Instacart, a leading online grocery platform, has recently launched Griffin 2.0, an advanced machine learning (ML) platform aimed at streamlining the development and deployment of ML applications. This new version builds upon the success of the first-generation Griffin platform while addressing its limitations.
One of the key improvements in Griffin 2.0 is the shift to a service-oriented architecture, moving away from Git-based tools and a command-line interface (CLI). This architectural change, combined with an intuitive web user interface, provides a seamless experience for Machine Learning Engineers (MLEs). Additionally, the Griffin SDK can be integrated with tools like BentoLM and Instacart’s cloud-based development environment for machine learning notebooks.
Griffin 2.0 boasts three major subsystems in its backend: the Model Training Platform (MLTP), Model Service Platform (MLSP), and Feature Store. The MLTP leverages Ray to provide a scalable computing environment and supports various training backend platforms. The MLSP automates model artifact storage, deployments, and provisioning of inference services, ensuring the availability of ML models at scale. The Feature Store facilitates the computation, ingestion, discovery, and sharing of features, simplifying the configuration workflow.
This new version of Griffin also introduces a centralized feature and metadata management system, enabling distributed computation and standardized serving mechanisms. These enhancements position Griffin 2.0 as an ideal platform for advanced applications such as training and serving Large Language Models (LLMs). The user-friendly UI-based workflow and data validation further streamline the creation of new feature sources and computation, improving the quality of generated features.
While Griffin 2.0 represents significant progress, the researchers behind the platform are actively seeking feedback to further enhance its functionality. They aim to foster adoption and continually improve the user experience, scalability, and capabilities of Griffin, aligning with their vision for the future of the platform.
Overall, Instacart’s introduction of Griffin 2.0 marks a significant step forward in their ML research and application development. By addressing limitations and introducing new features, the platform aims to reshape the domain of ML applications, providing a more efficient and user-friendly experience for developers and engineers.