Is bitnet_b1_58-large 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 31.1x smaller, because of how its attention layers cache.

Model comparison

bitnet_b1_58-large 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 31.1× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	bitnet_b1_58-large	embeddinggemma-300m
Parameters	729M	303M
Architecture	bitnet	gemma-embedding
Layers	24	24
Native context	2,048	2,048
Mixture of experts	no	no
Quantizations published	1	10
Smallest quantization	0.20 GiB	0.26 GiB
Q4_K_M	—	—
Licence	mit	—

KV cache by context

the term that decides long-context viability

Context	bitnet_b1_58-large	embeddinggemma-300m	Ratio
4,096	0.56 GiB	0.04 GiB	16.00×
8,192	1.13 GiB	0.05 GiB	22.15×
16,384	2.25 GiB	0.08 GiB	27.43×
32,768	4.50 GiB	0.14 GiB	31.14×
65,536	9.00 GiB	0.27 GiB	33.39×
131,072	18.00 GiB	0.52 GiB	34.65×