lostmsu

13 ms·

481.
by lostmsu 7mo ago
Not exactly what you asked for but try https://pelicans.borg.games/
482.
by lostmsu 7mo ago
In US downloading copyrighted data is not illegal AFAIK
483.
by lostmsu 7mo ago
You can download Qwen 3.5 under Apache 2.0
484.
by lostmsu 7mo ago
How many years did it take for you to write a Photoshop clone eating Adobe's lunch?
485.
by lostmsu 7mo ago
Couldn't tell if the first part was a sarcasm or not. The second part gave it away though.
486.
by lostmsu 7mo ago
> Bigger size means more knowledge. Qwen 3.5 27B beats StepFun 3.5 Flash on GPQA Diamond too, so probably no.
487.
by lostmsu 7mo ago
Tuned Qwen 3.5 27B beats Step 3.5 on almost all benchmarks, so the point about the size class is moot.
488.
by lostmsu 7mo ago
> all orthogonal concepts That's my point exactly. I just also assume the % of liars in 30s is the same as in general population by default.
489.
by lostmsu 7mo ago
> No, because smart people realize they are playing an iterated game and that behaving in a way that people identify as Machiavellian is actually suboptimal in the long run. Even if you are right coincidentally (which I wouldn't b
490.
by lostmsu 7mo ago
I would argue that flattery without calculation is just poorly calculated flattery. Same applies to many other traits in the list. Low achievements people lie right and left just as well. Are cynical when convenient, yada yada. Basically, t
491.
by lostmsu 7mo ago
In large providers KV caches are the main bottleneck, no?
492.
by lostmsu 7mo ago
Last time I checked Codex didn't have that option for $20 plan
493.
by lostmsu 7mo ago
I also really liked Foundation on Apple TV
494.
by lostmsu 7mo ago
> Because for devs especially Are you not a dev? If not, what would you use a coding tool for? They still require handholding for anything largeish. Still much cheaper than outsource.
495.
by lostmsu 7mo ago
If dollar crumbles, the debt becomes lighter, no?
496.
by lostmsu 7mo ago
The page doesn't mention what interface the 160GiB card uses. Quick Googling doesn't either.
497.
by lostmsu 7mo ago
Nothing like Crossfire/SLI? Not possible to efficiently connect multiple cards for one large model?
498.
by lostmsu 7mo ago
Is PCI-E too much to ask?
499.
by lostmsu 7mo ago
> small fp8 local model with almost 100k token context Would not fit Qwen3.5 27B would it? That's the SOTA
500.
by lostmsu 7mo ago
> voluntary scanning What is that? A setting in OS?
501.
by lostmsu 7mo ago
That's one of the tricks. The other trick is to vote in universal right for encrypted communication once and for all.
502.
by lostmsu 7mo ago
On a related note I miss LLaMA spelling.
503.
by lostmsu 7mo ago
It's worse, because there are actually integrated SoCs that include NPU, which I would say are real "AI accelerators".
504.
by lostmsu 7mo ago
But if mmap already works why would there be any interest? Besides, discussions are for users. He didn't open PRs or issues.
505.
by lostmsu 7mo ago
llama.cpp afaik does not run portion of the model on the CPU. --cpu-moe just offloads weights to RAM, but they are still loaded to GPU for compute.
506.
by lostmsu 7mo ago
Why would llama with --mmap crash?
507.
by lostmsu 7mo ago
How's the reproducibility of the results? Like avg score of 10 runs vs original.
508.
by lostmsu 7mo ago
> Mine was from websites 7 days I think. Randomly stopped booting a month ago (bsod after updates). This happened to mine too. I suspect this might be the real cause for the blog post in question.
509.
by lostmsu 7mo ago
To be fair Everything is heavy. 400MB on my current machine.
510.
by lostmsu 7mo ago
> MoE models via expert sharding with zero cross-node inference traffic This makes the whole project questionable

More ›