TPU support¶

Lightning supports running on TPUs. At this moment, TPUs are available on Google Cloud (GCP), Google Colab and Kaggle Environments. For more information on TPUs watch this video.

TPU Terminology¶

A TPU is a Tensor processing unit. Each TPU has 8 cores where each core is optimized for 128x128 matrix multiplies. In general, a single TPU is about as fast as 5 V100 GPUs!

A TPU pod hosts many TPUs on it. Currently, TPU pod v2 has 2048 cores! You can request a full pod from Google cloud or a “slice” which gives you some subset of those 2048 cores.

How to access TPUs¶

To access TPUs, there are three main ways.

Using Google Colab.
Using Google Cloud (GCP).
Using Kaggle.

Kaggle TPUs¶

For starting Kaggle projects with TPUs, refer to this kernel.

Colab TPUs¶

Colab is like a jupyter notebook with a free GPU or TPU hosted on GCP.

To get a TPU on colab, follow these steps:

Go to https://colab.research.google.com/.
Click “new notebook” (bottom right of pop-up).
Click runtime > change runtime settings. Select Python 3, and hardware accelerator “TPU”. This will give you a TPU with 8 cores.
Next, insert this code into the first cell and execute. This will install the xla library that interfaces between PyTorch and the TPU.
```
!pip install cloud-tpu-client==0.10 https://storage.googleapis.com/tpu-pytorch/wheels/torch_xla-1.8-cp37-cp37m-linux_x86_64.whl
```
Once the above is done, install PyTorch Lightning (v 0.7.0+).
```
!pip install pytorch-lightning
```
Then set up your LightningModule as normal.

DistributedSamplers¶

Lightning automatically inserts the correct samplers - no need to do this yourself!

Usually, with TPUs (and DDP), you would need to define a DistributedSampler to move the right chunk of data to the appropriate TPU. As mentioned, this is not needed in Lightning

Note

Don’t add distributedSamplers. Lightning does this automatically

If for some reason you still need to, this is how to construct the sampler for TPU use

import torch_xla.core.xla_model as xm

def train_dataloader(self):
    dataset = MNIST(
        os.getcwd(),
        train=True,
        download=True,
        transform=transforms.ToTensor()
    )

    # required for TPU support
    sampler = None
    if use_tpu:
        sampler = torch.utils.data.distributed.DistributedSampler(
            dataset,
            num_replicas=xm.xrt_world_size(),
            rank=xm.get_ordinal(),
            shuffle=True
        )

    loader = DataLoader(
        dataset,
        sampler=sampler,
        batch_size=32
    )

    return loader

Configure the number of TPU cores in the trainer. You can only choose 1 or 8. To use a full TPU pod skip to the TPU pod section.

import pytorch_lightning as pl

my_model = MyLightningModule()
trainer = pl.Trainer(tpu_cores=8)
trainer.fit(my_model)

That’s it! Your model will train on all 8 TPU cores.

TPU core training¶

Lightning supports training on a single TPU core or 8 TPU cores.

The Trainer parameters tpu_cores defines how many TPU cores to train on (1 or 8) / Single TPU to train on [1].

For Single TPU training, Just pass the TPU core ID [1-8] in a list.

Single TPU core training. Model will train on TPU core ID 5.

trainer = pl.Trainer(tpu_cores=[5])

8 TPU cores training. Model will train on 8 TPU cores.

trainer = pl.Trainer(tpu_cores=8)

Distributed Backend with TPU¶

The accelerator option used for GPUs does not apply to TPUs. TPUs work in DDP mode by default (distributing over each core)

TPU Pod¶

To train on more than 8 cores, your code actually doesn’t change! All you need to do is submit the following command:

$ python -m torch_xla.distributed.xla_dist
--tpu=$TPU_POD_NAME
--conda-env=torch-xla-nightly
-- python /usr/share/torch-xla-0.5/pytorch/xla/test/test_train_imagenet.py --fake_data

See this guide on how to set up the instance groups and VMs needed to run TPU Pods.

16 bit precision¶

Lightning also supports training in 16-bit precision with TPUs. By default, TPU training will use 32-bit precision. To enable 16-bit, set the 16-bit flag.

import pytorch_lightning as pl

my_model = MyLightningModule()
trainer = pl.Trainer(tpu_cores=8, precision=16)
trainer.fit(my_model)

Under the hood the xla library will use the bfloat16 type.

Performance considerations¶

The TPU was designed for specific workloads and operations to carry out large volumes of matrix multiplication, convolution operations and other commonly used ops in applied deep learning. The specialization makes it a strong choice for NLP tasks, sequential convolutional networks, and under low precision operation. There are cases in which training on TPUs is slower when compared with GPUs, for possible reasons listed:

Too small batch size.
Explicit evaluation of tensors during training, e.g. tensor.item()
Tensor shapes (e.g. model inputs) change often during training.
Limited resources when using TPU’s with PyTorch Link
XLA Graph compilation during the initial steps Reference
Some tensor ops are not fully supported on TPU, or not supported at all. These operations will be performed on CPU (context switch).
PyTorch integration is still experimental. Some performance bottlenecks may simply be the result of unfinished implementation.

The official PyTorch XLA performance guide has more detailed information on how PyTorch code can be optimized for TPU. In particular, the metrics report allows one to identify operations that lead to context switching.

About XLA¶

XLA is the library that interfaces PyTorch with the TPUs. For more information check out XLA.

Guide for troubleshooting XLA