Models
Gemma
Google's lightweight open-weight models. Gemma is a family of open-weight models from Google, built from the same research as Gemini but sized to run on a single GPU, a laptop or a phone. There are general instruction-tuned models and specialised variants. Teams use Gemma where a small, cheap and private model is enough, such as classification, extraction and on-device features.
When it is the right choice
- You need a small model that runs on limited hardware or at the edge
- Your task is narrow enough that a compact model, possibly fine-tuned, does it well
- You want low per-request cost for high-volume automation
- Not the right choice for complex multi-step reasoning, where a larger model does noticeably better
How we use it
- Run Gemma with Ollama or vLLM for internal tools and automation pipelines
- Fine-tune small Gemma models on your labelled data for classification and extraction
- Compare against larger models on your evaluation set to justify the smaller footprint
- Review the Gemma terms of use as part of the design
Ollama
vLLM
Hugging Face
GeminiFrom the blog
From Idea to AI Feature in Six WeeksHow an AI project runs end to end in six weeks: discovery, design, prototype, infrastructure, rollout, then measuring and improving.
CI/CD for Models and PromptsEvaluation gates, model registries, canary releases and rollback: applying CI/CD discipline to prompts and models in LLM applications.
Fine-Tuning or Prompting? A Decision Guide for Small TeamsWhen prompting, RAG or fine-tuning is the right way to adapt a language model, and what each one costs a small engineering team.
Related articles
From Idea to AI Feature in Six WeeksHow an AI project runs end to end in six weeks: discovery, design, prototype, infrastructure, rollout, then measuring and improving.CI/CD for Models and PromptsEvaluation gates, model registries, canary releases and rollback: applying CI/CD discipline to prompts and models in LLM applications.Fine-Tuning or Prompting? A Decision Guide for Small TeamsWhen prompting, RAG or fine-tuning is the right way to adapt a language model, and what each one costs a small engineering team.Questions
Gemma: frequently asked
How is Gemma different from Gemini?
Gemini is Google's hosted model family that you reach through an API. Gemma models are smaller, and you download and run the weights yourself.
Can Gemma run without a GPU?
The smallest Gemma models can run on CPU, especially when quantised, though slowly. For production traffic a GPU is usually worth it.
Running Gemma, or deciding whether to? Tell us what you have; the first consultation is free.
Let's talk »
