Model comparison
Llama3.2-30B-A3B-II-Dark-Champion-INSTRUCT-Heretic-Abliterated-Uncensored 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 2.4× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| Llama3.2-30B-A3B-II-Dark-Champion-INSTRUCT-Heretic-Abliterated-Uncensored | gemma-4-12B-it-qat-q4_0-unquantized | |
|---|---|---|
| Parameters | 30.0B | 12.0B |
| Architecture | llama | gemma4 |
| Layers | 47 | 48 |
| Native context | 131,072 | 262,144 |
| Mixture of experts | yes, 8 experts | no |
| Quantizations published | 34 | 2 |
| Smallest quantization | 6.07 GiB | 6.50 GiB |
| Q4_K_M | 17.11 GiB | — |
| Licence | apache-2.0 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | Llama3.2-30B-A3B-II-Dark-Champion-INSTRUCT-Heretic-Abliterated-Uncensored | gemma-4-12B-it-qat-q4_0-unquantized | Ratio |
|---|---|---|---|
| 4,096 | 0.73 GiB | 0.72 GiB | 1.02× |
| 8,192 | 1.47 GiB | 0.97 GiB | 1.52× |
| 16,384 | 2.94 GiB | 1.47 GiB | 2.00× |
| 32,768 | 5.88 GiB | 2.47 GiB | 2.38× |
| 65,536 | 11.75 GiB | 4.47 GiB | 2.63× |
| 131,072 | 23.50 GiB | 8.47 GiB | 2.77× |