Function profile

Executes the provided function f() and returns a promise that resolves with information about the function's memory use:

• newBytes: the number of new bytes allocated
• newTensors: the number of new tensors created
• peakBytes: the peak number of bytes allocated
• kernels: an array of objects for each kernel involved that reports their input and output shapes, number of bytes used, and number of new tensors created.
• kernelNames: an array of unique strings with just the names of the kernels in the kernels array.

const profile = await tf.profile(() => {
  const x = tf.tensor1d([1, 2, 3]);
  let x2 = x.square();
  x2.dispose();
  x2 = x.square();
  x2.dispose();
  return x;
});

console.log(`newBytes: ${profile.newBytes}`);
console.log(`newTensors: ${profile.newTensors}`);
console.log(`byte usage over all kernels: ${profile.kernels.map(k =>
k.totalBytesSnapshot)}`);
Copy

Doc