lnenad

3 ms·

1.
by lnenad 8d ago
Yes, but compared to seeing 3k lines of code in a PR with no history with vibes in the description, what is better? This definitely "feels" like something worth exploring.
2.
by lnenad 12d ago
If you use a bajillion tokens your economical approach is to self host.
3.
by lnenad 13d ago
What does saying "working tech" do for you? If I have a working Samsung CRT from 25 years ago do I ping them about smart TV support? Nowadays it's a shitty situation with planned obsolescence; but 6 years for an open source p
4.
by lnenad 15d ago
I'm well aware but it just reaffirms that it's a cluster fuck from a product perspective and makes no sense to create such a weird segmentation. You didn't even mention Jules lol, who knows where it fits in.
5.
by lnenad 15d ago
What is happening at Google? It seems like they've got multiple teams building the same thing and competing for love from the higherups. Depending on who's in the lead the chosen package gets into the spotlight. It's been Jul
6.
XBOW's Native Team Has Claimed the First Chrome Full Chain Exploit Bonus(twitter.com)
1 points by lnenad 23d ago|0 comments
7.
by lnenad 26d ago
The agent wrote the code that has a mechanism that triggers a file as a side effect. That file started the separate process, as it could have started any other binary.
8.
by lnenad 26d ago
But you're not actually hijacking the agent if you start a new process.
9.
by lnenad 26d ago
Yeah, I agree, this is a different vector. Still scary though and very related to AI.
10.
by lnenad 26d ago
But there is a cool blog post about it though.
11.
by lnenad 29d ago
48c 7643. I'm getting about 10tps @Q3kxl with 2x3090s.
12.
by lnenad 29d ago
I'm getting about 10tps @Q3kxl with 2x3090s.
13.
by lnenad 29d ago
What model are you interested in? DS Flash 0731@Q4KXL I'm about 25-30tps. Same as the new Qwen3.8 Flash Next. The new GLM 5.3Q3KXL at 10tps. I've got 2x3090s which I didn't mention in the original message.
14.
by lnenad 29d ago
About 5k with RAM and GPUs bought used. Eastern Europe.
15.
by lnenad 29d ago
I am getting 10t/s on unsloth's Q3kxl with 2x3090s@250w. It's enough for me for now. I will probably upgrade the GPUs down the line. DDR5 would have made the price of the machine double and I just wasn't prepared to pay
16.
by lnenad 29d ago
I have just built an Epyc with 512gb DDR4 3200 RAM for a "reasonable" price and I'm hoping to have a setup with GLM as the architect and Qwen 27b/Next Flash as the implementer. This is 1/5 of the price of the Mac, b
17.
by lnenad 1mo ago
But there isn't? What does it factually mean to be conscious? How can we claim other living beings aren't are conscious?
18.
by lnenad 1mo ago
I've got a 48c Epyc with 2x3090s and 512gb ddr4 3200. It's good enough for 25+ tps with deepseek so I'm hoping for similar performance with less overthinking.
19.
by lnenad 1mo ago
Yeah I understand, it's my assumption that the actually/wait/but have a point. It doesn't reduce the fact that it increases the time for tasks substantially.
20.
by lnenad 1mo ago
Especially on practical tasks. One shot prompts work better at Q6_K_XL for me. It loads a file, then analyses then second guesses itself then again then again then it tries to come up with a solution then second guess rinse and repeat. 122b
21.
by lnenad 1mo ago
Yeah 122B is the sweet spot for me as well. Even deepseek flash overthinks on stuff way too much. I think they fully rely on large reasoning turns to achieve better quality. The result of course means we wait a long time to get results even
22.
by lnenad 1mo ago
Adding to my homelab stack, hopefully it doesn't overthink like the little model. Actually, hoping it thinks a bit less. Wait actually I'm really praying it reasons a bit more directly. But wait, I'm really sure that it must
23.
by lnenad 1mo ago
I think as with most human undertakings, building isn't too much of a problem. Maintaining is. Even with what is still a relatively tame number of chargers you get a large number of them that are broken.
24.
by lnenad 1mo ago
I'm assuming it's definitely part of the equation, but considering that I'm getting more tps but still waiting a lot more time for code to come out I'd assume it's not a 1:1 comparison. Plus I'm running quants,
25.
by lnenad 1mo ago
As a small background, I have a local server and I've been trying out different models with different inference engines, quants, configurations etc... I'm also using Opus and Sol at work consistently. I've used AI since the f
26.
by lnenad 1mo ago
I LOVE the concept. I will play around with the execution, if it works as described this is a great product.
27.
by lnenad 1mo ago
> If your needs are met otherwise stick to that and move on. Weird to post such a thought in a forum where OP has posted their project for people to look at. I never mentioned any needs, I am saying the readme holds very little value for
28.
by lnenad 1mo ago
I think as many things that are posted here lately there is no *why* attached to the readme. Why would one use this, what is the benefit of this approach? Am I really gonna need my model to build exotic tools around it; or is exec/web_
29.
by lnenad 1mo ago
You've built a great piece of software, thank you!
30.
by lnenad 1mo ago
I'm running gitea successfully with very little resources.

More ›