Foundational models (FMs) are revolutionizing machine learning (ML) and artificial intelligence (AI), enabling faster development and adaptability to various tasks and applications. By combining IBM watsonx data and AI platform with edge computing, enterprises can run AI workloads at the operational edge, resulting in near-real-time predictions while complying with data sovereignty and privacy regulations. This allows for scaling AI deployments at the edge, reducing deployment time and costs.
Foundational models (FMs) are trained on large amounts of unlabeled data, making them versatile and applicable to multiple tasks and domains. They provide a foundation for AI applications and overcome the challenges of limited labeled data and labor-intensive annotation processes. FMs unlock the potential of massive amounts of unlabeled data in enterprises to drive valuable insights.
Large language models (LLMs) are a type of FM that utilizes layers of neural networks trained on vast amounts of unlabeled data. These models excel in natural language processing (NLP) tasks and mimic how humans use language. LLMs have become a prominent force in the field of AI.
To scale the impact of AI, IBM offers the IBM watsonx platform, consisting of watsonx.ai, watsonx.data, and watsonx.governance. These tools empower enterprises to leverage the power of FMs and multiply the impact of AI across their organization.
The importance of computing at the enterprise edge is growing, allowing for real-time analysis of data generated in various locations like manufacturing floors and retail stores. Edge AI enables near-real-time predictions while maintaining data sovereignty and privacy. It reduces latency, safeguards sensitive data, and minimizes data transfer costs. However, scaling AI deployments at the edge comes with challenges such as deployment time and management complexity.
IBM has developed an edge architecture that addresses these challenges by using an integrated hardware/software (HW/SW) appliance model. It offers policy-based, zero-touch provisioning, continuous system monitoring, and centralized management for software updates. A hub-and-spoke architecture, where a central cloud acts as the hub and edge-in-a-box appliances act as spokes, enables scalable AI deployments across hybrid cloud and edge environments.
The pre-training of base FMs can be performed in the cloud, utilizing its abundant compute resources and large data storage. Fine-tuning and inference for downstream tasks can be done at the enterprise edge, reducing the need for GPU resources and ensuring the safety and cost-effectiveness of sensitive data. Deploying AI models using a full-stack approach streamlines the development lifecycle and reduces latency associated with data processing.
Overall, the combination of edge computing and FMs allows for faster, scalable, and cost-effective AI deployments with near-real-time predictions and compliance with data regulations. An exemplar use-case of deploying an FM model for defect-detection using drone imagery illustrates the value proposition of FM finetuning and inference at the operational edge.