SYLIN

IFM

Maker · 3 entries

Everything from IFM

NameWhat to knowPrice or memoryRanks
K2 HorizonModelReport compares four model sizes trained on the same 22T tokens and describes a decoding adapter claimed to give 2.2x speedup~2 GB
PANModelpromising Open video world model people like, but it needs very large GPUs
K2 Horizon 7B UnoModelDiffusion-adapter LoRA on K2-Horizon-7B for lossless speedups via discrete diffusion; Apache-2.0