- name
SKILL.md
技能檔案保留在發布者的儲存庫中。
在終端機中貼上複製的指令。它會將這個 SKILL.md 安裝到對應的代理中。
npx skills add matlab/matlab-agentic-toolkit --skill matlab-deploy-embedded-ai -a cursor -y在終端機中貼上複製的指令。它會將這個 SKILL.md 安裝到對應的代理中。
npx skills add matlab/matlab-agentic-toolkit --skill matlab-deploy-embedded-ai -a claude-code -y在終端機中貼上複製的指令。它會將這個 SKILL.md 安裝到對應的代理中。
npx skills add matlab/matlab-agentic-toolkit --skill matlab-deploy-embedded-ai -a codex -y在終端機中貼上複製的指令。它會將這個 SKILL.md 安裝到對應的代理中。
npx skills add matlab/matlab-agentic-toolkit --skill matlab-deploy-embedded-ai -a grok -y在 Grok Bot 聊天中傳送複製的訊息。Bot 會從 GitHub 取得這個 SKILL.md,並加入你所有 Bot 共用的技能中。
Install this skill: https://github.com/matlab/matlab-agentic-toolkit/tree/HEAD/skills-catalog/ai-and-statistics/matlab-deploy-embedded-ai來自發布者
Deploy AI models to embedded hardware using MATLAB® and Simulink®. This skill is written specifically for MATLAB R2026a and uses APIs, functions, and workflows introduced in that release. It covers the complete lifecycle: model creation or import, verification, compression, system-level simulation, and code generation for resource-constrained targets.
Requires MATLAB R2026a or newer. Core toolboxes: Deep Learning Toolbox, Statistics and Machine Learning Toolbox, MATLAB Coder, Embedded Coder, Simulink, and Fixed-Point Designer. Workflow-specific support packages are checked during Environment Discovery. The MATLAB and Simulink Agentic Toolkits must be available so the agent can drive a live MATLAB and Simulink session through MCP tools.
/matlab-import-external-ai-model)loadPyTorchExportedProgram, importNetworkFromPyTorch, importNetworkFromONNX, importNetworkFromTensorFlow, dlquantizer, exportNetworkToSimulink, or Embedded Coder with AI modelsfitcnet/fitrnet neural network models ARE covered by this skillThis skill uses two deployment patterns:
dlnetwork, optionally
compress it, then generate C/C++ via MATLAB Coder or export to Simulink.
See references/pattern1/workflow.md.references/pattern2/workflow.md.Primary discriminator for external models: deployment capabilities + hardware class.
Q1: Where does the AI model come from?
|
+-- Trained in MATLAB, or requires training in MATLAB -------> Pattern 1
|
+-- External framework (PyTorch, TF, ONNX, Keras) --> Q2
|
Q2: Does the deployment need any of these?
| - Quantization (INT8 via dlquantizer)
| - Pruning or projection
| - Weight inspection / modification
| - exportNetworkToSimulink integration
|
+-- YES --> Pattern 1 (import as dlnetwork)
|
+-- NO ---> Q3
|
Q3: What is the deployment target?
|
+-- Cortex-M: Pattern 1
| (compression and Simulink verification typically needed)
|
+-- x86 / GPU:
| +-- PyTorch (.pt2) or LiteRT (.tflite) --> Pattern 2
| +-- ONNX, TF, Keras --> Pattern 1 (convert to ONNX recommended)
|
+-- Cortex-A/R:
+-- Small model --> Pattern 1
| (import as dlnetwork, then codegen)
+-- Large model:
+-- PyTorch (.pt2) or LiteRT (.tflite) --> Pattern 2
+-- ONNX, TF, Keras --> Pattern 1 (convert to ONNX recommended)
Pattern 2's generated C is portable to any target, but Cortex-M deployments typically require Pattern 1 capabilities (compression, Simulink verification).
Import step (Pattern 1): For PyTorch/ONNX/Keras/TensorFlow model import, use /matlab-import-external-ai-model. This skill takes over after import for the compression, Simulink integration, and code generation phases.
Rule of thumb: Choose Pattern 1 when you need to compress, quantize, inspect
weights, or use exportNetworkToSimulink — or when the model is already a
dlnetwork, or when targeting Cortex-M. Choose Pattern 2 when the model is already
in PyTorch (.pt2) or LiteRT (.tflite) format and you want the shorter path to C code
without compression.
Stats/ML models (fitrnet/fitcnet): These follow Pattern 1 but have their own
Simulink integration path. Use the RegressionNeuralNetwork Predict block (for
fitrnet) or ClassificationNeuralNetwork Predict block (for fitcnet) from
the Statistics and Machine Learning Toolbox library — do NOT use
exportNetworkToSimulink (which is for dlnetwork only). Configure simulation
programmatically with Simulink.SimulationInput.
