final

Author: vmoens

docs/source/reference/data_specs.rst

@@ -1,5 +1,7 @@
 .. currentmodule:: torchrl.data
 
+.. _ref_specs:
+
 TensorSpec System
 =================
 

docs/source/reference/data_storage.rst

@@ -1,5 +1,7 @@
 .. currentmodule:: torchrl.data.replay_buffers
 
+.. _checkpoint-rb:
+
 Storage Backends
 ================
 

docs/source/reference/llms.rst

@@ -204,7 +204,7 @@ such as :class:`~torchrl.envs.llm.ChatEnv`.
 An example of a tool transform is the :class:`~torchrl.envs.llm.transforms.PythonInterpreter` transform, which is used
 to execute Python code in the context of the LLM. The PythonInterpreter can optionally use a shared 
 :class:`~torchrl.envs.llm.transforms.PythonExecutorService` for efficient resource usage across multiple environments.
-See :ref:`ref_services` for more details on the service registry system.
+See :doc:`services` for more details on the service registry system.
 
     >>> from torchrl.envs.llm.transforms import PythonInterpreter
     >>> from torchrl.envs.llm import ChatEnv

docs/source/reference/modules.rst

@@ -4,6 +4,7 @@ torchrl.modules package
 =======================
 
 .. _ref_modules:
+.. _tdmodules:
 
 TorchRL offers a comprehensive collection of RL-specific neural network modules built on top of
 :class:`tensordict.nn.TensorDictModule`. These modules are designed to work seamlessly with

docs/source/reference/services.rst

@@ -590,7 +590,7 @@ For complete examples, see:
 See Also
 --------
 
-- :ref:`ref_llms` - LLM API documentation
+- :doc:`llms` - LLM API documentation
 - :ref:`ref_collectors` - Collector documentation
 - `Ray Documentation <https://docs.ray.io/>`_ - Ray distributed framework documentation
 

knowledge_base/HABITAT.md

@@ -29,7 +29,7 @@ Instructions can be found on [habitat github repo](https://github.com/facebookre
 
 ### Installing TorchRL
 
-Follow the instructions on the [README.md](../README.md).
+Follow the instructions on the [README.md](README.md).
 
 ### Using Habitat
 To get the list of available Habitat envs, simply run the following command:

knowledge_base/MUJOCO_INSTALLATION.md

@@ -176,7 +176,7 @@ issues when running `import mujoco_py` and some troubleshooting for each of them
     ```
 
     _Solution_: Link conda to the right `libOpenGL.so` file (replace `/path/to/conda` and `mujoco_env` with the proper paths and names):
-    ```shelf
+    ```shell
     conda install -y -c conda-forge libglvnd-glx-cos7-x86_64 --force-reinstall
     conda install -y -c conda-forge xvfbwrapper --force-reinstall
     conda env config vars set LD_PRELOAD=/path/to/conda/envs/mujoco_env/x86_64-conda-linux-gnu/sysroot/usr/lib64/libGLdispatch.so.0

knowledge_base/PRO-TIPS.md

@@ -102,7 +102,7 @@ Errors to look for that may be related to this misconception are the following:
 
 ## Common bugs
 - For bugs related to mujoco (incl. DeepMind Control suite and other libraries),
-  refer to the [MUJOCO_INSTALLATION](MUJOCO_INSTALLATION.md) file.
+  refer to the [MUJOCO_INSTALLATION](./MUJOCO_INSTALLATION.md) file.
 - `ValueError: bad value(s) in fds_to_keep`: this can have multiple reasons. One that is common in torchrl
   is that you are trying to send a tensor across processes that is a view of another tensor.
   For instance, when sending the tensor `b = tensor.expand(new_shape)` across processes, the reference to the original

torchrl/collectors/weight_update.py

@@ -360,9 +360,9 @@ class VanillaWeightUpdater(WeightUpdaterBase):
 
     Keyword Args:
         weight_getter (Callable[[], TensorDictBase], optional): a callable that returns the weights from the server.
-            If not provided, the weights must be passed to :meth:`~.update_weights` directly.
+            If not provided, the weights must be passed to ``push_weights`` directly.
         policy_weights (TensorDictBase): a TensorDictBase containing the policy weights to be updated
-            in-place.
+            in-place. Use ``push_weights`` to update the weights.
     """
 
     @classmethod

torchrl/envs/llm/transforms/browser.py

@@ -106,7 +106,7 @@ class BrowserTransform(SimpleToolTransform):
     This transform allows LLM agents to interact with web pages through a browser,
     supporting actions like navigation, clicking, typing, and extracting content.
 
-    For a complete example of how to use this transform, see the :ref:`llm_tools` tutorial.
+    For a complete example of how to use this transform, see the LLM Tools tutorial in the documentation.
 
     Args:
         allowed_domains (list[str], optional): List of allowed domains. If None, all domains are allowed.