ttflee

It is really a difficult choice in front of me: DeepSeek v4 flash 0731@q2 vs Qwen 3.{6,8} 27B@fp8 on a 96GB VRAM server.

But Swift has a stable ABI which neither Zig nor Rust could provide.

Why not using Swift?

He is autistic, and has ADHD.

Perhaps we could do it in Python in the first pass for validation purpose. And then vibe rewrite it in Haskell.

Hope that the `metered connection` option was toggled on.

> Swift 6.3 includes the first official release of the Swift SDK for Android.

I guess the day that a corporate AI could easily fake all my online existence is drawing nigh.

`perldoc perlrun`

TLDR: The video got VAEed, and we are the discriminators being fooled.

In Chinese, it always added something like "For study/research purpose only. Please delete after 48 hours." This is what those volunteers added in subtitles of (pirated) movies/shows.

Every few years, new progresses might remind me of this talk by Gary Bernhardt:

https://www.destroyallsoftware.com/talks/the-birth-and-death...

thanks to this, https://xkcd.com/1838/

Looks like a variant of iron-air battery project to me.

The model card for Yi-34B included benchmark scores in several Chinese-oriented evaluation sets, CMMLU, C-eval and GAOKAO.

Yeah. I guess most of the Chinese propagandists received training imported from Russia.

ChatGPT told me:

> Yes, Nazi propaganda did often refer to the Allies, particularly the UK and the US, using terms such as “Anglo-American” or “Anglo-Saxon” in a pejorative sense. This language reflected the Nazis’ racist ideology, in which they posited a hierarchy of races with Aryans at the top and viewed others, including Anglo-Saxons, as lesser or enemies. This also served to foster animosity and dehumanize the Allies among the German populace. However, such language was not universally used by all within the Nazi regime, and the terminology often varied depending on the context and specific propaganda objectives at the time.

Fun fact: hardcore xenophobic narratives in China has been using the term Anlgo-Saxons (昂撒), in recent years.

Thinking what Apple sells is a GPU bundled with a large amount of VRAM and a free ARM CPU makes things easier to accept.

NVIDIA never sold a GPU with expandable memory, either.

It was kinda pro-trumpish to say that before Joe nailed the presidency.

Now for an encoding in real life:

https://en.wikipedia.org/wiki/UTF-7

When you are coding a dynamic library, and providing consistent ABI between versions is a thing, you could appreciate this.

Perhaps you may also appreciate PImpl/D_PTR etc.

You need `-pix_fmt yuv420p` for an iPhone to recognize a ffmpeg-encoded video file.

I always wonder whether blockchain was designed to survive this.

How many steps did you perform?

I tried some and found no major differences after 16 steps or so with given random seed.

On intel MacBookPro 2020, CPU-only, the original one[1] using pytorch utilized one core only. A tensorflow implementation[2] with oneDNN support which utilized most of the cores ran at ~11sec/iteration. Another OpenVINO based implementation[3] ran at ~6.0sec/iteration.

[1] https://github.com/CompVis/stable-diffusion/

[2] https://github.com/divamgupta/stable-diffusion-tensorflow/

[3] https://github.com/bes-dev/stable_diffusion.openvino/

If Mother Nature changes, do not fight. Adapt to it.

FYI, Celestial grace in his letter meant the grace of Qing Emperor, and Celestial Court the Qing Government. Lin's perspective was pretty Qing-centric which was of course orthodox in his ideology. The letter reads so supercilious as if it were written by someone much worse than D. Trump. I doubt if the messenger had the guts to deliver or just burnt it silently.

There is nothing could not be solved by adding another layer of abstraction, except for too many layers of abstraction.