Model comparison

HiDream-I1-Full vs Krea-2-Turbo

At Q4_K_M, Krea-2-Turbo is the smaller download — 7,216,993,376 bytes against 11,481,430,592.

From the file· summed bytes, KV per layer

Side by side

HiDream-I1-FullKrea-2-Turbo
Parameters17.1B12.8B
Architecturepigqwen_image
Layers
Native context
Mixture of expertsnono
Quantizations published3133
Smallest quantization6.11 GiB2.78 GiB
Q4_K_M10.69 GiB6.72 GiB
Licencemitother

KV cache by context

the term that decides long-context viability
ContextHiDream-I1-FullKrea-2-TurboRatio
4,096
8,192
16,384
32,768
65,536
131,072
HiDream-I1-Full vs Krea-2-Turbo — size, memory and hardware fit — ossmodeldb