Is granite-4.0-micro-base or Llama-3.1-8B-Instruct smaller?

granite-4.0-micro-base is smaller at Q4_K_M: 1.96 GiB against 4.58 GiB.

Which uses less memory at long context?

granite-4.0-micro-base — its KV cache at 32K context is 1.6x smaller, because of how its attention layers cache.

Model comparison

granite-4.0-micro-base vs Llama-3.1-8B-Instruct

At Q4_K_M, granite-4.0-micro-base is the smaller download — 2,099,496,064 bytes against 4,920,739,168. At long context the gap widens: granite-4.0-micro-base's KV cache at 32K is 1.6× smaller, which usually matters more than the difference in weights.

From the file· summed bytes, KV per layer

Side by side

	granite-4.0-micro-base	Llama-3.1-8B-Instruct
Parameters	3.4B	8.0B
Architecture	granite	llama
Layers	40	32
Native context	131,072	131,072
Mixture of experts	no	no
Quantizations published	15	45
Smallest quantization	1.28 GiB	2.02 GiB
Q4_K_M	1.96 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-micro-base	Llama-3.1-8B-Instruct	Ratio
4,096	0.31 GiB	0.50 GiB	1.60×
8,192	0.63 GiB	1.00 GiB	1.60×
16,384	1.25 GiB	2.00 GiB	1.60×
32,768	2.50 GiB	4.00 GiB	1.60×
65,536	5.00 GiB	8.00 GiB	1.60×
131,072	10.00 GiB	16.00 GiB	1.60×