Model comparison
Mellum2-12B-A2.5B-Instruct vs gemma-4-12B-it-qat-q4_0-unquantized
These two publish different quantization sets; the table below has the exact sizes. At long context the gap widens: Mellum2-12B-A2.5B-Instruct's KV cache at 32K is 4.9× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| Mellum2-12B-A2.5B-Instruct | gemma-4-12B-it-qat-q4_0-unquantized | |
|---|---|---|
| Parameters | 12.1B | 12.0B |
| Architecture | mellum | gemma4 |
| Layers | 28 | 48 |
| Native context | 131,072 | 262,144 |
| Mixture of experts | yes, 64 experts | no |
| Quantizations published | 2 | 2 |
| Smallest quantization | 7.52 GiB | 6.50 GiB |
| Q4_K_M | 7.52 GiB | — |
| Licence | apache-2.0 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | Mellum2-12B-A2.5B-Instruct | gemma-4-12B-it-qat-q4_0-unquantized | Ratio |
|---|---|---|---|
| 4,096 | 0.12 GiB | 0.72 GiB | 6.18× |
| 8,192 | 0.17 GiB | 0.97 GiB | 5.67× |
| 16,384 | 0.28 GiB | 1.47 GiB | 5.24× |
| 32,768 | 0.50 GiB | 2.47 GiB | 4.95× |
| 65,536 | 0.94 GiB | 4.47 GiB | 4.77× |
| 131,072 | 1.81 GiB | 8.47 GiB | 4.67× |