This section shows what changed in {{ ml-platform-full-name }}.
{% note tip %}
To keep up to date with the latest changes and updates, subscribe to our {{ ml-platform-full-name }} Community news channel on Telegram.
{% endnote %}
Added a new configuration, gt4i.1 (1 GPU NVIDIA T4).
Discontinued supporting foundation model tuning in {{ ml-platform-name }}. Previously tuned models will no longer be supported along with their base models in accordance with their life cycle.
Use the {{ foundation-models-full-name }} tools to tune models.
Now you can use a service agent to work with {{ yandex-cloud }} services from {{ ml-platform-name }} notebooks, e.g., issue authorization tokens. To enable this feature in a community, follow this guide. For more information on how service agents work, see the {{ iam-name }} documentation.
- Added examples of operations with {{ yandexart-name }} and open-source foundational models to initial notebooks.
- When working with {{ dataproc-full-name }} using a Spark connector, you can now synchronize the environment
- Fixed some bugs and added minor performance improvements.
- Added a feature to create communities in different availability zones:
{{ region-id }}-a
and{{ region-id }}-b
. - You can now connect to your nodes an additional disk of 10 to 4,096 GB.
- Fixed some bugs and added minor performance improvements.
- {{ ml-platform-name }} projects now have a new type of resources: Spark connectors for integration with {{ dataproc-full-name }}.
- Improved creating nodes.
- Improved linking a billing account to a community.
- Fixed some bugs and added minor performance improvements.
- Now you can deploy node instances in different availability zones:
{{ region-id }}-a
and{{ region-id }}-b
. - Now you can rerun jobs.
- Python 3.7 is no longer supported.
- Fixed some bugs and added minor performance improvements.
- Updated configurations of {{ dataproc-name }} temporary clusters.
- Now you can use XGBoost and LightGBM models to deploy nodes from models.
- Added delivering input variables in fulfillment APIs.
- Improved creating nodes from Docker images.
- Fixed some bugs and added minor performance improvements.
Model tuning in {{ ml-platform-name }} now works based on the new {{ gpt-pro }} model.
The Serverless mode is no longer supported.
- Added the option to run a notebook in Dedicated mode to the API.
- Improved logs and metrics for nodes.
- Fixed bugs and added minor improvements in platform performance.
- Updated the NVIDIA driver to version 535.
- Added support for multi-login to multiple organizations in various federations.
- Added the option to pause and resume a running node.
- Fixed bugs and added minor improvements in platform performance.
- Added self-service problem-solving tools to the project page.
- In {{ ml-platform-name }} Jobs, now you can use your project resources: secrets, S3 connectors, environment dockers, datasets, and project disk.
- Fixed bugs and added minor improvements in platform performance.
- Added new configuration, gt4.1 (1 GPU NVIDIA T4).
- The g2.mig configuration (1 GPU MIG NVIDIA Ampere A100) is obsolete.
- A new node type from the model resource is available.
- Selecting a configuration in {{ dd }} mode will display its current availability.
- Fixed bugs and added minor improvements in platform performance.
- You can test fine-tuned {{ yagpt-name }} models right in {{ ml-platform-name }}. {{ yagpt-name }} Playground in {{ ml-platform-name }} is available after fine-tuning to users with access to {{ yagpt-full-name }}.
- You can now connect your {{ ml-platform-name }} project to {{ jlab }}Lab from a local IDE.
- Fixed bugs and added minor improvements in platform performance.
- With {{ ml-platform-name }} Jobs, cloud computing resources in {{ ml-platform-name }} can now be used from a user's local environment.
- {{ ml-platform-name }} projects now have a new type of resources: Models.
- Optimized JupyterLab 3 (available in dedicated mode) by adding new extensions.
- {{ yagpt-name }} model tuning is now available at the Preview stage.
- Fixed bugs and added minor improvements in platform performance.
- A new DS Default (Python 3.10) system image is used by default.
- Community administrators can now manage permissions to use the functionality.
- Improved working with community and project lists.
- Fixed bugs and added minor improvements in platform performance.
- Updated the {{ jlab }}Lab extension for working with GIT.
- Community administrators can now manage permissions to use computing resources.
- Community and project members can now be added before they accept an invitation to join an organization.
- Improved the Docker image build editor.
- Added an example of operations with {{ yagpt-full-name }} to initial notebooks.
- The process of starting a project is now more obvious and transparent.
- Fixed bugs and added minor improvements in platform performance.
- Added a page with a [list of all user projects]({{ link-datasphere-main }}projects).
- Updated initial notebooks.
- Fixed bugs and added minor improvements in platform performance.
- {{ ml-platform-name }} now supports a new {{ dd }} operation mode.
- In the {{ dd }} mode, the IDE version is updated to JupyterLab 3.5.3.
- You can now select an organization in an optimized way.
- Operations with community and project members are now easier to perform.
- Fixed bugs and added minor improvements in platform performance.
- You can now work with labels to label resources.
- Fixed bugs and added minor improvements.
- Added a tool to migrate projects to the new {{ ml-platform-name }} version.
- Fixed bugs and added minor improvements.
- You can now use the new {{ ml-platform-name }} version via the API.
- Fixed bugs and added minor improvements.
- The service now displays inherited roles of community and project members.
- Optimized the advanced settings for projects.
- Updated the snippets for working with S3, Yandex Disk, and Google Drive.
- Fixed bugs and added minor improvements.
Greatly improved the Apache Spark™ cluster functionality:
- {{ ml-platform-name }} now has a new type of resources: {{ dataproc-name }} templates.
- You can now configure a livy session when using {{ dataproc-name }} clusters.
- {{ ml-platform-name }} now supports the Spark SQL library.
Meet our large {{ ml-platform-name }} update: new interface, communities, resources, and many other features for ML development.
{% include old-releases %}