Is Gemma-2-9B-It-SPPO-Iter3 or llama-3-youko-8b smaller?

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

Which uses less memory at long context?

llama-3-youko-8b — its KV cache at 32K context is 1.5x smaller, because of how its attention layers cache.

Model comparison

Gemma-2-9B-It-SPPO-Iter3 vs llama-3-youko-8b

These two publish different quantization sets; the table below has the exact sizes. At long context the gap widens: llama-3-youko-8b'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

	Gemma-2-9B-It-SPPO-Iter3	llama-3-youko-8b
Parameters	9.2B	8.0B
Architecture	gemma2	llama
Layers	42	32
Native context	8,192	8,192
Mixture of experts	no	no
Quantizations published	67	2
Smallest quantization	2.22 GiB	5.34 GiB
Q4_K_M	5.37 GiB	—
Licence	gemma	llama3

KV cache by context

the term that decides long-context viability

Context	Gemma-2-9B-It-SPPO-Iter3	llama-3-youko-8b	Ratio
4,096	1.31 GiB	0.50 GiB	2.63×
8,192	2.05 GiB	1.00 GiB	2.05×
16,384	3.36 GiB	2.00 GiB	1.68×
32,768	5.99 GiB	4.00 GiB	1.50×
65,536	11.24 GiB	8.00 GiB	1.40×
131,072	21.74 GiB	16.00 GiB	1.36×