Is Tinyllama-2-1b-miniguanaco or Llama-3.2-1B-Instruct smaller?

Tinyllama-2-1b-miniguanaco is smaller at Q4_K_M: 0.62 GiB against 0.75 GiB.

Which uses less memory at long context?

Tinyllama-2-1b-miniguanaco — its KV cache at 32K context is 1.5x smaller, because of how its attention layers cache.

Model comparison

Tinyllama-2-1b-miniguanaco vs Llama-3.2-1B-Instruct

At Q4_K_M, Tinyllama-2-1b-miniguanaco is the smaller download — 667,822,976 bytes against 807,690,688. At long context the gap widens: Tinyllama-2-1b-miniguanaco'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

	Tinyllama-2-1b-miniguanaco	Llama-3.2-1B-Instruct
Parameters	1.1B	1.2B
Architecture	llama	llama
Layers	22	16
Native context	2,048	131,072
Mixture of experts	no	no
Quantizations published	12	39
Smallest quantization	0.45 GiB	0.39 GiB
Q4_K_M	0.62 GiB	0.75 GiB
Licence	other	—

KV cache by context

the term that decides long-context viability

Context	Tinyllama-2-1b-miniguanaco	Llama-3.2-1B-Instruct	Ratio
4,096	0.09 GiB	0.13 GiB	1.45×
8,192	0.17 GiB	0.25 GiB	1.45×
16,384	0.34 GiB	0.50 GiB	1.45×
32,768	0.69 GiB	1.00 GiB	1.45×
65,536	1.38 GiB	2.00 GiB	1.45×
131,072	2.75 GiB	4.00 GiB	1.45×