📌 The traditional networking OSI model does not directly apply to AI infrastructure, but its layered abstraction is a useful framework for understanding the complex components of modern AI systems.
Several conceptual models for an "AI OSI model" have been proposed to help structure and troubleshoot AI development and deployment.
This model is a conceptual tools to help engineers and developers design, manage, and troubleshoot complex AI systems by breaking down functionality into manageable, modular layers.
This model maps closer to the actual hardware and software stack for AI compute, often utilized by infrastructure engineers:
Layer 1: Physical The atomic layer including GPUs, ASICs, network fabrics, power, and cooling.
Layer 2: Link System software that connects thousands of chips and enables compute scaling.
Layer 3: Neural Network The architecture layer (e.g., Transformers, RAG, LoRA) that forms the "brain".
Layer 4: Context Manages prompting, tokenization, and working memory.
Layer 5: Agent LLMs equipped with memory, tool use, and external APIs, where autonomy begins.
Layer 6: Orchestrator Manages and deploys agents across various environments.
Layer 7: Application The final user-facing layer (e.g., chatbots, copilots).
cc:
@bigpanda|
@ScienceLogic |
@IBM |
@elastic |
@newrelic
#AIOps