pferdone

1) It would still run on the Mac's GPU.

2) Since it's unified memory, you won't have 96GB available.

3) I offered a solution that is usually recommended to the "gpu poor", if he's concerned with how much memory he would need.

4) I stated, that people already pointed out how he should be fine and that "gpu poor" doesn't apply to him.

5) "gpu poor" depends on what model you are trying to use. If you want to run Kimi or GLM you are still "gpu poor" even if you have an RTX Pro 6000 with 96GB of VRAM.

a friend told me there's a reddit called: unstable diffusion

you should have a look at https://github.com/deepbeepmeep/Wan2GP which is the goto tool for "gpu poor", although as people below already pointed out you should be fine with comfyui's standard setup aswell

Public doesnt know about Huggingface. ChatGPT (OpenAI) says it‘s dangerous. They must know.

You didn't even read your own article.

What does identity or sexuality have to do with it you fucking moron? These are people! People play games.

But do I have the option to run it 'text only'?

slop

The consensus right now is that Qwen3.6 in its 27B and 35B-A3B versions is better for coding whereas Gemma4 is stronger when it comes to OCR, audio transcription and the likes. Margins are slim though and the harness at these model sizes is the most important factor.

That‘s so cool man! Congrats!

First video with the guy walking the mountain in snow has consistency issues with the cave entrance. Which is "expected" at this model size?!

Who wrote your comment?

I can see that and I don't know your setup, but there are people pushing >70t/s with MTP on a single 3090, with big contexts still >50t/s. 64k is not a lot for agentic coding, and IIRC 128k with turboquant and the likes should be possible for you. r/LocalLLM/ and r/LocalLLaMA/ are worth a visit IMO.

EDIT: just found this recipe repo, may wanna give it a go: https://github.com/noonghunna/club-3090

EDIT-2: this can also shave off a lot of context need for tool calling -> https://github.com/rtk-ai/rtk

pi.dev as well

> unlike almost all qwen models

Almost all means there have been ones before that were not open. So, no contradiction there.

They said in the last paragraph[0]:

"[...] In the coming days, we will also open-source smaller-scale variants, reaffirming our commitment to accessibility and community-driven innovation. [...]"

[0] https://qwen.ai/blog?id=qwen3.6#summary--future-work

It‘s mainly due to system requirements that Flux.2-dev doesn’t get same usage as Z-Image. A 5090 needs about a minute to generate an image with a basic workflow with Flux.2-dev. But prompt adherence and scene/character consistency in edit mode is (way) ahead of Qwen-Edit-2509 if you ask me.

I also have a HUD in my car and I can read it just fine, even in bright sunlight.

I feel tired reading about him and it doesn’t even phase me anymore. It‘s just another thing I add on to the pile. Maybe it’s part of the plan to go numb to everything he does.

I mean they make a bold statement up top just to paddle back a little bit further down with: "[…] In terms of Chinese and Cantonese recognition, the SenseVoice-Small model has advantages."

It feels dishonest to me.

[0] https://github.com/FunAudioLLM/SenseVoice?tab=readme-ov-file...

only next frame

The Guardian, especially for their podcasts, is the only news website I am paying and have ever payed for. And I pay more willingly than any other newspaper would get from me for their paywall stuff. It‘s that valuable for me to support this approach.

Where have you heard that? Because Typescript is the defacto standard right now. Projects like bun and deno cement that status further.

Language learning via hearing comprehension of content not produced in the target language is almost impossible, because the subtitles never match.

However there‘s s difference between CC (close captions) and subtitles, with the former being the verbatim representation (including sfx, music etc.) in my experience.

I already commented [0] on this 2 years ago.

[0] https://news.ycombinator.com/item?id=27420959#27435311

I dont know anyone who thought input == output. Diets like keto clearly show the opposite.

> the US cares about you not endangering other people

Through their rigorous license process to operate road vehicles…yeah right.

It looks pretty good, but especially with the Spiderman comparison you can notice a little smudging and it basically loses the pattern on the suit.

I mean good for them for improving their user‘s experience. I am just surprised that maintaining your own webpack config seems like such challenge. Enabling chunks to split code to improve loading times or dynamic imports, it‘s not exactly rocket science. Yes DX and all improved as well, but CRA is such a bad choice for production code. It gets you going fast(er), but as soon as you have to tweak it, it seems like devs jump to the "next" framework that already has those tweaks builtin…until they hit the next roadblock. Thus you never learn what actually makes it all work.

You are right, it isn‘t of any value to me on HN. Additional information on the topic or a discussion with arguments is. Now you know what is.

I couldn’t hide my dislike for it and commented. Maybe it was unnecessary to draw comparisons to reddit and it’d be better to just state HN rules. I will try to consider this next time. Thank you.