# TensorFlow Nightly GPU VENV

This is a dockerized version of a TensorFlow virtual environment setup that works with a GPU.
This container and its documentation was made by [tiyn](https://github.com/tiyn) it is based on the
official [TensorFlow image](https://hub.docker.com/layers/tensorflow/tensorflow/nightly-gpu).

## Setup

Navigate into the `src` directory and build the image using the following command.

```sh
docker build . -t tf-nightly-gpu-venv
```

Then create a `rebuild.sh` file.
Change the settings according to your needs and run `./rebuild.sh` afterward.

### Environment-variables

Set the following variables with the -e tag.

| Name          | Usage                                               | Default     |
| ------------- | --------------------------------------------------- | ----------- |
| `VIRTUAL_ENV` | Specifies the path to the virtual environment       | `/opt/venv` |

### Volumes

Set the following volumes with the -v tag.

| Outside mount/volume name | Container mount  | Description                           |
| ------------------------- | ---------------- | ------------------------------------- |
| `$(pwd)`                  | `/workspace/src` | directory containing the source files |

### rebuild.sh

```sh
docker stop tf-nightly-gpu-venv
docker rm tf-nightly-gpu-venv
docker run --name tf-nightly-gpu-venv --rm --gpus all -it -v "$(pwd):/workspace/src" -w /workspace/src -d tf-nightly-gpu-venv
```