Is deepseek-coder-1.3b-base or Llama-3.2-3B-Instruct smaller?

deepseek-coder-1.3b-base is smaller at Q4_K_M: 0.81 GiB against 1.88 GiB.

Which uses less memory at long context?

Llama-3.2-3B-Instruct — its KV cache at 32K context is 1.7x smaller, because of how its attention layers cache.

Model comparison

deepseek-coder-1.3b-base vs Llama-3.2-3B-Instruct

At Q4_K_M, deepseek-coder-1.3b-base is the smaller download — 873,453,568 bytes against 2,019,377,440. At long context the gap widens: Llama-3.2-3B-Instruct's KV cache at 32K is 1.7× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	deepseek-coder-1.3b-base	Llama-3.2-3B-Instruct
Parameters	1.3B	3.2B
Architecture	llama	llama
Layers	24	28
Native context	16,384	131,072
Mixture of experts	no	no
Quantizations published	49	39
Smallest quantization	0.40 GiB	0.85 GiB
Q4_K_M	0.81 GiB	1.88 GiB
Licence	other	—

KV cache by context

the term that decides long-context viability

Context	deepseek-coder-1.3b-base	Llama-3.2-3B-Instruct	Ratio
4,096	0.75 GiB	0.44 GiB	1.71×
8,192	1.50 GiB	0.88 GiB	1.71×
16,384	3.00 GiB	1.75 GiB	1.71×
32,768	6.00 GiB	3.50 GiB	1.71×
65,536	12.00 GiB	7.00 GiB	1.71×
131,072	24.00 GiB	14.00 GiB	1.71×