Skip to content

Latest commit

 

History

History
 
 

cloud_tpu_colabs

README.md

JAX Cloud TPU Preview

JAX now runs on Cloud TPUs! This is a preview, and we're still working on it. Help us out by kicking the tires, and letting us know if you run into any problems (see the Reporting issues section below).

Example Cloud TPU notebooks

The following notebooks showcase how to use and what you can do with Cloud TPUs on Colab:

Pmap Cookbook

A guide to getting started with pmap, a transform for easily distributing SPMD computations across devices.

Lorentz ODE Solver

Contributed by Alex Alemi (alexalemi@)

Solve and plot parallel ODE solutions with pmap.

Wave Equation

Contributed by Stephan Hoyer (shoyer@)

Solve the wave equation with pmap, and make cool movies! The spatial domain is partitioned across the 8 cores of a Cloud TPU.

JAX Demo

An overview of JAX presented at the Program Transformations for ML workshop at NeurIPS 2019 and the Compilers for ML workshop at CGO 2020. Covers basic numpy usage, grad, jit, vmap, and pmap.

Performance notes

The guidance on running TensorFlow on TPUs applies to JAX as well, with the exception of TensorFlow-specific details. Here we highlight a few important details that are particularly relevant to using TPUs in JAX.

Padding

One of the most common culprits for surprisingly slow code on TPUs is inadvertent padding:

  • Arrays in the Cloud TPU are tiled. This entails padding one of the dimensions to a multiple of 8, and a different dimension to a multiple of 128.
  • The matrix multiplication unit performs best with pairs of large matrices that minimize the need for padding.

bfloat16 dtype

By default*, matrix multiplication in JAX on TPUs uses bfloat16 with float32 accumulation. This can be controlled with the precision keyword argument on relevant jax.numpy functions (matmul, dot, einsum, etc). In particular:

  • precision=jax.lax.Precision.DEFAULT: uses mixed bfloat16 precision (fastest)
  • precision=jax.lax.Precision.HIGH: uses multiple MXU passes to achieve higher precision
  • precision=jax.lax.Precision.HIGHEST: uses even more MXU passes to achieve full float32 precision

JAX also adds the bfloat16 dtype, which you can use to explicitly cast arrays to bfloat16, e.g., jax.numpy.array(x, dtype=jax.numpy.bfloat16).

* We might change the default precision in the future, since it is arguably surprising. Please comment/vote on this issue if it affects you!

Running JAX on a Cloud TPU from a GCE VM

Creating a Cloud TPU involves creating the user GCE VM and the TPU node.

To create a user GCE VM, run the following command from your GCP console or your computer terminal where you have gcloud installed.

export ZONE=us-central1-c
gcloud compute instances create $USER-user-vm-0001 \
   --machine-type=n1-standard-1 \
   --image-project=ml-images \
   --image-family=tf-1-14 \
   --boot-disk-size=200GB \
   --scopes=cloud-platform \
   --zone=$ZONE

To create a larger GCE VM, choose a different machine type.

Next, create the TPU node, following these guidelines to choose a <TPU_IP_ADDRESS>.

export TPU_IP_ADDRESS=<TPU_IP_ADDRESS>
gcloud compute tpus create $USER-tpu-0001 \
      --zone=$ZONE \
      --network=default \
      --accelerator-type=v2-8 \
      --range=$TPU_IP_ADDRESS \
      --version=tpu_driver_nightly

Now that you have created both the user GCE VM and the TPU node, ssh to the GCE VM by executing the following command:

gcloud compute ssh $USER-user-vm-0001

Once you are in the VM, from your ssh terminal session, follow the example below to run a simple JAX program.

Install jax and jaxlib wheels:

pip install --user jax jaxlib

Create a program, simple_jax.py:

IMPORTANT: Replace <TPU_IP_ADDRESS> below with the TPU node’s IP address. You can get the IP address from the GCP console: Compute Engine > TPUs.

from jax.config import config
from jax import random

# The following is required to use TPU Driver as JAX's backend.
config.FLAGS.jax_xla_backend = "tpu_driver"
config.FLAGS.jax_backend_target = "grpc://<TPU-IP-ADDRESS>:8470"

key = random.PRNGKey(0)
x = random.normal(key, (10,))
print(x)

Run the program:

python simple_jax.py

Reporting issues

If you believe you’re experiencing a problem specific to using Cloud TPUs, please create an issue in the Cloud TPU issue tracker. If you’re unsure whether it’s problem with Cloud TPUs, JAX, Colab, or anything else, feel free to create an issue in the JAX issue tracker and we'll triage it appropriately.

If you have any other questions or comments regarding JAX on Cloud TPUs, please email [email protected]. We’d like to hear from you!