Gemma 2 2B

by Googleopen weights

Gemma family

Parameters

Context window

8K tokens

Released

2024-07-31

Input price / 1M tok

—

Output price / 1M tok

—

Modality

text

About Gemma 2 2B

Google's tiny open model, optimised for on-device inference (phone, browser, laptop). Beats GPT-3.5 on benchmarks at 2B parameters. Runs at 30+ tok/s on modern laptops via llama.cpp or ONNX.

Strengths

Runs on-device
Beats GPT-3.5 despite tiny size
Cheap to fine-tune
Permissive licence

Weaknesses

Not for hard reasoning
8K context
No vision
No tool use

Best for

On-device chatEdge classificationMobile apps

Vendor page →HuggingFace →