Release Note - NHN Cloud User Guide

Machine Learning > AI EasyMaker > Release Notes

Improved endpoint feature
- Improved so that you can set resource allocation values directly.
Added support for Hugging Face model
- Added the Hugging Face model to AI EasyMaker to serve as an endpoint, batch inference.

Added ML pipeline feature
- ML Pipeline is a feature for managing and executing portable and scalable machine learning workflows.
- For more information, see the ML pipeline guide documentation.
Improved endpoint functionality
- Optimized resource allocation.
Support for serving PyTorch no-archive model
- Enabled PyTorch no-archive models to be registered in AI EasyMaker to serve as endpoints.

Added the batch inference feature
- Provides an environment to make batch inferences from an AI EasyMaker model and view inference results in statistics.
- For more information, see the Batch Inference Guide article.
Added the resource search feature
- You can search for resources from the console screen, and navigate to other resource screens via links.
Added the feature to change NAS of notebooks
- You can change the NHN Cloud NAS connection settings for running notebooks.
Scikit-learn serving support
- Enabled Scikit-learn models to be registered in AI EasyMaker to serve as endpoints.
Enable notebook shared memory
- Enabled more than 64 MB of shared memory to be available.
- The size depends on the instance type you selected when creating the notebook.
Removed the save_steps hyperparameter from the NHN Cloud-provided algorithms
- Removed the hyperparameter save_steps related to saving checkpoints.
- The algorithm automatically calculates the appropriate number of save_steps and saves up to three.

Notebooks and training with private images
- User-personalized container images can be used to power notebooks, training, and hyperparameter tuning.
- By registering private image and registry account, you can easily select private images to create resources.
Dashboard
- See your overall resource utilization, top 3 endpoint service monitoring, and top 3 CPU/GPU utilization on one page.
Endpoint > Autoscaler
- You can dynamically manage the number of nodes by setting policies to scale up/scale down endpoint nodes.

Ubuntu 22.04 version provided
- The new Ubuntu 22.04 version is provided. Ubuntu 18.04 version is no longer available, and existing customers can use the service as it is now.
Monitoring feature provided
- You can check system monitoring metrics for notebook, training, and endpoint.
- You can view API call metrics for each API resource path in the endpoint.
Basic algorithm for hyperparameter tuning
- Through hyperparameter tuning, you can optimize the hyperparameters of the basic algorithm provided by AI EasyMaker.
Endpoint > Serving multiple models
- You can serve multiple training models on one endpoint stage.
Parallel training for hyperparameter tuning
- You can optimize the performance of hyperparameter tuning by adjusting the number of parallel trainings.

Added hyperparameter tuning feature
- Hyperparameter tuning is the feature to automate repetitive experiments to find optimized hyperparameters to improve the predictive accuracy and performance of machine learning models.
- For more information, please see the Hyperparameter Tuning Guide.
Added 3 basic algorithms provided by NHN Cloud AI EasyMaker
- For more information, please see the guide document of each algorithm.
- Image Classification Guide
- Object Detection Guide
- Semantic Segmentation Guide

AI EasyMaker is an AI platform for environment, training and advancement, and endpoint services for machine learning development.