Custom Runtime allows you to customize the container to have the software and dependency that you need to run your model.

Here is a sample YAML file that you can write to import. You can follow the required structure of the build. You can specify the System as well as the Python packages that you need to run the model.

cuda_version (Optional):

By default, it will use CUDA 12.1.1. Above are the available options : “12.4.1” / “12.1.1” / “11.8.0”

system_packages :

These are any libraries that you need for the model to run, for eg: “opencv” for image processing or like “ffmpeg” to open the audio files.

python_packages :

These are pip-based libraries that you need in your Python runtime for example torch is required to load PyTorch-based models

run :

These are shell commands executed while building the runtime for example symlink is created here. All commands are executed Sequentially. If you have a package that requires step by step installation you can use run

build:
  system_packages:
    - "libssl-dev"
    - "opencv"
    - "ffmpeg"
  python_packages:
    - "transformers==4.29.0"
    - "torch==2.0.1"
    - "numpy==1.23.5"
    - "pandas==2.0.1"
  run:
    - "ln -s /usr/local/lib/python3.10/site-packages/torch/lib/lib{nv,cu}* /usr/lib"

Runtime Versions

You can edit the runtime in UI or patch the runtime using CLI to create a new version of the runtime. All your models will remain in the older version you previously deployed with unless you explicitly update them in Model Settings.

You can see all the versions of the runtime by clicking the ‘View Runtime Versions’.

You can look at the packages in the version by clicking on “View”

To update the Runtime of a model you can go to the Model Card and navigate to the Settings Page and select the version of the runtime you want to deploy and click on update.

Method A: Create a new Runtime using Inferless Platform

  1. Select on Runtime Tab in the left navigation and click on Create Runtime

  1. Upload the Yaml file created with the required dependencies.

How to use Custom Runtime for the Model

  1. Go to the Model Import wizard. To use the custom runtime, select it in the Setup Environment step.

  1. Select the Runtime that you have created. Now you all access to all the required software libraries while running the model

Method B: Create a new Runtime using Inferless CLI

Using the Inferless-CLI, inferless runtime command allows users to list, select and upload the runtimes.

Creating the runtime file

You can create the runtime file in multiple ways:

  1. You can create the requirements.txt file and then run the command inferless init, continue the process and select the requirements.txt file. This will create the inferless-runtime-config.yaml, which you can update according to your software and Python libraries requirements.

  2. You can create the config.yamlfile and mention all the software packages and Python libraries required for the model inference. And update the runtime file name on the inferless.yaml.

Upload the Runtime

Once you have created your runtime file, you can now run the command inferless runtime upload to start the uploading process. First, you are required to pass the file name, and then you need to set a name to your runtime. Update the same to the inferless.yaml file.

Now your runtime is ready to use!