Is AMD-OLMo-1B-SFT-DPO or Llama-3.2-1B-Instruct smaller?

AMD-OLMo-1B-SFT-DPO is smaller at Q4_K_M: 0.68 GiB against 0.75 GiB.

Which uses less memory at long context?

Llama-3.2-1B-Instruct — its KV cache at 32K context is 4.0x smaller, because of how its attention layers cache.

Model comparison

AMD-OLMo-1B-SFT-DPO vs Llama-3.2-1B-Instruct

At Q4_K_M, AMD-OLMo-1B-SFT-DPO is the smaller download — 733,520,640 bytes against 807,690,688. At long context the gap widens: Llama-3.2-1B-Instruct's KV cache at 32K is 4.0× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	AMD-OLMo-1B-SFT-DPO	Llama-3.2-1B-Instruct
Parameters	1.2B	1.2B
Architecture	olmo	llama
Layers	16	16
Native context	2,048	131,072
Mixture of experts	no	no
Quantizations published	18	39
Smallest quantization	0.53 GiB	0.39 GiB
Q4_K_M	0.68 GiB	0.75 GiB
Licence	apache-2.0	—

KV cache by context

the term that decides long-context viability

Context	AMD-OLMo-1B-SFT-DPO	Llama-3.2-1B-Instruct	Ratio
4,096	0.50 GiB	0.13 GiB	4.00×
8,192	1.00 GiB	0.25 GiB	4.00×
16,384	2.00 GiB	0.50 GiB	4.00×
32,768	4.00 GiB	1.00 GiB	4.00×
65,536	8.00 GiB	2.00 GiB	4.00×
131,072	16.00 GiB	4.00 GiB	4.00×