Model comparison

Delphi-25B-SimpleRL-Math vs gemma-4-E4B-it

At Q4_K_M, gemma-4-E4B-it is the smaller download — 4,977,171,584 bytes against 15,160,698,624. At long context the gap widens: gemma-4-E4B-it's KV cache at 32K is 66.2× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

Delphi-25B-SimpleRL-Mathgemma-4-E4B-it
Parameters25.0B8.0B
Architectureqwen3gemma4
Layers5142
Native context16,384131,072
Mixture of expertsnono
Quantizations published3436
Smallest quantization5.31 GiB3.30 GiB
Q4_K_M14.12 GiB4.64 GiB
Licenceapache-2.0

KV cache by context

the term that decides long-context viability
ContextDelphi-25B-SimpleRL-Mathgemma-4-E4B-itRatio
4,0964.18 GiB0.12 GiB34.00×
8,1928.37 GiB0.18 GiB47.08×
16,38416.73 GiB0.29 GiB58.29×
32,76833.47 GiB0.51 GiB66.16×
65,53666.94 GiB0.94 GiB70.96×
131,072133.88 GiB1.82 GiB73.62×