replace memory related APIs

Signed-off-by: Kunshang Ji <kunshang.ji@intel.com>

Author: jikunshang

vllm/utils/__init__.py

@@ -2553,10 +2553,10 @@ def __post_init__(self):
 
     def measure(self):
         # we measure the torch peak memory usage via allocated_bytes,
-        # rather than `torch.cuda.memory_reserved()` .
-        # After `torch.cuda.reset_peak_memory_stats()`,
-        # `torch.cuda.memory_reserved()` will keep growing, and only shrink
-        # when we call `current_platform.empty_cache()` or OOM happens.
+        # rather than `current_platform.memory_reserved()` .
+        # After `current_platform.reset_peak_memory_stats()`,
+        # `current_platform.memory_reserved()` will keep growing, and only
+        # shrink when we call `current_platform.empty_cache()` or OOM happens.
         from vllm.platforms import current_platform
 
         self.torch_peak = current_platform.memory_stats().get(
@@ -2565,7 +2565,7 @@ def measure(self):
         self.free_memory, self.total_memory = current_platform.mem_get_info()
         self.cuda_memory = self.total_memory - self.free_memory
 
-        # torch.cuda.memory_reserved() is how many bytes
+        # current_platform.memory_reserved() is how many bytes
         # PyTorch gets from cuda (by calling cudaMalloc, etc.)
         # this is used to measure the non-torch memory usage
         self.torch_memory = current_platform.memory_reserved()