the 6B actives should easily fit into an 8GB card
That’s not how MoE models work. There are many “expert’ models and there is a static router model (dense) which determines which “expert” models to route the tokens through. What you want at a minimum is the dense portion of the model to be on VRAM and all of the weights to be in RAM/VRAM for best performance.
Would having a duress password at all be what they need to prove intent? Did he go out of his way ahead of time to set it up or is it part of the setup process of the OS?
Video doesn’t show what lead up to it, so how are you basing that opinion?
Fuck Elon and all that, but don’t let your hatred blind you to the tech and the problem of autonomous vehicles. Isn’t this the technology community?
I also didn’t say it wasn’t the cars fault. I said we need to see what lead up to this because it’s really hard to tell what the actual issue is.
Really a stretch to call that a crash. How did it get in there, that’s what I’d like to know. Poor road design and controls is likely the culprit here and I’d be surprised if human drivers haven’t done it. Hard to say not seeing the lead up to the little trap area.
Blow the dust out of the case. Redo the heat sink grease. Same for the GPU. Still does it, try a different power supply. Also check the RAM with a bootable mem86+ tool.
That’s why I didn’t have any issues. Added the entry to my llama-server model ini and it Just worked. Great job and much appreciate your work!
If you customize nanorc then you can also point and click with the mouse and highlight text and copy/paste like it was a GUI from the terminal. Most people shiting on Nano just never learned it because they learned the editor they are familiar with. Same reason I shit on vim.
I had the exact same experience and reaction. It would have been better if I could just skip all the BS interactions but that wasn’t an option. Or if they did more with written things strewn about or other message type objects (stack of bones or something).
Other than em dashes (and not used how LLMs typically use them), what makes you so sure this wasn’t written by a human?
What are you wanting to do with the model?
Agentic development with that setup, you can easily run a good quant of Qwen3.6-27B at full context. unsloth/Qwen3.6-27B-MTP Q5_K_XL and use a fixed template
Creative writing? Probably need a different model. I’ve heard good things about Gemma 4 31B.
Both of those are dense models so won’t be as fast as Qwen3.6-35B-A3B. Also play around with MTP, quants, KV cache quant, KV caching, etc.
The RAM size will limit you on larger models unless you stream from storage.
I didn’t saw it was dangerous only that it was similar to MS GDID, and it is.
Wow that went dark quickly. Not sure any of that was warranted
I’m not technically inclined enough to harden against that, nor do ai really have the inclination. But here is a thread on that topic where I even heard about it.
See discussion below. I’m making it up obviously
That’s fair. I didn’t mean MS was the threat actor, just the seed from which this grew. I definitely take that back after reading up more on where machine-id grew out of.
However, I don’t concede that machine-id is theoretical for user fingerprinting. Any software installed can read it, especially a browser. That’s the vector for tracking across the web. Do I have instances of this occurring, no. Seems plausible and the more we can sandbox things, especially “web browsers” (untrusted app runners more like it) the better.
If you started a project, wouldn’t you list your server up top? The fact they changed it should tell you all you need to know. Also it’s their instance, they can do with it what they want. That’s like saying someone painted their house a color that is questionable. Block it and move on.
Sorry I just don’t think it’s right to pass judgement on people for doing what they want with their project or instance especially when they designed it in a way that anyone can set up their own instance do do what THEY want with their own.
Lemmy.ml is like 8th in the list on join-Lemmy.org. Doesn’t seem to be pushing?
So this was the Ox-Alpha model on OpenRouter. That’s cool.