SDK Installation

After purchasing the SDK, you will receive the installer and dependency library files.

Placeholder Values

Commands and code below use placeholders in angle brackets — <model-size>, <high/mid>, <version> — which are not valid to run as-is. Replace them with the actual values for the SDK build you purchased (e.g. your sales representative or the delivered wheel filename will tell you the exact package name and version to use).

Prepare the Python Environment

High/MidLow

The SDK is tested with Python 3.10 (Ubuntu 22.04) and Python 3.12 (Ubuntu 24.04).

The SDK is tested on Jetson Orin with JetPack 6.x and Python 3.10.

We recommend installing the SDK in a virtual environment. All commands on this page assume a virtual environment and work as-is.

Installing into the system Python environment

Ubuntu 22.04: no extra option is needed. Do not add --break-system-packages; the pip shipped with Ubuntu 22.04 does not support it.
Ubuntu 23.04 and later: pip blocks system-wide installs by default (PEP 668, externally-managed-environment error). Add --break-system-packages to each pip install command. This bypasses Ubuntu's package protection and may conflict with apt-managed Python packages. Use it only if you accept this risk.

Do not use --break-system-packages inside a virtual environment.

1. Verify your Python version

This only displays the version and does not change your system.

python3 --version

Expected output: Python 3.10.x or Python 3.12.x.

2. Install the venv module (first time only)

This installs Python's venv module system-wide and requires sudo.

sudo apt install -y python3-venv

Expected result: the command completes without errors.

3. Create and activate a virtual environment

This creates a .venv directory in the current directory. Activation applies only to the current terminal; run source .venv/bin/activate again in every new terminal before using the SDK.

python3 -m venv .venv
source .venv/bin/activate

Expected result: your prompt starts with (.venv).

4. Upgrade pip

This upgrades pip inside the virtual environment only.

python3 -m pip install --upgrade pip

Expected result: Successfully installed pip-<version> or Requirement already satisfied.

Install Dependency Libraries

Install the dependencies before the SDK. The SDK and FlashAttention require the PyTorch build installed in this step. This step downloads several GB and may take several minutes.

Third-Party Libraries

Certain features in this SDK depend on libraries maintained by independent third parties. While the SDK has been tested with specific versions for compatibility, ongoing updates and security maintenance are handled by each library's own maintainers. We recommend reviewing the official documentation for these libraries and installing secure, supported versions that meet your project's requirements.

High/MidLow

Create requirements.txt:

accelerate==1.10.1
av==15.1.0
decord==0.6.0
imageio==2.37.0
opencv-python==4.12.0.88
timm==1.0.19
transformers==4.56.1
torch==2.9.1
torchvision==0.24.1
wheel==0.46.2
vllm==0.14.1
scipy==1.15.3

python3 -m pip install -r requirements.txt

Expected result: ends with Successfully installed ..., including torch-2.9.1 and vllm-0.14.1.

Create requirements.txt:

accelerate==1.11.0
av==15.1.0
numpy<2
timm==0.6.13
transformers==4.51.3
torch==2.8.0
torchvision==0.23.0

Install from the Jetson AI Lab index to get Orin-compatible PyTorch builds:

python3 -m pip install -r requirements.txt --index-url https://pypi.jetson-ai-lab.io/jp6/cu126

Expected result: ends with Successfully installed ..., including torch-2.8.0 and torchvision-0.23.0.

Install SDK

Install the wheel files as delivered, without renaming them.

High/MidLow

FlashAttention is optional. It is built from source, may take a long time, and may not be compatible with every environment. A compatibility-tested prebuilt version is also included in the bundle. (See flash-attention-prebuild-wheels for reference.)

python3 -m pip install ./wave_<model-size>_<high/mid>-<version>-py3-none-any.whl
python3 -m pip install ./codemeter_api-9.0.500-py3-none-any.whl
# Optional: FlashAttention
python3 -m pip install flash-attn --no-build-isolation

Expected result: Successfully installed wave-<model-size>-<high/mid>-<version> and Successfully installed codemeter-api-9.0.500.

FlashAttention is not supported on this tier.

python3 -m pip install ./wave_<model-size>_low-<version>-py3-none-any.whl
python3 -m pip install ./codemeter_api-9.0.500-py3-none-any.whl

Expected result: Successfully installed wave-<model-size>-low-<version> and Successfully installed codemeter-api-9.0.500.

Verify Installation

Run the following in Python inside the activated virtual environment:

High/MidLow

from woven.vision.ai.wave_<model-size>.vllm_client import VLLMClient
client = VLLMClient()

Replace <model-size> in the import path based on your installer file name. The class name is always VLLMClient.

from woven.vision.ai.wave_<model-size>.client import Wave<model-size>Client
client = Wave<model-size>Client()

The Low tier uses a different class from High/Mid. Replace <model-size> in both the import path and the class name based on your installer file name. For example, if the file name is wave_1b_low-..., use from woven.vision.ai.wave_1b.client import Wave1BClient.

On first import, you will be prompted to acknowledge the End User License Agreement (EULA). This confirmation is required only once.

← Previous: License Activation

Next: API Usage Examples →