Distributed Training Made Easy with PyTorch-Ignite
Published June 28, 2021
Writing agnostic distributed code that supports different platforms, hardware configurations (GPUs, TPUs) and communication frameworks is tedious. In this blog, we will discuss how PyTorch-Ignite solves this problem with minimal code change.
- 🔥 Pytorch-Ignite Unified Distributed API
- Running Distributed Code
- Closing Remarks
This blog assumes you have some knowledge about:
- PyTorch's distributed package, the backends and collective functions it provides. In this blog, we will focus on distributed data parallel code.
- PyTorch-Ignite. Refer to this blog for a quick high-level overview.
Thus, you will now be able to run the same version of the code across all supported backends seamlessly:
- backends from native torch distributed configuration: nccl, gloo, mpi
- Horovod framework with
- XLA on TPUs via pytorch/xla
In this blog post we will compare PyTorch-Ignite's API with torch
native's distributed code and highlight the differences and ease of use
of the former. We will also show how Ignite's
automatically make your code compatible with the aforementioned
distributed backends so that you only have to bring your own model,
optimizer and data loader objects.
Code snippets, as well as commands for running all the scripts, are provided in a separate repository.
Then we will also cover several ways of spawning processes via torch
torch.multiprocessing.spawn and also via multiple distributed
launchers in order to highlight how Pytorch-Ignite's
idist can handle
it without any changes to the code, in particular:
More information on launchers experiments can be found here.
🔥 Pytorch-Ignite Unified Distributed API
We need to write different code for different distributed backends. This
can be tedious especially if you would like to run your code on
different hardware configurations. Pytorch-Ignite's
idist will do all
the work for you, owing to the high-level helper methods.
🔍 Focus on the helper
This method adapts the logic for non-distributed and available distributed configurations. Here are the equivalent code snippets for distributed model instantiation:
Additionally, it is also compatible with NVIDIA/apex
model, optimizer = amp.initialize(model, optimizer, opt_level=opt_level)model = idist.auto_model(model)
and Torch native AMP
model = idist.auto_model(model)with autocast(): y_pred = model(x)
This method adapts the optimizer logic for non-distributed and available distributed configurations seamlessly. Here are the equivalent code snippets for distributed optimizer instantiation:
This method adapts the data loading logic for non-distributed and available distributed configurations seamlessly on target devices.
auto_dataloader() automatically scales the batch size
according to the distributed configuration context resulting in a
general way of loading sample batches on multiple devices.
Here are the equivalent code snippets for the distributed data loading step:
idistprovides collective operations like
broadcastthat can be used with all supported distributed frameworks. Please, see our documentation for more details.
The code snippets below highlight the API's specificities of each of
the distributed backends on the same use case as compared to the
API. PyTorch native code is available for DDP, Horovod, and for XLA/TPU
PyTorch-Ignite's unified code snippet can be run with the standard
PyTorch backends like
nccl and also with Horovod and XLA
for TPU devices. Note that the code is less verbose, however, the user
still has full control of the training loop.
The following examples are introductory. For a more robust, production-grade example that uses PyTorch-Ignite, refer here.
The complete source code of these experiments can be found here.
PyTorch-Ignite - Torch native Distributed Data Parallel - Horovod - XLA/TPUs
- You can also mix the usage of
idistwith other distributed APIs as below:
dist.init_process_group(backend, store=..., world_size=world_size, rank=rank)rank = idist.get_rank()ws = idist.get_world_size()model = idist.auto_model(model)dist.destroy_process_group()
Running Distributed Code
idist also unifies the distributed codes launching
method and makes the distributed configuration setup easier with the
This context manager has the capability to either spawn
nproc_per_node (passed as a script argument) child processes and
initialize a processing group according to the provided backend or use
initializing the processing group given the
backend argument only in
a general way.
In this case
idist Parallel is using the native torch
torch.multiprocessing.spawn method under the hood in order to run the
distributed configuration. Here
nproc_per_node is passed as a spawn
- Running multiple distributed configurations with one code. Source: ignite_idist.py:
# Running with gloopython -u ignite_idist.py --nproc_per_node 2 --backend gloo# Running with ncclpython -u ignite_idist.py --nproc_per_node 2 --backend nccl# Running with horovod with gloo controller ( gloo or nccl support )python -u ignite_idist.py --backend horovod --nproc_per_node 2# Running on xla/tpupython -u ignite_idist.py --backend xla-tpu --nproc_per_node 8 --batch_size 32
With Distributed launchers
idist Parallel context manager is also compatible
with multiple distributed launchers.
Here we are using the
torch.distributed.launch script in order to
spawn the processes:
python -m torch.distributed.launch --nproc_per_node 2 --use_env ignite_idist.py --backend gloo
horovodrun -np 4 -H hostname1:2,hostname2:2 python ignite_idist.py --backend horovod
- In order to run this example and to avoid the installation procedure,
you can pull one of PyTorch-Ignite's docker image with pre-installed
It will include Horovod with
docker run --gpus all -it -v $PWD:/project pytorchignite/hvd-vision:latest /bin/bashcd project
The same result can be achieved by using
slurm without any
modification to the code:
srun --nodes=2 --ntasks-per-node=2 --job-name=pytorch-ignite --time=00:01:00 --partition=gpgpu --gres=gpu:2 --mem=10G python ignite_idist.py --backend nccl
sbatch script.bash with the script file
#!/bin/bash#SBATCH --job-name=pytorch-ignite#SBATCH --output=slurm_%j.out#SBATCH --nodes=2#SBATCH --ntasks-per-node=2#SBATCH --time=00:01:00#SBATCH --partition=gpgpu#SBATCH --gres=gpu:2#SBATCH --mem=10Gsrun python ignite_idist.py --backend nccl
As we saw through the above examples, managing multiple configurations and specifications for distributed computing has never been easier. In just a few lines we can parallelize and execute code wherever it is while maintaining control and simplicity.
- idist-snippets: complete code used in this post.
- why-ignite: examples with distributed data parallel: native pytorch, pytorch-ignite, slurm.
- CIFAR10 example of distributed training on CIFAR10 with muliple configurations: 1 or multiple GPUs, multiple nodes and GPUs, TPUs.
- If you want to learn more about PyTorch-Ignite or have any further queries, here is our GitHub, documentation and Discord.
- PyTorch-Ignite is currently maintained by a team of volunteers and we are looking for more contributors. See CONTRIBUTING.md for how you can contribute.
- Keep updated with all PyTorch-Ignite news by following us on Twitter and Facebook.