matlab-deploy-ai-model

Name: matlab-deploy-ai-model
Author: matlab

815stars105forks

How to install matlab-deploy-ai-model?

matlab/matlab-agentic-toolkit/matlab-deploy-ai-model

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

Installs into the current project.

Summary

Generate C/C++ or CUDA code from an AI model (PyTorch, LiteRT) using MATLAB Coder or GPU Coder. Use when the user wants to integrate an AI model into an application with code generation as the end goal — generating MEX, CUDA MEX, static library, dynamic library, or executable. Cu

Files · 1

View on GitHub

SKILL.md

1	# Generate C/C++/CUDA Code from an AI Model
2
3	Generate deployable C/C++ or CUDA code from an AI model using MATLAB Coder or
4	GPU Coder. The workflow follows a common pattern regardless of model framework:
5	load, inspect, write entry-point, generate MEX, verify, then generate production code.
6
7	## When to Use
8
9	- User wants to generate C/C++/CUDA code from an AI model (PyTorch, LiteRT)
10	- User has a model file (.pt2, .tflite) and wants to load it into MATLAB
11	- User wants MEX acceleration for an AI model
12	- User wants to generate CUDA code or GPU-accelerated MEX from an AI model
13	- User wants to deploy an AI model to hardware
14	- User wants to verify AI model numerics between the source framework and MATLAB
15
16	## When NOT to Use
17
18	- General MATLAB Coder usage (codegen syntax, config tuning, writing codegen-ready code)
19	- Editable dlnetwork for Deep Learning Toolbox workflows (quantization, compression, transfer learning) — use `importNetworkFromPyTorch` which returns a `dlnetwork` for PyTorch models
20	- Training or fine-tuning — this skill is for inference code generation only
21
22	## Supported Frameworks
23
24	\| Framework \| Model format \| Load function \| Status \|
25	\|-----------\|-------------\|---------------\|--------\|
26	\| PyTorch \| `.pt2` \| `loadPyTorchExportedProgram` \| Supported (R2026a+) \|
27	\| LiteRT / TFLite \| `.tflite` \| `loadLiteRTModel` \| Supported (R2026a+) \|
28
29	For PyTorch-specific details (API routing, entry-point pattern, export workflow,
30	data layout, common mistakes): see `references/pytorch-workflow.md`.
31
32	## Generic Workflow
33
34	The code generation workflow follows the same steps for any framework:
35
36	### 1. Load and Inspect
37
38	Load the model and check its input/output specifications to determine expected
39	shapes and types.
40
41	### 2. Write Entry-Point Function
42
43	Create a codegen-compatible entry-point function that:
44	- Loads the model from a file path
45	- Runs inference on an input
46	- Returns the output
47
48	The model file path must be wrapped with `coder.Constant` so it's known at
49	compile time.
50
51	### 3. Verify Numerics
52
53	Compare MATLAB inference output against the source framework to confirm correct
54	loading. Use the same input data in both environments and compare with tolerance.
55
56	### 4. Generate MEX (First!)
57
58	Always generate MEX before lib/exe to verify on the host machine:
59
60	CPU MEX:
61	```matlab
62	cfg = coder.config("mex");
63	codegen -config cfg -args {coder.Constant("model_file"), input} entryPoint
64	```
65
66	CUDA MEX (GPU acceleration):
67	```matlab
68	cfg = coder.gpuConfig("mex");
69	codegen -config cfg -args {coder.Constant("model_file"), input} entryPoint
70	```
71
72	### 5. Verify MEX Output
73
74	Compare MEX output against MATLAB reference using `matlab.unittest` with
75	tolerance:
76
77	```matlab
78	refOut = entryPoint("model_file", input);
79	mexOut = entryPoint_mex("model_file", input);
80	testCase = matlab.unittest.TestCase.forInteractiveUse;
81	testCase.verifyThat(mexOut, matlab.unittest.constraints.IsEqualTo(refOut, ...
82	'Within', matlab.unittest.constraints.AbsoluteTolerance(single(1e-5))));
83	```
84
85	### 6. Generate Library/Executable
86
87	Once MEX is verified, generate production code:
88
89	```matlab
90	cfgLib = coder.config("lib");
91	codegen -config cfgLib -args {coder.Constant("model_file"), input} entryPoint
92	```
93
94	For DLL: `coder.config("dll")`. For executable: `coder.config("exe")`.
95
96	CUDA variants: Replace `coder.config` with `coder.gpuConfig`.
97
98	### 7. Deploy to Hardware (Optional — requires Embedded Coder)
99
100	For embedded deployment, use the same entry-point function with an Embedded Coder
101	configuration. See the `matlab-deploy-embedded-code` skill for ERT config,
102	hardware settings, PIL/SIL verification, and target-specific options.
103	Ask the user to install the skill if it is not installed
104
105	## Key Functions
106
107	\| Function \| Purpose \| Package \| Since \|
108	\|----------\|---------\|---------\|-------\|
109	\| `coder.Constant` \| Make argument a compile-time constant \| MATLAB Coder \| R2011a \|
110	\| `coder.gpuConfig` \| Create GPU (CUDA) code generation config \| GPU Coder \| R2017b \|
111	\| `codegen` \| Generate code \| MATLAB Coder \| R2011a \|
112	\| `loadPyTorchExportedProgram` \| Load .pt2 into MATLAB \| MATLAB Coder Support Package for PyTorch and LiteRT Models \| R2026a \|
113	\| `loadLiteRTModel` \| Load .tflite into MATLAB \| MATLAB Coder Support Package for PyTorch and LiteRT Models \| R2026a \|
114
115	## Conventions

