Model comparison
NVIDIA-Nemotron-Nano-9B-v2 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: gemma-4-12B-it-qat-q4_0-unquantized's KV cache at 32K is 2.8× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| NVIDIA-Nemotron-Nano-9B-v2 | gemma-4-12B-it-qat-q4_0-unquantized | |
|---|---|---|
| Parameters | 8.9B | 12.0B |
| Architecture | nemotron_h | gemma4 |
| Layers | 56 | 48 |
| Native context | 131,072 | 262,144 |
| Mixture of experts | no | no |
| Quantizations published | 34 | 2 |
| Smallest quantization | 4.62 GiB | 6.50 GiB |
| Q4_K_M | 6.08 GiB | — |
| Licence | — | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | NVIDIA-Nemotron-Nano-9B-v2 | gemma-4-12B-it-qat-q4_0-unquantized | Ratio |
|---|---|---|---|
| 4,096 | 0.88 GiB | 0.72 GiB | 1.22× |
| 8,192 | 1.75 GiB | 0.97 GiB | 1.81× |
| 16,384 | 3.50 GiB | 1.47 GiB | 2.38× |
| 32,768 | 7.00 GiB | 2.47 GiB | 2.84× |
| 65,536 | 14.00 GiB | 4.47 GiB | 3.13× |
| 131,072 | 28.00 GiB | 8.47 GiB | 3.31× |