diff --git a/torchrec/distributed/types.py b/torchrec/distributed/types.py
index f12db2ca3..058b07ef9 100644
--- a/torchrec/distributed/types.py
+++ b/torchrec/distributed/types.py
@@ -30,7 +30,7 @@
 
 from torch.autograd.profiler import record_function
 from torchrec.tensor_types import UInt2Tensor, UInt4Tensor
-from torchrec.types import ModuleNoCopyMixin
+from torchrec.types import DataType, ModuleNoCopyMixin
 
 try:
     # For python 3.6 and below, GenericMeta will be used by
@@ -108,29 +108,6 @@ def _tabulate(
     return "\n".join(rows)
 
 
-# moved DataType here to avoid circular import
-# TODO: organize types and dependencies
-@unique
-class DataType(Enum):
-    """
-    Our fusion implementation supports only certain types of data
-    so it makes sense to retrict in a non-fused version as well.
-    """
-
-    FP32 = "FP32"
-    FP16 = "FP16"
-    BF16 = "BF16"
-    INT64 = "INT64"
-    INT32 = "INT32"
-    INT8 = "INT8"
-    UINT8 = "UINT8"
-    INT4 = "INT4"
-    INT2 = "INT2"
-
-    def __str__(self) -> str:
-        return self.value
-
-
 class ShardingType(Enum):
     """
     Well-known sharding types, used by inter-module optimizations.
diff --git a/torchrec/modules/embedding_configs.py b/torchrec/modules/embedding_configs.py
index 11d6dc65d..cf2c52019 100644
--- a/torchrec/modules/embedding_configs.py
+++ b/torchrec/modules/embedding_configs.py
@@ -14,7 +14,7 @@
 import torch
 from fbgemm_gpu.split_embedding_configs import SparseType
 from fbgemm_gpu.split_table_batched_embeddings_ops_training import PoolingMode
-from torchrec.distributed.types import DataType
+from torchrec.types import DataType
 
 
 @unique
diff --git a/torchrec/types.py b/torchrec/types.py
index 6788cbe2a..391cf2ca5 100644
--- a/torchrec/types.py
+++ b/torchrec/types.py
@@ -6,6 +6,7 @@
 # LICENSE file in the root directory of this source tree.
 
 from abc import abstractmethod
+from enum import Enum, unique
 
 import torch
 from torch import nn
@@ -35,3 +36,26 @@ class ModuleNoCopyMixin(CopyMixIn):
     def copy(self, device: torch.device) -> nn.Module:
         # pyre-ignore [7]
         return self
+
+
+# moved DataType here to avoid circular import
+# TODO: organize types and dependencies
+@unique
+class DataType(Enum):
+    """
+    Our fusion implementation supports only certain types of data
+    so it makes sense to retrict in a non-fused version as well.
+    """
+
+    FP32 = "FP32"
+    FP16 = "FP16"
+    BF16 = "BF16"
+    INT64 = "INT64"
+    INT32 = "INT32"
+    INT8 = "INT8"
+    UINT8 = "UINT8"
+    INT4 = "INT4"
+    INT2 = "INT2"
+
+    def __str__(self) -> str:
+        return self.value