Courtesy: Intel
Applications powered by artificial intelligence are some of the most popular pieces of software being developed, especially on cloud computing platforms, which can provide easy access to specified hardware and accelerators at a low startup cost with the option to scale effortlessly. A popular cloud service provider, Google Cloud Platform* (GCP), contains a suite of cloud computing services that provide a variety of tools to develop, analyze, and manage data and applications. GCP also includes tools specific to AI and machine learning development, such as the AI Platform, the Video Intelligence API, and the Natural Language API. Using a platform like GCP for your AI projects can simplify your development while gaining access to powerful hardware that meets your specific needs.
Further enhancements to model efficiency can be accomplished with pre-built software optimizations tailored for diverse applications. By implementing these software optimizations, developers can see models deploy and infer faster and with fewer resources. However, the process of discovering and integrating these optimizations into workflows can be time-consuming and demanding. Accessing comprehensive guides and documentation packaged in an open-source environment empowers developers to overcome challenges by incorporating new optimizing architectures, facilitating the effortless enhancement of their models’ performance.
What are Intel Cloud Optimization Modules?
The Intel Cloud Optimization Modules consist of open-source codebases that feature codified Intel AI software optimizations designed specifically for AI developers working in production environments. These modules provide a set of cloud-native reference architectures to enhance the capabilities of AI-integrated cloud solutions. By incorporating these optimization solutions, developers can boost the efficiency of their workloads and ensure optimal performance on Intel CPU and GPU technologies.
These cloud optimization modules are available on several highly popular cloud platforms, including GCP. The modules utilize specifically built tools and end-to-end AI software and optimizations that enhance workloads on GCP and increase performance. These optimizations can increase machine learning models for a variety of use cases, such as Natural Language Processing (NLP), transfer learning, and computer vision.
Within each module’s content package is an open-source GitHub repository that includes all the relevant documentation: a whitepaper with more information on the module and what it relates to, a cheat sheet that highlights the most relevant code for each module, and a video series with hands-on walkthroughs on how to implement the architectures. There is also an option to attend office hours for specific implementation questions.
Intel Cloud Optimization Modules for GCP
Intel Cloud Optimization Modules are available for GCP, including optimizations for generative pre-trained transformer (GPT) models and Kubeflow pipelines. You can learn more about these optimization modules available for GCP below:
nanoGPT Distributed Training
Large Language Models (LLMs) are becoming popular in Generative AI (GenAI) applications, but it is often sufficient to use smaller LLMs in many use cases. Using a GPT model, such as nanoGPT (124M parameter), can result in better model performance, as smaller models are quicker to build and easier to deploy. This module teaches developers how to fine-tune a nanoGPT model on a cluster of Intel Xeon CPUs on GCP and demonstrates how to transform a standard single-node PyTorch training scenario into a high-performance distributed training scenario. This module also integrates software optimizations and frameworks like the Intel Extension for PyTorch* and oneAPI Collective Communications Library (oneCCL) to accelerate the fine-tuning process and boost model performance in an efficient multi-node training environment. This training results in an optimized LLM on a GCP cluster that can efficiently generate words or tokens suitable for your specific task and dataset.
XGBoost on Kubeflow Pipeline
Kubeflow is a popular open-source project that helps make deployments of machine learning workflows on Kubernetes simple and scalable. This module guides you through the setup of Kubeflow on GCP and provides optimized training and models to predict the probability of client loan default. By completing this module, you will learn how to enable Intel Optimization for XGBoost and Intel daal4py in a Kubeflow pipeline. You’ll also learn to set up and deploy a Kubeflow cluster using Intel Xeon CPUs on GCP with built-in AI acceleration through Intel AMX. Developers also have the option to bring and build their own Kubeflow pipelines and learn how these optimizations can help improve the pipeline workflow.
Elevate your AI initiatives on GCP with Intel Cloud Optimization Modules. These modules can help you leverage Intel software optimizations and containers for popular tools to develop accelerated AI models seamlessly with your preferred GCP services and enhance the capabilities of your projects. See how you can take AI to the next level through these modules, and sign up for office hours if you have any questions about your implementation!
We encourage you to check out Intel’s other AI Tools and Framework optimizations and learn about the unified, open, standards-based oneAPI programming model that forms the foundation of Intel’s AI Software Portfolio. Also, check out the Intel Developer Cloud to try out the latest AI hardware and optimized software to help develop and deploy your next innovative AI projects!
