65a

7 ms·

31.
by 65a 2y ago
I refuse to say mebibyte or whatever alternative unit. 1024 bytes is one kilobyte, and 1000 kilobytes is not a useful unit (and so on). As far as being a conspiracy by hard drive manufacturers, Western Digital did settle the case rather tha
32.
by 65a 3y ago
Only with additional hardening between the container and the kernel and hardware itself.
33.
by 65a 3y ago
Other than process specifics, they could, and that's exactly what happened when the 1980s US Government got worried about a single CPU supplier. The result was (to some extent) modern cheap computing.
34.
by 65a 3y ago
I agree if we extend it to mathematics and the use of simple machines.
35.
by 65a 3y ago
There would be an ironic twist if nvidia (who first competed with AMD nee ATI) was forced to license their IP to intel for national security reasons, but it would rhyme with history (x86 licenses to AMD) and benefit the consumer.
36.
by 65a 3y ago
Disgusting abuse of the democratic process to halt scientific and technological progress in the name of making one sketchy man rich.
37.
by 65a 3y ago
It's somewhat regional, and it means to hunt down the target at the expense of everything else, as a dedicated hunting dog might.
38.
by 65a 3y ago
Is there a proper reverse engineering of the payload yet?
39.
by 65a 3y ago
Isn't u-root also basically this?
40.
by 65a 3y ago
Pointers are "nullable types", but if you're seeing that a lot, it generally indicates an architectural problem.
41.
by 65a 3y ago
I used many of these as test cases: http://wiki.franklinheath.co.uk/index.php/Enigma/Sample_Mess...
42.
by 65a 3y ago
CWOWD CZKZO DLUML TPRHR KXFHH MHIUS TMXON CHQXQ DFMBI APKZV DRFIU SEFLV ZSSPK ESGNE BNRQA QBVHH YMAZ just to not be cruel, it's an m3 without UKW-D, 3 rotors and a plugboard :) EDIT: I think I answered my question looking at the code.
43.
by 65a 3y ago
Ollama does a nice job of looking at how much VRAM the card has and tuning the number of gpu layers offloaded. Before that, I mainly just had to guess. It's still a heuristic, but I thought that was neat. I'm mainly just using lla
44.
by 65a 3y ago
Yep, and it deserves the credit! He who writes the cuda kernel (or translates it) controls the spice. I had wrapped this and had it working in Ollama months ago as well: https://github.com/ollama/ollama/pull/8
45.
by 65a 3y ago
Go has a clear() function now, but I haven't checked to see how it actually behaves in terms of secret abatement in memory. https://tip.golang.org/ref/spec#Clear
46.
by 65a 3y ago
By chunking and binning the data by user contribution count, you are structuring it as described. Pretraining is often accomplished on similar chunked text. k-anonymity is not perfect though, or state of the art, the wikipedia explains seve
47.
by 65a 3y ago
https://en.wikipedia.org/wiki/K-anonymity Basically keeping only chunks of text which have are not unique and contributed by many users.
48.
by 65a 3y ago
The inference side (adding something * something else) to every layer seems a lot like what happens with a LoRA? If so is it possible to encode a Control Vector as a LoRA for the purposes of using this with existing inference frameworks wit
49.
by 65a 3y ago
ROCm is preferred over vulkan for AMD GPUs, performance wise. Using OpenCL or Vulkan should only be for older cards or weird setups.
50.
by 65a 3y ago
I found it much easier to make ROCm/AMD work for AI (including on an laptop) than getting nvidia work with Xorg on an optimus laptop with an intel iGPU/nvidia dGPU. I swore off nvidia at that point.
51.
by 65a 3y ago
llama.cpp has had ROCm support for a long time
52.
by 65a 3y ago
I assume that terminals are authenticated, and stolen/diverted terminals can be disabled, so outside willful negligence or failure to report stolen/diverted terminals, it seems like it should be obvious who is accessing their mode
53.
by 65a 3y ago
But...why?
54.
by 65a 3y ago
I'd rather intelligence agencies spent their time fucking instead of killing, I guess. You're describing an ideal that has not been realized historically, so while I agree, I also completely understand the fact that tradecraft ten
55.
by 65a 3y ago
Look, I get what you are trying to say. But in a democracy, do what you can with what you have. Not voting or not participating leaves the entire game board to whoever pissed you off this much.
56.
by 65a 3y ago
40s, I'd say Fritz-X was a good example. https://en.wikipedia.org/wiki/Fritz_X
57.
by 65a 3y ago
Go works fine for C FFI, all of its problems there are caused by its innovation wrt dynamic stack sizes and having a garbage collector. I'd rather write multithreaded Go FFI than deal with JNI again, anyway. There isn't really a l
58.
by 65a 3y ago
Not resiliant, because in the 10-20 year lifespan of a home electrical panel, at least one critical vulnerability is going to allow complete remote pwnage of that device. It's stupid IOT garbage that belongs in ewaste, especially with
59.
by 65a 3y ago
Yes, it sounds like SNES implements debounce by just scanning slowly, but if you want to feed USB reports at maximum (or above maximum) specification, you will want to check the state more frequently, and you'll need an alternate mit
60.
by 65a 3y ago
The physical switches may actually bounce, so you might need some debounce logic, if it wasn't the shift register glitching out. It's pretty common for buttons to need this, often seen in keyboard firmware as well for similar reas

More ›