Model comparison
Marco-Mini-Instruct vs gemma-4-E4B-it
At Q4_K_M, gemma-4-E4B-it is the smaller download — 4,977,171,584 bytes against 10,505,424,448. At long context the gap widens: gemma-4-E4B-it's KV cache at 32K is 6.9× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| Marco-Mini-Instruct | gemma-4-E4B-it | |
|---|---|---|
| Parameters | 17.3B | 8.0B |
| Architecture | qwen3moe | gemma4 |
| Layers | 28 | 42 |
| Native context | 32,768 | 131,072 |
| Mixture of experts | yes, 256 experts | no |
| Quantizations published | 35 | 36 |
| Smallest quantization | 3.33 GiB | 3.30 GiB |
| Q4_K_M | 9.78 GiB | 4.64 GiB |
| Licence | apache-2.0 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | Marco-Mini-Instruct | gemma-4-E4B-it | Ratio |
|---|---|---|---|
| 4,096 | 0.44 GiB | 0.12 GiB | 3.56× |
| 8,192 | 0.88 GiB | 0.18 GiB | 4.92× |
| 16,384 | 1.75 GiB | 0.29 GiB | 6.10× |
| 32,768 | 3.50 GiB | 0.51 GiB | 6.92× |
| 65,536 | 7.00 GiB | 0.94 GiB | 7.42× |
| 131,072 | 14.00 GiB | 1.82 GiB | 7.70× |