gradient_clip.py

# We must calculate the mean of each gradient. Note that this is the
# synchronization point across all towers.
grads_and_vars = average_gradients(t_grads)

# Optionally perform gradient clipping
if config.max_norm_gradient > 0:
    grads, variables = zip(*grads_and_vars)
    grads_clipped, _ = tf.clip_by_global_norm(grads, clip_norm=config.max_norm_gradient)
    grads_and_vars = zip(grads_clipped, variables)

# Apply the gradients to adjust the shared variables.
apply_gradient_op = optimizer.apply_gradients(grads_and_vars, global_step=global_step)