Lesson 4.6 — Moving to the Mac Mini

Introduction to AI · Module 4: Local LLMs · About 60 min · Mac Mini (Apple Silicon), Terminal

Goal: run local models on the Mac Mini, rerun your 4 tasks from Lesson 4.1, and compare speed with the 4090.

Before you start

• Copy t1.txt–t4.txt and your 4.1 score table from the PC to the Mac (USB stick, Google Drive or AirDrop)
• Open Terminal (Cmd+Space → type Terminal → Enter)

Step 1 · How much memory? (⏱ 0:00–0:05)

sysctl -n hw.memsize | awk '{print $1/1073741824 " GB"}'

You should see e.g. 32 GB. Pick from this table:

• My memory: GB → model:

Step 2 · Install Ollama (⏱ 0:05–0:15)

Download Ollama for macOS from https://ollama.com/download, open the file and drag Ollama into Applications. Open it once from Applications, then in Terminal:

ollama --version

You should see ollama version is 0.40.x (or newer).

• Ollama installed

Step 3 · Pull your model (⏱ 0:15–0:30)

ollama pull qwen3:14b
ollama list

(Use the model from your Step 1 choice.) While it downloads, put your task files in a folder:

mkdir -p ~/llm-test && open ~/llm-test

Drag t1.txt–t4.txt into that folder.

• Model listed, tasks in ~/llm-test

Step 4 · Speed check (⏱ 0:30–0:35)

ollama run qwen3:14b --verbose "Explain VRAM in 2 sentences"
ollama ps

You should see eval rate: ... tokens/s and 100% GPU in ollama ps (on a Mac, GPU means Apple's Metal).

• Mac speed: tokens/s

Step 5 · Rerun the 4 tasks (⏱ 0:35–0:50)

This is the Mac (zsh) version of the 4.1 loop. Paste it all:

cd ~/llm-test
M=qwen3:14b
for t in 1 2 3 4; do
  echo "Running task $t..."
  start=$(date +%s)
  ollama run "$M" "$(cat t$t.txt)" > "mac-${M//:/_}-t$t.md"
  echo "  done in $(( $(date +%s) - start )) s"
done
echo "ALL DONE"

You should see 4 "done in ... s" lines, then ALL DONE.

• 4 answers saved

Step 6 · Compare (⏱ 0:50–1:00)

Write: "On the Mac I'll use for ; heavy jobs go to ."

• Table and decision done

✅ Pass check

• Speed recorded for both machines side by side
• A decision on which model to use on the Mac
• Told Skills Coach so 4.6 (and Module 4!) gets marked done 🎉

Troubleshooting