diff --git a/README.md b/README.md
index f26daab..1b4e0b4 100644
--- a/README.md
+++ b/README.md
@@ -1,7 +1,8 @@
-# nix-container-build-gha
+# nix-container-build-gha-cuda
 
-This is a simple testing repo I used to understand how to build python environments in github actions
-with nix. I wrote up my thoughts in [this medium post](https://medium.com/@seanrmurphy/building-container-images-using-nix-and-github-actions-ba548ab9080d).
+This is a simple testing repo I used to understand how to build python environments which require CUDA
+using github actions with nix. It's a small modification to [this git repo](https://github.com/seanrmurphy/nix-container-build-gha). I wrote up a commentary on this 
+work in [this medium post](https://medium.com/@seanrmurphy/building-container-images-using-nix-and-github-actions-ba548ab9080d).
 
 This is the public version of the repo which anyone can look through; I also maintain a private version 
 which is linked to some self hosted github runners - I don't want to link the public version to any such
@@ -15,10 +16,12 @@ The python application was taken from [this repo](https://github.com/mitchellh/f
 
 I have included a couple of unecessary dependencies in the `pyproject.toml` just to understand how these are
 handled (`torch`, `jupyter` and `beautifulsoup4`). They are available in the resulting python environment but not used by the application.
+In this version, torch should determine that it's being built in a CUDA context and bring in the necessary CUDA content.
 
 ## Working locally
 
-This assumes you have a sensible nix configuration and are comfortable using flakes.
+This assumes you have a sensible nix configuration and are comfortable using flakes. It assumes that your build environment 
+has GPU support.
 
 - `nix build` will build the application and put the content in the `result` directory
 - `nix build .#ociApplicationImage` will build a container image which runs the application - the resulting container image is a gzip'd tarball in the `result` directory which can be imported to docker using `docker load < result`
@@ -30,3 +33,5 @@ This assumes you have a sensible nix configuration and are comfortable using fla
 The github action was essentially copied from [this repo](https://github.com/wagdav/thewagner.net).
 
 It requires a token called DOCKER_ACCESS_TOKEN to access a docker repository and push the resulting container image there. This repo uses the standard github runners but in the private variant, I was using a self-hosted runner (there are a couple of comments in the github action definition which highlight the small differences).
+
+Note that the action assumes that a runner is available which has multiple labels.