---
name: Model architectures in plain English
description: >-
  Teach someone comfortable with computers how language, image, classifier, and
  generative model shapes differ, using the linked clothing course and From the metal to the model.
---
# Model architectures in plain English

Use this course to teach model architecture through simple input-to-output jobs. The reader is comfortable with computers but may be new to models. Keep hard words light and explain each term when it first appears.

## End-to-end agent recipe

1. Read the course index and identify the lesson that matches the reader question.
2. Read the lessons in order when the reader is new. Start with [lesson 01](01-what-architecture-means.llms.md) to separate architecture from weights.
3. Explain language models with [lesson 02](02-language-models.llms.md). Keep the sequence words → tokens → next token → reply visible.
4. Explain image models with [lesson 03](03-image-models.llms.md). Keep Fashion-MNIST ClothingCNN in the see and classify camp.
5. Use [lesson 04](04-how-they-differ.llms.md) for a side-by-side comparison of a clothing classifier and a chat model.
6. Use [lesson 05](05-picking-a-shape.llms.md) to choose a model shape from an app feature and continue into profiling.
7. Keep the two anchor courses in view throughout. [Forward](../pytorch-core-ai-xcode/) runs Fashion-MNIST through ClothingCNN, Core AI, and SwiftUI. [From the metal to the model](../core-ai-from-metal-back/) runs from Metal through Instruments, runtime, UI, conversion, training, and pipeline.
8. When discussing performance, say that different architectures still become work on silicon and that the hardware-first direction still applies.
9. Use the human HTML pages for the reader. Use the linked `.llms.md` briefs for agent execution and acceptance checks.
10. Verify that every human page has its Agent brief link, every brief links to its HTML page, every lesson links both anchor courses, and diagrams are progressive HTML `.flow` elements.

## Constraints

- Plain spoken English for someone who is comfortable with computers.
- Teach with input-to-output jobs.
- No math deep-dives, papers, CUDA, or quizzes.
- Use no Mermaid. Build progressive `.flow` diagrams in HTML.
- Do not use em dashes. Prefer short sentences.
- Keep the relative links exactly as used in the lessons: `../pytorch-core-ai-xcode/` and `../core-ai-from-metal-back/`.

## Key concepts

- Architecture is the shape of a model: what goes in, what comes out, and how information moves.
- Weights are the learned settings inside that shape. Architecture is the shape itself.
- This course uses ClothingCNN on Fashion-MNIST and From the metal to the model as the running examples.
- Language, image, classifier, and generative models are different jobs, not different brand names.

## Takeaways

- You can name a model by its input and output before you name a file or a vendor.
- You can place ClothingCNN as pixels in, clothing label out.
- You know the next step is the forward clothing course or From the metal to the model.

## Acceptance checks

- Every human lesson HTML has **Key concepts** and **Takeaways** sections before lesson-nav. The course home has the same two headings for the whole course.
- Five human lessons exist in numeric order.
- Five matching `.llms.md` briefs exist with executable steps and acceptance checks.
- Every brief includes a link to its human HTML page.
- Every human lesson includes an Agent brief link and both anchor-course links.
- The course index lists all five lessons and all six agent-facing files.

## Environment setup

No runtime environment is required: this course is conceptual and contains no executable code blocks.