본문으로 건너뛰기
버전: docs v25.02

AI Conductor Release Notes

v2.1.3​

April. 23, 2025

Improvements​

Change the saved version method

  • OLD : the versions are saved in DB as a manual.
  • NEW : the version is deployed using HELM.

When request train, improve checking the data required for train

  • Improving time efficiency in verifying Large datasets of train

v2.1.2​

Mar. 21, 2025

Bug Fixes​

Fixed an issue with the page transition logic that occurred when only one workspace existed.

Resolved a bug in the user verification step for LDAP-based user invitations, where certain special characters were not allowed.


v2.1.1​

Mar. 14, 2025

Improvements​

Delete Workspace

  • You can't delete the workspace while running a train
  • Delete DB data (HARD delete) for the workspace
  • Delete the data registered in AI Solution and generated by training

Update LDAP certificate

  • Updating the Root CA Certificate Due to LDAP Certificate Expiration

Bug Fixes​

Fixed a UI bug related to creating LDAP user accounts.

Fixed the bug in the “remember me” feature that stores the user ID upon login.

Fixed a responsive behavior issue in the Description section of the AI Solution.

Changed the row selection in the AI Solution list to allow only single selection instead of multiple selection.


v2.1.0​

Feb. 12, 2025

New Features​

Run pipeline of train in SageMaker

  • You can run a training pipeline into SageMaker using the same request method as before
  • NOTE : ALO v3 compatible. (ALO v2 is not available)
  • NOTE : Resource monitoring for Train is not yet supported.

Apply the Helm of deployment instead of kustomization

  • It easy to maintain deployment
  • Deployment consistency with other components(Edge Conductor, Edge App)

Add an overview page

  • Display the most registered information of AI Solution
  • Dispaly recent updated AI Solution
  • Display recent Train Run
  • Display account information
  • Display Train Resources

Improvements​

Create Workspace

  • You can check the resource of EC2 when inserting 'Instance type'
  • Kubeflow : Make sure that a nodegroup is created
  • SageMaker : Make sure that a instance is available

Train Failure Reason

  • You can see why the train failed.
  • Failed-Train: Train algorithm failues
  • Failed-OOM : Memory overflow
  • Failed-Quota : EC2 limits exceeded

Check metadata v1.2 of AI Solution

  • Check container_uri and dataset_uri of Train
  • Check container_uri and dataset_uri of Inference
  • Check detail of Description

Bug Fixes​

Fixed a bug for setting workspaces

  • Exclude a session error in Kubeflow associated with the deleted workspace

v2.0.0​

Oct. 07, 2024

New Features​

Monitor Resource for train

  • You can view the CPU and memory usage for the running train.
  • You can check the CPU and memory usage for the system software.

Stop train

  • Train can be stopped by Edge Conductor

Update the password of static account

  • Only the static account owner can change the password

Add LDAP accounts (only LGE)

  • ADMIN can add an account affiliated with LGE as an LDAP account.

Add a page for workspace creation

  • ADMIN can create a Workspace on the condition that the necessary infrastructure has been set up in the Workspace.

Add the Edge Conductor username of REST API in both Stream and Streamhistory(Train)

  • To display the actual username of the Edge Conductor, add 'updator' of REST API in both Stream and Streamhistory(Train)

Improvements​

Remove Instance

  • 'Stream' is created with reference to 'Solution Version' in the Edge Conductor
  • Request the Train with the Train Resource configured in the Edge Conductor.
  • Update the path changes for train_dataset_uri and train_artifact_uri

Update the description of AI Solution

  • Overview : The content defined in the description of the metadata
  • Information : ID, Updator, updated, uuid
  • metadata : version, content, content version, pipeline, label

Remove the concept of public

  • Remove the 'scope_ws' variable in Solution**
  • Remove the concept of public from S3 and ECR, and retain only the concept of private

Verify the workspace by workspace_id

  • old : Verify the workspace by workspace_name

v1.9.4​

July 09, 2024

Improvements​

Update the name of Wokspaces

  • Remove '-ws' from Workspace name

Cannot delete Stream while Training is in running

Change the visible members when adding account to the workspace

  • the visible members : Account not assigned to the workspace
  • the visible members : Account without ADMIN privileges

Bug Fixes​

Fix the bug in Train Resource output

  • Apply the Train Resource with the fixed output bug to Inf Model page

v1.9.3​

June 28, 2024

New Features​

Add REST APIs of Workspace Exectuion Spec (Train Resource)

  • Retrieve Workspace Exectuion Spec Information
  • Add Workspace Exectuion Spec
  • Update Workspace Exectuion Spec
  • Delete Workspace Exectuion Spec

Validation Report​

Change Inactivity Period for User Lockout

  • Developer (Manager, Member) User : 90 days
  • Admin User : 30 days

Improvements​

Fix for cases where an error occurs when writing a solution description

  • The modal does not close even if an error occurs.
  • Previously written content will remain the same.

