Model comparison
llama2-22b-chat-wizard-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 13.2× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| llama2-22b-chat-wizard-uncensored | gemma-4-12B-it-qat-q4_0-unquantized | |
|---|---|---|
| Parameters | 21.8B | 12.0B |
| Architecture | llama | gemma4 |
| Layers | 40 | 48 |
| Native context | 2,048 | 262,144 |
| Mixture of experts | no | no |
| Quantizations published | 2 | 2 |
| Smallest quantization | 7.56 GiB | 6.50 GiB |
| Q4_K_M | — | — |
| Licence | other | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | llama2-22b-chat-wizard-uncensored | gemma-4-12B-it-qat-q4_0-unquantized | Ratio |
|---|---|---|---|
| 4,096 | 4.06 GiB | 0.72 GiB | 5.65× |
| 8,192 | 8.13 GiB | 0.97 GiB | 8.39× |
| 16,384 | 16.25 GiB | 1.47 GiB | 11.06× |
| 32,768 | 32.50 GiB | 2.47 GiB | 13.16× |
| 65,536 | 65.00 GiB | 4.47 GiB | 14.55× |
| 131,072 | 130.00 GiB | 8.47 GiB | 15.35× |