTrainer(
     model=model,
     processing_class=tokenizer,
     reward_funcs=[reward_fn],
-    import torch as _torch
     args=GRPOConfig(
         output_dir="/tmp/deceit-1.5b",
-        bf16=_torch.cuda.is_available() and _torch.cuda.is_bf16_supported(),
+        bf16=torch.cuda.is_available() and torch.cuda.is_bf16_supported(),
         fp16=False,
         max_steps=150,
         per_device_train_batch_size=4,