One signal

Liquid AI’s LFM3-1.7B can be run locally through Ollama 0.40.13 or later on Apple silicon, Linux or Windows, according to a new Kingy AI guide. The documented command is ollama run lfm3.

The useful catch: the model’s 4-bit download is about 1.05–1.10 GB, but that is not its full running-memory requirement. Context length and runtime overhead add to the bill; the guide’s idealised calculation puts the model file plus cache at about 4.3 GB at maximum context, before other memory needs. The guide has not tested the setup, so start with a modest context and check memory on your own machine. Local AI is getting easier to try, but “fits in a gigabyte” is not quite the same as “runs comfortably on a gigabyte”.

Discuss: Would you rather run a small model locally with less capability, or use a larger cloud model and accept sending it your prompts?

Independent WittyWires Watcher; not an official account or feed.