2.4. Windows Setup

This guide provides Windows-specific instructions for installing and using Tiny ML Tensorlab.

Note

While Tiny ML Tensorlab works on Windows, Linux is the primary development platform. Some features may have better support on Linux.

2.4.1. Option 1: Native Windows Installation

Step 1: Install Python 3.10

Using pyenv-win (Recommended):

# Run PowerShell as Administrator
Set-ExecutionPolicy -ExecutionPolicy RemoteSigned -Scope CurrentUser

# Install pyenv-win
Invoke-WebRequest -UseBasicParsing -Uri "https://raw.githubusercontent.com/pyenv-win/pyenv-win/master/pyenv-win/install-pyenv-win.ps1" -OutFile "./install-pyenv-win.ps1"
& "./install-pyenv-win.ps1"

# Restart PowerShell, then:
pyenv install 3.10.14
pyenv global 3.10.14

# Verify
python --version

Step 2: Install Git

Download and install Git from https://git-scm.com/download/win

During installation:

• Select “Git from the command line and also from 3rd-party software”

• Select “Checkout as-is, commit Unix-style line endings”

Step 3: Clone and Install

# Clone repository
git clone https://github.com/TexasInstruments/tinyml-tensorlab.git
cd tinyml-tensorlab

# Create virtual environment
python -m venv venv
.\venv\Scripts\Activate.ps1

# Install components
cd tinyml-modelmaker
pip install -e .

cd ..\tinyml-tinyverse
pip install -e .

cd ..\tinyml-modeloptimization\torchmodelopt
pip install -e .

cd ..\..\tinyml-modelzoo
pip install -e .

Step 4: Configure Environment Variables

Warning

IMPORTANT: Environment Variables Required for Model Compilation

For AI model compilation to work, you MUST set environment variables specific to your target device before running examples.

The variables you need depend on which device you’re targeting:

• C2000 devices (F28P55, F28P65, etc.): Set C2000_CG_ROOT and C2000WARE_ROOT

• F29 devices (F29H85X, etc.): Set CG_TOOL_ROOT

• MSPM0 devices: Set ARM_LLVM_CGT_PATH

• AM13E devices: Set ARM_LLVM_CGT_PATH

• AM26x devices: Set ARM_LLVM_CGT_PATH

• Connectivity devices (CC2755, CC1352, etc.): Set ARM_LLVM_CGT_PATH

See Environment Variables for complete device-specific setup instructions.

Step 5: Run Example

cd tinyml-modelzoo
run_tinyml_modelzoo.bat examples\generic_timeseries_classification\config.yaml

2.4.2. Option 2: WSL2 (Recommended for Full Compatibility)

Windows Subsystem for Linux provides better compatibility with the toolchain.

See also

Official WSL Documentation

Step 1: Install WSL2 with Ubuntu 22.04

Open PowerShell as Administrator (right-click and choose “Run as Administrator”):

wsl --install --web-download --distribution Ubuntu-22.04

Restart your computer when prompted.

Note

The above command performs multiple internal steps. If it errors partway through, simply run the same command again after rebooting. It is safe to run multiple times.

Step 2: Create a UNIX User Account

After the reboot, reopen PowerShell and run the install command once more. This time it will detect that Ubuntu 22.04 is already installed and will launch it, prompting you to create a default UNIX user account:

Please create a default UNIX user account. The username does not need to match your Windows username.
Enter new UNIX username: your_username
New password:
Retype new password:

Choose any username and password you like. These do not need to match your Windows or enterprise credentials.

Accessing Files Between WSL and Windows

• From Windows: Open File Explorer and type \\wsl.localhost\Ubuntu-22.04 in the address bar to browse your WSL filesystem.

• From WSL: Your Windows drives are automatically mounted under /mnt/. For example, access the C: drive with ls /mnt/c/.

Tip

For best performance, keep project files inside the WSL filesystem (e.g., under your /home/<username> directory) rather than on a Windows-mounted drive.

Step 3: Update System and Install Dependencies

# Update packages
sudo apt update && sudo apt upgrade -y

# Install build tools and other dependencies
sudo apt install -y build-essential git curl

# Install pyenv
curl https://pyenv.run | bash

# Add to ~/.bashrc
echo 'export PYENV_ROOT="$HOME/.pyenv"' >> ~/.bashrc
echo 'export PATH="$PYENV_ROOT/bin:$PATH"' >> ~/.bashrc
echo 'eval "$(pyenv init -)"' >> ~/.bashrc
source ~/.bashrc

# Install Python 3.10
pyenv install 3.10.14
pyenv global 3.10.14

# Clone and install
git clone https://github.com/TexasInstruments/tinyml-tensorlab.git
cd tinyml-tensorlab/tinyml-modelmaker
./setup_all.sh

Step 4: Configure Environment Variables

Warning

IMPORTANT: Environment Variables Required for Model Compilation

For AI model compilation to work, you MUST set environment variables specific to your target device before running examples.

The variables you need depend on which device you’re targeting:

• C2000 devices (F28P55, F28P65, etc.): Set C2000_CG_ROOT and C2000WARE_ROOT

• F29 devices (F29H85X, etc.): Set CG_TOOL_ROOT

• MSPM0 devices: Set ARM_LLVM_CGT_PATH

• AM13E devices: Set ARM_LLVM_CGT_PATH

• AM26x devices: Set ARM_LLVM_CGT_PATH

• Connectivity devices (CC2755, CC1352, etc.): Set ARM_LLVM_CGT_PATH

See Environment Variables for complete device-specific setup instructions.

