Keras · Module

model.get_weights

Model.get_weights() retrieves a model's current weights as a flat list of NumPy arrays. It is useful for inspecting or transferring weight values in memory.

How it works

model.get_weights() reads the model's tracked weights and returns their current values as a flat list of NumPy arrays. The arrays correspond to the model's weights in Keras' established weight ordering and can be inspected or transferred to another compatible object.

Syntax

Model.get_weights()

Returns

A flat list of NumPy arrays containing the model's weight values.

Key points

It reads and returns the model's current weight values.
The return value is a one-dimensional Python list whose elements are NumPy arrays.
For a layer, the method exposes the values represented by its tracked weight variables.
Pair it with set_weights() to copy in-memory weight values between compatible Keras objects.

Common mistakes

The result is a flat list, so each weight array is an individual element rather than a nested structure grouped by layer.

Example: Retrieve model weights

import numpy as np
import keras

model = keras.Sequential([
    keras.layers.Dense(3, input_shape=(2,)),
    keras.layers.Dense(1),
])

weights = model.get_weights()
print(len(weights))
print([w.shape for w in weights])

Creates a model, retrieves its weights as NumPy arrays, and prints the number and shape of each array.

When to use it

Examine learned weights

Inspecting parameters

Copy weights between compatible objects

In-memory transfer

Check initialized weight arrays

Debugging initialization

Interview questions

QWhat does Model.get_weights() return?

It returns the current values of the model's weights as a flat list of NumPy arrays. The list is suitable for inspecting or transferring model state in memory.

Follow-up: How would you restore those values? Use set_weights(weights) on a compatible model or layer, with matching weight structure and shapes.

QHow does get_weights() differ from accessing weights?

The method returns values rather than variable objects. Consequently, the result is convenient for NumPy-based inspection and for passing weight values to a compatible Keras object.

Follow-up: What distinction matters when inspecting variables directly? Layer properties such as weights expose weight variables, while get_weights() returns their values as NumPy arrays.

QWhen would you use get_weights() instead of saving weights?

Call it after the model has been built when you need the model's initialized or learned parameters. The returned arrays represent the model's current state at the time of the call.

Follow-up: How can the values be persisted? Use the documented weights-only saving APIs when storage on disk is required rather than keeping arrays only in memory.

Model.set_weights
Layer.get_weights
Layer.set_weights
Model.save_weights
Model.load_weights

Source: Keras documentation: Weights-only saving & loading. Checked 2026-10-15.