Model comparison

tinygemma3_cifar vs LFM2.5-230M

These two publish different quantization sets; the table below has the exact sizes. At long context the gap widens: tinygemma3_cifar's KV cache at 32K is 3.0× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

tinygemma3_cifarLFM2.5-230M
Parameters39M230M
Architecturegemma3lfm2
Layers814
Native context131,072128,000
Mixture of expertsnono
Quantizations published125
Smallest quantization0.04 GiB0.10 GiB
Q4_K_M0.14 GiB
Licencewtfplother

KV cache by context

the term that decides long-context viability
Contexttinygemma3_cifarLFM2.5-230MRatio
4,0960.06 GiB0.05 GiB1.33×
8,1920.08 GiB0.09 GiB1.22×
16,3840.09 GiB0.19 GiB2.02×
32,7680.12 GiB0.38 GiB3.02×
65,5360.19 GiB0.75 GiB4.02×
131,0720.31 GiB1.50 GiB4.82×