Model comparison

ERNIE-Image vs Krea-2-Turbo

At Q4_K_M, ERNIE-Image is the smaller download — 2,146,489,664 bytes against 7,216,993,376.

From the file· summed bytes, KV per layer

Side by side

ERNIE-ImageKrea-2-Turbo
Parameters8.0B12.8B
Architecturewanqwen_image
Layers
Native context
Mixture of expertsnono
Quantizations published2933
Smallest quantization2.00 GiB2.78 GiB
Q4_K_M2.00 GiB6.72 GiB
Licenceapache-2.0other

KV cache by context

the term that decides long-context viability
ContextERNIE-ImageKrea-2-TurboRatio
4,096
8,192
16,384
32,768
65,536
131,072