Uncle Brent

[Uncle Brent @RealCrowMagnum]

A Monkey's Tale https://t.co/nXiCIDJfVR

[Uncle Brent @RealCrowMagnum]

@0xBakeer Let me know when you are doing the 2x spark version, I decided to switch qwen3.8-flash to fp8 the way I landed on running 3.6 and 3.8 27b. Something always feels off with the 4 bit quants and kV not at 16 bit for me.

[Uncle Brent @RealCrowMagnum]

@sudoingX What recipe?

[Uncle Brent @RealCrowMagnum]

@0xBakeer Thank you. Mia's is a bit too optimized for speed and high context for my liking, but blaze seems solid so far, and back on the RadixArk that your recipe uses that I had become familiar with.

[Uncle Brent @RealCrowMagnum]

@0xBakeer I'm running your qwen3.8-flash vllm single spark recipe, I really appreciate your work. From your benchmarking would you recommend a different recipe?

[Uncle Brent @RealCrowMagnum]

@chamath *second wife

[Uncle Brent @RealCrowMagnum]

@ritchie_live - Ferris Bueller

[Uncle Brent @RealCrowMagnum]

@QwenDevs Too soon head shake (you know the two I'm talking about)

[Uncle Brent @RealCrowMagnum]

@michaeloburns

[Uncle Brent @RealCrowMagnum]

@Plinz They thought recursive self improvement could be owned, despite it being fundamental

[Uncle Brent @RealCrowMagnum]

@LLMJunky Much easier to solar power a spark

[Uncle Brent @RealCrowMagnum]

@bijanbowen @JulianGoldieSEO @Alibaba_Qwen "This is on me" is the em dash of agents responding to getting called out for their mistakes

[Uncle Brent @RealCrowMagnum]

@steipete @JustSurajan @openclaw Hopefully one of the pushes fixes vllm served model context compaction issue

[Uncle Brent @RealCrowMagnum]

@TheAhmadOsman Quality > Speed

[Uncle Brent @RealCrowMagnum]

@darvasch @UnslothAI @Alibaba_Qwen I'm seeing significant reliability/capability decrease of this model relative to the FP8 running in openclaw, running on gx10 as well, modified eugr recipe. Have you done any evals beyond basic token/second?

[Uncle Brent @RealCrowMagnum]

@richroll I thought he was Mads Mikkelsen

[Uncle Brent @RealCrowMagnum]

Promoting agents to prompt agents that prompt agents...

[Uncle Brent @RealCrowMagnum]

@thdxr People who think they are "logical" have no awareness of where their thoughts come from, get in touch with your intuition folks, and stop getting in its way

[Uncle Brent @RealCrowMagnum]

@websloprat @thdxr Plum sauce of course

[Uncle Brent @RealCrowMagnum]

@michaeloburns YouTube face?

[Uncle Brent @RealCrowMagnum]

@DeepPsycho_HQ @HeidiPriebe1 The gems should be internal for the one on the right, changing to steps is the same thing as the singular goal just on a different scale