Is embeddinggemma-300m or LFM2.5-1.2B-Instruct 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 2.6x smaller, because of how its attention layers cache.

Model comparison

embeddinggemma-300m vs LFM2.5-1.2B-Instruct

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

From the file· summed bytes, KV per layer

Side by side

	embeddinggemma-300m	LFM2.5-1.2B-Instruct
Parameters	303M	1.2B
Architecture	gemma-embedding	lfm2
Layers	24	16
Native context	2,048	128,000
Mixture of experts	no	no
Quantizations published	10	23
Smallest quantization	0.26 GiB	0.45 GiB
Q4_K_M	—	0.68 GiB
Licence	—	other

KV cache by context

the term that decides long-context viability

Context	embeddinggemma-300m	LFM2.5-1.2B-Instruct	Ratio
4,096	0.04 GiB	0.05 GiB	1.33×
8,192	0.05 GiB	0.09 GiB	1.85×
16,384	0.08 GiB	0.19 GiB	2.29×
32,768	0.14 GiB	0.38 GiB	2.59×
65,536	0.27 GiB	0.75 GiB	2.78×
131,072	0.52 GiB	1.50 GiB	2.89×