DogroseMK

[DogroseMK @dogrosemk]

@Biggest Broski sorry for my wodden engrish, but for real, can you push for Vulkan support there :)))

[Brian Armstrong @brian_armstrong]

We were able to fine tune an open weight model (Qwen3.5-9B) on proprietary internal data at @coinbase, and it out performed frontier models, at much lower cost and speed, for a specialized task (fraud prevention in this case).

I think we'll see more of this pattern inside companies. The frontier models are amazing generalists. But increasingly you'll see specialist models out perform on more narrow tasks, either working on teams, or by themselves.

We're making it easy to fine tune models internally for our employees. We even have a skill for it in our internal harness that makes it easy to spin these up and start training them:

"/fine-tune-and-serve-llm I want to fine-tune Qwen for payments risk review. help me get started"

I imagine we'll have some backend services that could be replaced entirely by this new architecture.

[Niko Veit @coldniko]

@Biggest Wrong… Claude, Codex, DeepSeek + OpenCode 🫡

[DogroseMK @dogrosemk]

It is not much, but it is honest work. Waiting for an agents to finish with GLM5.3 Flash Q4 and Qwen3.8 Flash Next Q8 inference tuning on the monster and I move towards Strata in Pascal mode :-)

[Niko Veit @coldniko]

Did I go a bit overboard?

[DogroseMK @dogrosemk]

@axitek @coldniko If you have good cpu most probably you will end with 200t/s :-)

[DogroseMK @dogrosemk]

@axitek @coldniko Same :-)

[DogroseMK @dogrosemk]

Strata and dual gpu 3090, on xeon E5 + 96GB ram _S version. Some startup optimizations :-)

[Vlad @VladTheInflator]

Here we come

[DogroseMK @dogrosemk]

Lel 🥳

[Niko Veit @coldniko]

@Tono_Ken3 QwenImage support comes in very very soon!! ;)

[DogroseMK @dogrosemk]

✨🥳🏴‍☠️

[Niko Veit @coldniko]

Be ready - October 14th

[Niko Veit @coldniko]

New version is now live with another performance boosts and improvements on Strata

https://t.co/jCQGsoSOb5

[DogroseMK @dogrosemk]

@gospaceport Pin the workers per core, you should go towards 200 :-)

[AI Search @aisearchio]

Huge database of decompiled games

https://t.co/UbkgkKw8M6

[DogroseMK @dogrosemk]

@gospaceport CPU and RAM? With little polishing 180+t/s will be possible

[Financelot @FinanceLancelot]

Anthropic's annualized revenue run rate was just leaked, showing it peaked back in July 2026.

That month sounds familiar doesn't it? 🤨
That's exactly when the psyop Jacob Coxon was deployed across all mainstream media, scaring everyone that AI would destroy the world.

Now we know why AI companies are panicking. Their revenue has peaked but their costs keep rising.

They've run out of funding because flat revenue = no more "investment" loans, yet their cash burn rate is still skrocketing.

They need a taxpayer bailout to continue spending a.k.a. "regulation".

[zerohedge @zerohedge]

*NVIDIA IN TALKS TO ACQUIRE REFLECTION AI: FT

Jensen going all-in open models.

[DogroseMK @dogrosemk]

Except gpu everything else that manner for Local AI/SI 😂 is on par with 3090! 3090 is the last piece of perfect hw that was shipped towards masses by acident. 🤷‍♂️

[Rino🚀 @RinoTheBouncer]

RUMOR: NVIDIA is claimed to be end RTX 5090 production and replace it RTX 5080 16GB with 24GB🚀

✅GB202 GPUs will no longer be used for new RTX 5090 production

✅Newly produced GB202 chips will be allocated to RTX PRO lineup

✅This affects standard RTX 5090 along with its Chinese variants

✅NVIDIA to reserve GB202 for professional products, leaving GeForce lineup without a GB202-based model in active production

✅ RTX 5080 24GB will replace the current 16GB version and become the highest-end GeForce card

Source: @Zed__Wang & @hongxing2020 via @VideoCardz