change transpose to non-inplace op

Author: jerryzh168

tutorials/developer_api_guide/my_dtype_tensor_subclass.py

@@ -194,7 +194,7 @@ def dequantize(self, output_dtype=None):
         block_size = (1, int_data.shape[-1])
         if hasattr(self.layout_tensor, "transposed") and self.layout_tensor.transposed:
             transposed = True
-        res = dequantize_affine(int_data, block_size, scale, None, int_data.dtype)
+        res = dequantize_affine(int_data, block_size, scale, None, int_data.dtype, output_dtype=output_dtype)
         if transposed:
             res = res.t()
         return res
@@ -331,8 +331,7 @@ def __torch_dispatch__(cls, func, types, args, kwargs):
             elif dim == 1:
                 return PlainMyDTypeLayout(aten.slice.Tensor(self.int_data, dim, start, end, step), self.scale.view(-1, 1), self.transposed, self.layout_type)
         elif func is aten.t.default:
-            args[0].transposed = not args[0].transposed
-            return return_and_correct_aliasing(func, args, kwargs, args[0])
+            return return_and_correct_aliasing(func, args, kwargs, PlainMyDTypeLayout(args[0].int_data, args[0].scale, not args[0].transposed, args[0].layout_type))
 
         raise NotImplementedError(
             f"PlainMyDTypeLayout dispatch: attempting to run {func}, this is not supported"

tutorials/developer_api_guide/tensor_parallel.py

@@ -57,10 +57,8 @@ def _(func, types, args, kwargs):
 @implements(aten.t.default)
 def _(func, types, args, kwargs):
     tensor = args[0]
-    print("before transpose, ", tensor.shape)
     shape = tensor.shape[::-1]
     new = tensor.__class__(tensor.layout_tensor.t(), shape, tensor.dtype)
-    print("after transpose:", new.shape)
     return return_and_correct_aliasing(func, args, kwargs, new)
 
 @implements(aten.addmm.default)
@@ -80,8 +78,7 @@ def _(func, types, args, kwargs):
         args[1],
         None
     )
-    print("mm input tensor shape:", input_tensor.shape)
-    print("mm weight tensor shape:", weight_tensor.shape)
+    print("mm weight transposed:", weight_tensor.layout_tensor.transposed)
     weight_tensor = weight_tensor.dequantize()
     return aten.mm(input_tensor, weight_tensor)
 
@@ -172,6 +169,7 @@ def rowwise_shard(m: torch.nn.Module, mesh: DeviceMesh) -> torch.nn.Module:
 
     # Shard the models
     d_up = colwise_shard(q_up, mesh)
+    print("d_up weight shape:", d_up.linear.weight.shape)
     d_dn = rowwise_shard(q_dn, mesh)
 
     # We need to turn inputs into DTensor form as well -- just a format change
@@ -188,10 +186,10 @@ def rowwise_shard(m: torch.nn.Module, mesh: DeviceMesh) -> torch.nn.Module:
     # [rank0]: torch._dynamo.exc.TorchRuntimeError: Failed running call_function <built-in function linear>(*(DTensor(local_tensor=FakeTensor(..., device='cuda:0', size=(128, 1024)), device_mesh=DeviceMesh('cuda', [0, 1,
     # 2, 3]), placements=(Replicate(),)), DTensor(local_tensor=MyDTypeTensorTP(data=FakeTensor(..., device='cuda:0', size=(128, 1024)), shape=torch.Size([1024, 1024]), device=cuda:0, dtype=torch.float32, requires_grad=False), device_mesh=DeviceMesh('cuda', [0, 1, 2, 3]), placements=(Shard(dim=0),)), None), **{}):
     # [rank0]: a and b must have same reduction dim, but got [128, 1024] X [128, 1024].
-    c_up = torch.compile(d_up, backend="eager")
+    c_up = torch.compile(d_up)
     y_up = c_up(input_dtensor)
     print("y_up:", y_up.shape)
-    c_dn = torch.compile(d_dn, backend="eager")
+    c_dn = torch.compile(d_dn)
     y_dn = c_dn(y_up)
     print("y_dn:", y_dn.shape)
     print("compiled result:", y_dn)