⚡Serverless Runs

Run (and sweep) any private or public Github repository.


Run (and sweep) any private or public Github repository on the cloud.

# clone repo
git clone https://github.com/williamFalcon/hello
# start the sweep
cd hello
grid run hello.py --number "[1, 2]" --food_item "['pizza', 'hotdog']"

⚡️⚡️Forget about infrastructure ⚡️⚡️

Runs are serverless which means you only pay for the time your scripts are actually running. When running on your own infrastructure this amounts to massive cost savings as well.

1 minute overview

In this video we're going to run an arbitrary model (from the pytorch examples github repo) across 4 GPUs (4 experiments each on 2 GPUs)

Product Tour

Click here for a 2-minute tour of RUN

Option 1: Run via the CLI

RUN any GitHub file with Grid in 4 steps:

# 1. clone the repo
git clone https://github.com/pytorch/examples
# 2. find the file to run
cd examples/dcgan
# 3. verify it works locally (optional)
python main.py --dataset cifar10 --lr 0.0002 --dataroot .
# 4. run on a cloud instance via grid
grid run main.py --dataset cifar10 --lr 0.0002 --dataroot .

Grid offers advanced syntax for starting a run. With this code:

grid run hello.py --number "[1, 2]" --food_item "['pizza', 'hotdog']"

Grid will run the script 4 times... these are the 4 equivalent script calls (we call each script call an experiment)

python hello.py --number 1 --food_item 'pizza'
python hello.py --number 2 --food_item 'pizza'
python hello.py --number 1 --food_item 'hotdog'
python hello.py --number 2 --food_item 'hotdog'

A RUN is a collection of experiments (the run has 4 experiments in this example).

Option 2: Start via the web UI