Harlequin 🇷🇴

[Harlequin 🇷🇴 @AndreiColgiu]

@lvsdro Singurul partid care are o pozitie rezervata fata de occident: SOS ✌️

[Harlequin 🇷🇴 @AndreiColgiu]

@NicusorDanRO "Păstrarea direcției pro-occidentale este o condiție de la care nu ne vom abate." - doar Diana Sosoaca are curajul sa declare ca exista o alternativa (mai buna in opinia mea): BRICS. Restul sunt doar slugi pentru 2 grupări occidentale care si ele servesc aceeasi stapani.

[Harlequin 🇷🇴 @AndreiColgiu]

@TateTheTalisman It is extremely unlikely the two are in any way related. The Covid vaccines were simply way less effective than they were presented and most likely they generated more adverse reactions, but not cancer. Your stance if being fully "against" vaccination was wrong. Just addmit it!

[Harlequin 🇷🇴 @AndreiColgiu]

@EuropeMacApps @GPTWare Better in your opinion than qwen 3.6 35B A3B - assuming same quantization level?

[Harlequin 🇷🇴 @AndreiColgiu]

@MarioNawfal the drone after the failed attack.
Obviously, something doesn't add up here!

[Harlequin 🇷🇴 @AndreiColgiu]

@MarioNawfal There are two main red alerts about this official story:
1) Romanian authorities were only notified 15m -20m before the self detonation and only after the drone got stuck
2) Considering Ukr-NATO alliance, there was absolutely NO reason for Ukr to hide their attack or to detonate

[Harlequin 🇷🇴 @AndreiColgiu]

@Dan_Tanasa Eu sunt curios la ce se refera acest program de cyberprotection (?) bazat pe IA? Bugetul e unul foarte generos pentru un proiect de genul acesta. Sa intelegem ca se doreste o varianta de supraveghere sociala (sub pretextul "protectiei") avansata, bazata pe IA?

[Harlequin 🇷🇴 @AndreiColgiu]

@bobbydtiktok In locul lui Putin acceptam, cu conditia ca Zele sa vb doar in rusa 😂😂😂

[Harlequin 🇷🇴 @AndreiColgiu]

@lmstudio Excellent, now bring tq support to your runtimes 👍💪✌️ You may even try to go for both weights and kv cache.

[Harlequin 🇷🇴 @AndreiColgiu]

@KuittinenPetri Finally, a fair point of view. I would also add that we still don't know the final price range on rtx spark configurations, but historically, Nvidia is far from being "cheap".

[Harlequin 🇷🇴 @AndreiColgiu]

@brave_romania EU is an unstoppable cancer imo. Nobody can stop it from spreading and destroying each European nation. Even the mighty Russian Federation will eventually succumb to the this horrible corruption! Give it a few decades at most, it will happen!

[Harlequin 🇷🇴 @AndreiColgiu]

@KoAtti77 @ivan_8848 This version is more historically accurate, indeed. However, won't happen, there is way too much western greed involved 🤫 and the target of the western elites remain the Russian Federation wealth - it's a long term plan!

[Harlequin 🇷🇴 @AndreiColgiu]

@mog_russEN Pretty sure i spotted a cybertruck in the background? That's actually interesting!

[Harlequin 🇷🇴 @AndreiColgiu]

@Ex0byt Does nvfp4 have llama.cpp support?

[Harlequin 🇷🇴 @AndreiColgiu]

@oana_toiu Wow, citesc comentariile si-mi dau seama cat de populara este actuala conducere a țării... Referitor la subiect, raman la părerea lui nea Ceașcă, aia cu "neamestecul in 'treburili' interne"... Ne dam noi cu părerea la ce fac rușii in... Rusia?!

[Harlequin 🇷🇴 @AndreiColgiu]

@TateTheTalisman @GiorgiaMeloni Say what!?

[Harlequin 🇷🇴 @AndreiColgiu]

@selogerkkk @ItsmeAjayKV It highly depends on the cpu/mb/ddr performance as well.

[Harlequin 🇷🇴 @AndreiColgiu]

@witcheer I think you should try with ud-iq3_s and/or ud-q3_k_xl. For the 2nd variant I'd suggest a llama fork that supports turbo quants for kv cache.

[Harlequin 🇷🇴 @AndreiColgiu]

@andreibasti @RomaniaMare1918 Care "noi"? Romania nu câștigă decat un procent minuscul din exploatarea resurselor (de toate felurile). Daca nu sunteti cumva in spatele companiilor exploatatoare, nici sa nu va ganditi ca veti avea vreo urma de beneficiu din aceste exploatari de resurse.

[Harlequin 🇷🇴 @AndreiColgiu]

@Maor_Elkarat Any idea if llama.cpp forks that implemented turbo3 for kv cache can also work with iq quantization for model? I can run ud-iq3_s perfectly in lm studio, but not in the llama.cpp fork. I'm running ud-q3_k_xl with 204800 context (19 cmoe) on my 5070 though.