Krex

[Krex @krexillo]

@Morpheu29440305 @bscholl the x goes at the end, NvidiaX

[Krex @krexillo]

@semichenkko @AIMind_Ai thats the point of using ai isnt it? geting things straight

[Krex @krexillo]

@backtoqueue @theo why fable high when fable low is already better then opus?

[Krex @krexillo]

@davidvgilmore can this be used inside claude code for autorouting?

[Krex @krexillo]

@shinboson they are giving 30hours of 32vram gpu on kaggle just by confirming your phone number

[Krex @krexillo]

@satellitedown now do this
and lets see what crazy amounts of tps you get lol

https://t.co/MVTDR4tomz

[𝞍 Shin Megami Boson 𝞍 @shinboson]

I updated my quantization compiler to use e-means as a replacement for k-means in codebook generation and got Qwen3.5-4B from 8.41 GB to 1.23 GB while retaining 91.9% of original performance which I think is approximately SOTA, at least comparable to bonsai 2 but I think we could go lower if someone wanted to donate some GPU time since this is just a hobby for me.

at this size, it's small enough that memory bandwidth isn't the bottleneck to decode speed so I'm interested in seeing how fast we can get larger models on consumer hardware. link below.

[Krex @krexillo]

@haider1 where is low effort on cost per task?

[Krex @krexillo]

@Machinedelusion 80gb gpu to run this btw

[Krex @krexillo]

@DeemosTech beautiful

[Krex @krexillo]

@TheRealPOJ @SafwanAlghazal this would be good for gamers but not for the usual customer who wants an ez ui

[Krex @krexillo]

@kevinace talk with @HerbalistChief nature heal bro

[Krex @krexillo]

@ajaeger74 @sfxnz right now the power is on retail.

treat them bad and they going to change your saas for some chinese option.

[Krex @krexillo]

@aisearchio whats the point of this when personaplex exists?

[Krex @krexillo]

@romankhrupa @liquidai why not yolo? i think is way lighter

[Krex @krexillo]

@painn_x and they made it fable 5.2 instead of 5.5 just to int openai hahaha

[Krex @krexillo]

@DeemosTech @UnrealEngine how do you make sure it doesnt distort the mesh?

[Krex @krexillo]

@TensorSlay yes something like that

[Krex @krexillo]

@Da7_Tech what model can this be compared to?

[Krex @krexillo]

@morganlinton @davidasinclair thanks