People who liked this also viewed

6 picks

Type

matlab-deploy-ai-model

815stars105forks

View on GitHub

How to install matlab-deploy-ai-model?

matlab/matlab-agentic-toolkit/matlab-deploy-ai-model

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

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Generate C/C++/CUDA Code from an AI Model
2
3	Generate deployable C/C++ or CUDA code from an AI model using MATLAB Coder or
4	GPU Coder. The workflow follows a common pattern regardless of model framework:
5	load, inspect, write entry-point, generate MEX, verify, then generate production code.
6
7	## When to Use
8
9	- User wants to generate C/C++/CUDA code from an AI model (PyTorch, LiteRT)
10	- User has a model file (.pt2, .tflite) and wants to load it into MATLAB
11	- User wants MEX acceleration for an AI model
12	- User wants to generate CUDA code or GPU-accelerated MEX from an AI model
13	- User wants to deploy an AI model to hardware
14	- User wants to verify AI model numerics between the source framework and MATLAB
15
16	## When NOT to Use
17
18	- General MATLAB Coder usage (codegen syntax, config tuning, writing codegen-ready code)
19	- Editable dlnetwork for Deep Learning Toolbox workflows (quantization, compression, transfer learning) — use `importNetworkFromPyTorch` which returns a `dlnetwork` for PyTorch models
20	- Training or fine-tuning — this skill is for inference code generation only
21
22	## Supported Frameworks
23
24	\| Framework \| Model format \| Load function \| Status \|
25	\|-----------\|-------------\|---------------\|--------\|
26	\| PyTorch \| `.pt2` \| `loadPyTorchExportedProgram` \| Supported (R2026a+) \|
27	\| LiteRT / TFLite \| `.tflite` \| `loadLiteRTModel` \| Supported (R2026a+) \|
28
29	For PyTorch-specific details (API routing, entry-point pattern, export workflow,
30	data layout, common mistakes): see `references/pytorch-workflow.md`.
31
32	## Generic Workflow
33
34	The code generation workflow follows the same steps for any framework:
35
36	### 1. Load and Inspect
37
38	Load the model and check its input/output specifications to determine expected
39	shapes and types.
40
41	### 2. Write Entry-Point Function
42
43	Create a codegen-compatible entry-point function that:
44	- Loads the model from a file path
45	- Runs inference on an input
46	- Returns the output
47
48	The model file path must be wrapped with `coder.Constant` so it's known at
49	compile time.
50
51	### 3. Verify Numerics
52
53	Compare MATLAB inference output against the source framework to confirm correct
54	loading. Use the same input data in both environments and compare with tolerance.
55
56	### 4. Generate MEX (First!)
57
58	Always generate MEX before lib/exe to verify on the host machine:
59
60	CPU MEX:
61	```matlab
62	cfg = coder.config("mex");
63	codegen -config cfg -args {coder.Constant("model_file"), input} entryPoint
64	```
65
66	CUDA MEX (GPU acceleration):
67	```matlab
68	cfg = coder.gpuConfig("mex");
69	codegen -config cfg -args {coder.Constant("model_file"), input} entryPoint
70	```
71
72	### 5. Verify MEX Output
73
74	Compare MEX output against MATLAB reference using `matlab.unittest` with
75	tolerance:
76
77	```matlab
78	refOut = entryPoint("model_file", input);
79	mexOut = entryPoint_mex("model_file", input);
80	testCase = matlab.unittest.TestCase.forInteractiveUse;
81	testCase.verifyThat(mexOut, matlab.unittest.constraints.IsEqualTo(refOut, ...
82	'Within', matlab.unittest.constraints.AbsoluteTolerance(single(1e-5))));
83	```
84
85	### 6. Generate Library/Executable
86
87	Once MEX is verified, generate production code:
88
89	```matlab
90	cfgLib = coder.config("lib");
91	codegen -config cfgLib -args {coder.Constant("model_file"), input} entryPoint
92	```
93
94	For DLL: `coder.config("dll")`. For executable: `coder.config("exe")`.
95
96	CUDA variants: Replace `coder.config` with `coder.gpuConfig`.
97
98	### 7. Deploy to Hardware (Optional — requires Embedded Coder)
99
100	For embedded deployment, use the same entry-point function with an Embedded Coder
101	configuration. See the `matlab-deploy-embedded-code` skill for ERT config,
102	hardware settings, PIL/SIL verification, and target-specific options.
103	Ask the user to install the skill if it is not installed
104
105	## Key Functions
106
107	\| Function \| Purpose \| Package \| Since \|
108	\|----------\|---------\|---------\|-------\|
109	\| `coder.Constant` \| Make argument a compile-time constant \| MATLAB Coder \| R2011a \|
110	\| `coder.gpuConfig` \| Create GPU (CUDA) code generation config \| GPU Coder \| R2017b \|
111	\| `codegen` \| Generate code \| MATLAB Coder \| R2011a \|
112	\| `loadPyTorchExportedProgram` \| Load .pt2 into MATLAB \| MATLAB Coder Support Package for PyTorch and LiteRT Models \| R2026a \|
113	\| `loadLiteRTModel` \| Load .tflite into MATLAB \| MATLAB Coder Support Package for PyTorch and LiteRT Models \| R2026a \|
114
115	## Conventions

People who liked this also viewed

6 picks

Type