Track 2 — Builder 构建 · M10 · 30 min · Live today
Self-hosting in China
Ubuntu or Windows 11 GPU box, Qwen via Ollama, local database, no cross-border egress required. Same build, configurable model source.
Prerequisites: M8 · Run safety: budgets & kill switch (W4)
After this module you can…
- Describe a fully local deployment: model source, database, egress.
- Pin a model id so behaviour is reproducible.
Key points
- The chat model is resolved from environment, so hosted and local share one build.
- Runbooks cover both Ubuntu and Windows 11.
Hands-on exercise: Local dry run
- Install Ollama and pull a Qwen model.
- Point the environment at the local model and run one governed task.
Evidence of success: A governed task completed with zero cross-border calls.
Knowledge check
Getting it wrong is fine — the explanation tells you why.
1. Does a local box need cross-border access?
2. Why pin the model id?
Module 10 of 20 · progress is stored on this device.