Additional feature improvements and bug fixes have been updated since the official release of Backend.AI 22.09 in September.
Backend.AI Core & UI (22.09)
- Enhanced session scheduler stability
- Added feature to allocate NUMA cores used by accelerators supporting NUMA priority during session creation
- Fine-grained task authorization for each storage host
- Support for granting or revoking separate permissions for folder creation/deletion, downloading/uploading, and shared folder-related permissions
- Support for downloading session logs as files
- Support for direct download of the desktop app
- (Admin panel only) Added Shared Folder page
- Administrators can easily manage shared folder details
- Permission granting/revocation function for shared folders
Backend.AI Forklift
- Supports simple/advanced mode differentiation to allow for simplified or detailed configuration of session images according to their purpose
Backend.AI FastTrack
FastTrack 22.09 version has been released. Based on the Backend.AI cluster, FastTrack provides a flexible user interface that allows customization at the task level, enabling users to easily compose a series of linked tasks from model training to inference and deployment in a DAG (Directed Acyclic Graph) style. In particular, tasks can be created or configured on the visualization graph using interactions such as drag-and-drop, and task execution status and results are displayed on the graph for intuitive task monitoring and management. For more information, please see the Backend.AI FastTrack post.
Development and Research Framework Support
- Supports TensorFlow 2.11
- Supports PyTorch 1.13
- Supports NGC Tensorflow/PyTorch 22.11 versions
- Supports the November 2022 versions of TensorFlow images from NGC
- Supports the November 2022 versions of PyTorch images from NGC
- Note: PyTorch image's Python Conda-based image is version 3.9
More features are coming in the future!