Release notes
ZEBLOK COMPUTATIONAL AI-MICROCLOUD RELEASE NOTES
Version 1.2.6 (December 2023)
SaaS
Model Serving of Large Language models
LAMACPPserves quantized models
Facebook 125M model using vLLM serving
Served using command line interface as an API
PaaS
ZBL CLI
vLLM serving of models as APIs
LLAMA CPP serving of models as APIs
Capability of multipart upload of larger model sizes
COMP APP
Refactoring of the whole code base in comp-app in order to reduce SDK communication crashing and improve Error Handling and API interaction handling
IAM roles and privileges updated for more User privilege and easy segregation of access in organisation's System admins
Generic Image for Enterprise edition which allows the use of a single image that would be compatible with all our environments.
UI Improvements for AI-Microcloud
The support button alignment has been updated
Once the secret keys are generated they're always available in the session for the user.
Copy button has been added for the keys
View password has been also added in the profile page and login page has context menu
Microservice and Workstations card alignment.
Removal of the Zeblok Ai-WorkSpace text
Segregation of Manage account page
User role should also get displayed on Manage your account page.
Rename get data-lake key and get api credentials button to Generate Datalake key and Generate API credentials.
View password button in the register page
Padding fixes in the UI
Role table fixes
COMP AUTH
Complete code refactoring for making the app more resilient and avoid server crashing
Client libraries for object storage(minIO) have been updated which allow on the fly bucket creation and policy creation on minIO simultaneously on user registration.
Microservices and Workstations
Introduction of new and updated microservices and workstations into the Microcloud for more dynamic and extensive usability e.g. StreamLIT UI, Sceenscape, ChromaDB, etc.
The new versions for docker images have been provided on the seeding part of AI-MicroCloud.
Monitoring
Event Exporter: Dependency on the third party Kubernetes event exporter has been removed and usage of Kube-API is introduced, this has improved event caching and improved reliability, ensuring we are not missing out on released events
CaaS
CaaS is made available for concurrent users by the help of kubernetes jobs in return with no wait time for any user .
The containerization as a service has been totally redesigned to help improve the usability and to enhance the user experience.The user can have the CaaS service invoked with custom plans as required according to them and also be able to see the logs for the whole process.
Improvements
Big chunks of data is now being able to be fed and processed while model serving .
Any size of models can be served and packaged as an API using our CaaS service
SDK
Refactoring of the whole code base for validation processes and changes authentication mechanism and packaging for seamless user experience.
IaaS
Hardware
Intel AMX support now available with Basic and Enterprise SKU of AI-Microcloud
Basic Baremetal Launcher
Our basic launcher now provides the user to create AI platform into the Bare-metal environment automatically within few minutes by using the launcher filling up some details and providing access credentials(For Bare-Metal connectivity) which are taken in as encrypted data, With all the information taken the scripts to run in the background and providing the URL in the Launcher to access the AI-Microcloud all deployed and ready to use.
Enterprise Azure Launcher
The enterprise launcher allows the user to connect their 3 hubs with any amount of bare metal servers and the user can easily establish the environment automatically within a few minutes by using the launcher filling up some details.
Platform upgrade
Our Applications(MicroCLoud, Launcher and OTF) are now compatible for the MONGO Atlas integration, it’s been tested and ready for use
Website
Playground
Fixes and Improvements
Enterprise Launcher fixes
Bugs on User Interface now resolved
Removed the cross symbol from the success message. Instead display a message on new window
User will land on Marketplace tab when log inside the AI-Microcloud
Launcher script-Token fixed inside the script
Hash is now being received on email
User input are taken interactively while running the script
Launcher user registration email verification is now enabled
The instance type in the form t2xlarge is fixed
Masters name removed
User now lands on Request page on launcher once he logs in
Version 1.2.5 (September 2023)
Hardware:
AIM now runs on Intel Sapphire-Rapid Processor
BASIC:
Storage for Azure Cloud: AZURE Blob storage from UI for uploading viewing and download of datasets
AZURE Basic version integrated with Orchestration addon as a usability(Ray usability) which enable the user to use the HPC usability of the platform for training and developing complex models and running immense computations.
Multi-master cluster on Bare Metal servers supported
Version 1.2.4 (June 2023)
Enterprise version:
Software Updates: All the software packages are updated to latest version which covers bug fixes and security updates.
Security:
User can select and implement the Minimum base line security standard(MBSS) over edge server using UI of Enterprise launcher.
User can restrict the SSH access to the edge server using UI of Enterprise launcher.
Hardware:
Supermicro IoT SuperServer SYS-510D-10C-FN6P is certified for use as edge server.
Supermicro IoT SuperServer SYS-110D-20C-FRAN8TP is certified for use as Hub and edge server.
