Model comparison

Trinity-Nano-Preview vs gemma-4-E4B-it

At Q4_K_M, Trinity-Nano-Preview is the smaller download — 3,786,957,088 bytes against 4,977,171,584.

From the file· summed bytes, KV per layer

Side by side

Trinity-Nano-Previewgemma-4-E4B-it
Parameters6.1B8.0B
Architectureafmoegemma4
Layers5642
Native context131,072131,072
Mixture of expertsyes, 128 expertsno
Quantizations published5036
Smallest quantization1.94 GiB3.30 GiB
Q4_K_M3.53 GiB4.64 GiB
Licenceapache-2.0apache-2.0

KV cache by context

the term that decides long-context viability
ContextTrinity-Nano-Previewgemma-4-E4B-itRatio
4,0960.16 GiB0.12 GiB1.28×
8,1920.21 GiB0.18 GiB1.19×
16,3840.32 GiB0.29 GiB1.12×
32,7680.54 GiB0.51 GiB1.07×
65,5360.98 GiB0.94 GiB1.04×
131,0721.85 GiB1.82 GiB1.02×