Increasing the length of the description content of the AI Solution

  • From 500bytes to 5000bytes

Display the Train records in chronological order on page Inf. Model

  • Sort in chronological order based on "Trained Date".

v1.9.2​

May 30, 2024

Improvements​

Exposed system users (edge) belonging to the Workspace

  • If not exposed, system users cannot be deleted when the Workspace is deleted
  • Only admin can delete system users

Bug Fixes​

Fixed the bug in the Workspace creation REST API


v1.9.1​

May 23, 2024

Improvements​

Update admin authority

  • Admin has access to all workspaces.
  • Admin can perform CRUD operations in all workspaces

Update status of "Stream Coupling Status" in Instance page

  • Coupling : exist Stream based on Instance
  • Decoupling : not exist Stream based on Instance

v1.9.0​

May 10, 2024

Improvements​

Individual EC2 allocation during model training execution

  • Users can use the memory provided by EC2, excluding 2GB. (2GB : system default size)
  • When all EC2s are in use, additional model training requests wait until an available EC2 occurs.

Improved status column data in stream history

  • Provides trained data, train duration, and related logs for stream history whose status is failed or error

Change in permission policy related to workspace member management

  • A workspace manager cannot manage workspace members

Update the status of result of model training

  • Succeeded : Model training completed and model creation completed
  • Failed : Model training failed

v1.8.0​

Apr 08, 2024

New Features​

Major assessments are all completed

  • Personal information impact assessment
  • Cloud impact assessment
  • Penetration testing

Improvements​

Workspace manager is not allowd to add/delete/update members in the workspace

  • Only admin could manage members in a workspace as requested from AIDX project

Same AI Solution name is allowed in different workspaces

  • It should be unique in all workspaces but not anymore
  • Managements for instance name and solution name should also be updated later

Update pipeline status by a exit event instead of polling

Change columne name on instance listing UI and show tooltip for the values(Coupled, Decoupled)

  • colume name : Edge status -> Stream coupling status

Unified account UI as the feature requested

  • disabling account UI on Customer Service Index and IOX
  • enabling account UI on Mellerikat

Internal users like Admin and Edge are not listed on member list UI


v1.7.1​

Apr 02, 2024

Bug Fixes​

Fix log viewer in Train History

  • Synchronize the name of log file generated by execution of train pipeline.

v1.7.0​

Mar 29, 2024

New Features​

On the Train History table, the "Train Resource" has been added

  • "Train Resource" shows what resource the training was executed with.

Add hyperlink to Solution Description

  • When you enter a URL, the description is converted to a hyperlink and displayed.

Deletion related Solution and Instance have been updated.

  • The solution can only be deleted if the connected instance is deleted.
  • Deleting an instance is possible only after deleting the related stream.

Applying log viewer in Train History

  • pipeline.log generated during asset operation of AI Contents.
  • process.log generated when executing Train pipeline.

Displaying the reason for login failure

  • Invalid ID
  • Invalid PW (failed password attempts)
  • Lock : There have been more than 5 failed login attempts for this account.
  • Lock : This account has been inactive for a long time.

Improvements​

On the Instance table, the conditions for updating the "Operation Status" and "Last Trained Date"

  • The "In-Use" conditions of "Operation Status" is being utilized by the Edge Conductor.
  • The date condition of "Last Trained Date" has been executed as a training request by the Edge Conductor.

Apply regulation logger

  • Applied regulation logger for easier debugging.
  • Applied regulation logger for data and user activity tracking.

Applied Redis encryption for security

  • Set the configuration of transit encryption to be enabled in Redis
  • Use Boltdb instead of redis in kubeflow

Applying Kubeflow Pipeline ExitHandler

  • The method of checking if the Kubeflow Pipeline is complete has changed from Polling to ExitHandler.

Unified response for Solution and SolutionVersion REST API


v1.6.0​

Mar 13, 2024

New Features​

Display the Version of AI Conductor at the UI of AI Conductor

  • The Version of AI Conductor is displayed at the login page and the account detail.

