Code & Cloud Chronicle

Google Gemma 4 OSS edge + Scion testbed

Google Gemma 4 OSS edge + Scion testbed

Key Questions

What is Gemma 4's edge deployment capability?

Gemma 4 offers INT4 quantized models on Hugging Face via IntelAI, with 2B/4B sizes for edge devices like offline phones performing agentic tasks.

Can Gemma 4 run offline on phones?

Yes, Gemma 4 runs on phones without internet, enabling local agentic tasks; Google AI Edge Eloquent uses it for offline dictation.

What is Google's Scion testbed?

Scion is an open-source experimental multi-agent orchestration testbed for managing concurrent agents in Kubernetes or local compute environments.

What is VoxCPM 2?

VoxCPM 2 is an open-source TTS model from China, live on platforms like Hugging Face, advancing voice capabilities alongside Gemma developments.

How is Gemma 4 sourced from Gemini?

Gemma 4 are open models derived from Gemini 3, launched under Apache 2.0 with playgrounds and mobile tools for deployment.

What datasets support open-source agents?

Calls for building datasets for frontier open-source agents, alongside tools like HF traces, emphasize community efforts for agent development.

Are Meta's new AI models going open-source?

Reports indicate Meta plans to open-source versions of its next AI models under new leadership, following Llama precedents.

What tools run Gemma-like models?

Docker Model Runner enables running open-source AI models compatibly with OpenAI protocols; browser tools like TurboQuant-WASM support vector quantization.

Gemma4 INT4 quantized HF (IntelAI); offline phones agentic tasks incl. Eloquent dictation app (local processing/privacy); 2B/4B edge, Scion multi-agent K8s/local; VoxCPM2 TTS, HF traces.

Sources (20)
Updated Apr 8, 2026
What is Gemma 4's edge deployment capability? - Code & Cloud Chronicle | NBot | nbot.ai