Correction of documentation on KeyjaggedTensor implied batch size (#3197)

Summary:
Pull Request resolved: #3197

TorchRec's "concepts" page incorrectly declared that the implied batch size for a `KeyJaggedTensor` is the number of features divided by the length of the `lengths` tensor.

It is in fact the lenght of `lenghts` divded by the number of keys. This diff corrects that.

Reviewed By: iamzainhuda

Differential Revision: D78375136

fbshipit-source-id: 10bd20f4328c2deae0a65aa059c38853b2f3b4d1

Author: rodrigodesalvobraz

docs/source/concepts.rst

@@ -96,9 +96,10 @@ is the data type used in ``forward`` of ``EmbeddingBagCollection`` and
 ``EmbeddingCollection`` as they are used to represent multiple features
 in a table.
 
-A ``KeyedJaggedTensor`` has an implied batch size, which is the number
-of features divided by the length of ``lengths`` tensor. The example
-below has a batch size of 2. Similar to a ``JaggedTensor``, the
+A ``KeyedJaggedTensor`` has an implied batch size, which is the length
+of ``lengths`` tensor divided by the number of keys. The example
+below has a batch size of 2 (4 lengths divided by 2 keys).
+Similar to a ``JaggedTensor``, the
 ``offsets`` and ``lengths`` function in the same manner. You can also
 access the ``lengths``, ``offsets``, and ``values`` of a feature by
 accessing the key from the ``KeyedJaggedTensor``.