|
4 | 4 | Cloud Training |
5 | 5 | ############## |
6 | 6 |
|
7 | | -Lightning has a native solution for training on AWS/GCP at scale. |
8 | | -Go to `grid.ai <https://www.grid.ai/>`_ to create an account. |
| 7 | +Lightning makes it easy to scale your training, without the boilerplate. |
| 8 | +If you want to train your models on the cloud, without dealing with engineering infrastructure and servers, you can try `Grid.ai <https://www.grid.ai/>`_. |
9 | 9 |
|
10 | | -We've designed Grid to work seamlessly with Lightning, without needing to make ANY code changes. |
| 10 | +Developed by the creators of `PyTorch Lightning <https://www.pytorchlightning.ai/>`_, Grid is a platform that allows you to: |
11 | 11 |
|
12 | | -To use Grid, replace ``python`` in your regular command: |
13 | 12 |
|
14 | | -.. code-block:: bash |
| 13 | +- **Scale your models to multi-GPU and multiple nodes** instantly with interactive sessions |
| 14 | +- **Run Hyperparameter Sweeps on 100s of GPUs** in one command |
| 15 | +- **Upload huge datasets** for availability at scale |
| 16 | +- **Iterate faster and cheaper**, you only pay for what you need |
| 17 | + |
| 18 | + |
| 19 | +**************** |
| 20 | +Training on Grid |
| 21 | +**************** |
| 22 | + |
| 23 | +.. raw:: html |
15 | 24 |
|
16 | | - python my_model.py --learning_rate 1e-6 --layers 2 --gpus 4 |
| 25 | + <video width="50%" max-width="400px" controls |
| 26 | + poster="https://grid-docs.s3.us-east-2.amazonaws.com/grid.png" |
| 27 | + src="https://pl-bolts-doc-images.s3.us-east-2.amazonaws.com/pl_docs/grid.mp4"></video> |
17 | 28 |
|
18 | | -To use the ``grid run`` command: |
| 29 | +| |
| 30 | +
|
| 31 | +You can launch any Lightning model on Grid using the Grid `CLI <https://pypi.org/project/lightning-grid/>`_: |
19 | 32 |
|
20 | 33 | .. code-block:: bash |
21 | 34 |
|
22 | | - grid run --gpus 4 my_model.py --learning_rate 'uniform(1e-6, 1e-1, 20)' --layers '[2, 4, 8, 16]' |
| 35 | + grid run --instance_type v100 --gpus 4 my_model.py --gpus 4 --learning_rate 'uniform(1e-6, 1e-1, 20)' --layers '[2, 4, 8, 16]' |
| 36 | +
|
| 37 | +You can also start runs or interactive sessions from the `Grid platform <https://platform.grid.ai>`_, where you can upload datasets, view artifacts, view the logs, the cost, log into tensorboard, and so much more. |
| 38 | + |
23 | 39 |
|
24 | | -The above command will launch (20 * 4) experiments, each running on 4 GPUs (320 GPUs!) - by making ZERO changes to |
25 | | -your code. |
| 40 | +********** |
| 41 | +Learn More |
| 42 | +********** |
26 | 43 |
|
27 | | -The ``uniform`` command is part of our new expressive syntax which lets you construct hyperparameter combinations |
28 | | -using over 20+ distributions, lists, etc. Of course, you can also configure all of this using yamls which |
29 | | -can be dynamically assembled at runtime. |
| 44 | +`Sign up for Grid <http://platform.grid.ai>`_ and receive free credits to get you started! |
30 | 45 |
|
31 | | -*************** |
32 | | -Grid Highlights |
33 | | -*************** |
| 46 | +`Grid in 3 minutes <https://docs.grid.ai/#introduction>`_ |
34 | 47 |
|
35 | | -* Run any public or private repository with Grid, or use an interactive session. |
36 | | -* Grid allocates all the machines and GPUs you need on demand, so you only pay for what you need when you need it. |
37 | | -* Grid handles all the other parts of developing and training at scale: artifacts, logs, metrics, etc. |
38 | | -* Grid works with the experiment manager of your choice, no code changes needed. |
39 | | -* Use Grid Datastores- high-performance, low-latency, versioned datasets. |
40 | | -* Attach Datastores to a Run so you don't have to keep downloading datasets |
41 | | -* Use Grid Sessions for fast prototyping on a cloud machine of your choice |
42 | | -* For more information check the `grid documentation <https://docs.grid.ai/>`_ |
| 48 | +`Grid.ai Terms of Service <https://www.grid.ai/terms-of-service/>`_ |
0 commit comments