Update

[ghstack-poisoned]

tensordict/tensorclass.py

@@ -351,28 +351,41 @@ def is_non_tensor(obj):
 
 
 class _tensorclass_dec:
-    def __new__(cls, autocast: bool = False, frozen: bool = False):
+    def __new__(
+        cls, autocast: bool = False, frozen: bool = False, nocast: bool = False
+    ):
         if not isinstance(autocast, bool):
             clz = autocast
             self = super().__new__(cls)
-            self.__init__(autocast=False, frozen=False)
+            self.__init__(autocast=False, frozen=False, nocast=False)
             return self.__call__(clz)
         return super().__new__(cls)
 
-    def __init__(self, autocast: bool = False, frozen: bool = False):
+    def __init__(
+        self, autocast: bool = False, frozen: bool = False, nocast: bool = False
+    ):
+        if autocast and nocast:
+            raise ValueError("autocast is exclusive with nocast.")
         self.autocast = autocast
         self.frozen = frozen
+        self.nocast = nocast
 
     @dataclass_transform()
     def __call__(self, cls: T) -> T:
         clz = _tensorclass(cls, frozen=self.frozen)
         clz.autocast = self.autocast
+        clz.nocast = self.nocast
         return clz
 
 
 @dataclass_transform()
 def tensorclass(
-    cls: T = None, /, *, autocast: bool = False, frozen: bool = False
+    cls: T = None,
+    /,
+    *,
+    autocast: bool = False,
+    frozen: bool = False,
+    nocast: bool = False,
 ) -> T | None:
     """A decorator to create :obj:`tensorclass` classes.
 
@@ -381,30 +394,44 @@ def tensorclass(
     indexing, item assignment, reshaping, casting to device or storage and many
     others.
 
-    Args:
+    Keyword Args:
         autocast (bool, optional): if ``True``, the types indicated will be enforced when an argument is set.
-            Defaults to ``False``.
+            Thie argument is exclusive with ``autocast`` (both cannot be true at the same time). Defaults to ``False``.
         frozen (bool, optional): if ``True``, the content of the tensorclass cannot be modified. This argument is
             provided to dataclass-compatibility, a similar behavior can be obtained through the `lock` argument in
             the class constructor. Defaults to ``False``.
+        nocast (bool, optional): if ``True``, Tensor-compatible types such as ``int``, ``np.ndarray`` and the like
+            will not be cast to a tensor type. Thie argument is exclusive with ``autocast`` (both cannot be true
+            at the same time). Defaults to ``False``.
 
     tensorclass can be used with or without arguments:
+
     Examples:
         >>> @tensorclass
         ... class X:
-        ...     y: torch.Tensor
-        >>> X(1).y
-        1
+        ...     y: int
+        >>> X(torch.ones(())).y
+        tensor(1.)
         >>> @tensorclass(autocast=False)
         ... class X:
-        ...     y: torch.Tensor
+        ...     y: int
+        >>> X(torch.ones(())).y
+        tensor(1.)
+        >>> @tensorclass(autocast=True)
+        ... class X:
+        ...     y: int
+        >>> X(torch.ones(())).y
+        1
+        >>> @tensorclass(nocast=True)
+        ... class X:
+        ...     y: Any
         >>> X(1).y
         1
-        >>> @tensorclass(autocast=True)
+        >>> @tensorclass(nocast=False)
         ... class X:
-        ...     y: torch.Tensor
+        ...     y: Any
         >>> X(1).y
-        torch.tensor(1)
+        tensor(1)
 
     Examples:
         >>> from tensordict import tensorclass
@@ -456,7 +483,7 @@ def tensorclass(
     """
 
     def wrap(cls):
-        return _tensorclass_dec(autocast, frozen)(cls)
+        return _tensorclass_dec(autocast, frozen, nocast)(cls)
 
     # See if we're being called as @tensorclass or @tensorclass().
     if cls is None:
@@ -1665,7 +1692,10 @@ def _is_castable(datatype):
         elif (
             issubclass(value_type, torch.Tensor)
             or _is_tensor_collection(value_type)
-            or issubclass(value_type, (int, float, bool, np.ndarray))
+            or (
+                not cls.nocast
+                and issubclass(value_type, (int, float, bool, np.ndarray))
+            )
         ):
             return set_tensor()
         else:

test/test_tensorclass.py

@@ -2155,6 +2155,54 @@ class AutoCastTensor:
     anything: Any
 
 
+class TestNoCasting:
+    def test_nocast_int(self):
+        @tensorclass(nocast=False)
+        class X:
+            a: int  # type is irrelevant
+
+        assert isinstance(X(1).a, torch.Tensor)
+
+        @tensorclass(nocast=True)
+        class X:
+            a: int  # type is irrelevant
+
+        assert isinstance(X(1).a, int)
+
+    def test_nocast_np(self):
+        @tensorclass(nocast=False)
+        class X:
+            a: int  # type is irrelevant
+
+        assert isinstance(X(np.array([1])).a, torch.Tensor)
+
+        @tensorclass(nocast=True)
+        class X:
+            a: int  # type is irrelevant
+
+        assert isinstance(X(np.array([1])).a, np.ndarray)
+
+    def test_nocast_bool(self):
+        @tensorclass(nocast=False)
+        class X:
+            a: int  # type is irrelevant
+
+        assert isinstance(X(True).a, torch.Tensor)
+
+        @tensorclass(nocast=True)
+        class X:
+            a: int  # type is irrelevant
+
+        assert isinstance(X(False).a, bool)
+
+    def test_exclusivity(self):
+        with pytest.raises(ValueError, match="exclusive"):
+
+            @tensorclass(nocast=True, autocast=True)
+            class X:
+                a: int  # type is irrelevant
+
+
 class TestAutoCasting:
     @tensorclass(autocast=True)
     class ClsAutoCast: