Model Architectures · Practitioner
Gemma: Google's Open Models and When Self-Hosting Actually Wins
Gemma is the open-weight family built from the same research as Gemini. Learn where it runs on Google Cloud, and the two conditions under which hosting your own model beats a managed API call.
- 1.2 hours lesson
- intermediate level
- Google Cloud AI Track
- Narrated presentation
What you’ll learn
- Gemma as open-weight models from the same research lineage as Gemini
- Variants: instruction-tuned text, multimodal, CodeGemma, PaliGemma, ShieldGemma, EmbeddingGemma
- Deployment options: Model Garden endpoint, Cloud Run with GPU (scale to zero), GKE, local
- Fixed cost vs variable cost, and where the real crossover point sits
- The two genuine self-hosting cases: perimeter isolation and sustained high volume
- Quantisation as an accelerator-tier reduction with modest accuracy loss
- Versioning base model, tuning data and adapter together
Inside this lesson
A narrated, slide-by-slide presentation with 1 prediction exercise, 1 knowledge check, 3 videos. You answer as you go, and the platform checks your work.