qskousen

I have a pet project I have been working away on for some time that involves building GPU backends for various cards in Zig, lots of complex stuff in it. Lately I mostly use Opus 5, it can pretty reliably plug away at things but it does mess stuff up occasionally. For this codebase, Fable 5.1 was noticeably better at getting things right and doing things in a good reliable way. Of course, I can only use Fable for a bit before I hit the usage cap for the week, so I save it for the tougher things. That said, I absolutely abhor the way recent Anthropic models write prose, especially comments.

I recently tried doing a fairly normal task for this codebase with codex, as I have seen a lot of people talking it up on here. A single task running for ~1-2 hours burned through over half of my usage for the week on the $125/month plan, not on a top model (I don't remember which one specifically I used). It struggled to get the basics done, then got absolutely stuck on a follow up. Handed it over to Claude and it 1-shot it.

I've had several bun-related crashes in Claude code in the last week, which I don't remember ever seeing before then.

it does have telemetry, enabled by default, that sends metrics to tracking.miui.com, including what model you are using. it can be turned off by environment variable (MIMOCODE_ENABLE_ANALYSIS=false), and yes it still has all the normal OpenCode provider logic so it will work with other/local models. it also automatically looks for updates and fetches a mimo model list, including when the telemetry is off, though those can also be disabled.

telemetry enabled by default and named "analysis" is not great.

Not sure if it was a typo or not but "colleagues" is a great word, I'm going to start using it.

Sorry this is a bit of a tangent, but I noticed you also released UD quants of ERNIE-Image the same day it released, which as I understand requires generating a bunch of images. I've been working to do something similar with my CLI program ggufy, and was curious of you had any info you could share on the kind of compute you put into that, and if you generate full images or look at latents?

My father just closed his local small shop, as it was no longer paying the bills. It's harder and harder to compete with the internet.

I've experimented with this with diffusion models with a safetensors - gguf tool I wrote. even with relatively few sample images (~10k, still enough to keep my 3090 spinning for days straight) the benefits are quite noticeable - a smaller file with overall better results.

Wendell at level 1 techs built one for Steve of gamers nexus! It was interesting. I was going to link the video but couldn't find it, sorry.

I've done it with a 6800XT, which should be similar. It's a little trickier than with an Nvidia card (because everything is designed for CUDA) but doable.

It seems like you are saying the AI features don't work if you don't have a GPU, if I understood correctly, but I have my install on a server with no GPU and the object search and facial recognition features work fine. Probably slower to generate the embeddings, but I don't have any comparison to make.

My number one reason for moving away from using LXD in production after this change is that LXD is only available through snap, which caused multiple downtimes in the cluster because of the forced updates.

It does - the inference speed is much slower than a consumer video card. The draw for the Spark and systems like it are the massive amounts of memory available to the GPU.