Is Nanbeige4.2-3B-heretic or llama-3-youko-8b smaller?

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

Which uses less memory at long context?

Nanbeige4.2-3B-heretic — its KV cache at 32K context is 1.5x smaller, because of how its attention layers cache.

Model comparison

Nanbeige4.2-3B-heretic vs llama-3-youko-8b

These two publish different quantization sets; the table below has the exact sizes. At long context the gap widens: Nanbeige4.2-3B-heretic's KV cache at 32K is 1.5× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	Nanbeige4.2-3B-heretic	llama-3-youko-8b
Parameters	4.2B	8.0B
Architecture	nanbeige	llama
Layers	22	32
Native context	262,144	8,192
Mixture of experts	no	no
Quantizations published	4	2
Smallest quantization	2.40 GiB	5.34 GiB
Q4_K_M	2.40 GiB	—
Licence	apache-2.0	llama3

KV cache by context

the term that decides long-context viability

Context	Nanbeige4.2-3B-heretic	llama-3-youko-8b	Ratio
4,096	0.34 GiB	0.50 GiB	1.45×
8,192	0.69 GiB	1.00 GiB	1.45×
16,384	1.38 GiB	2.00 GiB	1.45×
32,768	2.75 GiB	4.00 GiB	1.45×
65,536	5.50 GiB	8.00 GiB	1.45×
131,072	11.00 GiB	16.00 GiB	1.45×