← sources

Downloaded 217GB of glm- 5.3 onto a 128GB Mac and it technically ran

source: reddit (reddit:ollama)published 2026-10-27 13:42ingested 2026-10-28 02:38

retained text

I do the books for a few dozen small businesses, and the weekend after the October 15 extension deadline was the first one in about a month where I did not open a client file. I do not care much about privacy for my own stuff, but client bank statements cannot go anywhere near a cloud model, which is why I bought a 128GB Mac Studio back in April, right after tax season. I remember thinking 128 was overkill.

I run everything through Ollama so I looked there first. There is one tag and it is glm-5.3:cloud. Not what I was after.

So I found the smallest GGUF on Hugging Face, the 1-bit one at 217GB, and let it download overnight. It ate most of what was left on the drive. In the morning I pointed llama.cpp at it and it would not load until I turned the GPU layers off. After that it was reading the model off the SSD the whole time, so I asked it to summarize a one page letter from the IRS and went and made eggs. It was still going when I came back and I killed it around lunch.

I put a smaller model on the statements instead and start it before bed. It feels like the cloud models did a couple of years ago, good enough for sorting transactions. I use the cloud one for the scripts, the ones that split the PDFs and hand the lines to the local model. It rewrote my PDF splitter, the one I had been patching for two years, and it worked first try, which was a bit annoying. I checked first that Ollama hosts the model in the US and Europe, but it is still not on my desk, and that was the reason I bought the Mac.

Twice now I have caught myself about to paste a real line from a statement into it to show it the format. Both times around eleven at night.

The GGUF is still on the drive. I keep meaning to delete it.

document fetched 2026-10-28 00:41 · canonical https://reddit.com/r/ollama/comments/1xgsvea/downloaded_217gb_of_glm_53_onto_a_128gb_mac_and