Regardless of pattern, always begin with these two prerequisite steps before entering the pattern-specific phases (which start at Phase 1):
references/shared/environment-setup.mdreferences/shared/project-discovery.mdProject Discovery determines the workflow pattern via the decision tree above.
For legacy import functions (importONNXNetwork, importKerasNetwork, etc.), see
/matlab-import-external-ai-model which handles all model import workflows.
Distinguish between two modes based on the user's intent:
detect_matlab_toolboxes and support packages via matlabshared.supportpkg.getInstalled before any workflow stepsingle precision for all inference inputs.m script file and execute it with run_matlab_file or evaluate_matlab_code. Do NOT run ad-hoc MATLAB commands without first writing the script file. If a script needs changes, edit the script file and re-run it. This gives users full visibility into what code is being executed and enables reproducibility. IMPORTANT: run_matlab_file sets the working directory to the script's folder. Always use absolute paths (via fullfile) for model files, data, and saved outputs — never rely on pwd or relative paths.deepNetworkDesigner(net)) so the user can visually inspect the architecture. Announce this action and wait for user acknowledgment before proceeding.references/pattern1/compression-decision.md — hardware target + Simulink availability, primary goal, retraining tolerance. The answers determine the compression techniques and the code-replacement library to use. Even if the hardware target is already stated, you MUST still ask about the primary deployment goal (flash, SRAM, latency, accuracy) and present the user with the recommended recipe for confirmation before executing any compression step.DAGNetwork, SeriesNetwork, or network objectscompression-decision.md[C × T] array with format "CBT" to a sequence model — always reshape to [C × 1 × T] for single-sequence inferencedlnetwork to prepareNetwork — in R2026a the function takes a dlquantizer object (prepareNetwork(quantObj)) and mutates it in place. The legacy net = prepareNetwork(net) form is no longer definedThis skill covers C/C++ code generation for CPU targets (Cortex-M, Cortex-A/R,
x86) via MATLAB Coder and Embedded Coder, and GPU targets (CUDA/TensorRT) via
Pattern 2's DeepLearningConfig options. The following deployment targets use
different toolchains and are NOT covered:
If the user's target falls into one of these categories, inform them that this skill does not cover that workflow and suggest the relevant toolchain.
MATLAB and Simulink are registered trademarks of The MathWorks, Inc. See www.mathworks.com/trademarks for a list of additional trademarks. Other product or brand names may be trademarks or registered trademarks of their respective holders.
Copyright 2026 The MathWorks, Inc.
>
---name: matlab-deploy-embedded-aidescription: >---# skills-catalog/ai-and-statistics/matlab-deploy-embedded-ai/SKILL.md常見問題
從發布者下載 skills-catalog/ai-and-statistics/matlab-deploy-embedded-ai/SKILL.md,或複製路徑後開啟儲存庫。PlugMCP 只收錄這項技能,不代管檔案。
發布者將它存放在 https://github.com/matlab/matlab-agentic-toolkit。
Claude、Cursor、Codex 和 Grok Build 可以讀取你從發布者儲存庫加入的 SKILL.md。在聊天中傳送技能的 GitHub 資料夾連結,Grok Bot 就會安裝它。
fprintfdisphelp <function> output and on the function's reference page. Use those as the source of truth instead of any inline parameter table in this skill — inline tables go stale across releases and burn context. This skill only flags name-value arguments that materially change the recipe (e.g., ValidationThreshold for accuracy-budgeted pruning). Lookup procedure:
help <function> in the live MATLAB session. Fast and reflects the actually-installed release of the toolbox or support package.help returns only a stub like "Run doc for more information." — common for support-package functions whose help redirects to the browser doc — fall back to the agent's web browsing of the online reference page at https://www.mathworks.com/help/<product>/ref/<funcname>.html (lower-case function name). Extract every name-value argument with its default value, formatted as a markdown table, quoting defaults verbatim.which <func> returns "not found") on a system that has the relevant support package installed, the support package is likely on a stale build. Ask the user to update via Add-On Explorer rather than working around the missing function.references/pattern1/compression-decision.md and walk the user through the question flow (hardware + Simulink availability, primary goal, retraining tolerance). Pick the compression and code generation path based on the answers. Compression is not mandatory and the optimal combination of pruning, projection, and quantization depends on the goal — for example, on Cortex-M with a latency-bound LSTM model, the float32 path with CMSIS-DSP outperforms the quantized path because CMSIS-NN provides no INT8 kernel for recurrent layers.