[rfoo]

Or, you can say, OpenAI has some real technical advancements on stuff besides attn architecture. GQA8, alternating SWA 128 / full attn do all seem conventional. Basically they are showing us that "no secret sauce in model arch you guys just sucks at mid/post-training", or they want us to believe this.

The model is pretty sparse tho, 32:1.

[liuliu]

Kimi K2 paper said that the model sparsity scales up with parameters pretty well (MoE sparsity scaling law, as they call, basically calling Llama 4 MoE "done wrong"). Hence K2 has 128:1 sparsity.

[throwdbaaway]

I thought Kimi K2 uses 8 active experts out of 384? Sparsity should be 48:1. Indeed Llama4 Maverick is the only one that has 128:1 sparsity.

[liuliu]

You are right. I mis-remembered the sparsity part of K2. The "done wrong" part I was thinking about how the scout -> maverick -> behemoth doesn't scale sparsity according to any formula (less sparse -> sparse -> less sparse).

[throwdbaaway]

> how the scout -> maverick -> behemoth doesn't scale sparsity according to any formula (less sparse -> sparse -> less sparse)

Ah I see. I didn't notice that behemoth has the same sparsity as scout. That seems quite random indeed.

[nxobject]

It's convenient to be able to attribute success to things only OpenAI could've done with the combo of their early start and VC money – licensing content, hiring subject matter experts, etc. Essentially the "soft" stuff that a mature organization can do.