Is Llama-3.2-1B-Instruct-heretic 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 6.9x smaller, because of how its attention layers cache.

Model comparison

Llama-3.2-1B-Instruct-heretic 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 6.9× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	Llama-3.2-1B-Instruct-heretic	embeddinggemma-300m
Parameters	1.2B	303M
Architecture	llama	gemma-embedding
Layers	16	24
Native context	131,072	2,048
Mixture of experts	no	no
Quantizations published	36	10
Smallest quantization	0.37 GiB	0.26 GiB
Q4_K_M	0.75 GiB	—
Licence	llama3.2	—

KV cache by context

the term that decides long-context viability

Context	Llama-3.2-1B-Instruct-heretic	embeddinggemma-300m	Ratio
4,096	0.13 GiB	0.04 GiB	3.56×
8,192	0.25 GiB	0.05 GiB	4.92×
16,384	0.50 GiB	0.08 GiB	6.10×
32,768	1.00 GiB	0.14 GiB	6.92×
65,536	2.00 GiB	0.27 GiB	7.42×
131,072	4.00 GiB	0.52 GiB	7.70×