Is granite-4.0-7B-A1B-Creative-v0.1 or Llama-3.1-8B-Instruct smaller?

granite-4.0-7B-A1B-Creative-v0.1 is smaller at Q4_K_M: 3.77 GiB against 4.58 GiB.

Which uses less memory at long context?

granite-4.0-7B-A1B-Creative-v0.1 — its KV cache at 32K context is 16.0x smaller, because of how its attention layers cache.

Model comparison

granite-4.0-7B-A1B-Creative-v0.1 vs Llama-3.1-8B-Instruct

At Q4_K_M, granite-4.0-7B-A1B-Creative-v0.1 is the smaller download — 4,049,883,328 bytes against 4,920,739,168. At long context the gap widens: granite-4.0-7B-A1B-Creative-v0.1's KV cache at 32K is 16.0× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	granite-4.0-7B-A1B-Creative-v0.1	Llama-3.1-8B-Instruct
Parameters	6.7B	8.0B
Architecture	granitehybrid	llama
Layers	40	32
Native context	131,072	131,072
Mixture of experts	yes, 62 experts	no
Quantizations published	36	45
Smallest quantization	1.29 GiB	2.02 GiB
Q4_K_M	3.77 GiB	4.58 GiB
Licence	apache-2.0	llama3.1

KV cache by context

the term that decides long-context viability

Context	granite-4.0-7B-A1B-Creative-v0.1	Llama-3.1-8B-Instruct	Ratio
4,096	0.03 GiB	0.50 GiB	15.96×
8,192	0.06 GiB	1.00 GiB	15.98×
16,384	0.13 GiB	2.00 GiB	15.99×
32,768	0.25 GiB	4.00 GiB	16.00×
65,536	0.50 GiB	8.00 GiB	16.00×
131,072	1.00 GiB	16.00 GiB	16.00×