Package settings - Amazon Lookout for Vision

Package settings

Use the following information to decide the package settings for your model packaging job.

To create a model packaging job, see Packaging your model (Console) or Packaging your model (SDK).

Target hardware

You can choose a target device or target platform for your model, but not both. For more information, see Tested devices, chip architectures, and operating systems.

Target device

The target device for the model, such as NVIDIA® Jetson AGX Xavier. You don't need to specify compiler options.

Target platform

Amazon Lookout for Vision supports the following platform configurations:

  • X86_64 (64-bit version of the x86 instruction set) and Aarch64 (ARMv8 64-bit CPU) architectures.

  • Linux operating system.

  • Inference using NVIDIA or CPU accelerators.

You need to specify the correct compiler options for your target platform.

Compiler options

Compiler options allow you to specify the target platform for your AWS IoT Greengrass Version 2 core device. Currently you can specify the following compiler options.

NVIDIA accelerator
  • gpu-code — Specifies the gpu code of the core device that runs the model component.

  • trt-ver — Specifies the TensorRT version in x.y.z. format.

  • cuda-ver — Specifies the CUDA version in x.y format.

CPU accelerator
  • (Optional) mcpu — specifies the instruction set. For example core-avx2. If you don't provide a value, Lookout for Vision uses the value core-avx2.

You specify the options in JSON format. For example:

{"gpu-code": "sm_75", "trt-ver": "7.1.3", "cuda-ver": "10.2"}

For more examples, see Tested devices, chip architectures, and operating systems.

Component settings

The model packaging job creates a model component that contains your model. The job creates artifacts that AWS IoT Greengrass V2 uses to deploy the model component to the core device.

You can't create a model component with the same component name and component version as an existing component.

Component name

A name for the model component that Lookout for Vision creates during model packaging. The component name you specify is displayed in the AWS IoT Greengrass V2 console. You use the component name in the recipe that you create for the client application component. For more information, see Creating the client application component.

Component description

(Optional) A description for the model component.

Component version

A version number for the model component. You can accept the default version number or choose your own. The version number must follow the semantic version number system – major.minor.patch. For example, version 1.0.0 represents the first major release for a component. For more information, see Semantic Versioning 2.0.0. If you don't provide a value, Lookout for Vision uses the version number of your model to generate a version for you.

Component location

The Amazon S3 location where you want the model packaging job to save the model component artifacts. The Amazon S3 bucket must be in the same AWS Region and AWS account in which you use AWS IoT Greengrass Version 2. To create an Amazon S3 bucket, see Creating a bucket.

Tags

You can identify, organize, search for, and filter your components by using tags. Each tag is a label consisting of a user-defined key and value. The tags are attached to the model component when the model packaging job creates the model component in Greengrass. A component is an AWS IoT Greengrass V2 resource. The tags aren't attached to any of your Lookout for Vision resources, such as your models. For more information, see Tagging AWS resources.