Alauda AI

Release Notes

TOC

AI 1.5.0

Features and Enhancements

Langflow Agent Development

Leverage Langflow's upper-layer UI orchestration (powered by LangChain's foundational framework) to rapidly develop AI Agents for intelligent applications like expert Q&A, RAG-powered search, and workflow automation.

Featureform Feature Store Integration

Featuform serves as a dedicated feature store, resolving performance inconsistencies in AI model training-inference pipelines caused by divergent data processing logic through centralized feature definition, version-controlled storage, and real-time feature serving.

Evidently Data Drift Detection Capability Integration

Evidently, as a AI model monitoring component, provides real-time data drift detection functionality. It continuously analyzes statistical discrepancies between online feature distributions and training data to automatically identify potential risks of data distribution shifts.

Workbench Plugin Enhances

The legacy version of the Workbench plugin provided foundational support for model development environments but required users to manually configure environment definitions (e.g., Jupyter Notebook/VS Code container images) and development environment permission policies, resulting in a cumbersome operational workflow.

The new version introduces targeted optimizations for this scenario:

1、Pre-integrated Standardized Development Environments Built-in, ready-to-use Jupyter Notebook environment templates eliminate manual image downloading and synchronization steps. 2、Default Permission Management Configuration Provides predefined baseline permission policies for development environments while supporting customization of permission scopes via configuration files.

Fine-tuning Plugin Reconstruction and Stability Enhancement

The new Fine-tuning plugin has undergone an architectural-level reconstruction, with a primary focus on optimizing the stability of training workflows, introducing distributed training capabilities, and adjusting the built-in template strategy:

1、The built-in finetune-text-generation-llamafactory template supports fine-tuning tasks for mainstream text-generation models. 2、The three template types previously provided in older versions image classification, text classification, and text-to-image are no longer included as built-in options. Users can implement these functionalities through custom template solutions.

Training Plugin Reconstruction and Stability Enhancement

The new Training plugin has undergone an architectural-level reconstruction, with a primary focus on optimizing the stability of training workflows, introducing distributed training capabilities

Deprecated and Removed Features

Deprecation of Image Management Functionality in Model Repository

In Alauda AI v1.5.0, the original Image Management(model image building and image-based inference service)capabilities will be officially deprecated. These features were initially provided as experimental functionalities in earlier releases but, due to identified stability risks and usability concerns, have been evaluated and determined to no longer warrant ongoing maintenance.

Fixed Issues

  • On the Model Repository details page of the Alauda AI product, files under a folder are not displayed in the file preview on the left side of the File Management Tab page. This prevents users from previewing files in the Model Repository folder.
  • When the platform is configured with other access addresses and Workspace is accessed in a browser using the other access address, a 403 error is encountered. This is because the OIDC Issuer address configured by OAuth2-Proxy does not match the browser's access address, which causes the permission check to fail. This issue has been fixed in the new version.
  • After updating the Alauda AI name in the Platform Parameters submenu of the Platform Settings in the Administrator view, the name of the Alauda AI platform is not correctly changed.
  • When switching from "Standard Mode" to "Advanced Mode" after creating an inference service, a "deployment not found" event occurs because the hpa resource has not been deleted. This phenomenon does not affect the update of the inference service.

Known Issues

  • In the Alauda AI product inference service feature, the branch drop-down box for selecting the model repository in the Create Inference Service form can only display 20 pieces of data, making it impossible to select the newly created branch. Temporary solution: Delete useless branches from the model repository if the model is frequently modified.
  • With Aluada AI deployed on ACP 4.2.0, the monitoring data is not displayed on the monitoring details page of the reasoning service for reasoning services created using Alauda AI due to ACP updating the default security policy. This issue will be fixed in a new release of ACP.