What's new in ML Runtimes version 2025.01.1

Review the major features and updates for ML Runtimes.

This release is available with ML Runtimes version 2025.01.1.

New Features

ML Runtimes' base image has been updated from Ubuntu 20.04 to Ubuntu 24.04. This update enables many new features, a better experience for end users, faster and better fixes of CVEs and enhanced security to prevent malicious activities.
note

Due to the extensive changes with SSL between Ubuntu 20.04 and Ubuntu 24.04, we recommend testing your workflows if they contain any https/tls/ssl/... operations.
The new ML Runtimes are not drop-in replacements of the previously released ML Runtimes. If you install any additional software into Cloudera AI projects, it might result in issues and might require reinstallation or recompilation.
CUDA images are based on nvidia/cuda:12.5.1-devel-ubuntu24.04. The CUDA version on the Nvidia GPU Runtimes has also been also updated.
All Runtime images are now PBJ-based (Powered by Jupyter); Workbench-based Runtime images are no longer released.
note
You can no longer use any of the new ML Runtimes going forward with Cloudera Data Science Workbench. (CDSW support ended in May 2024)

Behavioral changes

Python changes
- Workbench-based Python images have been removed, only PBJ Workbench Python images and JupyterLab Runtime images will be released going forward.
- Python 3.7- and 3.8-based images are no longer supported.
  
  note
  
  Note, that only Python 3.7 Runtime is compatible with Spark 2. If you want to use Python with Spark 2, you need to install older ML Runtimes to your Cloudera AI Workbenches.
- All ML Runtime images have updated PIP (Python package manager) package versions.
JupyterLab changes
- JupyterLab has been migrated to PBJ (Powered by Jupyter).
- node.js has been removed.
R changes
- Non-PBJ Workbench Runtime images are no longer available.
- Python has been updated to Python version 3.12.
Scala changes
- Scala has been migrated to PBJ (Powered by Jupyter).
- Python has been updated to Python 3.12.
- Scala image has been updated to support Spark 3.x.
- Spark version 2.x is no longer supported (starting a Spark 2.x session will fail with an error).
- Scala has been updated to version 2.12. The Scala update might bring backward compatibility issues.
- When building a custom PBJ Scala Runtime image either use the Cloudera PBJ Scala Runtime or start from the Cloudera-provided public Dockerfiles. The custom PBJ Scala Runtime image provided by Cloudera contains necessary updates, such as the Scala chunker on the Runtime image and other necessary enhancements.
Conda changes
- Conda has been migrated to PBJ (Powered by Jupyter).
- Python has been updated to 3.12.

To ensure the compatibility of AMPs with ML Runtimes 2025.01.1 using JupyterLab, you need to proceed with either of the following options:

Switch to PBJ Workbench in the AMPs’ .project-metadata.yaml file.:
```
runtimes:
  - editor: PBJ Workbench
```

Use jobs instead of sessions for automated tasks.

Sessions are intended for interactive use, while AMP setup steps are executed in a programmed manner or automatically. Instead of running a session, change a session-based task into a job, create and execute a job.

Instead of running a session:

tasks:
  - type: run_session
    name: Install Dependencies
    script: scripts/install_dependencies.py
    memory: 2
    cpu: 1

Use the following example entry in the .project-metadata.yaml file for creating and executing a job:

tasks:
  - type: create_job
    name: Install Dependencies
    entity_label: install_deps
    script: scripts/install_dependencies.py
    arguments: None
    cpu: 1
    memory: 2
    short_summary: Create job to install project dependencies.
    environment:
      TASK_TYPE: CREATE/RUN_JOB

  - type: run_job
    entity_label: install_deps
    short_summary: Running install dependencies job.
    long_summary: >-
      Running the job to install dependencies.

The cdsw Python module has been replaced by the cml module.

Improvements

Cloudera Copilot enhancements:
- Support for default embedding models.
- Support for additional Amazon Bedrock models: Claude v3.5 Haiku, Claude v3.5 Sonnet, Llama 3.1 405b Instruct, and Mistral Large 24.02.
- Increase in maximum tokens limit to 2048 for Cloudera AI Inference model's responses.
Python maintenance versions have been upgraded as follows:
- 3.9.19 -> 3.9.21
- 3.10.14 -> 3.10.16
- 3.11.9 -> 3.11.11
- 3.12.4 -> 3.12.8

Fixed issues

R workloads' .Libpaths() changed
Earlier the order of R workloads' .Libpaths() was:
1. Cloudera AI project
2. CMLAddon
3. Runtime image
Now R workloads that are running PBJ Runtimes have the .Libpaths() set up in this order:
1. Cloudera AI project
2. Runtime image
3. CMLAddon
Reinstall your R packages in your projects if the above detailed change of order in R workload's library path might cause any issues.