[Minor] Improve code style (#2422)

python/sglang/srt/utils.py

@@ -169,7 +169,7 @@ def calculate_time(show=False, min_cost_ms=0.0):
     return wrapper
 
 
-def get_available_gpu_memory(device, gpu_id, distributed=False):
+def get_available_gpu_memory(device, gpu_id, distributed=False, empty_cache=True):
     """
     Get available memory for cuda:gpu_id device.
     When distributed is True, the available memory is the minimum available memory of all GPUs.
@@ -184,7 +184,8 @@ def get_available_gpu_memory(device, gpu_id, distributed=False):
                 "which may cause useless memory allocation for torch CUDA context.",
             )
 
-        torch.cuda.empty_cache()
+        if empty_cache:
+            torch.cuda.empty_cache()
         free_gpu_memory, _ = torch.cuda.mem_get_info(gpu_id)
 
     elif device == "xpu":
@@ -196,7 +197,9 @@ def get_available_gpu_memory(device, gpu_id, distributed=False):
                 f"WARNING: current device is not {gpu_id}, but {torch.xpu.current_device()}, ",
                 "which may cause useless memory allocation for torch XPU context.",
             )
-        torch.xpu.empty_cache()
+
+        if empty_cache:
+            torch.xpu.empty_cache()
         used_memory = torch.xpu.memory_allocated()
         total_gpu_memory = torch.xpu.get_device_properties(gpu_id).total_memory
         free_gpu_memory = total_gpu_memory - used_memory