Conda Installation¶

CUDA 12.1 Bug

Due to a known bug with CUDA versions greater than 12.1, executing compiled models on Orin Jetpack 5.X may fail.

Cross Compilation Support

Cross compilation allows users to compile for a aarch64 device (say Jetson or Raspberry Pi) on an x86_64 host device (say Intel or AMD CPU). If you wish to compile for aarch64 you will need to install g++-aarch64-linux-gnu. You can use APT to install this on Ubuntu systems:

sudo apt-get install -y g++-aarch64-linux-gnu

1. Access Latent AI's Container Repository¶

In order to pull containers or install packages from Latent AI, you'll need to create a personal access token. To do so, follow these steps:

Click the Sign in link in the upper right.
Select Sign In with SSO.
Enter your access credentials.

ii. Create your Personal Access Token¶

Click on your profile in the upper right.
Select User Token on the left navigation.
Select the Access user token button.
View your user token name and user token pass code.

You can export the token name and token pass code as environment variables:

export REPOSITORY_TOKEN_NAME=<user_token_name>
export REPOSITORY_TOKEN_PASSCODE=<user_token_pass_code>

2. Export License Key¶

You should have received this via email.

Tip

You can persist variables like LICENSE_KEY across bash sessions by placing them in your ~/.bashrc file.

export LICENSE_KEY=key/<license key>

3. Create and Activate conda Environment¶

Copy or download the following environment.yml:

name: latentai
channels:
  - nvidia/label/cuda-12.1.0
  - conda-forge
  - defaults
dependencies:
  - jupyter
  - libstdcxx-ng
  - nvidia::cuda-tools
  - python=3.10
  - pip
  - pip:
      - --extra-index-url https://${REPOSITORY_TOKEN_NAME}:${REPOSITORY_TOKEN_PASSCODE}@repository.latentai.com/repository/pypi/simple
      - leip-design

Then run the following command to create the conda environment:

conda env create -f environment.yml

A new environment will be created in your conda installation called latentai which you can activate with the following command:

conda activate latentai