Local Models via Ollama
Advanced · Technical & Agentic track · ~35 min hands-on + readings and quiz
← Back to the Technical & Agentic track
When privacy, cost, or offline use matter, a local model on your machine can be the right tool.
What you’ll be able to do
- Run an open model locally with Ollama
- Judge when local beats cloud (privacy, cost, offline)
- Know the trade-offs in capability and setup
Overview
Tools like Ollama run open-weight models on your own computer — nothing leaves the machine. That’s attractive for sensitive-adjacent work, cost control, or offline settings. The trade-off is capability and setup: local models are smaller and need a capable machine.
Local is not automatically “approved” for sensitive data — confirm with Drexel before using any tool, local or cloud, with protected information.
Practice activities
Activity 1 · Advanced — Run a model locally
Time ~20 min · Tools Ollama + a capable laptop
Goal. Run an open model offline and judge the tradeoff.
Setup. Install Ollama (ollama.com) and pull a small model (e.g., a 7–8B model).
Steps.
Run a basic task entirely offline — summarize a public document:
ollama run <model>then paste the text.Run the same task on a cloud model (ChatGPT Edu) and compare quality and speed.
Note where the local model is good enough and where it falls short.
Expected result. A working local run plus a quality/speed comparison.
Check your work. Confirm nothing left your machine (that’s the point) — and that Drexel rules still apply for any sensitive data.
Common pitfalls. Local is not automatically approved for sensitive data. Local models are smaller and weaker — set expectations.
Stretch (optional). Write the conditions under which you’d choose local over cloud.
Activity 2 · Advanced — Decide local vs. cloud for a real task
Time ~12 min · Tools ChatGPT Edu
Goal. Make a defensible tool choice.
Setup. A task where privacy, cost, or offline use matters.
Steps.
List the task’s constraints: privacy, cost, connectivity, capability needed.
Argue both sides:
Given these constraints, argue local vs. cloud for this task. What would change the decision?
Decide, and note the data-rule check.
Expected result. A reasoned local-vs-cloud decision.
Check your work. Did you confirm the data classification regardless of where the model runs?
Common pitfalls. “It’s local, so it’s fine” is not a data-rule exemption.
Stretch (optional). Estimate the capability gap for your specific task by testing both.
Check your readiness
Answer these, then check — your score suggests whether to dive in or skim the readings first.
Recommended readings
Available in the shared OneDrive folder Staff Faculty AI Workshop → Readings, and online where linked:
- Running Local Models Is Good Now (Vicki Boykis) — a practical look at the current state of local models.
Useful resources
- Ollama — run open models locally.
- Hugging Face — open models and tools.
- ChatGPT Edu (Drexel AI Tools) — the approved tool, with data rules.