zorgmonkey

I largely agree with you, but funnily enough the very same blog has a great post on the x86 decoding myth https://chipsandcheese.com/2021/07/13/arm-or-x86-isa-doesnt-...

I wish their was a better option, but people can at least get the IEEE 754 standard by putting the DOI for it into sci-hub.

I'm not an expert in postgres performance, but I would think that excessive WAL flushing would also slow down workloads with constant writes.

Technically something like this is relevant, but most people would assume that in practice it is going to be way too much work to get it implemented on GPUs and in such a way that it doesn't add too much overhead.

SawStop works by detecting electrical conductance, and there are many reports of it misfiring when attempting to cut wood that isn't fully dry (i.e., there is moisture inside the wood, increasing its electrical conductance).

Feel free to mess with it, his tweak to softmax was actually supported by pytorch before the article was written, but off by default. Maybe it needs to be more widely used though, after all good ideas are often independently discovered multiple times. Details are in this tweet https://twitter.com/SamuelMullr/status/1683582347793530884 or if you don't like twitter the option is add_zero_attn for pytorch MultiheadAttention.

It looks like someone already uploaded the code to the internet archive with git history https://archive.org/download/portal64_202401/portal64-repo.z... for stuff like this it is often better to search the manual uploads to internet archive instead of by URL.

It looks like the reason it was submitted today is a new feature added "Added support for Stable Diffusion XL Turbo 1.0! (thanks to @AeroX2)", from the news section of the README.

I think you're about the goal, but in practice it looks more line they are making code that is readable in 0 languages instead of at least 1 language. It looks like in practice people are unsurprisingly coming up with workarounds for this by adding some boilerplate code, these are the first two lines from the python example above (also all the local variables in the function are in English too):

  def Z10591(Z10591K1, Z10591K2):
    word, is_ra_plural = Z10591K1, Z10591K2

I like the idea a lot, but I had two main UX problems:

* it is hard to know which of the green blocks can be expanded by clicking, maybe a different color or border for the ones that can be expanded.

* I kept accidentally clicking the text to go to github, but I did realize that if you aim for the edge it works a lot more reliably

It has been a while since I last tried, but I never had very good results when I tried inpainting with SDXL in comparison to the SD1.5 inpainting models.

If you really care about being able to search github without logging in https://grep.app/ is pretty good, I would often use it instead of the old github search cause I found the results to be better.

The author didn't say it, but I'm pretty sure for-loop was meant to be pseudocode to help the reader understand what it does and not the actual implementation.

For what it is worth google has fortunately abandonded the WEI proposal, a quote from the README of https://github.com/RupertBenWiser/Web-Environment-Integrity/

  NOTE: This proposal is no longer pursued.
  Thank you for all the constructive feedback and engagement on the topic. An Android-specific API that does not target the open web is being considered here.

The android specific proposal is only adding support for it to WebView, which developers actually could already by combining the WebView and play integrity APIs, so that as much as I don't love it that doesn't seem too terrible if it is just saving developers from writing some boilerplate code to connect the two. Here is the recent discussion about the WebView changes https://news.ycombinator.com/item?id=38118627

Google fi phone numbers are not VOIP they operate as a MVNO. Google voice numbers are VOIP.

It is very easy to tweak this to generate images quickly on a nvidia GPU:

* after `pip install -r requirements.txt` do `pip3 install torch torchvision torchaudio xformers --index-url https://download.pytorch.org/whl/cu121`

* on line 17 of main.py change torch.float32 to torch.float16 and change mps:0 to cuda:0

* add a new line after 17 `model.enable_xformers_memory_efficient_attention()`

The xFormers stuff is optional, but it should make it a bit faster. For me this got it generating images in less than second [00:00<00:00, 9.43it/s] and used 4.6GB of VRAM.

Google has actually audited using cargo-vet every crate that chromiumos and fucshia depend on that have unsafe in it. They also have some additional rules related to cryptographic algorithms. I'm pretty surprised they haven't done the same for rust usage in android. https://github.com/google/rust-crate-audits

PoC code: https://github.com/UT-Security/gpu-zip

Preprint paper: https://www.hertzbleed.com/gpu.zip/GPU-zip.pdf

I've never used it, but if memory serves they are, like many other companies, using a fork clang. So I expect it will be about as fast if not slower then clang.

How does the performance with Vulkan compare to the ROCm performance on the same hardware?

If popular models are still making this mistake then it still seems noteworthy and making a blog post or paper to increase awareness definitely seems worthwhile. Also multiple independent discovery of good ideas is quite common.

They include versions without the ligatures for exactly that reason. It is the files with NL in the name.

My understanding is that the downside of those downgrade devices is HDCP 1.4 will leave you stuck without 4k and HDR for content that uses HDCP.

Looks like a cool project, but it would be great if you could add a license to the project since under US copyright lack of a license implies all rights are reserved and not open source. If you don't know much about software licenses https://choosealicense.com/ is a helpful site.

ROCm does not compile PTX IR to AMD assembly, if it did you'd be able to run nearly any compiled CUDA program on and AMD GPU without the source. ROCm is a source level compiler for CUDA (technically HIP is the actual compiler, but whatever), it allows you to a substantial fraction of the CUDA APIs. This notably also means that any Nvidia open source library that uses inline PTX assembly won't work, but fortunately AMD does have alternatives to many of the Nvidia libraries.

The website (https://functiontrace.com/) mentions that dual licensing is available at the bottom of the page, so my best guess would be that you need to contact the author for pricing details.

Here is one on ebay for $600, realistically you'd need a nice rework station and the skills to reball it to actually use it. FPGA pricing for on places like mouser and digikey is notorious for being an outright lie, the high volume prices are way less then 50k (don't get me wrong they're still expensive chips, just less expensive than that).

https://www.ebay.com/itm/364213117135?hash=item54cccb20cf:g:...

You don't swear anything with a DMCA, the law says this in 3.A.vi: > A statement that the information in the notification is accurate, and under penalty of perjury, that the complaining party is authorized to act on behalf of the owner of an exclusive right that is allegedly infringed.

https://www.law.cornell.edu/uscode/text/17/512

Google will never touch Citus because it is AGPL licensed[0], I wouldn't be surprised if the other cloud providers have similar policies.

[0] https://opensource.google/documentation/reference/using/agpl...