Google JAX

JAX

Google JAX logo
Developer(s)	Google
Preview release	v0.4.31 / 30 July 2024 (2024-07-30)
Repository	github.com/google/jax
Written in	Python, C++
Operating system	Linux, macOS, Windows
Platform	Python, NumPy
Size	9.0 MB
Type	Machine learning
License	Apache 2.0
Website	jax.readthedocs.io/en/latest/

Google JAX is a machine learning framework for transforming numerical functions. It is described as bringing together a modified version of autograd (automatic obtaining of the gradient function through differentiation of a function) and TensorFlow's XLA (Accelerated Linear Algebra). It is designed to follow the structure and workflow of NumPy as closely as possible and works with various existing frameworks such as TensorFlow and PyTorch. The primary functions of JAX are:

1. grad: automatic differentiation
2. jit: compilation
3. vmap: auto-vectorization
4. pmap: SPMD programming

grad

Main article: Automatic differentiation

The below code demonstrates the grad function's automatic differentiation.

# imports
from jax import grad
import jax.numpy as jnp

# define the logistic function
def logistic(x):  
    return jnp.exp(x) / (jnp.exp(x) + 1)

# obtain the gradient function of the logistic function
grad_logistic = grad(logistic)

# evaluate the gradient of the logistic function at x = 1 
grad_log_out = grad_logistic(1.0))   
print(grad_log_out)

The final line should outputː

0.19661194

jit

The below code demonstrates the jit function's optimization through fusion.

# imports
from jax import jit
import jax.numpy as jnp

# define the cube function
def cube(x):
    return x * x * x

# generate data
x = jnp.ones((10000, 10000))

# create the jit version of the cube function
jit_cube = jit(cube)

# apply the cube and jit_cube functions to the same data for spreed comoparion
cube(x)
jit_cube(x)

The computation time for jit_cube (line #17) should be noticeably shorter than that for cube (line #16). Increasing the values on line #7, will further exacerbate the difference.

vmap

Main article: Array programming

The below code demonstrates the vmap function's vectorization.

# imports
from jax import vmap partial
import jax.numpy as jnp

# define function
def grads(self, inputs):
    in_grad_partial = jax.partial(self._net_grads, self._net_params)
    grad_vmap = jax.vmap(in_grad_partial)
    rich_grads = grad_vmap(inputs)
    flat_grads = np.asarray(self._flatten_batch(rich_grads))
    assert flat_grads.ndim == 2 and flat_grads.shape[0] == inputs.shape[0]
    return flat_grads

The GIF on the right of this section illustrates the notion of vectorized addition.

Illustration video of vectorized addition

pmap

The below code demonstrates the pmap function's parallelization for matrix multiplication.

# import pmap and random from JAX; import JAX NumPy
from jax import pmap, random
import jax.numpy as jnp

# generate 2 random matrices of dimensions 5000 x 6000, one per device
random_keys = random.split(random.PRNGKey(0), 2)
matrices = pmap(lambda key: random.normal(key, (5000, 6000)))(random_keys)

# without data transfer, in parallel, perform a local matrix multiplication on each CPU/GPU
outputs = pmap(lambda x: jnp.dot(x, x.T))(matrices)

# without data transfer, in parallel, obtain the mean for both matrices on each CPU/GPU separately
means = pmap(jnp.mean)(outputs)
print(means)

The final line should print the valuesː

[1.1566595 1.1805978]

See also

• NumPy
• TensorFlow
• PyTorch
• CUDA

External links

• Documentationː jax.readthedocs.io
• Colab (Jupyter/iPython) Quickstart Guideː colab.research.google.com/github/google/jax/blob/main/docs/notebooks/quickstart.ipynb
• TensorFlow's XLAː www.tensorflow.org/xla (Accelerated Linear Algebra)
• YouTube TensorFlow Channel "Intro to JAX: Accelerating Machine Learning research": www.youtube.com/watch?v=WdTeDXsOSj4
• Original paperː mlsys.org/Conferences/doc/2018/146.pdf