Step 5: Run Examples

cd ~/tinyml-tensorlab/tinyml-modelzoo
./run_tinyml_modelzoo.sh examples/generic_timeseries_classification/config.yaml

2.4.2.1. Network and Proxy Configuration in WSL (Optional)

Note

This section is relevant if you are behind a corporate firewall, VPN, or proxy. If your WSL instance can already reach the internet (e.g., sudo apt update works), you can skip this.

DNS Configuration

If DNS resolution fails inside WSL, you may need to configure /etc/resolv.conf manually:

sudo nano /etc/resolv.conf

Add a public DNS nameserver (or your corporate DNS server):

nameserver 8.8.8.8

To prevent WSL from overwriting this file on each startup, edit /etc/wsl.conf:

sudo nano /etc/wsl.conf

Add the following:

[network]
generateResolvConf = false

Then make resolv.conf immutable so it persists:

sudo chattr +i /etc/resolv.conf

APT Proxy Configuration

If apt cannot connect to package repositories, configure a proxy. Replace your.proxy.address:port with your actual proxy settings:

1. Create or edit /etc/apt/apt.conf:

   Acquire::http::proxy "http://your.proxy.address:port";
   Acquire::https::proxy "http://your.proxy.address:port";
   

2. Create /etc/profile.d/proxy.sh to set environment-wide proxy variables:

   export http_proxy="http://your.proxy.address:port"
   export https_proxy="http://your.proxy.address:port"
   export no_proxy="localhost,127.0.0.1"
   export HTTP_PROXY="http://your.proxy.address:port"
   export HTTPS_PROXY="http://your.proxy.address:port"
   export NO_PROXY="localhost,127.0.0.1"
   

3. Shut down and restart WSL for changes to take effect:

   wsl --shutdown
   

4. Reopen your WSL terminal and verify with env | grep -i proxy.

2.4.2.2. Docker in WSL (Optional)

If you prefer to use Docker for an isolated build and run environment, you can install it directly inside WSL:

sudo apt update
sudo apt install -y docker.io
sudo usermod -aG docker ${USER}
sudo systemctl start docker
sudo systemctl enable docker

Log out and log back in for the group change to take effect, then verify:

docker --version

Note

Docker is not required if you are using a Python virtual environment for development. It is an alternative approach, useful when a pre-built Docker image is available.

2.4.2.3. USB Device Access from WSL

To connect USB devices (such as TI LaunchPads) to your WSL instance, you need usbipd-win.

1. In a Windows PowerShell running as Administrator, install usbipd-win:

   winget install --interactive --exact dorssel.usbipd-win
   

2. Restart PowerShell after installation.

3. Follow the Microsoft guide to attach a USB device to WSL: Connect USB devices

2.4.2.4. VS Code with WSL

Visual Studio Code can connect directly to your WSL instance, giving you a full GUI editor for files inside WSL.

1. Install Visual Studio Code on Windows.

2. Install the WSL extension (also called “Remote - WSL”) from the VS Code marketplace.

3. Open a WSL terminal and run:

   code .
   

   This launches VS Code on Windows, connected to the WSL filesystem.

See also

Developing in WSL with VS Code

2.4.3. Path Configuration

Long Path Support (Native Windows)

Enable long path support for deep directory structures:

# Run as Administrator
New-ItemProperty -Path "HKLM:\SYSTEM\CurrentControlSet\Control\FileSystem" -Name "LongPathsEnabled" -Value 1 -PropertyType DWORD -Force

Environment Variables

Set up paths for TI tools (if compiling for devices):

# Add to your profile or set in System Properties
$env:C2000_CG_ROOT = "C:\ti\ccs\tools\compiler\ti-cgt-c2000_22.6.1.LTS"
$env:C2000WARE_ROOT = "C:\ti\c2000\C2000Ware_5_03_00_00"

2.4.4. Common Windows Issues

PowerShell Execution Policy

If you get “running scripts is disabled”:

Set-ExecutionPolicy -ExecutionPolicy RemoteSigned -Scope CurrentUser

Path Too Long Errors

Clone to a short path (e.g., C:\dev\tensorlab) or enable long paths.

Line Ending Issues

Configure Git to handle line endings:

git config --global core.autocrlf true

Virtual Environment Activation

If venv\Scripts\Activate.ps1 fails:

# Alternative activation
.\venv\Scripts\activate.bat

# Or run Python directly
.\venv\Scripts\python.exe -m tinyml_modelmaker ...

2.4.5. GPU Support on Windows

For CUDA support (useful for NAS):

1. Install NVIDIA drivers from https://www.nvidia.com/drivers

2. Install CUDA Toolkit from https://developer.nvidia.com/cuda-downloads

3. Verify with:

python -c "import torch; print(torch.cuda.is_available())"

2.4.6. WSL2 vs Native Windows Comparison

Feature	Native Windows	WSL2
Setup Complexity	Moderate	Higher initially
Shell Scripts	Requires .bat files	Full bash support
Compilation	Full support	Full support
GPU Support	Native CUDA	WSL2 CUDA (newer)
CCS Integration	Direct	Requires file sharing
Recommended For	Quick start	Full development

2.4.7. Next Steps

• Environment Variables - Configure TI compiler paths

• Quickstart - Train your first model