Added landing page for torcheval (#49)

Summary:
Pull Request resolved: #49

Added landing page with simple example and installation instructions

Reviewed By: ninginthecloud

Differential Revision: D39712441

fbshipit-source-id: bc438a2f89d0570f9cdfe36f2ae4c3993c0d38f7

docs/source/index.rst

@@ -1,19 +1,79 @@
 TorchEval
 ===========================================
 
-A library that contains a rich collection of performant PyTorch model metrics, a simple interface to create new metrics, a toolkit to facilitate metric computation in distributed training and tools for PyTorch model evaluations.
+A library with simple and straightforward tooling for model evalutations and a delightful user experience. At a high level TorchEval:
 
-
-`Installation instructions <https://github.com/pytorch-labs/torcheval#installing-torcheval>`_
+1. Contains a rich collection of high performance metric calulcations out of the box. We utilize vectorization and GPU acceleration where possible via PyTorch.
+2. Integrates seemlessly with distributed training and tools using `torch.distributed <https://pytorch.org/tutorials/beginner/dist_overview.html>`_
+3. Is designed with extensibility in mind: you have the freedom to easily create your own metrics and leverage our toolkit.
+4. Provides tools for profiling memory and compute requirements for PyTorch based models.
 
 TorchEval Tutorials
 -------------------
 .. toctree::
    :maxdepth: 2
    :caption: Contents:
 
-   metric_example.rst
+QuickStart
+===========================================
+
+Installing
+-----------------
+
+TorchEval can be installed from PyPi via
+
+.. code-block:: console
+
+   pip install torcheval
+
+or from github
+
+.. code-block:: console
+
+   git clone https://github.com/pytorch-labs/torcheval
+   cd torcheval
+   pip install -r requirements.txt
+   python setup.py install
+
+Usage
+-----------------
 
+TorchEval provides two interfaces to each metric. If you are working in a single process environment, it is simplest to use metrics from the ``functional`` submodule. These can be found in ``torcheval.metrics.functional``.
+
+.. code-block:: python
+
+   from torcheval.metrics.functional import binary_f1_score
+   predictions = model(inputs)
+   f1_score = binary_f1_score(predictions, targets)
+
+We can use the same metric in the class based route, which provides tools that make computation simple in a multi-process settings. On a single device, you can use the class based metrics as follows:
+
+.. code-block:: python
+
+   from torcheval.metrics import BinaryF1Score
+   predictions = model(inputs)
+   metric = BinaryF1Score()
+   metric.update(predictions, targets)
+   f1_score = metric.compute()
+
+In a multi-process setting, the data from each process must be synchronized to compute the metric across the full dataset. To do this, simply replace ``metric.compute()`` with ``sync_and_compute(metric)``:
+
+.. code-block:: python
+
+   from torcheval.metrics import BinaryF1Score
+   from torcheval.metrics.toolkit import sync_and_compute
+   predictions = model(inputs)
+   metric = BinaryF1Score()
+   metric.update(predictions, targets)
+   f1_score = sync_and_compute(metric)
+
+Read more about the class based method in the distributed example.
+
+Further Reading
+-----------------
+* Check out the guides exaplaining the compute example
+* Check out the distributed example
+* Check out how to make your own metric
 
 TorchEval API
 -----------------