add docs about local dev and AWS dev

ghstack-source-id: 1a32ca2f7e97d7d2b178d3b1a1b73fa6879bf742
Pull Request resolved: #32

torchrecipes/paved_path/README.md

@@ -0,0 +1,60 @@
+# paved path project
+
+**This project is currently in Prototype. If you have suggestions for improvements, please open a GitHub issue. We'd love to hear your feedback.**
+
+## Local development
+1. Install dependencies
+```bash
+pip install -r requirements.txt
+```
+
+2. Train a model
+```bash
+python charnn/main.py
+```
+
+3. Generate text from a model
+```bash
+python charnn/main.py charnn.task="generate" charnn.phrase="hello world"
+```
+
+4. [Optional] train a model with torchx
+```bash
+torchx run  -s local_cwd dist.ddp -j 1x2 --script charnn/main.py
+```
+* NOTE: 
+    * `-j 1x2` means single node with 2 GPUs. Learn more about torchx [here](https://pytorch.org/torchx/latest/)
+
+## Development in AWS
+### Setup environment
+1. Launch an EC2 instance following [EC2 GetStarted](https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/EC2_GetStarted.html)
+2. Install docker and nvidia driver if not already installed
+
+You can use EC2 for [Local development](#Local-development). However, you may need to a cluster and scheduler to manage resources(GPU, CPU, RAM, etc.) more efficiently. There are various options like [Slurm](https://slurm.schedmd.com/documentation.html), [kubernetes](https://kubernetes.io/), etc. AWS provides a fully managed [Batch](https://aws.amazon.com/batch/) that is easy to get started. We will use it as the default scheduler in this example. With torchx, the job launching CLI will be similar for all supported schedulers.
+
+### Create a container image on AWS ECS
+Before launching a job in Batch, we need to create a docker image containing the executable(`charnn/main.py` and its dependencies). Please follow [docker/README.md](https://github.com/facebookresearch/recipes/tree/main/torchrecipes/paved_path/docker).
+
+### AWS Batch
+1. Create Batch through Wizard: https://docs.aws.amazon.com/batch/latest/userguide/Batch_GetStarted.html
+  * NOTE: 
+    * Configure Compute Environment and Job Queue(named it as "torch-gpu"). Do not need to Define Job if launch with torch.x
+2. Setup env variables
+```bash
+export REGION="us-west-2"  # or any region in your case
+export JOB_QUEUE="torchx-gpu"  # must match the name of your Job Queue
+export ECR_URL="YOUR_AWS_ACCOUNT_ID.dkr.ecr.YOUR_REGION.amazonaws.com/charnn"  # defined in docker/README.md
+```
+3. Launch a model training job with torchx
+```bash
+torchx run --workspace '' -s aws_batch \
+        -cfg queue=$JOB_QUEUE,image_repo=$ECR_URL/charnn dist.ddp \
+        --script charnn/main.py --image $ECR_URL/charnn:latest \
+        --cpu 8 --gpu 2 -j 1x2 --memMB 20480
+```
+Note that it will output a URL like "aws_batch://torchx/..." that is used to track the job status.
+4. Check job status
+```bash
+torchx status "aws_batch://torchx/..."
+```
+