Model comparison

ERNIE-Image vs Z-Image-Turbo

At Q4_K_M, ERNIE-Image is the smaller download — 2,146,489,664 bytes against 4,981,532,736.

From the file· summed bytes, KV per layer

Side by side

ERNIE-ImageZ-Image-Turbo
Parameters8.0B6.2B
Architecturewanlumina2
Layers
Native context
Mixture of expertsnono
Quantizations published2930
Smallest quantization2.00 GiB2.41 GiB
Q4_K_M2.00 GiB4.64 GiB
Licenceapache-2.0apache-2.0

KV cache by context

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