Model comparison

TinyMistral-248M-v2-Instruct vs tinygemma3_cifar

These two publish different quantization sets; the table below has the exact sizes.

From the file· summed bytes, KV per layer

Side by side

TinyMistral-248M-v2-Instructtinygemma3_cifar
Parameters248M39M
Architecturellamagemma3
Layers128
Native context32,768131,072
Mixture of expertsnono
Quantizations published61
Smallest quantization0.10 GiB0.04 GiB
Q4_K_M0.14 GiB
Licenceapache-2.0wtfpl

KV cache by context

the term that decides long-context viability
ContextTinyMistral-248M-v2-Instructtinygemma3_cifarRatio
4,0960.06 GiB
8,1920.08 GiB
16,3840.09 GiB
32,7680.12 GiB
65,5360.19 GiB
131,0720.31 GiB