Model comparison

CapybaraHermes-2.5-Mistral-7B vs gemma-4-E4B-it

At Q4_K_M, CapybaraHermes-2.5-Mistral-7B is the smaller download — 4,368,450,688 bytes against 4,977,171,584.

From the file· summed bytes, KV per layer

Side by side

CapybaraHermes-2.5-Mistral-7Bgemma-4-E4B-it
Parameters7.2B8.0B
Architecturellamagemma4
Layers3242
Native context32,768131,072
Mixture of expertsnono
Quantizations published1236
Smallest quantization2.53 GiB3.30 GiB
Q4_K_M4.07 GiB4.64 GiB
Licenceapache-2.0apache-2.0

KV cache by context

the term that decides long-context viability
ContextCapybaraHermes-2.5-Mistral-7Bgemma-4-E4B-itRatio
4,0960.12 GiB
8,1920.18 GiB
16,3840.29 GiB
32,7680.51 GiB
65,5360.94 GiB
131,0721.82 GiB