Merge branch 'master' into HeyangQin/enable_hpz_nograd

Author: awan-10

deepspeed/runtime/zero/partition_parameters.py

@@ -250,9 +250,11 @@ def wrapped_fn(*args, **kwargs) -> Tensor:
 
 def get_new_tensor_fn_for_dtype(dtype: torch.dtype) -> Callable:
 
-    def new_tensor(cls, *args) -> Tensor:
+    def new_tensor(cls, *args, **kwargs) -> Tensor:
         device = torch.device(get_accelerator().device_name(os.environ["LOCAL_RANK"]))
-        tensor = _orig_torch_empty(0, device=device).new_empty(*args)
+        if not args:
+            args = (0, )
+        tensor = _orig_torch_empty(0, device=device).new_empty(*args, **kwargs)
         if tensor.is_floating_point():
             tensor = tensor.to(dtype)