Is LFM2.5-Embedding-350M or embeddinggemma-300m-qat-q8_0-unquantized smaller?

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

Which uses less memory at long context?

embeddinggemma-300m-qat-q8_0-unquantized — its KV cache at 32K context is 2.6x smaller, because of how its attention layers cache.

Model comparison

LFM2.5-Embedding-350M vs embeddinggemma-300m-qat-q8_0-unquantized

These two publish different quantization sets; the table below has the exact sizes. At long context the gap widens: embeddinggemma-300m-qat-q8_0-unquantized'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

	LFM2.5-Embedding-350M	embeddinggemma-300m-qat-q8_0-unquantized
Parameters	354M	303M
Architecture	lfm2	gemma-embedding
Layers	16	24
Native context	128,000	2,048
Mixture of experts	no	no
Quantizations published	23	1
Smallest quantization	0.15 GiB	0.31 GiB
Q4_K_M	0.21 GiB	—
Licence	other	gemma

KV cache by context

the term that decides long-context viability

Context	LFM2.5-Embedding-350M	embeddinggemma-300m-qat-q8_0-unquantized	Ratio
4,096	0.05 GiB	0.04 GiB	1.33×
8,192	0.09 GiB	0.05 GiB	1.85×
16,384	0.19 GiB	0.08 GiB	2.29×
32,768	0.38 GiB	0.14 GiB	2.59×
65,536	0.75 GiB	0.27 GiB	2.78×
131,072	1.50 GiB	0.52 GiB	2.89×