Model comparison

JOSIE-2-4B-Preview vs Qwen3-8B

At Q4_K_M, JOSIE-2-4B-Preview is the smaller download — 2,783,447,232 bytes against 5,027,783,488. At long context the gap widens: JOSIE-2-4B-Preview's KV cache at 32K is 4.5× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

JOSIE-2-4B-PreviewQwen3-8B
Parameters4.7B8.2B
Architectureqwen35qwen3
Layers3236
Native context262,14440,960
Mixture of expertsnono
Quantizations published2751
Smallest quantization1.82 GiB2.12 GiB
Q4_K_M2.59 GiB4.68 GiB
Licencemitapache-2.0

KV cache by context

the term that decides long-context viability
ContextJOSIE-2-4B-PreviewQwen3-8BRatio
4,0960.13 GiB0.56 GiB4.50×
8,1920.25 GiB1.13 GiB4.50×
16,3840.50 GiB2.25 GiB4.50×
32,7681.00 GiB4.50 GiB4.50×
65,5362.00 GiB9.00 GiB4.50×
131,0724.00 GiB18.00 GiB4.50×