Intel® Data Center GPU Flex 140 is tested and certified for use at Hub and edge server
Monitoring: User can view the edge nodes attached to the cluster and check their online status whether active or not.
Basic version:
Auto scaling plans on AWS: Users can now define auto-scaling plans in Ai-MicroCloud® which makes the platform capable to scale the infrastructure as the need of resource increases while running the AI workloads.
Storage: Ai-MicroCloud® datalake now uses S3 as backend in AWS environment.
Security: Encryption layer provided onto the Basic Launcher while taking in sensitive data from the user which allows the system to be more secured from foreign attack
HPC: Users can spawn Ray clusters using orchestration add-on feature of Ai-MicroCloud® which will enable them to train and develop models with speed.
BUG FIXES:
Removal of the Datalake button on the user interface
Flickering of the Ai-MicroCloud® login page
Error handling, error message will be shown at the bottom of the webpage
Support button fixed with proper ticket creation
Alerting integrated into the platform to know the status of the environment in case of downtime
CROSS cloud integration resolved
Automatic redirection to the next/previous page once action completed on current
All mandatory fields highlighted in forms for the users
ML-flow integration on the ZBL CLI
Version 1.2.2 (JANUARY 2023)
Edge Monitoring:
SuperAdmin can monitor CPU and Memory for edge servers and deployed microservices with a more detailed overview which allows them to gather detailed information about individual Edge servers or particular microservices.
Detailed Graphical representation of the utilization of server resources.
CICD:
Automatic cyclic updation without taking down the whole environment i.e. only changing the git version which will directly incorporate changes into a pre-deployed AI-M ecosystem
Automated deployment of the whole ecosystem from a single point of contact just by using variables that are collected from the user irrespective of the platform/environment to be deployed upon.
Edge Security:
Minimum Baseline security(MBSS) has been incorporated into our ecosystem for providing standardized security according to the global regulations which makes the whole platform very secure to outside attacks.
Access control on Edge provides SSH and BMC console only from specific subnets and on specific interfaces.
Data encryption on local storage is incorporated into the system to make it more secure.
Version 1.2.1 (SEPTEMBER 2022)
Improved Microservice Configuration: Spawning MicroService workflow now has a more flexible and dynamic configuration system for environment variables, arguments, ports, etc.
Simultaneous EDGE deployments of AI-API and MicroServices.
Zeblok Cli: Zeblok Cli now provides an option to deploy your Ai-Models to multiple edge nodes simultaneously in an Edge Datacenter as an Ai-API.
Dynamic plans (CPU and Memory): Spawning Ai-WorkStation, MicroService and Ai-API workflows have been added with an option to attach customized plans (CPU and RAM according to user’s needs) where the resources will be within the limits set by the real-time resource metrics recommendation system.
Resource manager with monitoring: Resource Manager Tab in Ai-MicroCloud now gives the superadmin and admins an ability to view the monitoring graphs of each resource like Ai-API, Ai-Workstation and MicroServices.
Crosscloud integration: Ai-Models and Ai-Pipelines have been enabled to show their security scan logs along with previously enabled Ai-WorkStations and MicroServices which gives the users a better understanding of vulnerabilities in their images.
LAUNCHER APP(Edge):
New GUI for easy and efficient setup of the Microcloud on bare metal servers
Zeblok Launcher App is the newest member in the Ai-MicroCloud installation ecosystem. This feature seamlessly installs the whole Ai-MicroCloud edge ecosystem from scratch in a bare-metal environment efficiently.
Version 1.2.0 (JUNE 2022)
Zeblok Python SDK now has the feature to deploy simultaneous multi-datacenter Ai-Pipelines with also the ability to integrate the Ai-Runs output with open source tools.
Version 1.1.0 (MARCH 2022)
Openvino Model Serving capabilities are now supported in the Ai-API Engine.
Openvino toolkit distribution has been upgraded to version 2022.4.1
Ai-WorkStations have been upgraded with Nvidia Cuda Version 11.4
JupyterLab has been upgraded to the latest version along with Python v3.9.
Version 1.0.0 (DECEMBER 2021)
Zeblok Ai-MicroCloud is now available at Azure Cloud Environment with all the latest features and power giving the same experience of Zeblok AI-MicroCloud on the Azure Platform.
Zeblok has established Ai-MicroClouds for two private baremetal turned cloud environments: Advantech Labs and Intel Dev Cloud.
Version 0.0.9 (SEPTEMBER 2021)
PaaS Updates:
Admin analytics dashboard has been updated with more precise and accurate data along with bug fixes providing a more precise idea of the usage by the users.
Zeblok Authentication MicroService has been updated with more secure tokens and authorization logics while making the user’s authentication and signup process easier than before.
Zeblok Ai-MicroCloud introduces a brand new workflow for microservices where you can run pre-uploaded and modified images in an isolated and an independent environment in just a few clicks like MongoDB, Intel DLWorkBench, etc.
