Is embeddinggemma-300m-qat-q8_0-unquantized or bge-m3 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 20.8x smaller, because of how its attention layers cache.

Model comparison

embeddinggemma-300m-qat-q8_0-unquantized vs bge-m3

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

From the file· summed bytes, KV per layer

Side by side

	embeddinggemma-300m-qat-q8_0-unquantized	bge-m3
Parameters	303M	567M
Architecture	gemma-embedding	bert
Layers	24	24
Native context	2,048	8,194
Mixture of experts	no	no
Quantizations published	1	1
Smallest quantization	0.31 GiB	0.59 GiB
Q4_K_M	—	—
Licence	gemma	mit

KV cache by context

the term that decides long-context viability

Context	embeddinggemma-300m-qat-q8_0-unquantized	bge-m3	Ratio
4,096	0.04 GiB	0.38 GiB	10.67×
8,192	0.05 GiB	0.75 GiB	14.77×
16,384	0.08 GiB	1.50 GiB	18.29×
32,768	0.14 GiB	3.00 GiB	20.76×
65,536	0.27 GiB	6.00 GiB	22.26×
131,072	0.52 GiB	12.00 GiB	23.10×