Sign up to save tools and stay up to date with the latest in AI
bg
bg
1

Release 2024.2.0 · openvinotoolkit/openvino

Jun 18, 2024 - github.com
The markdown data highlights the major features and improvements of the OpenVINO toolkit 2024.2 release. These include more Gen AI coverage, Llama 3 optimizations for CPUs and GPUs, support for Phi-3-mini AI models, Python Custom Operation in OpenVINO, and expansion of notebooks. The release also offers broader Large Language Model (LLM) support, GPTQ method for 4-bit weight compression in NNCF, and significant LLM performance improvements. Other enhancements include Model Serving Enhancements, support for Intel Atom Processor X Series, and Intel Xeon 6 processor.

The data also provides information on support change and deprecation notices. Several features and components, including Intel Gaussian & Neural Accelerator, OpenVINO C++/C/Python 1.0 APIs, and certain tools like Deployment Manager and Accuracy Checker, will be discontinued in 2024.0. The OpenVINO Development Tools package and Model Optimizer will be removed from installation options and distribution channels beginning with OpenVINO 2025.0. The document also acknowledges contributions from the OpenVINO developer community.

Key takeaways:

  • OpenVINO has introduced more Gen AI coverage and framework integrations, Llama 3 optimizations for improved performance, and support for Phi-3-mini AI models.
  • Python Custom Operation is now enabled in OpenVINO, allowing Python developers to code their custom operations more easily. There's also an expansion of notebooks for better coverage for new models.
  • There are significant improvements in Large Language Model (LLM) support, including GPTQ method for 4-bit weight compression, performance improvements, and reduced latency for GPUs.
  • Several features and components are set to be deprecated or discontinued in the future, including the OpenVINO Development Tools package, Model Optimizer, and certain OpenVINO Model Server components.
View Full Article

Comments (0)

Be the first to comment!