YellowjacketGames (Ben Kelly)

liked a dataset 1 day ago

nvidia/simready-assets

Viewer • Updated about 8 hours ago • 222 • 393k • 510

liked a dataset 3 days ago

scenario-labs/3d-collection

Viewer • Updated 7 days ago • 1.34k • 20.3k • 272

replied to danielhanchen's post 8 days ago

If you were to build a hardware rig to serve this at Q8, what would you rack up? It seems like "a ton of 96gb blackwells" is the answer for the uber-rich, but I'm wondering what other configurations might get the job done.

liked a model 25 days ago

BadToBest/EchoMimicV3

2B • Updated Jan 23 • 56

upvoted a paper about 1 month ago

UniMaia: Steering Chess Policies with Language for Human-like Play

Paper • 2605.27767 • Published May 26 • 2

liked a model about 1 month ago

shermansiu/maia2-blitz

Other • 23.3M • Updated Jul 23 • 27 • 2

New activity in Quazim0t0/neural-x86-doom 2 months ago

What sorcery is this?

1

#1 opened 2 months ago by

YellowjacketGames

liked a Space 2 months ago

Neural DOOM

💀

5

Play Doom in your browser with a neural‑verified CPU

New activity in Edge0/Audio8-TTS-Preview-0.6b 2 months ago

We use Fishaudio already and are curious about this

4

#6 opened 3 months ago by

YellowjacketGames

upvoted a collection 2 months ago

Harness Science & Governed Self-Evolution

Collection

An agent's harness as a measurable, evolvable object: ablation with CIs, governed self-evolution, and a Bayesian fitness surrogate. • 5 items • Updated Jun 13 • 2

liked 2 models 2 months ago

ggml-org/gemma-4-12B-GGUF

Any-to-Any • 12B • Updated 8 days ago • 1.3k • 5

nivvis/Qwen3.5-9B-EQ-v5.1-GGUF

Text Generation • 9B • Updated Apr 17 • 876 • 3

liked 2 models 3 months ago

unsloth/Qwen3.5-4B-GGUF

Image-Text-to-Text • 4B • Updated Mar 2 • 1.24M • 517

microsoft/Fara1.5-27B

Image-Text-to-Text • 27B • Updated Jul 27 • 2.51k • • 297

updated a collection 3 months ago

[models] RTX a6000 48gb

Collection

Models that run well on a *standalone* RTX a6000's 48gb of VRAM. • 17 items • Updated Jul 31 • 1

replied to danielhanchen's post 3 months ago

That's almost exactly what I'm estimating based on Deepseek R1 pulling about 1 or 2 TPS. Offloading a few layers to some A6000s didnt really help much either.

replied to danielhanchen's post 3 months ago

I have enough DDR4 to put this on a 64 core threadripper, but I'm worried about how brutally slow it will be :)

liked a Space 3 months ago

Chess LLM

♟

77

Play chess between two AI models

New activity in UofTCSSLab/Maia3-79M 4 months ago

why AGPL?

#1 opened 4 months ago by

YellowjacketGames

upvoted a collection 4 months ago

Maia3

Collection

Maia-3 is the state-of-the-art in human chess move-matching accuracy across skill levels. • 8 items • Updated Jun 23 • 10

Ben Kelly PRO

AI & ML interests

Recent Activity