matlab-import-external-ai-model

Name: matlab-import-external-ai-model
Author: matlab

815stars105forks

How to install matlab-import-external-ai-model?

matlab/matlab-agentic-toolkit/matlab-import-external-ai-model

$npx -y skills add matlab/matlab-agentic-toolkit --skill matlab-import-external-ai-model

Installs into the current project.

Summary

Import 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, importN

Files · 1

View on GitHub

SKILL.md

1	# Import Deep Learning Models into MATLAB
2
3	Import trained PyTorch, ONNX, or Keras 3 models into MATLAB as `dlnetwork`
4	objects and verify numerical correctness.
5
6	## When to Use
7
8	- User wants to import a deep learning model from PyTorch, ONNX, or Keras/TensorFlow
9	- User has `.pt2`, `.pt`, `.onnx`, or `.keras` files to bring into MATLAB
10	- User mentions `importNetworkFromPyTorch`, `importNetworkFromONNX`, `importNetworkFromKeras`, or `importNetworkFromTensorFlow`
11	- User mentions `torch.export.export`, `torch.jit.trace`, `PyTorchInputSizes`, `InputDataFormats`, or `matlabsaver`
12	- User encounters import errors, unsupported operator warnings, uninitialized networks, or 0 learnables after import
13	- User wants to validate that an imported model matches the source framework's outputs
14
15	## When NOT to Use
16
17	- Exporting MATLAB networks to ONNX/PyTorch (use `exportONNXNetwork` / `exportNetworkToPyTorch`)
18	- Training or fine-tuning after import — use `/matlab-train-network`
19	- Deploying to embedded hardware — use `/matlab-deploy-embedded-ai`
20	- Simulink integration after import (agent handles this well without guidance)
21
22	## Router: Which Framework?
23
24	```
25	Q: What format is the source model?
26	\|
27	+-- .pt2 (PyTorch exported program) ──────────> PYTORCH IMPORT below
28	+-- .pt (PyTorch traced model) ───────────────> PYTORCH IMPORT below
29	+-- .onnx ────────────────────────────────────> ONNX IMPORT below
30	+-- .keras / TensorFlow 2.16+ / matlabsaver ──> KERAS IMPORT below
31	+-- Unknown ("import my model") ──────────────> Ask: framework? file extension?
32	```
33
34	---
35
36	## PyTorch Import
37
38	Full pipeline: export from PyTorch → import into MATLAB → validate numerics.
39
40	### Determine Starting Point
41
42	\| User has \| Action \|
43	\|----------\|--------\|
44	\| PyTorch model (code or saved) \| Export as .pt2 first → see `references/pytorch-export-guidance.md` \|
45	\| `.pt2` file (exported program) \| Import directly (below) \|
46	\| `.pt` file (traced model) \| Import with input sizes (below) \|
47
48	Always prefer .pt2 over .pt. If user has a traced model, recommend re-exporting
49	with `torch.export.export` first. Only use traced path if re-export is not feasible.
50
51	### Import .pt2 (Exported Program)
52
53	```matlab
54	net = importNetworkFromPyTorch("model.pt2");
55	```
56
57	No input size argument needed — shape info is embedded in the .pt2 file.
58
59	### Import .pt (Traced Model)
60
61	```matlab
62	net = importNetworkFromPyTorch("model.pt", ...
63	PyTorchInputSizes=[1 3 224 224]);
64	```
65
66	`PyTorchInputSizes` is mandatory for traced models. Specify sizes in PyTorch
67	dimension ordering. For multiple inputs use a cell array: `{[1 3 256 256], [1 10]}`.
68
69	### Name-Value Arguments
70
71	\| Argument \| When to use \|
72	\|----------\|-------------\|
73	\| `PyTorchInputSizes` \| Required for traced models (.pt). Not needed for .pt2 \|
74	\| `Namespace` \| Control where auto-generated custom layer files are stored \|
75	\| `PreferredNestingType` \| Choose `"networklayer"` (default) or `"customlayer"` \|
76
77	### PyTorch Critical Mistakes
78
79	\| Mistake \| Correct Approach \|
80	\|---------\|-----------------\|
81	\| Using `InputShape` NV argument \| Does not exist — use `PyTorchInputSizes` for .pt, nothing for .pt2 \|
82	\| Using `PackageName` NV argument \| Deprecated — use `Namespace` \|
83	\| Not calling `model.to("cpu")` before export \| Always `model.to("cpu")` before export \|
84	\| Not checking PyTorch version before export \| Assert `torch.__version__` starts with "2.8" \|
85	\| Passing `PyTorchInputSizes` for .pt2 \| Unnecessary — .pt2 embeds shape info, omit it \|
86	\| Guessing input size for unknown models \| Always ask the user for exact input dimensions \|
87	\| Assuming `net.InputNames` matches `forward()` order \| Importer may reorder — always check `net.InputNames` \|
88
89	### PyTorch Conventions
90
91	- Always `model.to("cpu")` and `model.eval()` before export
92	- Always verify PyTorch version is 2.8 before exporting as .pt2
93	- Never guess input sizes — ask the user or inspect the model
94	- Use `Namespace` not `PackageName` for custom layer storage
95	- Prefer .pt2 over .pt — recommend `torch.export.export` over `torch.jit.trace`
96
97	### PyTorch References
98
99	- `references/pytorch-export-guidance.md` — Full Python-side export procedure
100	- `references/pytorch-import-guidance.md` — Deta

