fix colo parameter torch function (#1117)

colossalai/tensor/colo_parameter.py

@@ -7,6 +7,23 @@ from colossalai.tensor.param_op_hook import ParamOpHookManager
 from typing import Optional
 
 
+def filter_args(func, *args):
+    return [arg for arg in args if func(arg)]
+
+
+def unpack_args(*args):
+    if len(args) == 1:
+        return args[0]
+    return args
+
+
+def replace_args(args, kwargs, new_args):
+    args = new_args[:len(args)]
+    for k, v in zip(kwargs.keys(), new_args[len(args):]):
+        kwargs[k] = v
+    return unpack_args(args), kwargs
+
+
 class ColoParameter(ColoTensor, torch.nn.Parameter):
     r"""A kind of ColoTensor to be considered as a module parameter.
 
@@ -50,12 +67,13 @@ class ColoParameter(ColoTensor, torch.nn.Parameter):
     def __torch_function__(cls, func, types, args=..., kwargs=None):
         if ParamOpHookManager.has_hook():
             if not func.__name__.startswith('__'):
-                params = list(filter(lambda arg: isinstance(arg, ColoParameter), args))
+                if kwargs is None:
-                if kwargs is not None:
+                    kwargs = {}
-                    params.extend(list(filter(lambda arg: isinstance(arg, ColoParameter), kwargs.values())))
+                params = filter_args(lambda arg: isinstance(arg, ColoParameter), *args, *kwargs.values())
                 if len(params) > 0:
                     with torch._C.DisableTorchFunction():
-                        args = ParamOpHookManager.pre_op(params, *args)
+                        new_args = ParamOpHookManager.pre_op(params, *args, *kwargs.values())
+                    args, kwargs = replace_args(args, kwargs, new_args)
                     ret = super().__torch_function__(func, types, args, kwargs)
                     with torch._C.DisableTorchFunction():
                         ret = ParamOpHookManager.post_op(params, ret)