Add overview section to docs

docs/source/index.rst

@@ -15,6 +15,14 @@ specific to your environment. Then you can install OLMo-core from PyPI with:
 
     pip install ai2-olmo-core
 
+.. toctree::
+   :hidden:
+   :maxdepth: 2
+   :caption: Overview
+
+   overview/introduction.rst
+   overview/installation.rst
+
 .. toctree::
    :hidden:
    :maxdepth: 2

docs/source/overview/installation.rst

@@ -0,0 +1,9 @@
+Installation
+============
+
+Prior to installing OLMo-core you should install `PyTorch <https://pytorch.org>`_ according to the official instructions
+specific to your operating system and hardware.
+
+Then you can install OLMo-core from `PyPI <https://pypi.org/project/ai2-olmo-core/>`_ with::
+
+    pip install ai2-olmo-core

docs/source/overview/introduction.rst

@@ -0,0 +1,48 @@
+Introduction
+============
+
+OLMo-core represents a major rewrite of the original training and modeling code from `OLMo <https://github.com/allenai/OLMo>`_
+with a focus on performance and API stability.
+It aims to provide a standard set of robust tools that can be used by LLM researchers at `AI2 <https://allenai.org>`_ and other organizations
+to build their research projects on.
+
+The library is centered around a highly efficient, yet flexible, :class:`~olmo_core.train.Trainer` and a :mod:`~olmo_core.launch`
+module that handles all of the boilerplate of launching experiments on `Beaker <https://beaker.org>`_
+or other platforms. It also comes with a simple, yet optimized, :class:`~olmo_core.nn.transformer.Transformer`
+model and many other useful :class:`torch.nn.Module` implementations.
+
+Most users will likely follow a workflow that looks like this:
+
+1. Define the various components of an experiment through configuration classes.
+   For example::
+
+     model_config = TransformerConfig.llama2_7B(...)
+     optim_config = AdamWConfig(lr=1e-3, ...)
+     data_config = MemMapDatasetConfig(...)
+     trainer_config = TrainerConfig(...)
+
+2. Build the corresponding components within a ``main()`` function at runtime and then call :meth:`Trainer.fit() <olmo_core.train.Trainer.fit>`.
+   For example::
+
+     def main(model_config, optim_config, data_config, trainer_config):
+         model = model_config.build()
+         optim = optim_config.build()
+         dataset = data_config.build()
+         trainer = trainer_config.build(model, optim, dataset)
+
+         trainer.fit()
+
+     if __name__ == "__main__":
+         prepare_training_environment(seed=SEED)
+         try:
+             main()
+         finally:
+             teardown_training_environment()
+
+3. Launch their training script with a :mod:`~olmo_core.launch` config, like the :class:`~olmo_core.launch.beaker.BeakerLaunchConfig`.
+   For example::
+
+     launch_config = BeakerLaunchConfig(...)
+     launch_config.launch(follow=True)
+
+You can find a complete example of this workflow in the `Train a language model <examples/train.rst>`_ example.