People who liked this also viewed

6 picks

Type

matlab-import-external-ai-model

815stars105forks

View on GitHub

How to install matlab-import-external-ai-model?

matlab/matlab-agentic-toolkit/matlab-import-external-ai-model

$npx -y skills add matlab/matlab-agentic-toolkit --skill matlab-import-external-ai-model

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Import Deep Learning Models into MATLAB
2
3	Import trained PyTorch, ONNX, or Keras 3 models into MATLAB as `dlnetwork`
4	objects and verify numerical correctness.
5
6	## When to Use
7
8	- User wants to import a deep learning model from PyTorch, ONNX, or Keras/TensorFlow
9	- User has `.pt2`, `.pt`, `.onnx`, or `.keras` files to bring into MATLAB
10	- User mentions `importNetworkFromPyTorch`, `importNetworkFromONNX`, `importNetworkFromKeras`, or `importNetworkFromTensorFlow`
11	- User mentions `torch.export.export`, `torch.jit.trace`, `PyTorchInputSizes`, `InputDataFormats`, or `matlabsaver`
12	- User encounters import errors, unsupported operator warnings, uninitialized networks, or 0 learnables after import
13	- User wants to validate that an imported model matches the source framework's outputs
14
15	## When NOT to Use
16
17	- Exporting MATLAB networks to ONNX/PyTorch (use `exportONNXNetwork` / `exportNetworkToPyTorch`)
18	- Training or fine-tuning after import — use `/matlab-train-network`
19	- Deploying to embedded hardware — use `/matlab-deploy-embedded-ai`
20	- Simulink integration after import (agent handles this well without guidance)
21
22	## Router: Which Framework?
23
24	```
25	Q: What format is the source model?
26	\|
27	+-- .pt2 (PyTorch exported program) ──────────> PYTORCH IMPORT below
28	+-- .pt (PyTorch traced model) ───────────────> PYTORCH IMPORT below
29	+-- .onnx ────────────────────────────────────> ONNX IMPORT below
30	+-- .keras / TensorFlow 2.16+ / matlabsaver ──> KERAS IMPORT below
31	+-- Unknown ("import my model") ──────────────> Ask: framework? file extension?
32	```
33
34	---
35
36	## PyTorch Import
37
38	Full pipeline: export from PyTorch → import into MATLAB → validate numerics.
39
40	### Determine Starting Point
41
42	\| User has \| Action \|
43	\|----------\|--------\|
44	\| PyTorch model (code or saved) \| Export as .pt2 first → see `references/pytorch-export-guidance.md` \|
45	\| `.pt2` file (exported program) \| Import directly (below) \|
46	\| `.pt` file (traced model) \| Import with input sizes (below) \|
47
48	Always prefer .pt2 over .pt. If user has a traced model, recommend re-exporting
49	with `torch.export.export` first. Only use traced path if re-export is not feasible.
50
51	### Import .pt2 (Exported Program)
52
53	```matlab
54	net = importNetworkFromPyTorch("model.pt2");
55	```
56
57	No input size argument needed — shape info is embedded in the .pt2 file.
58
59	### Import .pt (Traced Model)
60
61	```matlab
62	net = importNetworkFromPyTorch("model.pt", ...
63	PyTorchInputSizes=[1 3 224 224]);
64	```
65
66	`PyTorchInputSizes` is mandatory for traced models. Specify sizes in PyTorch
67	dimension ordering. For multiple inputs use a cell array: `{[1 3 256 256], [1 10]}`.
68
69	### Name-Value Arguments
70
71	\| Argument \| When to use \|
72	\|----------\|-------------\|
73	\| `PyTorchInputSizes` \| Required for traced models (.pt). Not needed for .pt2 \|
74	\| `Namespace` \| Control where auto-generated custom layer files are stored \|
75	\| `PreferredNestingType` \| Choose `"networklayer"` (default) or `"customlayer"` \|
76
77	### PyTorch Critical Mistakes
78
79	\| Mistake \| Correct Approach \|
80	\|---------\|-----------------\|
81	\| Using `InputShape` NV argument \| Does not exist — use `PyTorchInputSizes` for .pt, nothing for .pt2 \|
82	\| Using `PackageName` NV argument \| Deprecated — use `Namespace` \|
83	\| Not calling `model.to("cpu")` before export \| Always `model.to("cpu")` before export \|
84	\| Not checking PyTorch version before export \| Assert `torch.__version__` starts with "2.8" \|
85	\| Passing `PyTorchInputSizes` for .pt2 \| Unnecessary — .pt2 embeds shape info, omit it \|
86	\| Guessing input size for unknown models \| Always ask the user for exact input dimensions \|
87	\| Assuming `net.InputNames` matches `forward()` order \| Importer may reorder — always check `net.InputNames` \|
88
89	### PyTorch Conventions
90
91	- Always `model.to("cpu")` and `model.eval()` before export
92	- Always verify PyTorch version is 2.8 before exporting as .pt2
93	- Never guess input sizes — ask the user or inspect the model
94	- Use `Namespace` not `PackageName` for custom layer storage
95	- Prefer .pt2 over .pt — recommend `torch.export.export` over `torch.jit.trace`
96
97	### PyTorch References
98
99	- `references/pytorch-export-guidance.md` — Full Python-side export procedure
100	- `references/pytorch-import-guidance.md` — Deta

People who liked this also viewed

6 picks

Type