Video Tutorial

How to use decorators

Learn about the different ways to use tracing and transpilation functions.

⚠️ If you are running this notebook in Colab, you will have to install Ivy and some dependencies manually. You can do so by running the cell below ⬇️

If you want to run the notebook locally but don’t have Ivy installed just yet, you can check out the Get Started section of the docs.

!pip install ivy

import ivy
import numpy as np
import tensorflow as tf
import torch

ivy.set_backend("tensorflow")

Trace

In the example below, the ivy.trace_graph function is called as a decorator.

@ivy.trace_graph
def normalize(x):
    mean = ivy.mean(x)
    std = ivy.std(x, correction=1)
    return ivy.divide(ivy.subtract(x, mean), std)

x = tf.random.uniform((10,))
normalize(x) # tracing happens here

<tf.Tensor: shape=(10,), dtype=float64, numpy=
array([-0.34431235,  0.51129461, -0.06686894, -0.36452447, -0.98795534,
        0.15493582, -0.91630631,  1.41939619,  1.78909753, -1.19475674])>

Likewise, the function can still be called either eagerly or lazily when calling as a decorator. The example above is lazy, whereas the example below is eager:

@ivy.trace_graph(args=(x,))
def normalize(x):
    mean = ivy.mean(x)
    std = ivy.std(x, correction=1)
    return ivy.divide(ivy.subtract(x, mean), std)

normalize(x) # already traced

<tf.Tensor: shape=(10,), dtype=float64, numpy=
array([-0.34431235,  0.51129461, -0.06686894, -0.36452447, -0.98795534,
        0.15493582, -0.91630631,  1.41939619,  1.78909753, -1.19475674])>

Graph Transpile

In the example below, the ivy.graph_transpile function is called as a decorator.

@ivy.graph_transpile(source="torch", to="tensorflow")
def normalize(x):
    mean = torch.mean(x)
    std = torch.std(x)
    return torch.div(torch.sub(x, mean), std)

normalize(x) # transpilation happens here

<tf.Tensor: shape=(10,), dtype=float64, numpy=
array([-0.34431235,  0.51129461, -0.06686894, -0.36452447, -0.98795534,
        0.15493582, -0.91630631,  1.41939619,  1.78909753, -1.19475674])>

The function can still be called either eagerly or lazily when calling as a decorator. The example above is lazy, whereas the example below is eager:

@ivy.graph_transpile(source="torch", to="tensorflow", args=(x,))
def normalize(x):
    mean = torch.mean(x)
    std = torch.std(x)
    return torch.div(torch.sub(x, mean), std)

normalize(x) # already transpiled

<tf.Tensor: shape=(10,), dtype=float64, numpy=
array([-0.34431235,  0.51129461, -0.06686894, -0.36452447, -0.98795534,
        0.15493582, -0.91630631,  1.41939619,  1.78909753, -1.19475674])>

Transpile 🚧

In the future, ivy.transpile will be able to be used as a decorator like so:

@ivy.transpile(source="torch", target="tensorflow")
def normalize(x):
    mean = torch.mean(x)
    std = torch.std(x)
    return torch.div(torch.sub(x, mean), std)

# use normalize as a tensorflow function
normalize(tf.random.uniform((10,)))