Model comparison
gemma-4-12B-it-qat-q4_0-unquantized vs Llama-3.2-8X3B-MOE-Dark-Champion-Instruct-uncensored-abliterated-18.4B
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 1.4× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| gemma-4-12B-it-qat-q4_0-unquantized | Llama-3.2-8X3B-MOE-Dark-Champion-Instruct-uncensored-abliterated-18.4B | |
|---|---|---|
| Parameters | 12.0B | 18.4B |
| Architecture | gemma4 | llama |
| Layers | 48 | 28 |
| Native context | 262,144 | 131,072 |
| Mixture of experts | no | yes, 8 experts |
| Quantizations published | 2 | 31 |
| Smallest quantization | 6.50 GiB | 6.56 GiB |
| Q4_K_M | — | 10.54 GiB |
| Licence | apache-2.0 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | gemma-4-12B-it-qat-q4_0-unquantized | Llama-3.2-8X3B-MOE-Dark-Champion-Instruct-uncensored-abliterated-18.4B | Ratio |
|---|---|---|---|
| 4,096 | 0.72 GiB | 0.44 GiB | 1.64× |
| 8,192 | 0.97 GiB | 0.88 GiB | 1.11× |
| 16,384 | 1.47 GiB | 1.75 GiB | 1.19× |
| 32,768 | 2.47 GiB | 3.50 GiB | 1.42× |
| 65,536 | 4.47 GiB | 7.00 GiB | 1.57× |
| 131,072 | 8.47 GiB | 14.00 GiB | 1.65× |