修改train_model_grpo.py文件-开启vLLM 观察能否解决损失率值0并且无变化问题

src/train_model_grpo_v2.py

@@ -123,7 +123,7 @@ class ModelTrainer:
             dtype=self.dtype,
             fast_inference = True, # Enable vLLM fast inference
             max_lora_rank = lora_rank,
-            gpu_memory_utilization=0.95, # 0.6 # Reduce if out of memory
+            gpu_memory_utilization=0.005, # 0.6 # Reduce if out of memory
         )
 
         # 将模型移动到设备上