Model comparison
Laguna-XS-2.1 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: Laguna-XS-2.1's KV cache at 32K is 1.8× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| Laguna-XS-2.1 | gemma-4-12B-it-qat-q4_0-unquantized | |
|---|---|---|
| Parameters | 33.4B | 12.0B |
| Architecture | laguna | gemma4 |
| Layers | 40 | 48 |
| Native context | 262,144 | 262,144 |
| Mixture of experts | yes, 256 experts | no |
| Quantizations published | 31 | 2 |
| Smallest quantization | 8.76 GiB | 6.50 GiB |
| Q4_K_M | 18.88 GiB | — |
| Licence | openmdw-1.1 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | Laguna-XS-2.1 | gemma-4-12B-it-qat-q4_0-unquantized | Ratio |
|---|---|---|---|
| 4,096 | 0.27 GiB | 0.72 GiB | 2.63× |
| 8,192 | 0.43 GiB | 0.97 GiB | 2.25× |
| 16,384 | 0.74 GiB | 1.47 GiB | 1.98× |
| 32,768 | 1.37 GiB | 2.47 GiB | 1.81× |
| 65,536 | 2.62 GiB | 4.47 GiB | 1.71× |
| 131,072 | 5.12 GiB | 8.47 GiB | 1.65× |