Model comparison

SmolVLM-256M-Instruct vs embeddinggemma-300m

These two publish different quantization sets; the table below has the exact sizes. At long context the gap widens: embeddinggemma-300m's KV cache at 32K is 4.9× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

SmolVLM-256M-Instructembeddinggemma-300m
Parameters256M303M
Architecturellamagemma-embedding
Layers3024
Native context8,1922,048
Mixture of expertsnono
Quantizations published210
Smallest quantization0.16 GiB0.26 GiB
Q4_K_M
Licenceapache-2.0

KV cache by context

the term that decides long-context viability
ContextSmolVLM-256M-Instructembeddinggemma-300mRatio
4,0960.09 GiB0.04 GiB2.50×
8,1920.18 GiB0.05 GiB3.46×
16,3840.35 GiB0.08 GiB4.29×
32,7680.70 GiB0.14 GiB4.86×
65,5361.41 GiB0.27 GiB5.22×
131,0722.81 GiB0.52 GiB5.41×