top of page

STADLE Model Orchestration

Manages multi-model orchestration and deployment scenarios, facilitating seamless model updates and tracking across different environments.

image.png

STADLE is constructed to revolutionize a real-world machine learning process with a CICD for AI Model Development with continuous and collaborative learning paradigm and has been used and validated by enterprise customers that desire to bring their AI to the next level

image.png

CI/CD (continuous integration, continuous deployment) is the de facto standard development and operations pipeline for software development.

image.png

Continuous Integration:

Developers work on different parts of the codebase simultaneously and push their changes asynchronously.

​

Version control software such as Git help developers to push their changes without worrying about interfering with someone else’s development.

​

Continuous Deployment:

Live software products can be updated in real time with the changes made by developers to significantly shorten the update cycle.

​

An automated pipeline is often developed to push the changes made by developers directly to production (e.g. when the version number has been increased).​

​​CI/CD pipelines greatly improve the efficiency of development teams and allow new changes to quickly be made customer-facing → development time is significantly reduced, customer satisfaction is maximized.

STADLE aims to bring the CI/CD concept from software engineering to AI model development.

image.png

Continuous Integration:

Agents (cloud instances, edge devices, etc.) all train a shared global model on different datasets, updating the global model with the new data (often simultaneously and asynchronously).

​

STADLE orchestrates the learning processes across the agents and uses unique algorithms to efficiently combine and apply model update.

​

Continuous Deployment:

AI models currently deployed for end applications (e.g. IoT device image recognition, chatbot LLM deployed on cloud) can receive and integrate model updates produced by the agents in real time.

​

STADLE also manages the sending and integration of model updates to already deployed models, either automatically or in a managed way (e.g. based on model performance).​

​​CI/CD for AI model development thus carries over similar benefits from the software engineering analogue - model training and deployment time can be significantly reduced, directly resulting in compute cost savings.

On the STADLE’s ModelOps Dashboard, you can check the how many aggregators (orchestrators) are running and how many ML agents are running. You can also check the aggregator endpoint and the base AI model information. The recent AI models uploaded to the STADLE can quickly be checked as well.

dashboard.png

Performance metrics of the uploaded local AI models for each aggregation round can be tracked on the Performance Tracking page. You can monitor the learning process for each metric that you define for your ML models.

performance-tracking.png

You can check, filter through, and sort out all the models uploaded and aggregated through STADLE.  Prior models can be downloaded as needed (e.g for one-time deployments or for use as a new base model for further training). Also you can pick up the best model and distribute the model to the devices that are connected to STADLE platform.

model-repo.png

You can manage how the models are aggregated on this page. You can select appropriate algorithm that fits your AI application and models and set up life cycle management of distributed learning, etc.

agg-mngt.png

You can manage the AI model training sacalability with our advanced infrastructure setup through Kubernetes. This way you do not have to worry about the large number of edge servers,  devices, or training environments - the STADLE platform optimize the scalability for you.

auto-scale.png
In a Meeting

Get In Touch

Thanks for submitting!

bottom of page