Gracana

9 ms·

151.
by Gracana 1y ago
AI-haters. It's an entire identity.
152.
by Gracana 1y ago
Yeah, the whole "let's come up with a slur for <blank>" thing entices people to build their fictional racism on real racism, and it just devolves from there. I saw "wirebacks" thrown around recently, among oth
153.
by Gracana 1y ago
https://www.servethehome.com/maxsun-intel-arc-pro-b60-dual-g... It's two B60 GPUs on a single x16 PCIe card. Nothing unified.
154.
by Gracana 1y ago
You might be able to do "large models slow" better than the spark with a 5090 and CPU offload, so long as you stick with MoE architectures. With the kv cache and shared parts of the model on GPU and all of the experts on CPU, it c
155.
by Gracana 1y ago
CPUs with stock cooling solutions will turbo boost up to max temp and stay there, that's completely normal and shouldn't cause a CPU to physically burn up, even if you do it for months.
156.
by Gracana 1y ago
Delaminating the PCB is absolutely wild. I'm really fascinated by the idea of doing meetings in Second Life, though. What did people use for avatars at work?
157.
by Gracana 1y ago
> I don't know how anyone can look at the $1B judgement and not decide There's more to the case than that number. > $10M of "lied about the kids" Ridiculous. I'm not going to waste any more of my time replying
158.
by Gracana 1y ago
Alex Jones made his money by convincing people that the parents of children murdered at Sandy Hook Elementary were actors in a conspiracy, which resulted in them bein harassed endlessly. He even continued to attack them during the defamatio
159.
by Gracana 1y ago
I looked at mine, and it only has entries from my current login session.
160.
by Gracana 1y ago
CelGenStudios and Usagi Electric are good channels for vintage computing stuff.
161.
by Gracana 1y ago
Great tip! Apparently that's readline's config file, so this will affect a lot of things. That's great news for me; after switching to zsh I got used to case-insensitive tab completion, and now it annoys me when other tools d
162.
by Gracana 1y ago
Yes, it frequently works "too well." Few-shot with good variance can help, but it's still a bit like a wish granted by the monkey's paw.
163.
by Gracana 1y ago
Example-based prompting is a good way to get specific behaviors. Write a system prompt that describes the behavior you want, write a round or two of assistant/user interaction, and then feed it all to the LLM. Now in its context it has
164.
by Gracana 1y ago
Seems simple enough, all the base models do is continuation, but you can do few-shot prompting to establish a pattern if you want. I will have to give it a try.
165.
by Gracana 1y ago
It seems somewhat popular for developers who want to avoid github. Gnome and KiCAD also use it.
166.
by Gracana 1y ago
I mean in terms of prompting. What methods do you employ to get useful results out of a model that is not tuned for particular form of response?
167.
by Gracana 1y ago
How do you use the base model?
168.
by Gracana 1y ago
Have you used any thinking models? I remember being surprised by QwQ-32B when I tried it. It would think about what I said and how it should respond, reiterate the behaviors I had assigned to it, and respond accordingly. That constant self-
169.
by Gracana 1y ago
Seems about right. $17/ea for a qty 10 order from JLCPCB. Rises to $25/ea if you do the minimum qty 5 order.
170.
by Gracana 1y ago
The context cache (or KV cache) is where intermediate results are stored. One for each output token. Its size depends on the model architecture and dimensions. KV cache size = 2 * batch_size * context_len * num_key_value_heads * head_dim *
171.
by Gracana 1y ago
> no high quantisation library supports two GPUs for example, and there seems to be no interest in implementating it by the guys behind the best inference tools I'm curious to hear what you're trying to run, because I haven&#x2
172.
by Gracana 1y ago
HF does this for ggufs, and it’ll show you what quantizations will work on the GPU(s) you’ve selected. Hopefully that feature gets expanded to support more model types.
173.
by Gracana 1y ago
That will always be valuable, but I don’t think that’s what most designers are doing. If AI can copy flat design and Corporate Memphis style, it’ll compete just fine with the average designer.
174.
by Gracana 1y ago
Have you tried abliterated models? I'm curious if the current de-censorship methods are effective in that area / at that level.
175.
by Gracana 1y ago
You can go look at one up close at the Udvar-Hazy Center in VA. I highly recommend a visit there if you are in the area and interested in aerospace stuff. They've got a ton of amazing exhibits.
176.
by Gracana 1y ago
I got the feeling that it was dead long ago. Sure there's plenty of readers and lots of sites "support" it, but what they tend to give you is a headline and article blurb feed with links to click through to see the ads. They
177.
by Gracana 1y ago
I was on the Microchip bandwagon until the PIC32. Little MIPS MCUs with onboard RAM and graphics, awesome! ...except they came with an errata that listed huge problems with every interesting peripheral, and it didn't improve for ages.
178.
by Gracana 1y ago
Oh, sure, I’m not proposing that the connector itself should have those features, rather that it shouldn’t be used without them present on the device.
179.
by Gracana 1y ago
Previous well-designed video cards used the technologies I described. Eliminating the sense circuits and fusing is a recent development. I do like the idea of just using big wires. It’d be so much cleaner and simpler. Also using 24 or 48V w
180.
by Gracana 1y ago
It still has no fusing, sensing, or load balancing for the individual wires. It is a fire waiting to happen.

More ›