[Doc] Add Tokenizer and auto-reset doc link

ghstack-source-id: 90f55b5
Pull Request resolved: #2754

Author: Vincent Moens

docs/source/reference/envs.rst

@@ -509,6 +509,8 @@ single agent standards.
 Auto-resetting Envs
 -------------------
 
+.. _autoresetting_envs:
+
 Auto-resetting environments are environments where calls to :meth:`~torchrl.envs.EnvBase.reset` are not expected when
 the environment reaches a ``"done"`` state during a rollout, as the reset happens automatically.
 Usually, in such cases the observations delivered with the done and reward (which effectively result from performing the
@@ -920,6 +922,7 @@ to be able to create this other composition:
     TargetReturn
     TensorDictPrimer
     TimeMaxPool
+    Tokenizer
     ToTensorImage
     TrajCounter
     UnaryTransform

torchrl/envs/common.py

@@ -317,6 +317,9 @@ class EnvBase(nn.Module, metaclass=_EnvPostInit):
             .. note:: The auto-resetting is achieved by the `EnvBase` metaclass. It does not appear in the
                 `__init__` method and is included in the keyword arguments strictly for type-hinting purpose.
 
+            .. seealso:: The :ref:`auto-resetting environments API <autoresetting_envs>` section in the API
+                documentation.
+
     Attributes:
         done_spec (Composite): equivalent to ``full_done_spec`` as all
             ``done_specs`` contain at least a ``"done"`` and a ``"terminated"`` entry