Saving weights only in TensorFlow

Introduction

Saving weights only lets you keep just the learned parts of a model. This saves space and lets you reuse the model structure later.

You want to save disk space by not saving the full model.

You plan to rebuild the model architecture in code and load saved weights.

You want to share only the learned parameters without the model code.

You want to continue training later from saved weights.

Syntax

TensorFlow

model.save_weights('path_to_file')

model.load_weights('path_to_file')

The file path can be a filename or a folder path depending on format.

You must have the model architecture defined before loading weights.

Examples

Save and load weights using HDF5 format file.

TensorFlow

model.save_weights('weights.h5')
model.load_weights('weights.h5')

Save and load weights using TensorFlow checkpoint format.

TensorFlow

model.save_weights('checkpoints/weights')
model.load_weights('checkpoints/weights')

Sample Model

This code trains a simple model, saves only its weights, then creates a new model with the same structure and loads the saved weights. Finally, it evaluates the new model to show the weights were loaded correctly.

TensorFlow

import tensorflow as tf

# Create a simple model
model = tf.keras.Sequential([
    tf.keras.layers.Dense(5, activation='relu', input_shape=(3,)),
    tf.keras.layers.Dense(2, activation='softmax')
])

# Compile the model
model.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy'])

# Dummy data
import numpy as np
x = np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9]], dtype=np.float32)
y = np.array([0, 1, 1], dtype=np.int32)

# Train the model
model.fit(x, y, epochs=2, verbose=0)

# Save weights only
model.save_weights('my_weights')

# Create a new model with the same architecture
new_model = tf.keras.Sequential([
    tf.keras.layers.Dense(5, activation='relu', input_shape=(3,)),
    tf.keras.layers.Dense(2, activation='softmax')
])

# Compile new model
new_model.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy'])

# Load saved weights
new_model.load_weights('my_weights')

# Evaluate new model on same data
loss, accuracy = new_model.evaluate(x, y, verbose=0)
print(f'Loss: {loss:.4f}, Accuracy: {accuracy:.4f}')

Important Notes

You must define the model architecture exactly the same before loading weights.

Saving weights only is smaller than saving the full model but needs the model code to rebuild.

Use .h5 extension for HDF5 format or no extension for TensorFlow checkpoint format.

Summary

Saving weights only stores just the learned parameters, not the model structure.

You need to recreate the model architecture before loading weights.

This method is useful for saving space and sharing learned parameters.

Practice

1. What does saving weights only in TensorFlow mean?

easy

A. Saving the entire model including architecture and optimizer

B. Saving just the learned parameters without the model structure

C. Saving only the model architecture without weights

D. Saving the training dataset along with the model

Solution

1. Step 1: Understand what weights represent

   Weights are the learned parameters of a model after training.

2. Step 2: Differentiate saving weights only vs full model

   Saving weights only means storing these parameters without saving the model's structure or optimizer state.

3. Final Answer:

   Saving just the learned parameters without the model structure -> Option B

4. Quick Check:

   Saving weights only = learned parameters only [OK]

Hint: Weights only means parameters, not architecture or optimizer [OK]

Common Mistakes:

• Confusing weights with full model saving
• Thinking architecture is saved with weights only
• Assuming dataset is saved with weights

2. Which of the following is the correct syntax to save only weights of a TensorFlow model named model to a file weights.h5?

easy

A. model.save_weights(weights='weights.h5')

B. model.save('weights.h5')

C. model.save_weights('weights.h5')

D. tf.save_weights(model, 'weights.h5')

Solution

1. Step 1: Recall TensorFlow method to save weights

   The method to save weights is model.save_weights(filepath).

2. Step 2: Check syntax correctness

   model.save_weights('weights.h5') uses correct method and argument format. model.save('weights.h5') saves full model, C is invalid TensorFlow syntax, A uses wrong argument name.

3. Final Answer:

   model.save_weights('weights.h5') -> Option C

4. Quick Check:

   Use model.save_weights(filepath) to save weights [OK]

Hint: Use model.save_weights('filename') to save weights only [OK]

Common Mistakes:

• Using model.save() instead of save_weights()
• Incorrect argument names or order
• Trying to use non-existent TensorFlow functions

3. Consider this code snippet:

import tensorflow as tf
model = tf.keras.Sequential([
  tf.keras.layers.Dense(5, input_shape=(3,))
])
model.save_weights('w.h5')

new_model = tf.keras.Sequential([
  tf.keras.layers.Dense(5, input_shape=(3,))
])
new_model.load_weights('w.h5')
print(new_model.get_weights()[0].shape)

What will be the printed output?

medium

A. (3, 5)

B. (5, 3)

C. (5,)

D. Error: weights shape mismatch

Solution

1. Step 1: Understand model architecture

   The model has one Dense layer with 5 units and input shape (3,), so weights shape is (input_dim, units) = (3, 5).

2. Step 2: Loading weights into identical architecture

   Since new_model has the same architecture, loading weights works and weights shape remains (3, 5).

3. Final Answer:

   (3, 5) -> Option A

4. Quick Check:

   Weights shape = (input_dim, units) = (3, 5) [OK]

Hint: Weights shape = (input features, output units) in Dense layers [OK]

Common Mistakes:

• Confusing input and output dimensions order
• Assuming error when architectures match
• Expecting bias shape instead of weights

4. You saved weights using model.save_weights('w.h5'). When loading weights with new_model.load_weights('w.h5'), you get an error about mismatched shapes. What is the most likely cause?

medium

A. TensorFlow does not support loading weights from .h5 files

B. The weights file is corrupted

C. You forgot to compile the new model before loading weights

D. The new model architecture is different from the original model

Solution

1. Step 1: Understand weight loading requirements

   Weights can only be loaded if the new model has the same architecture as the original model.

2. Step 2: Analyze error cause

   Mismatched shapes error usually means layer sizes differ between saved and new model.

3. Final Answer:

   The new model architecture is different from the original model -> Option D

4. Quick Check:

   Model architectures must match to load weights [OK]

Hint: Check model architectures match before loading weights [OK]

Common Mistakes:

• Assuming file corruption without checking architecture
• Thinking compile is needed before loading weights
• Believing .h5 files are unsupported for weights

5. You want to share your trained model's learned parameters with a friend, but not the full model code. Which approach is best to save and share only the weights, and what must your friend do to use them?

hard

A. Save weights with model.save_weights(); friend recreates model architecture then loads weights

B. Save full model with model.save(); friend loads full model directly

C. Save weights as a numpy file; friend loads weights without model code

D. Save model architecture as JSON; friend loads architecture and trains from scratch

Solution

1. Step 1: Identify saving weights only method

   Use model.save_weights() to save only learned parameters without model structure.

2. Step 2: Explain how friend uses weights

   Friend must recreate the exact model architecture in code, then load weights with load_weights().

3. Final Answer:

   Save weights with model.save_weights(); friend recreates model architecture then loads weights -> Option A

4. Quick Check:

   Save weights only + recreate model to load weights [OK]

Hint: Save weights only, friend rebuilds model then loads weights [OK]

Common Mistakes:

• Thinking weights can be loaded without model code
• Confusing full model save with weights only
• Assuming JSON architecture saves weights