model.save() and load_model() — Three Formats

~10 min · save, load, keras-format

Level 0Level 0

0 XP0/78 lessons0/17 achievements

0/100 XP to next level100 XP to go0% complete

Choose the format before you need to deploy

Training a model is expensive. Saving lets you reuse it without retraining — for inference, continued training, deployment, sharing, publishing. TF provides multiple formats; the wrong choice creates deployment headaches.

The format is determined by the file extension you save with:

Format	Architecture	Weights	Optimizer	Best for
`.keras`	✅	✅	✅	Development checkpoints, framework-agnostic Keras 3
SavedModel directory	✅	✅	✅	Production: TF Serving, TFLite, TF.js
`.h5` (legacy)	✅	✅	✅	Avoid for new projects
weights only (`.weights.h5`)	❌	✅	❌	When you have the architecture code already

Save to .keras during development for maximum fidelity. Export to SavedModel for production serving / TFLite conversion / TF.js. Knowing the format differences before you need to deploy saves hours of debugging later.

Code

저장 + 불러오기 — 4 가지 패턴·python

import tensorflow as tf

# 1. Keras 3 native format (recommended for development)
model.save('model.keras')
loaded = tf.keras.models.load_model('model.keras')

# 2. SavedModel format (TF deployment ecosystem)
model.save('saved_model_dir/')          # creates a directory
# equivalent low-level API:
tf.saved_model.save(model, 'saved_model_dir/')
loaded = tf.keras.models.load_model('saved_model_dir/')

# 3. Legacy HDF5 (avoid for new code)
model.save('model.h5')
loaded = tf.keras.models.load_model('model.h5')

# 4. Weights only — needs same architecture to load
model.save_weights('weights.weights.h5')
model.load_weights('weights.weights.h5')

External links

Save and load models guide

Progress

Progress is local-only — sign in to sync across devices.

← 📝 Sequences, RNNs & AttentionQuiz · 5 questions Next →SavedModel Format In Depth

Spotted a bug or have feedback on this page?Report an Issue

Comments 0

🔔 Reply notifications (sign in)

No comments yet — be the first.