Overview

TensorFlow 2.14, released on September 8, 2023, requires NumPy 1.24+ and improves tf.function performance.

Main Features

NumPy 1.24+ required

Compatibility with NumPy 1.24+ enables new types and improves array interoperability.

python 
import tensorflow as tf
import numpy as np

# Seamless NumPy <-> TensorFlow conversion
arr = np.array([1.0, 2.0, 3.0], dtype=np.float32)
tensor = tf.constant(arr)
result = tf.math.reduce_sum(tensor)
print(result.numpy())  # 6.0

Improved tf.function

tf.function tracing is faster and uses less memory during graph compilation.

python 
import tensorflow as tf

@tf.function(reduce_retracing=True)
def train_step(x, y):
    with tf.GradientTape() as tape:
        pred = model(x, training=True)
        loss = loss_fn(y, pred)
    grads = tape.gradient(loss, model.trainable_variables)
    optimizer.apply_gradients(zip(grads, model.trainable_variables))
    return loss

Sources