AI · Knowledge track

Understand
AI models.

Model names change quickly. The concepts underneath them change more slowly. This track explains what an AI model is, what parameter counts do and do not tell you, how context windows shape a run, and what open-weight or closed access means in practice.

Conceptual AI model anatomy showing parameters feeding an inference runtime that uses context and produces output tokens.
Start here

What is an AI model?

Separate the trained model from the application, runtime, API and surrounding tools.

Build the mental model →
Scale

Model parameters

Understand what billions of parameters represent — and why parameter count is not a quality score.

Understand parameters →
Runtime

Context window

Learn what occupies context, why usable context differs from a headline number and how long inputs affect a run.

Explore context →
Access

Open-weight vs closed

Compare weight access, local deployment, API dependence and the separate question of licensing.

Compare access models →
Execution

Model inference

Follow a prompt through tokenization, prefill, decoding, KV cache and hardware execution.

Understand inference →
Precision

Quantization

Understand lower-precision weights, scales, groups and the trade-offs behind bit labels.

Understand quantization →
Hardware bridge

Size vs memory

Connect parameters and precision to file size, RAM, VRAM, context state and runtime overhead.

Connect the numbers →
Runtime state

KV cache

See why inference stores attention state from prior tokens and why active context can add memory.

Understand the cache →
Interactive

Model Size & Memory Explorer

Compare theoretical weight payloads across precision classes and add a conventional KV-cache estimate.

Explore model size →
Knowledge map

From model architecture to real hardware

Use this path when you want to connect model concepts instead of reading isolated definitions.

1 · Parameters

Learned values define the weight scale.

2 · Precision

Representation changes weight storage.

3 · Inference

The runtime executes the model.

4 · Context

Active tokens add runtime state.

5 · Memory

Weights + state + overhead meet hardware.

Turn the concepts into a memory estimate with the VRAM Calculator →

MODEL ≠ PRODUCT
A model is one component of an AI system. Interfaces, retrieval, tools, safety layers, prompts and inference settings can substantially change the behavior users observe.