yano

I remember letting my computer run protein/gene folding back in 2010-2016... free spare cycles...

Sorry, probably slightly off topic.

Interesting and curious.

In my limited experience, asking for flaws in something i would only get surface level identification of flaws of why something wouldn't work.

This potentially may improve that a lot. Yeah wanting to build tower out of gold is a silly example, but i see a lot of movies and books where events happen 'so the plot can move forward', and making idiotic choices is one thing that has turned me off from a lot of media including books; Writers using something to identify and remove the braindead stupidities would certainly be a huge improvement in the long run.

for a logo... top on, 12th over (or 3rd from the right), the depth steam-punk like 3D look seems to stand out for me the most.

Otherwise any of them will work, but a lot of them look rather generic.

A number of models i wish i could get it to shut up on the thinking, in fact 'thinking' modes tends to turn me off on what would otherwise be good models.

Thinking generally is a summarizing of all the details, sometimes giving better insight on what has more emphasis, but usually the 'thinking' i see is useless, as half the output beyond being a summary it talks about staying within guidelines and not offending anyone. 99% of the time 'thinking' is just a waste of time.

Then again i'm not using the cloud versions so maybe those actually are much better at it.

Actually one thing 'thinking' might do for very very large models, is to gather appropriate data all in one block right before it answers. Thus it might use 2 modes, a very large context mode, and then a shorter faster one with the more recent context. That in theory may work...

Saw a piece a couple months back of a lawsuit against Facebook for their released model, which consistently defamed and gave outright lies regarding a public figure or two.

Though i've come to the conclusion anything the LLM gives you information about the real world needs to be checked and taken with a cup of salt (but i doubt everyone else has come to that conclusion too yet. People assume the AI output is factual.)

Well this is fun. Found a tutorial on using ComfyUI to get Qwen Image working and... i totally flubbed it. Wrong template trying to be used, files put in the wrong locations, missing files, etc.

Downsides of being self taught.

Anyways, i'll try to get this working later today and generate a few images correctly this time.

edit: and the universe hates me. No Qwen images for me.

Hmmm... curious. I've mentioned it elsewhere of an experimental compression that could make data get to 1.33-1.66 for effectively 2bit data. But the quantization process would have to decide which bits/values may need adjusting/flubbing to minimize/control the process. (decompressed it would be treated as some form of Q2)

Regardless, I've seen good results with Q4_K_M, decent results in Q3_M, and even a 235B model at Q2 which performed basically on par with a 70B model. I'd like to try a large 200B model at Q1 to see how it does, but we'll see. I tend to also get i1-Q6_K models i consider control/max quality rather than Q8_0.