Sample Collections

SiMa provides multiple sample sources to help developers explore, evaluate, and deploy models, pipelines, and tools across different stages of development.

Each source serves a distinct purpose and is accessed in a different way. This document explains what each collection is, when to use it, and how to access the full list of available samples.

The Model Zoo is a curated collection of precompiled and quantized models that are ready to run on SiMa devices.

Purpose

Evaluate model accuracy and performance
Avoid manual compilation and quantization
Use validated, production-ready model artifacts
Select models optimized for specific hardware targets

Access

The Model Zoo is accessed via the CLI.

user@host-machine:~$ sima-cli modelzoo list
user@host-machine:~$ sima-cli modelzoo get yolov5s
user@host-machine:~$ sima-cli modelzoo describe yolov5

The App Zoo provides a collection of prebuilt, end-to-end pipelines that can be deployed directly to a DevKit and executed.

Purpose

Rapid functional validation on hardware
End-to-end pipeline evaluation
Demonstration of complete applications
Deployment of runnable examples with minimal setup

App Zoo entries represent complete pipelines, including models, preprocessing, and postprocessing logic.

Access

The App Zoo is accessed via the CLI.

user@host-machine:~$ sima-cli appzoo list

Applications from App Zoo can typically be deployed and executed directly on a DevKit.

SiMa supports Large Language Models (LLMs) and Vision-Language Models (VLMs) sourced from Hugging Face, primarily for interactive evaluation and demonstration.

Purpose

Evaluate LLM and VLM capabilities on SiMa hardware
Run interactive demos and inference workloads
Experiment with generative AI use cases
Validate end-to-end LLM pipelines

These examples focus on runtime evaluation and interaction rather than precompiled Model Zoo artifacts.

Access

Hugging Face–based LLM/VLM samples are provided through the LLiMA demo.

To install the LLiMA sample package:

sima@modalix:~$ sima-cli install samples/llima -t select

During installation, the user will be prompted to select one or more models to download. If multiple models are available on the system, the user will be prompted to select a model when launching the LLiMA application.

For more information regarding LLiMa example, refer to this article.

SiMa maintains a collection of open source samples, tools, and reference implementations on GitHub under the sima-ai organization.

Purpose

Reference applications and developer tools
Automation and integration examples
Community collaboration and customization

These repositories provide fully transparent source code and may evolve independently from SDK releases.

Access

Browse available repositories on GitHub.