Document the legacy constructor for Tensor (#122625)
Fixes https://github.com/pytorch/pytorch/issues/122408
Signed-off-by: Edward Z. Yang <ezyang@meta.com>
Pull Request resolved: https://github.com/pytorch/pytorch/pull/122625
Approved by: https://github.com/albanD
diff --git a/docs/source/tensors.rst b/docs/source/tensors.rst
index 218c83d..7bfa870 100644
--- a/docs/source/tensors.rst
+++ b/docs/source/tensors.rst
@@ -212,6 +212,37 @@
(see :ref:`tensor-creation-ops`).
- To create a tensor with similar type but different size as another tensor,
use ``tensor.new_*`` creation ops.
+ - There is a legacy constructor ``torch.Tensor`` whose use is discouraged.
+ Use :func:`torch.tensor` instead.
+
+.. method:: Tensor.__init__(self, data)
+
+ This constructor is deprecated, we recommend using :func:`torch.tensor` instead.
+ What this constructor does depends on the type of ``data``.
+
+ * If ``data`` is a Tensor, returns an alias to the original Tensor. Unlike
+ :func:`torch.tensor`, this tracks autograd and will propagate gradients to
+ the original Tensor. ``device`` kwarg is not supported for this ``data`` type.
+
+ * If ``data`` is a sequence or nested sequence, create a tensor of the default
+ dtype (typically ``torch.float32``) whose data is the values in the
+ sequences, performing coercions if necessary. Notably, this differs from
+ :func:`torch.tensor` in that this constructor will always construct a float
+ tensor, even if the inputs are all integers.
+
+ * If ``data`` is a :class:`torch.Size`, returns an empty tensor of that size.
+
+ This constructor does not support explicitly specifying ``dtype`` or ``device`` of
+ the returned tensor. We recommend using :func:`torch.tensor` which provides this
+ functionality.
+
+ Args:
+ data (array_like): The tensor to construct from.
+
+ Keyword args:
+ device (:class:`torch.device`, optional): the desired device of returned tensor.
+ Default: if None, same :class:`torch.device` as this tensor.
+
.. autoattribute:: Tensor.T
.. autoattribute:: Tensor.H
