JAX Colab GPU Test¶

import jax
import jaxlib

!cat /var/colab/hostname
print(jax.__version__)
print(jaxlib.__version__)

from jaxlib import xla_extension
import jax
key = jax.random.PRNGKey(1701)
arr = jax.random.normal(key, (1000,))
device = arr.device_buffer.device()
print(f"JAX device type: {device}")
assert isinstance(device, xla_extension.GpuDevice), "unexpected JAX device type"

import jax
import numpy as np

# matrix multiplication on GPU
key = jax.random.PRNGKey(0)
x = jax.random.normal(key, (3000, 3000))
result = jax.numpy.dot(x, x.T).mean()
print(result)

import jax.numpy as jnp
import jax.random as rand

N = 10
M = 20
key = rand.PRNGKey(1701)

X = rand.normal(key, (N, M))
u, s, vt = jnp.linalg.svd(X)
assert u.shape == (N, N)
assert vt.shape == (M, M)
print(s)

@jax.jit
def selu(x, alpha=1.67, lmbda=1.05):
  return lmbda * jax.numpy.where(x > 0, x, alpha * jax.numpy.exp(x) - alpha)
x = jax.random.normal(key, (5000,))
result = selu(x).block_until_ready()
print(result)