gliptic

5 ms·

1.
by gliptic 2mo ago
The very wikipedia page you're linking says: > The term abliteration has been coined for the process of using ablation to uncensor large language models by modifying internal functions to completely eliminate refusal behaviors while
2.
by gliptic 2mo ago
https://en.wiktionary.org/wiki/abliterate
3.
by gliptic 2mo ago
Except we're talking about a _different_ word, a word that was _coined_ as a portmanteau of the word you're referring to and another word.
4.
by gliptic 2mo ago
The name for it _is_ abliterate. It's a portmanteau of ablate and obliterate.
5.
by gliptic 4mo ago
> Zig-native immediate-mode dvui?
6.
by gliptic 9mo ago
The borrow checker doesn't decide when things are dropped. It only checks reference uses and doesn't generate any code. This will work exactly the same as long as your program doesn't violate any borrowing rules.
7.
by gliptic 9mo ago
Camping is like.. regulatory capture? Stretching the analogy thin here.
8.
by gliptic 11mo ago
But that fine-tuning is done only on those 100-200 good samples. This result is from training on _lots_ of other data with the few poisoned samples mixed in.
9.
by gliptic 1y ago
This doesn't really make sense to me. Most cats I've known react to such reflections without ever having seen a laser pointer in their life, for the same reason they react to laser pointers.
10.
by gliptic 1y ago
Glyph advance or line spacing is not part of the bitmaps.
11.
by gliptic 1y ago
> I think Sweden was deploying the Drakken (Dragon) and later the Vigen (Lightning). The names are much less flashy, Draken (The kite, due to the shape) and Viggen (The tufted duck) :P.
12.
by gliptic 1y ago
`f(++i, ++i)` is/was indeed UB, but the example in munificent's comment was `foo(print(1), print(2))` which as far as I know is not even if both `print` calls read/write the same memory.
13.
by gliptic 1y ago
Yes, but that's just a subset of expressions where unspecified sequencing applied. For instance, the example with two `print()` as parameters would have a sequence point (in pre-C++11 terminology) separating any reads/writes insid
14.
by gliptic 1y ago
The author is using the term in the way that everyone else understands it. They are not aware of your unusual definition.
15.
by gliptic 1y ago
The evaluation order is _unspecified_, not undefined behaviour.
16.
by gliptic 1y ago
Recent developments on BB(6) previously posted here: https://scottaaronson.blog/?p=8972
17.
by gliptic 1y ago
It's really not. Looks like something like JS with almost identical syntax. Also, the compiler (and what else?) isn't even implemented [1]. [1] https://github.com/profullstack/smashlang/blob/master&#
18.
by gliptic 1y ago
> Dynamic Typing: Flexible type system with runtime type checking Uhm, what does this have to do with C/Zig?
19.
by gliptic 1y ago
Ah, FastML is an extremely overloaded name.
20.
by gliptic 1y ago
It's strange that someone from FastML can be confused about this, unless it's supposed to be a bad joke.
21.
by gliptic 1y ago
I'm not seeing how this is pull in any sense. Calling recv on the channel doesn't cause any result to be computed. The push of the previous operators will cause the compution to continue. EDIT: Ok, I guess because they are bounded
22.
by gliptic 1y ago
For pixelation you can use another technique invented for astronomy: drizzling [1]. [1] https://en.wikipedia.org/wiki/Drizzle_(image_processing)
23.
by gliptic 1y ago
You're probably underestimating what you can do with these.
24.
by gliptic 2y ago
Any that can run 70B at >5 t/s are >$2k as far as I know.
25.
by gliptic 2y ago
I guess it will be a bigger issue the longer it's been since they stopped making them, but most I've heard (including me) haven't had any issue. Crypto rigs don't necessarily break GPUs faster because they care about pow
26.
by gliptic 2y ago
I arbitrarily chose $1k as the "cheap" cut-off. Two 3090 is definitely the most bang for the buck if you can fit them.
27.
by gliptic 2y ago
That's 1.2 t/s for the 14B Qwen finetune, not the real R1. Unless you go with the GPU with the extra cost, but hardly anyone but Jeff Geerling is going to run a dedicated GPU on a Pi.
28.
by gliptic 2y ago
Your best bet for 33B is already having a computer and buying a used RTX 3090 for <$1k. I don't think there's currently any cheap options for 70B that would give you >5. High memory bandwidth is just too expensive. Strix Hal
29.
by gliptic 2y ago
So yes, it's a limitation of their own API at the moment, not a model limitation.
30.
by gliptic 2y ago
R1 is trained for a context length of 128K. Where are you getting 8K/32K? The model doesn't distinguish "thinking" tokens and "output" tokens, so this must be some specific API limitations.

More ›