Is Chocolatine-3B-Instruct-DPO-Revised 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 3.0x smaller, because of how its attention layers cache.

Model comparison

Chocolatine-3B-Instruct-DPO-Revised 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 3.0× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	Chocolatine-3B-Instruct-DPO-Revised	llama-3-youko-8b
Parameters	3.8B	8.0B
Architecture	phi3	llama
Layers	32	32
Native context	4,096	8,192
Mixture of experts	no	no
Quantizations published	1	2
Smallest quantization	2.23 GiB	5.34 GiB
Q4_K_M	2.23 GiB	—
Licence	mit	llama3

KV cache by context

the term that decides long-context viability

Context	Chocolatine-3B-Instruct-DPO-Revised	llama-3-youko-8b	Ratio
4,096	1.50 GiB	0.50 GiB	3.00×
8,192	3.00 GiB	1.00 GiB	3.00×
16,384	6.00 GiB	2.00 GiB	3.00×
32,768	12.00 GiB	4.00 GiB	3.00×
65,536	24.00 GiB	8.00 GiB	3.00×
131,072	48.00 GiB	16.00 GiB	3.00×