Is MiniCPM5-1B-Uncensored or Llama-3.2-1B-Instruct smaller?

MiniCPM5-1B-Uncensored is smaller at Q4_K_M: 0.64 GiB against 0.75 GiB.

Which uses less memory at long context?

MiniCPM5-1B-Uncensored — its KV cache at 32K context is 1.3x smaller, because of how its attention layers cache.

Model comparison

MiniCPM5-1B-Uncensored vs Llama-3.2-1B-Instruct

At Q4_K_M, MiniCPM5-1B-Uncensored is the smaller download — 688,066,944 bytes against 807,690,688. At long context the gap widens: MiniCPM5-1B-Uncensored's KV cache at 32K is 1.3× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	MiniCPM5-1B-Uncensored	Llama-3.2-1B-Instruct
Parameters	1.1B	1.2B
Architecture	llama	llama
Layers	24	16
Native context	131,072	131,072
Mixture of experts	no	no
Quantizations published	36	39
Smallest quantization	0.33 GiB	0.39 GiB
Q4_K_M	0.64 GiB	0.75 GiB
Licence	other	—

KV cache by context

the term that decides long-context viability

Context	MiniCPM5-1B-Uncensored	Llama-3.2-1B-Instruct	Ratio
4,096	0.09 GiB	0.13 GiB	1.33×
8,192	0.19 GiB	0.25 GiB	1.33×
16,384	0.38 GiB	0.50 GiB	1.33×
32,768	0.75 GiB	1.00 GiB	1.33×
65,536	1.50 GiB	2.00 GiB	1.33×
131,072	3.00 GiB	4.00 GiB	1.33×