Do other instances defederate with them, though?

None of my family will watch 3D in theatres. It makes them sick.

This is a huge factor.

This is what basically anyone in the ML research/dev industry will tell you, heh.

I’m in a similar boat, though I’ve been present for some time.

Dbzer0 seems like the best “fit” for me, but practically I just want the instance that’s not defederated/blocking other instances.

…Not sure which that is. But I’d look at Piefed before Lemmy, since they work together, but Piefed seems more desirable feature-wise.

Yeah. Exactly

I’m bristling at this because it’s exactly what Trump does here. Some local official screws up or does something bad, and it’s blown up on Twitter and generalized to all “Democrats.”

…no? From my perspective, that’s like saying:

Democrats and police worked to protect the slavers and stop the rescue of slaves from out of the province.

When it was really the city, or at most state-level government getting in trouble. I know China isn’t federalized to the extend the US is, and it’s technically true since all govt is CCP, but still.

It sounds like you are implicating national officials when there’s no mention in the source.

Party officials and police worked to protect the slavers…

Seems like local government. Right?

Concealed camera revealed that the local police refused to take action to rescue the slaves. Later the reporters were allowed into the illegal brickyards with the company of the local police. Concealed camera showed the police keeping them from rescuing children who were not from Henan which showed obvious local government protection for the illegal brickyards.

As the scandal received immediate media attention, it also caught the eyes of the major party and state leaders, including CCP General Secretary Hu Jintao and Chinese Premier Wen Jiabao. Governor Yu Youjun of Shanxi province offered an unprecedented self-criticism, took responsibility, and tendered his resignation on 30 August. He was replaced by Meng Xuenong, an official who had been sacked as Beijing mayor after the SARS outbreak.[8]

Not gonna lie. When I need a wank, I need a wank.

But “professional” porn (like pornhub largely hosts) always felt gross to me.

Sometimes. As a tool, not an outsourced human, oracle, or some transcendent companion con artists like Altman are trying to sell.

See how grounded this interview is, from a company with a model trained on peanuts compared to ChatGPT, and that takes even less to run:

…In 2025, with the launch of Manus and Claude Code, we realized that coding and agentic functions are more useful. They contribute more economically and significantly improve people’s efficiency. We are no longer putting simple chat at the top of our priorities. Instead, we are exploring more on the coding side and the agent side. We observe the trend and do many experiments on it.

https://www.chinatalk.media/p/the-zai-playbook

They talk about how the next release will be very small/lightweight, and more task focused. How important gaining efficiency through architecture (not scaling up) is now. They even touch on how their own models are starting to be useful utilities in their workflows, and specifically not miraculous worker replacements.

Heh, so does mine.

All our parents’ book hoarding may end up saving us. And the internet, if they become the new standard?

You joke, but that’s horrifying.

This is already an SEO technique, apparently, and I could see Amazon book sellers finding a way to fudge it: https://yoast.com/help/date-appears-search-results/

100%.

But I was pondering more what the general population might do. People are going to figure out slop recipes don’t work, but the question is what’s the next most accessible thing to replace it with?

…So are we going back to print cookbooks? Published before 2024?

Honestly, that feels like the practical solution.

Plantains are a fickle plant. Ripeness is a huge factor, and that aside, some are just fibery/less sweet and don’t cook as fast.

In theory, Google should fight all attempts at SEO.

But they infamously stopped doing that to bump some quarterly result (as sifting through them generates more clicks), and here we are.

deleted by creator

It’s mind boggling that Google is letting this happen.

Even if I pretend to be a stone cold Google accountant, and look at this… YouTube is a gigantic golden goose. It’s strategic, it’s a cultural lynchpin. And they’re going to let it wither just to boost next quarter and avoid saying “AI is a problem here”?

and more often than not, it’s content that is over two or three years old.

Bingo.

Friend, content that’s over a year old is all you can trust. There’s a lot of it, and its great.

The date filter is the slop filter.

Vllm is a bit better with parallelization. All the kv cache sits in a single “pool”, and it uses as many slots as will fit. If it gets a bunch of short requests, it does many in parallel. If it gets a long context request, it kinda just does that one.

You still have to specify a maximum context though, and it is best to set that as low as possible.

…The catch is it’s quite vram inefficient. But it can split over multiple cards reasonably well, better than llama.cpp can, depending on your PCIe speeds.

You might try TabbyAPI exl2s as well. It’s very good with parallel calls, thoughts I’m not sure how well it supports MI50s.

Another thing to tweak is batch size. If you are actually making a bunch of 47K context calls, you can increase the prompt processing batch size a ton to load the MI50 better, and get it to process the prompt faster.

EDIT: Also, now that I think about it, I’m pretty sure ollama is really dumb with parallelization. Does it even support paged attention batching?

The llama.cpp server should be much better, eg use less VRAM for each of the “slots” it can utilize.

I’ll save you the searching!

For max speed when making parallel calls, vllm: https://hub.docker.com/r/btbtyler09/vllm-rocm-gcn5

Generally, the built in llama.cpp server is the best for GGUF models! It has a great built in web UI as well.

For a more one-click RP focused UI, and API server, kobold.cpp rocm is sublime: https://github.com/YellowRoseCx/koboldcpp-rocm/

If you are running big MoE models that need some CPU offloading, check out ik_llama.cpp. It’s specifically optimized for MoE hybrid inference, but the caveat is that its vulkan backend isn’t well tested. They will fix issues if you find any, though: https://github.com/ikawrakow/ik_llama.cpp/

mlc-llm also has a Vulcan runtime, but it’s one of the more… exotic LLM backends out there. I’d try the other ones first.