APAI25-LAB07-PULP-Tiling-Part1

Assignment DEADLINE: 02/01/2026 (at 15:30)

Material

Assignment: here Slides: here Board setup: here

How to deliver the assignment:

Use Virtuale, upload only the assignment file named as follows:

LAB#_APAI_name.pdf

Part1: Deploying a CNN on GAP9 with NNTool

This first tutorial is under the Part1-NNTool folder. This task is done only in LAB1, on the GAP9 boards. To setup the board (ONLY IN LAB1!!), follow the instructions in board-setup.

Run the CNN on GAP9

First, connect the board to the PC and setup GAP-SDK following the setup guide.

To quantize the model, generate and compile its code and run it on GAP9 with:

• python nntool_generate_model.py

Part2: Deploying a Transformer model on a generic platform with Deeploy

This second tutorial is under the Part2-Deeploy folder. This setup requires your PC and the WSL2 machine you used for the previous labs.

Setting up your environment

This setup requires a different Docker image with respect to the one used in the previous labs. To this aim:

1. Open your WSL2 machine (Ubuntu24) that you already installed for the previous labs.
2. Clone and enter this repository with:

git clone https://github.com/EEESlab/APAI25-LAB10-End-to-End-Deployment.git
cd APAI25-LAB10-End-to-End-Deployment/

3. Open this folder in VSCode with:

code .

4. As the folder opens, click on "Re-open in container" to install and run the Docker image reserved to this lab.
5. Set up Deeploy with:

git config --global --add safe.directory /workspaces/APAI25-LAB10-End-to-End-Deployment/Part2-Deeploy
git submodule update --init --recursive
cd Part2-Deeploy && pip install -e . && cd ..

6. Enjoy the assignment :)

Doing the assignment and running the code

To this end, please follow the instructions in the assignment.