Is granite-embedding-107m-multilingual or tinygemma3_cifar smaller?

Both publish quantizations; compare the table for the exact sizes.

Which uses less memory at long context?

tinygemma3_cifar — its KV cache at 32K context is 2.3x smaller, because of how its attention layers cache.

Model comparison

granite-embedding-107m-multilingual vs tinygemma3_cifar

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 2.3× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	granite-embedding-107m-multilingual	tinygemma3_cifar
Parameters	107M	39M
Architecture	bert	gemma3
Layers	6	8
Native context	514	131,072
Mixture of experts	no	no
Quantizations published	19	1
Smallest quantization	0.11 GiB	0.04 GiB
Q4_K_M	0.11 GiB	—
Licence	apache-2.0	wtfpl

KV cache by context

the term that decides long-context viability

Context	granite-embedding-107m-multilingual	tinygemma3_cifar	Ratio
4,096	0.04 GiB	0.06 GiB	1.78×
8,192	0.07 GiB	0.08 GiB	1.10×
16,384	0.14 GiB	0.09 GiB	1.52×
32,768	0.28 GiB	0.12 GiB	2.27×
65,536	0.56 GiB	0.19 GiB	3.02×
131,072	1.13 GiB	0.31 GiB	3.61×