Model comparison
L3.2-Rogue-Creative-Instruct-Uncensored-Abliterated-7B 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 3.4× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| L3.2-Rogue-Creative-Instruct-Uncensored-Abliterated-7B | gemma-4-12B-it-qat-q4_0-unquantized | |
|---|---|---|
| Parameters | 7.5B | 12.0B |
| Architecture | llama | gemma4 |
| Layers | 67 | 48 |
| Native context | 131,072 | 262,144 |
| Mixture of experts | no | no |
| Quantizations published | 10 | 2 |
| Smallest quantization | 2.73 GiB | 6.50 GiB |
| Q4_K_M | 4.27 GiB | — |
| Licence | apache-2.0 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | L3.2-Rogue-Creative-Instruct-Uncensored-Abliterated-7B | gemma-4-12B-it-qat-q4_0-unquantized | Ratio |
|---|---|---|---|
| 4,096 | 1.05 GiB | 0.72 GiB | 1.46× |
| 8,192 | 2.09 GiB | 0.97 GiB | 2.16× |
| 16,384 | 4.19 GiB | 1.47 GiB | 2.85× |
| 32,768 | 8.38 GiB | 2.47 GiB | 3.39× |
| 65,536 | 16.75 GiB | 4.47 GiB | 3.75× |
| 131,072 | 33.50 GiB | 8.47 GiB | 3.96× |