---
title: "LLM: chat and code assistants"
lesson_id: "02"
---

# LLM: chat and code assistants

- **Lesson ID:** 02
- **Goal:** Teach the official `llm` presets as language-writing choices for replies, drafts, summaries, and code.
- **Human lesson:** [02-llms.html](02-llms.html)

## Prerequisites

- Lesson 01 and the difference between the three registry types.
- The language-model shape in [the architectures course](../model-architectures-plain/02-language-models.html).

## Inputs, outputs, and artifacts

- **Inputs:** Text context, app job, device target, and the supplied official LLM family list.
- **Outputs:** A simple map of every listed LLM family and the packing note.
- **Artifacts:** Progressive language-writing and device-choice `.flow` diagrams.

## Agent build steps

1. Explain that `llm` takes instructions and context and writes language.
2. Cover Qwen2.5 and Qwen3, including the exact 8bit-kv variants and coder entry.
3. Cover Gemma 3 and Gemma 3n using only the supplied names and variants.
4. Cover Mistral, Mixtral, and GPT-OSS, explaining Mixtral's MoE style: different parts of the job can go to different expert groups.
5. Cover Phi, Muse Glimmer with its drafter assistant, and SmolLM2 entries.
6. Explain model size as a simple trade of capacity versus footprint, and macOS versus iOS packing as the same chat job packed for different devices.
7. Link the architectures, forward, and From the metal to the model.

## Constraints

Do not add models or capabilities beyond the supplied catalog. Explain MoE in simple English. No math, CUDA, quizzes, Mermaid, or em dashes.

## Key concepts

- `llm` reads instructions and context and writes language.
- Named families include Qwen, Gemma, Mistral, Mixtral, GPT-OSS, Phi, Muse Glimmer, and SmolLM2.
- Many families have separate macOS and iOS compression presets.
- `8bit-kv` is another packing choice, not a different job.

## Takeaways

- You can pick `llm` when the app must write language, not label a photo.
- You can name official presets from the registry, not invented model names.
- You can treat macOS versus iOS packing as the same chat job packed for a Mac or a phone.

## Acceptance checks

- Every supplied LLM family and example is named exactly.
- `qwen3-coder-30b-a3b-instruct` and the Qwen `8bit-kv` variants are present.
- Mistral's macOS and iOS presets, Mixtral's MoE style, and Muse Glimmer's drafter assistant are explained.
- The device-packing explanation is present.
- Progressive `.flow` diagrams, prior-course links, and the Agent brief link are present.