Vemloir

Larping ML

[Vemloir @Vemloir]

You are acting like a total piece of shit on Twitter. Maybe one day you will realise that you're making yourself look like a fucking idiot to your readers. Rethink your freaking behaviour. Theo, fuck you❤️

[Theo - t3.gg @theo]

“Ivan is so nice, why did you block him??”

[Vemloir @Vemloir]

@S4AD_DXD @ravikiran_dev7 It doesn't matter at all

[Vemloir @Vemloir]

@DmitryMatveev @vladnineplusone В инференс движках или фреймворков для обучения, нацеленных на использование в домашних пк можно допустить небольшую потерую скорости compute, НО в случае с anthropic и подобными большими лабами такое отнош��ние к своему стеку обойдётся им в десятки миллионов долларов

[Vemloir @Vemloir]

@DmitryMatveev @vladnineplusone И как раз таки проблема в оптимизации под конкретные GPU

[Vemloir @Vemloir]

@DmitryMatveev @vladnineplusone GEMV и GEMM? Зто наименьшая часть пролбемы)

[Vemloir @Vemloir]

@uzair__afridi It is not much faster. Llama.cpp and vllm have the same speed on the same specs, except of offload scenario

[Vemloir @Vemloir]

@tysn_dev Pi.

[Vemloir @Vemloir]

@AdamHoltererer You need to be studied how llm works

[Sam Altman @sama]

@jestermolecule @HarshithLucky3 but only because...i actually do want the world to have tons of memory and people to put as much as they want in their computers, and it is fucking annoying that this was taken out of context.

i have been arguing for the world to ramp up memory production for a long time.

[Vemloir @Vemloir]

@habr_com 24576 экспертов и на каждый токен у тебя будет работать 480, а не 10. У каждого слоя qwen3.8 flash next 512 routed экспертов и из них выбирается 10 на каждом токене

[Vemloir @Vemloir]

really thought they were talking about Moonshot.

[Vemloir @Vemloir]

@1red2black

[Vemloir @Vemloir]

@elshayib_

[Vemloir @Vemloir]

@nisten

[Vemloir @Vemloir]

@plotarmordev Qwen3.8 max 0902 is proprietary

[Vemloir @Vemloir]

@itstxssu @radzimich @Kimi_Moonshot @KimiDevs Я в день бана сразу же купил ChatGPT Pro за 100)

[Vemloir @Vemloir]

@itstxssu @radzimich @Kimi_Moonshot @KimiDevs даваЛ. Сейчас почти каждого русского просят пройти верификацию и если паспорт русский - моментальный бан.

[Vemloir @Vemloir]

@morqon @CheekyPint Sama is missing two things here:
1. ​Hyperscalers like his own company are the ones that caused the DRAM shortage in the first place.
2. ​Consumer demand for local AI grows gradually, giving fabs time to adapt—unlike hyperscalers hoarding the entire supply overnight