Session page now also handles metrics for spawned microservices along with Ai-WorkStations.
User’s profile has two new tabs for Spawned MicroServices and one for their total runtime consumptions, both of which can be accessed by the SuperAdmin and the Organization’s Admin.
Periodic Backup in every 6 hours has been established in aws s3 buckets as a disaster management measure.
Superadmin now gets a new tab in their menu called “Resource Manager” where you can have a look at all and every kind of resource in the Ai-MicroCloud like Ai-WorkStation, MicroServices and Ai-API and also control them if required.
Version 0.0.8 (JUNE 2021)
Infrastructure Re-Architecture is done for better performance.
Jupyterhub is taken out and Kuberntetes APIs are Introduced for Orchestration giving a more stable infrastructure and debugging system.
This new Architecture allows more control over resources and gives a faster experience so that the users don’t have to wait for the spawning process for too long.
Dynamic Plans can be introduced now giving full access to the power of the infrastructure.
Istio Mesh Networking is introduced for Load Balancing and Networking among the pods for better connectivity along with a better security.
PaaS Updates:
Login and Signup pages have a new UI now which is also mobile responsive.
Update has been rolled out for the new Admin analytics dashboard with more analytical metrics for viewing the activities of specific users in an organization.
Session page has more attributes added for more in depth metering and analysis.
User’s profile has a new tab for durations of the Ai-WorkStations now.
The Organizations table has a new View button for viewing the users and their duration of work in the Ai-Workstations.
Server Side pagination is implemented for reducing the bandwidth usage for users.
We now introduce a new workflow in our Ai-MicroCloud called Ai-API Engine which can be accessed by going to the Ai-Api section in the menu and clicking on the Create Ai-API button. For more information, you can visit https://computationaldocs.zeblok.com/zeblok-computational-1/ai-api/introduction
A new separate Zeblok Authentication microservice has been added exclusively for enhanced Authentication Processes and validation and also giving users the ability to secure their account using 2 factor authentication.
The Login Method has been improved with the addition of Refresh Tokens and JWT Token for a better and new level of security.
Version 0.0.7 (MARCH 2021)
All the Zeblok benefits, now in AWS:
Automated/scripted the deployment of the Zeblok platform within AWS, which will help us serve our customers better. How?
They don’t have to move their data across datacenter or platforms, reducing a large amount of effort and billing, as they don’t have to engineer the data movement in and out of AWS, which also impacts data security. Less movement of data in and out of AWS saves bandwidth cost.
Usage can be customized as per individual organization needs. Doesn’t matter if the requirement is of 1 GPU or 10 or 100, the system can easily be scaled as and when required. Increasing impact or ROI and decreasing un-utilized cost.
Version 0.0.6 (DECEMBER 2020)
A new button of DATA LAKE is added besides the create workstation button where users can experience object storage provided via zeblok for storing their necessary files for their Ai-WorkStation.
Session Icon (Time Watch) is added to the listed spawned workstation (Active) on the landing page , where you can see your sessions.
Activity Logs
A new tab is added in the profile section where you can see all your activities and the resources which you have visited.
DashBoard
All users display more information like Allowed Notebook , Last Active etc are visible.
A new Section in View Button of Users is added to Activity Logs , where Super Admin can monitor users activities.
A quick summary of users is displayed and a “View More analytics” button is added for further resources monitoring page.
Users Section - In the users information display , on clicking “View” for a particular user , 2 new tabs of activities and spawnedImages are created to show its activities log and spawned Workstation by user.
Resources Monitoring
Super Admin can see following data
Total number of resources created like Workstations ,User Groups,Roles, Organizations,Plans etc.
Further re-distribution pie charts of various different sub categories of resource like -
Workstations : Data Science , Algorithms , Hyperconvergence .
Plans : CPU , GPU , Both
Data Center : Public , Private
A User Statistics Bar graph showing all new Users , all new Beta Users , all new Non Beta Users created along with that all active users , all active Beta users , all active non Beta users.
A Stacked Graph of users created each day is shown with categories of Beta and Non Beta User.
A pie chart showing Spawned Images data of users along with a table of further different categories of data.
Stacked Bar Graphs showing Workstations Created Day wise ,Workstation which are currently running as Day wise ,Number of CPU/GPU spawned with the notebooks on the respective date.
A stacked Bar graph showing the count of All Notebooks spawned with different plans.
For more detailed information , please refer to https://docs.google.com/document/d/1sv0aMzLhc63X5HKsAPfSufRn8kz8mH4eXnkTHMcPBEU/edit?usp=sharing
All the data can be filtered out between any time intervals. While the data which is related to any organization then it will be updated as per the organization.
Admin can use filters for both categories like Search by Organization , Search by Duration or Both.
