Is LightOnOCR-2-1B or embeddinggemma-300m smaller?

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

Which uses less memory at long context?

embeddinggemma-300m — its KV cache at 32K context is 24.2x smaller, because of how its attention layers cache.

Model comparison

LightOnOCR-2-1B vs embeddinggemma-300m

These two publish different quantization sets; the table below has the exact sizes. At long context the gap widens: embeddinggemma-300m's KV cache at 32K is 24.2× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	LightOnOCR-2-1B	embeddinggemma-300m
Parameters	1.0B	303M
Architecture	qwen3	gemma-embedding
Layers	28	24
Native context	16,384	2,048
Mixture of experts	no	no
Quantizations published	20	10
Smallest quantization	0.28 GiB	0.26 GiB
Q4_K_M	0.37 GiB	—
Licence	—	—

KV cache by context

the term that decides long-context viability

Context	LightOnOCR-2-1B	embeddinggemma-300m	Ratio
4,096	0.44 GiB	0.04 GiB	12.44×
8,192	0.88 GiB	0.05 GiB	17.23×
16,384	1.75 GiB	0.08 GiB	21.33×
32,768	3.50 GiB	0.14 GiB	24.22×
65,536	7.00 GiB	0.27 GiB	25.97×
131,072	14.00 GiB	0.52 GiB	26.95×