I tried out the Gemma AI models from Google, running locally on my AMD APU (Ryzen 7 Pro 7840U with Radeon 780M) and asked it some questions about ZFS send / receive.

[larvitz@burningboard.net]

I tried out the Gemma AI models from Google, running locally on my AMD APU (Ryzen 7 Pro 7840U with Radeon 780M) and asked it some questions about ZFS send / receive.

gemma-4-26B-A4B-Q4_K_M:
14.29 tok/sec . The information, it generated, was factually correct and well laid out. Not the fasted, but surprisingly good.

gemma-4-E4B-Q4-K_M:
26 tok/sec. The information was completely wrong BS with made up parameters. The presentation was confident and well laid out. But it generated it quickly
Bottom line: confidently incorrect at high speeds.

#ai #gemma #lmstudio #llm #generativeAI #amd #radeon

[tux@burningboard.net]

@Larvitz

You can now run this on your phone too:

https://the-decoder.com/googles-gemma-4-puts-free-agentic-ai-on-your-phone-and-no-data-ever-leaves-the-device

[tux@burningboard.net]

@Larvitz

I'm currently testing whether I can build a simple Mastodon CLI tool – with Gemma4 for spell checking.