Workstation
In the workstation table , a new column of licence is added to put licenses in algorithms and hpc’s.
IAM Menu
A new menu of IAM introduced which consists of all access management menus like roles , resources , menu and user groups from where superadmin can give or modify access for a specific role in the Ai-MicroCloud.
This helps to customize the side navigation menu also.
Version 0.0.5 (SEPTEMBER 2020)
An AlgoStore is now made available for the users to go and get detailed information about all the containers giving knowledge and also making the selection easier according to the needs.
Read More button on each algorithm provides information on capabilities, a youtube video link and links to other sources for more information.
Join Slack option provided for all users to join our dedicated Slack channel for any kind of clarification, bug posting and discussion with our team.
Documentation link is provided at the navbar to read and understand the design and workflows of the Ai-MicroCloud for a non ambiguous experience.
Dashboard
Can view all the users present on the platform segregated according to their roles, organizations for getting a view of the whole microcloud users.
Each user is categorized into a beta and non-beta user. Admin can update the user as beta or non-beta. Beta users have many more features unlocked for them than a non beta user.
Each user has a no. of allowed notebooks assigned to him. The user can only spawn that many number of notebooks if he is a beta user which depends on his subscriptions and resource management rules.
Beta users are the ones who are already present in the computational-beta environment. For these users they do not have to buy plans. Their billing would continue as it was happening before. They can spawn a limited number of notebooks as given access by the Zeblok Admin.
Workstations
All notebook containers are displayed here as a list in a table where the superadmin can perform CRUD operations on the notebooks.
Superadmin can edit the notebook image, name, description, restrict public access, add video hyperlink and other sources, and can add the stripe product ID for charging this workstation.
Superadmin can set the related plan and default plan for the container.
Spawned Images
Workstations spawned by the users are shown in Ai-WorkSpace along with the current status of their workstations and corresponding sessions.
All the activities of each workstation can be listed by clicking on their respective sessions icon giving a full in-depth overview of the activities done on the Ai-WorkStation.
Roles
Orgadmin role added giving full access to the organizational unit which consists of an orgadmin and all org users.
Orguser Role added as the role given to the users who are invited to the Ai-MicroCloud via an invite sent out by the OrgAdmin from the dashboard.
Plans
Plans can now be edited and deleted by the zeblok superadmin giving a more dynamic resource allocation system for the Ai-WorkStations.
Data Center
Data center requests can be seen here (Requests are by default given a Pending status).
It is the responsibility of zeblok admin to check and approve those data center requests and make them active.
Once approved these data centers would be visible to the users while creating workstations. (shown on the world map)
Version 0.0.4 (JUNE 2020)
Each Notebook has some predefined plans by the admin, and among them is a default plan which is recommended by us for the selected workstation to run without any resource limitations.
Organization Admin Role Added for better handling of access control within the organization so that the onboarded users can make the microcloud their own.
Organization Admin has been given the access to view/edit/update/delete all the users under his organization.
Version 0.0.3 (MARCH 2020)
The better, fresh and responsive user interface is now available.
Jupyterlab and Notebooks are updated to their latest Versions.
Each user will now have access to a dedicated GPU and from 4GB to 8GB of memory.
Each user will be able to spawn one server at a time.
Cull feature is now introduced. If a workstation is not utilized by users for two hours then it is automatically culled and resources released, if the browser session is killed. JupyterHub will run the culling process every ten minutes and will cull any user pods that have been inactive for more than one hour.
A license agreement is now required on signUp which is stored in the Database.
Zeblok Readme Notebook is now available in the environment showing tutorial instructions on retrieving compute, datalake and other platform features.
Nvidia Rapids example Notebooks are available on spawning the server using Nvidia Rapids Image.
Now the infrastructure has been moved within a VPN, making it more secure and easy to expand the infrastructure.
Slack community is added.
Version 0.0.2 (DECEMBER 2019)
A wide range of Ai-ML and DataScience Notebooks is added now as a library of notebooks giving users multiple options to spawn in JupyterHub like PySpark, TensorFlow and many more.
Ai-AppStore now includes trademarked notebooks and available for use upon agreeing the license agreement for Terms and usage. For example: Explainable AI by Akai Kaeru.
Notebooks have been CUDA optimized for the NVIDIA GPU with SLI support.
Version 0.0.1 (SEPTEMBER 2019)
Users are now able to spawn JupyterHub based notebooks as Ai-Workstation within CEWIT (Center of Excellence in Wireless and Information Technology) Datacenter cluster.
Users are now allowed to spawn basic minimal notebooks with cuda optimization for NVIDIA GPU cards (RTX6000) saving enormous time for users.
JWT Authentication layer is now available on top of our JupyterHub that authenticates only the valid users to spawn in a secure manner.
Last updated
Was this helpful?