Wildmimic

Hardware is a 5070 12GB VRAM, an AMD 5950x and 64G of DDR4 running on Nobara (Fedora spinoff).

I use 3 models for most things: Ternary Bonsai 27b or and Gemma 4 E4B Uncensored Q8_0 for everyday stuff, and Qwen Agentworld 35B A3B Q8_0 with CPU offloading for tasks that require to keep the model on track for long periods of time.

Everyday performance is best with Bonsai for me - it comfortably fits in my VRAM and is noticeably better with staying on task than similarily sized models. But if i want it to work overnight on a task without issue, the lower speed of the agentic Qwen is outweighed by its capability to do this.

They are allowed tool calls for websearch and reading webpages and have a sandboxed directory for R/W operations, but anything regarding command line and R/W access in other areas has to be manually authorized by me (i’m not willing to play with fire here).

They are currently not reachable from outside, but i am currently building a smaller machine as a home server, which will then be using a 3070Ti and 32G RAM to provide probably the Bonsai Model for me on the go per VPN.

The smaller models are here for quickly creating documents i do not care to do myself, RAG for technical documentation, creating quick python scripts for automating small tasks and teaching me RegEx and Python(mostly by analyzing the LLM output and playing around with it) and also some RP as NPC in a fantasy setting. The agentic model is currently working on cataloguing my games collection (around 36000 titles) by looking up games, their genres and descriptions and filling in blanks in the database. (Funnily, this task was refused by any model freely available because of the size of the database. Even the local models moan and bitch that it takes so long. No shit, this is why i outsource this!)

Might wanna try a ternary model, which are a lot denser in comparison.

Ternary Bonsai 27b is 7.17GB in size, which might fit on your 3070 depending on how much free VRAM you have (and with CPU offloading it should work in all cases). You can also try their edge-computing model designed to be working on smart phones here - that one is less than 4GB in size, which should fit in any case. In my experience they do work very well compared to other models in that size class.

Be aware that they need a specific llama.cpp build found here

Eh, people have to start somewhere to get into the material. My first contact with anime was Akira, and i did not understand a ton of it before watching it repeatedly, but it fascinated me. Evangelion could have taken that spot easily as a first - the themes are pretty accessible for people with some general media compentency.

Drivers of older BMWs and Audis have had a bad reputation for being reckless drivers in my country for at least the last 30 years. Volvo, that’s a new one for me.

The original studies authors were very selective about whp got their hands on that data, and

About 20 years ago, on April 20, 2006, one of the authors of the forthcoming replication, Kyle Hyndman, received the original data files in an email sent from ariely@mit.edu [1]. And about 3 years ago, on August 9, 2023, […] we received an out-of-the-blue email from Hyndman in which he sent those files to us. We performed quick analyses of the data, and then had a conversation with Hyndman and his co-author, Alberto Bisin. In that conversation, they told us they were going to conduct a replication […] upon reading Footnote 14 of their paper, we also learned this: … In October 2024, at the request of the editors, we shared with Dan Ariely an analysis of the contents from the file purportedly for their Study 2 and asked for permission to include a summary of it in the paper. Dan Ariely denied our request, arguing, among other things, that the files we received may not be the actual data. He did not subsequently provide us with any additional data from the original paper. Consequently, we are unable to supplement our replication exercise with any additional analysis of the files we received in 2006 or any other data.

Additionally, Dan Ariely (the main contributor to the studies in question) opted their website out of the Wayback Machine, and if Louis Rossman ever teached me something, it’s that people who do that very try to retcon the past.

Looking at what they found, it seems that the study authors themselves suffered from procrastination that much that they randomly created data that that fit their hypothesis about 10 minutes before the paper was due. They really phoned that in - if i actually attempt fraud, i would at least try to make the data look somewhat reasonable. Non-correlating values for liking a task and finding a task interesting, or noncorrelating performance in 3 nearly identical task is not even trying to be compelling lol

can’t argue about taste ¯\_(ツ)_/¯

my cat currently loves hitting the “Sleep"-button on my keyboard every time he jumps on my table -.- at least i can wake it again, but it IS annoying. Haven’t had the time yet to reconfigure that button to instead run a soundfile of me scolding my cat lol.

And this is the way bioinvasor species spread - most of the time they don’t come delivered inside of PSUs but simple crates, which improves their survivability lol

I have no formal training except 1 single year of a trade school where i learnt a bit of pascal and c++. But i have been in front of computers since my earliest childhood starting with ataris in the early 80, C64, Amiga 500, and then starting from 286 across every major consumer x86/x64 CPU architecture. All I know except logic and programming basics are self taught, i never worked in the field outside of giving tech support for ISP clients a.k.a help grandma configure her mailaccount or help someone reinstall their modem driver.

The slot that is currently in use is the slower one since it’s only Type: x1. Depending on what you use the slot currently (such as an M.2 drive), you might wanna switch slots so you can take advantage of the M.2 speed; the x1 slot is completely adequate for 1gbps ethernet if you don’t need more. Both slots are otherwise functionally the same, but make sure to look up in your mainboard documentation if there’s any limitation if you use both slots or a specific hardware configuration (mine deactivated the second M.2 slot after changing from an 2600x to an 5950x for example)

https://metager.org does the trick

That might introduce more issues than help. If high speed particles impact your shielding, you might get a “particle shower” from the impact on your electronics. Radiation Hardening is part of the design of the chips - mainly creating less dense structures with bigger transistors, because they don’t flip as easily as the very small gates on a H200. That’s also the reason why most space based computers have the processing power of a system around 2005.

Wanted to say that - Random shit does happen, even to the most stable systems. There’s a cutoff in consumer hardware where selecting for more stability simply isn’t worth the cost such as radiation hardening. Best you can do is ECC Ram.