BadWolf63

Those who fail to learn history are doomed to repeat it;
those who fail to learn history *correctly* - why they are simply doomed.
-- Achem Dro'hm

🚫DM🚫

[BadWolf63 @BadWolf_63]

@x
@elonmusk
@Xsupport
Ok, Fuck you X.
I got locked out 2 FUCKING MONTHS waiting for the "evaluation" of an appeal to an "harassing" post with just an image (pic 1).
After the first month i cancelled my subscription+, and now i am leaving.
Fuck you all X support people.

[BadWolf63 @BadWolf_63]

@ValentinaForUSA @olgloryandgrace @tebec58 @TMobile @Verizon @ATT @Mintmobile @PatriotMobile Well, i use very little data, so for me is good Mint Mobile, 5G data, Unlimited voice and text, $15/month

[BadWolf63 @BadWolf_63]

@sudoingX Happy 3090 owner here, but go on with what you can. I'll be just envious i can't spend $4k for a spark

[BadWolf63 @BadWolf_63]

@TeksEdge @SpaceTimeViking @nik_algo 🤣🤣🤣

Not particularly reassuring, "the price will be under $10M"...

$100k are just for the container and environmental protection.

[BadWolf63 @BadWolf_63]

@_1jiis Tiramisù land?😋😋😋

[BadWolf63 @BadWolf_63]

@hmmmmmm1458 @ofirg7 🤣🤣🤣

Have you seen the movie "The last starfighter"?

[BadWolf63 @BadWolf_63]

@DefiantLs 🤣🤣🤣

Sure, and Jill Biden could be "a great Surgeon General"...

[BadWolf63 @BadWolf_63]

@EshaAA33 They know where the door is, if they don't like the place.

Adapt to the environment or leave, it is not the environment duty to change itself to cuddle your needs.

[BadWolf63 @BadWolf_63]

@libsoftiktok Well, it's x
"Bulle*i*t in the head", so or they cannot write, or there is something wrong ...

[BadWolf63 @BadWolf_63]

@ElonMuskAOC Do you still have factories in California?

***RUN!!!***

[BadWolf63 @BadWolf_63]

@ElonMuskAOC ... And he sounds *Proud* of it, while at the same time he was receiving millions of $ from Nomisma or China...

But he is not elite, is he?

[BadWolf63 @BadWolf_63]

@mdancho84 Interesting for local LLM models, to see how the new data is saved after the modification (the whole model or just a LoRE like file).
For on-the-cloud models with millions of interactions daily it would be a nightmare if the model decides autonomously what to save.

[BadWolf63 @BadWolf_63]

@2030visiontech @TheAhmadOsman 3090 has 24GB VRAM.
Best model for me until now is Qwen3.6 35B A3B MoE Q4_K_M

[BadWolf63 @BadWolf_63]

@HarshGoel11 @TheAhmadOsman Qwen3.6 35B A3B MoE on single 3090 (Q4_K_M), 256k context, i get 2000-2500 token/s input and between 30 and 50 token/s generation depending on how much the context is full

[BadWolf63 @BadWolf_63]

@Miesiu01 @MuseumCommodore I am not the one saying it:

"The pilot "The Gathering" It was rendered by eight computers Amiga 2000 interconnected with Video Toaster boards that were connected to an IBM computer that stored the images in five gigabytes of memory."

[BadWolf63 @BadWolf_63]

@RightScopee Wasn't Buffett who said the debt wasn't a problem because we can always print money?

https://t.co/BWJhFndONl

[BadWolf63 @BadWolf_63]

@sudoingX @stevibe I don't know...
My RTX3090 runs Qwen 3.6 35B A3B (ok, it's quantized Q4_K_M) + video input mmoroj, embedder and image generation (all on CPU) with a context of 156k at a ~40T/s generation and ~2,200T/s input.

How is the price difference?

[BadWolf63 @BadWolf_63]

@learning_yohei Nope, but i could ask the same question if i eat sashimi drowned in soy sauce + wasabi in front of a japanese...

[BadWolf63 @BadWolf_63]

@GovPressOffice

[BadWolf63 @BadWolf_63]

@learning_yohei That would be saying like you ordered salmon and anyone said that you were eating fish.

Spaghetti is just one kind of pasta

[BadWolf63 @BadWolf_63]

@simplifyinAI Interesting, but at this point wouldn't it be better to tokenize the chain of thought and embed it vectorized in a temporary RAG?
In this way you totally free the context window, sure losing some time with the embedding phase, but it would save a lot of memory.