Update on "Adding uint4 dtype implementation"

Summary:
We have a lot of interest for int4 dtypes, and we'd like to add the dtype out of PyTorch core.
This PR added some preliminary support for uint4 through tensor subclass and we'll continue to iterate on this

Test Plan:
python test/dtypes/test_int4.py

Reviewers:

Subscribers:

Tasks:

Tags:

[ghstack-poisoned]

Author: jerryzh168

torchao/dtypes/uint4.py

@@ -107,11 +107,6 @@ def __hash__(self):
     def __eq__(self, other):
         return torch.equal(self.elem, other.elem)
 
-    # def __getattribute__(self, name):
-    #     if name == "dtype":
-    #         return torch.uint4
-    #     return super().__getattribute__(name)
-
     @classmethod
     def __torch_dispatch__(cls, func, types, args, kwargs=None):
         if func is torch.ops.aten.view.default:
@@ -272,10 +267,6 @@ def from_unpacked(cls, unpacked, scales):
 
     @classmethod
     def __torch_dispatch__(cls, func, types, args, kwargs=None):
-        # didin't hit this, there is a mysterious error if we try to go through this path
-        # torch._dynamo.exc.BackendCompilerFailed: backend='inductor' raised:
-        # Exception: Please convert all Tensors to FakeTensors first or
-        # instantiate FakeTensorMode with 'allow_non_fake_inputs'
         if func is torch.ops.aten.addmm.default:
             bias, x, weight = args
             x_view = x.view(-1, x.shape[-1])
@@ -285,7 +276,7 @@ def __torch_dispatch__(cls, func, types, args, kwargs=None):
                 y += bias
             return y
         elif func is torch.ops.aten.t.default:
-            # assert False, "transpose is not properly implemented currently"
+            # TODO: add proper support for transpose
             self, = args
             unpacked = unpack_uint4(self.elem)
             transposed = torch.ops.aten.t.default(unpacked)