nunodonato

what would be the viability of a model changing its weights and using genetic algorithms to try to achieve something novel, even without understanding exactly which weights need changing?

This is so nuts and pointless... I love it!

The kind of thing no AI would ever come up with :)

that's why local AI is so important, and open-weight models

maybe the significant difference is that this time robotics will not be limited by humans. AI can iterate faster and develop better, adaptable, robotics

My blog/website, my tiny code experiments, my ai agent.

isn't 27b too big (slow) for a decision model? I think they should be as fast as possible

I don't touch AWS. We rent from an european provider. Demand can be tricky for serverless/spot-pricing. We have an always-on server so, no issues.

we use that same model at our company, it powers not only our devs but also many business needs. We rent 1 gpu (B300), it costs around 10x less than the api costs

Last time I wrote about my productivity system was back in 2022. Today I felt like writing an update on how I’m managing it, because a lot of things have changed. Funny enough, that last article was published on the 18th of September 2022. Some sort of strange 4-year cycle going on!

Loved netscape navigator! That entire suite with mail, website builder (composer?)

Mosaic :P

yay! /s

Why are people still putting up with this kind of attitude, especially when there are so many good alternatives available?

ahh I missed that first table. thanks, fixed it :)

hey thanks a lot for the feedback!

1. works fine for me, are you sure you don't have any other filters active that might result in 0 models?

2. good idea, a few people have requested that. It becomes a little bit more difficult when models have engrams, but I will consider!

3. There are some tools to do that, I personally don't like them. I understand the convenience but I am staying away from that. There are many factors at play, not all models work the same way, even if they use the same VRAM. Not to speak of using quants and how each quant may affect a model differently

Hi folks! I love running local small models, but it's quite annoying to attempt to compare them using benchmarks. Not only are they scattered around, they often don't even share the same ones.

So I built Tiny League for those of use who care about small models. By small, I'm aiming at 3-200B, which is a range that I consider both capable and able to run at home (for those who have unified memory like mac, strix or spark)

I'm limiting the model list to models released after April and excluding models that are trained for specific use-cases.

Appreciate all sorts of feedback :)

that's genetic modification for you. nature doesn't make seedless fruits

yes, please. Especially now that xcancel is gone

climbing stairs was my first question. But looking at its size, I dont think it will be able to

Kimi K3 is amazing

wow, what kind of stuff does that script do? I've seen non-vision models analyze images, but mostly histograms, color averages etc. This one seems to actually understand the image itself and reproduce the layout, very impressive

this is really cool!

nice of you to think that by evening you get cooler air :D we have had days where the only time of the day where the outside temp was lower than indoor's was between 01-07. So either you stay up to open the windows, or wait till early morning... but then you just get 1-2 hrs before having to close everything again

yeah, I've been dying to do something with ESP32, but still waiting for that idea. Something with eInk displays :D

I love her gigglyness :D

unsloth is always here to save the day https://huggingface.co/unsloth/Laguna-S-2.1-GGUF

are these bad actors here in the room with us?