Model comparison

stable-diffusion-3.5-medium vs Krea-2-Raw

At Q4_K_M, stable-diffusion-3.5-medium is the smaller download — 1,787,064,768 bytes against 7,486,289,184.

From the file· summed bytes, KV per layer

Side by side

stable-diffusion-3.5-mediumKrea-2-Raw
Parameters2.5B12.8B
Architecturekrea2
Layers
Native context
Mixture of expertsnono
Quantizations published2129
Smallest quantization1.35 GiB2.34 GiB
Q4_K_M1.66 GiB6.97 GiB
Licenceotherother

KV cache by context

the term that decides long-context viability
Contextstable-diffusion-3.5-mediumKrea-2-RawRatio
4,096
8,192
16,384
32,768
65,536
131,072