I wonder if it really needs to be worse. I am playing with the idea of fine tuning a model on my exact stack and coding patterns. I suspect I could get better performance by training “taste” into a model rather than breadth.

You burn your lap?

Currently I'm testing something like this just to see what happens. I have an old laptop with 4GB of RAM. I attached a USB drive with Gemma 4 31B model (which is 32.6 GB). Currently the laptop is running llama.cpp and trying to respond to a prompt by streaming the model from disk.

The USB drive light is flickering, showing something is happening. It's been about 8 hours since I entered the prompt and I've gotten about 10 tokens back so far. I'm going to leave it running overnight and see what happens.

Nothing special?

I mean, inference engine might need to get some tweaks, to support whatever compute is available. But then, if you put a few terabytes of disk for swap, and replace RAM to bigger sticks if possible, it should work? Slowly, of course, but there is no reason it should not to.

You can if you have enough ram slots?

Not sure if this is exactly the scenario you envision but I run ComfyUI on an Acer Helio 300 laptop, from four years ago. Has 16GB RAM, NVIDIA GeForce RTX 2060 w/6144MiB of VRAM and have generated a few images using "NetaYumev35_pretrained_all_in_one.safetensors" @ 10.6GB checkpoint, (well beyond the 6GB capacity of the RTX 2060 card). That being said, it takes more than 10 minutes to complete the task. Of course, I have to turn off all other apps, and browser tabs or hibernate them. If I don't, the laptop's fans begin to spin up like an airplane propeller. It's worth mentioning that I've tried to do this with other IDEs and all seem to fail with some error or another, usually out of VRAM issue. I've only gotten it to work with ComfyUI.

I use an anaconda environment, though would have preferred an "uv" environment, on Linux and automate the startup sequence using the following script (start_comfy.sh) from the term rather than manually starting the environment from same said term:

#!/bin/bash

#

# temporary shell version

eval "$(conda shell.bash hook)"

conda activate comfy-env

comfy launch -- --lowvram --cpu-vae

Here are some of the images: https://imgbox.com/nqjYhdx3 https://imgbox.com/93vSWFic https://imgbox.com/qs1898dz

I'm hesitant to increase the sizes of the renders as that will surely stress my laptop's components.

I'm running opencode with qwen3.6-35b-a3b at a 3-bit quant. I also have qwen3.5-0.8b used for context compaction. I run with 128k context.

It's usable. I set it loose on the postgres codebase, told it to find or build a performance benchmark for the bloom filter index and then identify a performance improvement. It took a long time (overnight), but eventually presented an alternate hashing algorithm with experimental data on false positive rate, insertion speed and lookup speed. There wasn't a clear winner, but it was a reasonable find with rigorous data.

I'm playing with a tape drive for backups, so I asked a local model to rewrite LTFS ( https://github.com/LinearTapeFileSystem/ltfs ) in Go.

I gave it the reference C implementation, the LTFS spec from SNIA, and asked it to use the C implementation to verify the correctness of the Go code.

LTFS is a pretty straightforward spec, so it made a very reasonable port within about 2 days. It's now working on implementing the iSCSI initiator (client) to speak with my tape drive directly, without involving the kernel.

Edit: the model is Qwen3.6-35B

Hey man, you can just say "I'm lazy, so I'm staying with the cloud. if I wanted to use my brain, I wouldn't be using AI, gosh" - it's much shorter.