Distributed PyTorch Lightning Training on Ray
This library adds new PyTorch Lightning accelerators for distributed training using the Ray distributed computing framework.
These PyTorch Lightning Accelerators on Ray enable quick and easy parallel training while still leveraging all the benefits of PyTorch Lightning and using your desired training protocol, either PyTorch Distributed Data Parallel or Horovod.
Once you add your accelerator to the PyTorch Lightning Trainer, you can parallelize training to all the cores in your laptop, or across a massive multi-node, multi-GPU cluster with no additional code changes.
This library also comes with an integration with Ray Tune for distributed hyperparameter tuning experiments.
Installation
You can install the master branch of ray_lightning_accelerators like so:
pip install git+https://github.com/ray-project/ray_lightning_accelerators#ray_lightning
PyTorch Distributed Data Parallel Accelerator on Ray
The RayAccelerator provides Distributed Data Parallel training on a Ray cluster. PyTorch DDP is used as the distributed training protocol, and Ray is used to launch and manage the training worker processes.
Here is a simplified example:
import pytorch_lightning as ptl
from ray_lightning import RayAccelerator
# Create your PyTorch Lightning model here.
ptl_model = MNISTClassifier(...)
accelerator = RayAccelerator(num_workers=4, cpus_per_worker=1, use_gpu=True)
# If using GPUs, set the ``gpus`` arg to a value > 0.
# The actual number of GPUs is determined by ``num_workers``.
trainer = pl.Trainer(..., gpus=1, accelerator=accelerator)
trainer.fit(ptl_model)
Because Ray is used to launch processes, instead of the same script being called multiple times, you CAN use this accelerator even in cases when you cannot use the standard DDPAccelerator such as
- Jupyter Notebooks, Google Colab, Kaggle
- Calling
fitortestmultiple times in the same script
Horovod Accelerator on Ray
Or if you prefer to use Horovod as the distributed training protocol, use the HorovodRayAccelerator instead.
import pytorch_lightning as ptl
from ray.util.lightning_accelerators import HorovodRayAccelerator
# Create your PyTorch Lightning model here.
ptl_model = MNISTClassifier(...)
# 2 nodes, 4 workers per node, each using 1 CPU and 1 GPU.
accelerator = HorovodRayAccelerator(num_hosts=2, num_slots=4, use_gpu=True)
# If using GPUs, set the ``gpus`` arg to a value > 0.
# The actual number of GPUs is determined by ``num_slots``.
trainer = pl.Trainer(..., gpus=1, accelerator=accelerator)
trainer.fit(ptl_model)
Multi-node Distributed Training
Using the same examples above, you can run distributed training on a multi-node cluster with just 2 simple steps.
- Use Ray's cluster launcher to start a Ray cluster-
ray up my_cluster_config.yaml. - Execute your Python script on the Ray cluster-
ray submit my_cluster_config.yaml train.py. This willrsyncyour training script to the head node, and execute it on the Ray cluster.
You no longer have to set environment variables or configurations and run your training script on every single node.
Hyperparameter Tuning with Ray Tune
ray_lightning also integrates with Ray Tune to provide distributed hyperparameter tuning for your distributed model training. You can run multiple PyTorch Lightning training runs in parallel, each with a different hyperparameter configuration, and each training run parallelized by itself. All you have to do is move your training code to a function, pass the function to tune.run, and make sure to add the appropriate callback (Either TuneReportCallback or TuneReportCheckpointCallback) to your PyTorch Lightning Trainer.
Example using ray_lightning with Tune:
def train_mnist(config):
# Create your PTL model.
model = MNISTClassifier(config)
# Create the Tune Reporting Callback
metrics = {"loss": "ptl/val_loss", "acc": "ptl/val_accuracy"}
callbacks = [TuneReportCallback(metrics, on="validation_end")]
trainer = pl.Trainer(
max_epochs=4,
callbacks=callbacks,
accelerator=RayAccelerator(num_workers=4, use_gpu=False))
trainer.fit(model)
config = {
"layer_1": tune.choice([32, 64, 128]),
"layer_2": tune.choice([64, 128, 256]),
"lr": tune.loguniform(1e-4, 1e-1),
"batch_size": tune.choice([32, 64, 128]),
}
# Make sure to specify how many actors each training run will create via the "extra_cpu" field.
analysis = tune.run(
train_mnist,
metric="loss",
mode="min",
config=config,
num_samples=num_samples,
resources_per_trial={
"cpu": 1,
"extra_cpu": 4
},
name="tune_mnist")
print("Best hyperparameters found were: ", analysis.best_config)
FAQ
RaySGD already has a Pytorch Lightning integration. What's the difference between this integration and that?
The key difference is which Trainer you'll be interacting with. In this library, you will still be using Pytorch Lightning's Trainer. You'll be able to leverage all the features of Pytorch Lightning, and Ray is used just as a backend to handle distributed training.
With RaySGD's integration, you'll be converting your LightningModule to be RaySGD compatible, and will be interacting with RaySGD's TorchTrainer. RaySGD's TorchTrainer is not as feature rich nor as easy to use as Pytorch Lightning's Trainer (no built in support for logging, early stopping, etc.). However, it does have built in support for fault-tolerant and elastic training. If these are hard requirements for you, then RaySGD's integration with PTL might be a better option.
I see that
RayAcceleratoris based off of Pytorch Lightning'sDDPSpawnAccelerator. However, doesn't the PTL team discourage the use of spawn?
As discussed here, using a spawn approach instead of launch is not all that detrimental. The original factors for discouraging spawn were:
- not being able to use 'spawn' in a Jupyter or Colab notebook, and
- not being able to use multiple workers for data loading.
Neither of these should be an issue with the RayAccelerator due to Ray's serialization mechanisms. The only thing to keep in mind is that when using this accelerator, your model does have to be serializable/pickleable.
![[Tune] Ray Tune + Ray Lightning too many tasks warning](https://avatars.githubusercontent.com/u/21029745?v=4)
1 Jul 9, 2022
23 Nov 2, 2022
373 Dec 31, 2022
80 Dec 21, 2022
48 Sep 27, 2022
520 Dec 30, 2022
1.4k Jan 1, 2023
6 Mar 17, 2022
50 Dec 16, 2022
50 Dec 16, 2022
178 Dec 21, 2022
2 Oct 20, 2022
20 Dec 3, 2022
157 Dec 26, 2022
40 Sep 19, 2022
396 Dec 30, 2022
294 Dec 17, 2022
58 Nov 18, 2022