matlab/matlab-agentic-toolkit / skills-catalog/ai-and-statistics/matlab-import-external-ai-model
matlab-import-external-ai-model
VerifiedImport PyTorch, ONNX, or Keras 3 / TensorFlow 2.16+ deep learning models into MATLAB as dlnetwork objects. Use when importing .pt2 exported programs, traced .pt files, .onnx models, or Keras 3 models via matlabsaver. Covers importNetworkFromPyTorch, importNetworkFromONNX, importNetworkFromKeras, importNetworkFromTensorFlow, torch.export.export, PyTorchInputSizes, InputDataFormats, matlabsaver, tf_keras downgrade, numeric validation against PyTorch or ONNX Runtime, and placeholder/custom layer implementation. Applies when user mentions any of these functions, file formats, or encounters import errors, unsupported operator warnings, 0 learnables, or uninitialized networks.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 11.1 KB |
| manifest.yaml | config | 626 B |
| references/keras-matlabsaver-workflow.md | markdown | 4.0 KB |
| references/keras-tf-keras-downgrade.md | markdown | 3.8 KB |
| references/onnx-validation-workflow.md | markdown | 6.4 KB |
| references/pytorch-export-guidance.md | markdown | 3.6 KB |
| references/pytorch-import-guidance.md | markdown | 3.9 KB |
| references/pytorch-numeric-validation.md | markdown | 4.3 KB |
| references/pytorch-placeholder-guidance.md | markdown | 4.3 KB |
| scripts/validateImportedNetwork.m | other | 4.3 KB |
references/keras-tf-keras-downgrade.md
View on GitHub# tf_keras Downgrade Workflow
Use this path when the Keras 3 model uses only standard layers (no `keras.ops`,
no multi-backend features) and the target MATLAB release is older than R2026a.
## Prerequisites
- MATLAB R2023b or newer (for `importNetworkFromTensorFlow`)
- Python environment with TensorFlow 2.16+
- The model must NOT use Keras 3-specific features
## Step 1: Install tf_keras
```bash
pip install tf_keras
```
This installs Keras 2 as a standalone package that TensorFlow 2.16+ can use
in place of the bundled Keras 3.
## Step 2: Configure the Environment
The environment variable MUST be set before any TensorFlow import:
```python
import os
os.environ["TF_USE_LEGACY_KERAS"] = "1"
# NOW import TensorFlow / tf_keras
import tf_keras as keras
from tf_keras import layers
```
### Critical: Import Ordering
```python
# WRONG - env var has no effect after TF import
import tensorflow as tf
os.environ["TF_USE_LEGACY_KERAS"] = "1" # Too late!
# CORRECT - env var before any TF import
os.environ["TF_USE_LEGACY_KERAS"] = "1"
import tf_keras as keras
```
### Jupyter Notebooks
If working in Jupyter, restart the kernel after installing `tf_keras`. The old
kernel has Keras 3 cached in memory and will not pick up the environment
variable change without a fresh process.
## Step 3: Verify Keras 2 Is Active
```python
import tf_keras as keras
assert not hasattr(keras, "version"), (
"Still using Keras 3! Check TF_USE_LEGACY_KERAS and import ordering."
)
print(f"Using: {keras.__name__}") # Should print: tf_keras
```
The `version` attribute is Keras 3-specific. Its absence confirms the
downgrade worked.
## Step 4: Build and Save the Model
Use `tf_keras` APIs (they mirror Keras 2 exactly):
```python
import tf_keras as keras
from tf_keras import layers
inputs = keras.Input(shape=(784,), name="input")
x = layers.Dense(128, activation="relu", name="hidden")(inputs)
outputs = layers.Dense(10, activation="softmax", name="output")(x)
model = keras.Model(inputs, outputs)
model.compile(optimizer="adam", loss="sparse_categorical_crossentropy")
model.fit(xTrain, yTrain, epochs=10)
# Save as SavedModel (Keras 2 format with keras_metadata.pb)
model.save("savedModelFolder")
```
The saved folder will contain `keras_metadata.pb` — this is what
`importNetworkFromTensorFlow` needs to decompose the layer structure.
## Step 5: Import into MATLAB
```matlab
net = importNetworkFromTensorFlow("savedModelFolder");
summary(net)
```
## Step 6: Verify
```matlab
numLearnables = numel(net.Learnables.Value);
assert(numLearnables > 0, "Import failed: 0 learnables")
disp("Layers: " + numel(net.Layers) + ", Learnables: " + numLearnables)
```
## When NOT to Use This Path
Do not use `tf_keras` if the model relies on:
- `keras.ops.*` (the backend-agnostic operation API)
- `keras.Operation` subclasses
- Multi-backend execution (JAX, PyTorch backends)
- Keras 3-only layer types (check Keras 3 changelog)
- Any feature introduced after Keras 2.x was frozen
If any of these are present, the model will fail to build under `tf_keras`.
Use Path 3 (ONNX export) instead.
## Troubleshooting
| Symptom | Cause | Fix |
|---------|-------|-----|
| `hasattr(keras, 'version')` returns `True` | `TF_USE_LEGACY_KERAS` not set early enough | Move env var to first line of script, before all imports |
| `ModuleNotFoundError: tf_keras` | Package not installed | Run `pip install tf_keras` |
| Model builds but `model.save()` produces `.keras` file | Using `keras` (Keras 3) instead of `tf_keras` | Change import to `import tf_keras as keras` |
| Kernel restart didn't help (Jupyter) | Multiple Python environments | Check `sys.executable` matches the env where tf_keras is installed |
| 0 learnables after import | SavedModel missing `keras_metadata.pb` | Verify the saved folder contains `keras_metadata.pb`. If missing, the save used Keras 3 not tf_keras |
----
Copyright 2026 The MathWorks, Inc.
----
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |