update doc

TorchEnsemble-Community · Nov 13, 2022 · 8cc576c · 8cc576c · binunalex · Nov 30, 2022
1 parent 4afc38c
commit 8cc576c
Show file tree

Hide file tree

Showing 2 changed files with 16 additions and 0 deletions.
diff --git a/docs/advanced.rst b/docs/advanced.rst
@@ -0,0 +1,15 @@
+Advanced Usage
+==============
+
+The following sections outline advanced usage in :mod:`torchensemble`.
+
+Faster inference using functorch
+--------------------------------
+
+:mod:`functorch` has been integrated into Pytorch since the release of version 1.13, which is JAX-like composable function transforms for PyTorch. To enable faster inference of ensembles in :mod:`torchensemble`, you could use :meth:`vectorize` method of the ensemble to convert it into a stateless version (fmodel), and stacked parameters and buffers.
+
+The stateless model, parameters, along with buffers could be used to reduce the inference time using :meth:`vmap` in :mod:`functorch`. More details are available at `functorch documentation <https://pytorch.org/functorch/stable/notebooks/ensembling.html>`__. The code snippet below demonstrates how to pass :meth:`ensemble.vectorize` results into :meth:`functorch.vmap`.
+
+.. code:: python
+
+  from torchensemble import VotingClassifier  # voting is a classic ensemble strategy
diff --git a/docs/index.rst b/docs/index.rst
@@ -72,6 +72,7 @@ Content
   Guidance <guide>
   Experiment <experiment>
   API Reference <parameters>
+  Advanced Usage <advanced>
 
 .. toctree::
   :maxdepth: 1