Is Hy-MT2-1.8B or Llama-3.1-8B-Instruct smaller?

Hy-MT2-1.8B is smaller at Q4_K_M: 1.06 GiB against 4.58 GiB.

Which uses less memory at long context?

Hy-MT2-1.8B — its KV cache at 32K context is 2.0x smaller, because of how its attention layers cache.

Model comparison

Hy-MT2-1.8B vs Llama-3.1-8B-Instruct

At Q4_K_M, Hy-MT2-1.8B is the smaller download — 1,133,080,448 bytes against 4,920,739,168. At long context the gap widens: Hy-MT2-1.8B's KV cache at 32K is 2.0× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	Hy-MT2-1.8B	Llama-3.1-8B-Instruct
Parameters	2.0B	8.0B
Architecture	hunyuan-dense	llama
Layers	32	32
Native context	262,144	131,072
Mixture of experts	no	no
Quantizations published	3	45
Smallest quantization	1.06 GiB	2.02 GiB
Q4_K_M	1.06 GiB	4.58 GiB
Licence	apache-2.0	llama3.1

KV cache by context

the term that decides long-context viability

Context	Hy-MT2-1.8B	Llama-3.1-8B-Instruct	Ratio
4,096	0.25 GiB	0.50 GiB	2.00×
8,192	0.50 GiB	1.00 GiB	2.00×
16,384	1.00 GiB	2.00 GiB	2.00×
32,768	2.00 GiB	4.00 GiB	2.00×
65,536	4.00 GiB	8.00 GiB	2.00×
131,072	8.00 GiB	16.00 GiB	2.00×