digestVidSign in
summary

Teacher's Tech - Google Gemma 4 Tutorial - Run AI Locally for Free

Thumbnail of the video “Teacher's Tech - Google Gemma 4 Tutorial - Run AI Locally for Free”
Community upload. AI output has not been independently verified.
Author: Teacher's Tech
Categories: Technology Education
Tags: technology, education
Views: 2 · Last viewed: 2026-10-15 07:48 UTC

Short Summary

Main recommendation: Try Google's Gemma 4 locally with Ollama to run a free AI model entirely on your own computer, with no subscription, API key, or data sent to the cloud, and start with the E4B model for most users. Before installing, the video suggests testing Gemma 4 in Google AI Studio in your browser with no download needed. The video also covers model sizes, image understanding, and sample prompts to test the model.

Key Points

  • Gemma 4 is a family of Google AI models designed to run locally on hardware such as laptops, and even a Raspberry Pi, rather than on a cloud server.
  • Because it runs locally, your questions and data stay on your machine, which the speaker presents as a major privacy benefit.
  • The model is free: you download it once and use it without subscriptions, API keys, or usage limits.
  • Gemma 4 can read images, such as photos, screenshots, charts, and receipts, and the smaller E2B and E4B models can also process audio.
  • In testing, the speaker found the model strong at explanations, writing, image reading, and generating working HTML/JavaScript, but it struggled with a constrained optimization math problem.

Details Worth Noting

  • Concrete instructions:
    • Try it first without installing: go to aistudio.google.com, sign in with a Google account, open the model selector in the side panel, and choose a Gemma model (the speaker selected the 26B option).
    • Install Ollama: go to ollama.com and download the version for your system. On Windows, run the .exe installer and follow the prompts. On Mac, unzip the download and drag Ollama into Applications. On Linux, install with a single terminal command.
    • Download Gemma 4 in Ollama: open the model selector, search for "Gemma 4," and click download. Alternatively, open Command Prompt and run ollama pull gemma4 to get the default model (about 9.6 GB).
    • For a specific size, use ollama pull gemma4:31b (or the E2B or 26B variant names) instead of the default.
    • Run the model from Command Prompt by launching Gemma 4, entering prompts, and typing /bye to exit.
    • Model choice: the E2B models can run on about 5 GB of RAM; the 26B model is suggested for desktops with 16 to 20 GB of RAM; the 31B model needs about 20 GB of RAM or a dedicated GPU; the speaker recommends starting with E4B for most modern computers.
  • The 26B model uses a mixture-of-experts design, meaning it is large but activates only a small portion at a time.
  • Speed depends on hardware: a GPU makes responses faster, but the model still runs on a CPU, only more slowly.
  • The speaker's test machine was a Lenovo Legion T7 with 32 GB RAM and an RTX 4080 GPU, and said the default model runs fine on most computers with 8 GB RAM.
  • The optimization test (450 students, buses and vans, no empty seats) showed the model reasoning about cost but reporting the no-empty-seats condition as impossible, while the speaker noted a nine-bus, nine-van solution was possible and found that Gemini Pro also produced that answer.

Bottom Line

Gemma 4 is a practical free option for private, offline AI on your own hardware, and the video's advice is to test it in the browser first, then install Ollama and start with the E4B model.