Add the page of AI Solution at the UI of AI Conductor

  • You can view the information of the generated AI Solutions and their Versions

Add the page of Account at the UI of AI Conductor

  • The Account page is accessible only to MANAGERs
  • The MANAGER can manage the accounts by the UI of AI Conductor

Add page to register solution description

  • You can write a description for solution after first registering the solution.
  • Modifications can also be made after the initial write, and these are immediately reflected in the instance.

Locking long-term inactive accounts

  • Accounts that have been inactive for more than 90 days will be locked.
  • You can request to activate locked accounts to admin.

When there are 5 consecutive login failures, the account will be locked.

  • If there are 5 consecutive login failures, the account will be locked.
  • You can request to activate locked accounts from the admin.

Activate locked account

  • If your account is locked due to the above two reasons, you can request to activate it from the admin.

Apply formatted log formatting

  • Apply formatted logging using Python's logger.
  • You can view logs of the training results by downloading the Artifact at the Train History.

Improvements​

The description of AI Solution supports Markdown

  • The description of the AI Solution can be written in Markdown style.
  • You can view the description written in Markdown on the AI Solution page.

Improving the Information on the AI Solution Instance Page

  • You can check list of instances based on the use of Edge Conductor.
  • By checking the instance, You can view the Train History of that instance.

Constraints on Creating AI Solution Instance

  • Only one AI Solution Instance can be created by an AI Solution Version.

Add deletion permission to the MEMBER

  • You(MEMBER) can delete the AI Solution (Version) and Instance that you have created.

Increase in the length of AI Solution and Instance names

  • The length of names has been increased from 50 bytes to 100 bytes.

Restrictions on operations within the Workspace

  • To prevent misuse of AI Conductor, access to other Workspaces is restricted.

unified login

  • We have unified the previously segregated static and LDAP user logins.

Simultaneous Login Restriction

  • AI Conductor allows only one login per account.
  • An account cannot login AI Conductor simultaneously from multiple locations.

Bug Fixes​

The length(10000 bytes) limitation in the METADATA of AI Solution

  • The METADATA is passed AI Conductor->Kubeflow Pipeline->Train Container as an argument.
  • An error related to size length occurs in Kubeflow Pipeline.
  • To avoid this error, the argument is directly passed from AI Conductor to the Train Container.

Incomplete Training

  • The completion of Training is verified through Kubeflow Pipeline.
  • If the Kubeflow session expires, the verification becomes impossible.
  • If the Kubeflow session expires, AI Conductor can update automatically the session.

v1.5.0​

Feb 02, 2024

New Features​

Release the UI of AI Conductor

  • You can log in.
  • You can check AI Solutions.
  • You can check and create AI Solution Instances.
  • You can check Train History.

We have Creating / Listing / Deleting AI Solution Version

  • AI Solution has Versions, and these Versions are used for updates to the respective Solution.
  • You can create, view, and delete AI Solutions and their versions.

Supporting the LDAP login

  • You can log in with a account created through LDAP.

Grant a permission based on account role

  • The account has either MANAGER or MEMBER privileges.
  • The MANAGER can manage the accounts

Supporting Cluster Auto Scaler when Training

  • You can save cost during Training, By applying Cluster Auto Scaler.
  • It takes a maximum of 1 minute to wake up an EC2 instance for training.
  • If an EC2 instance remains unused for 10 minutes, it will be terminated.

v0.7.0​

Nov 30, 2023

New Features​

Supporting the static login (not including LDAP)

  • a user for admin created
  • a user for 고객지수 MLOps Manager created
  • a user for LG Magna MLOps Manager created

Supporting two Workspaces

  • a workspace for 고객지수 MLOps created
  • a workspace for LG Magna MLOps created

Supporting updating the name of the Stream

  • You can view the name of the Stream in the Train History of the AI Solution Instance.
  • The name of the Stream can be updated by the Edge Conductor.

Bug Fixes​

fixed the issue of not being able to retrieve data

  • "MySQL server has gone away" is generated when the issue occurs
  • Add the 'pool_pre_ping=True' option to the MySQL database for the issue.

v0.5.0​

Nov 17, 2023

New Features​

We have Creating / Listing / Deleting Workspace

  • Workspace is a space that manages AI solutions, instances, and training.
  • Only admin users can control the Workspace.

We have Creating / Listing / Deleting AI Solution

  • You can create, view, and delete AI Solutions

We have Creating / Listing / Deleting AI Solution Instance

  • AI Solution Instance can be created with AI Solution.
  • You can configure resources for training

We have Requesting / Listing Train

  • Train is requested by the AI Solution Instance.
  • You can access artifacts for the training results.
  • You can check the list of training results in the AI Solution Instance.