Model comparison
gemma-4-12B-it-qat-q4_0-unquantized vs WizardLM-Uncensored-SuperCOT-StoryTelling-30b
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 19.7× 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 | WizardLM-Uncensored-SuperCOT-StoryTelling-30b | |
|---|---|---|
| Parameters | 12.0B | 32.5B |
| Architecture | gemma4 | llama |
| Layers | 48 | 60 |
| Native context | 262,144 | 2,048 |
| Mixture of experts | no | no |
| Quantizations published | 2 | 12 |
| Smallest quantization | 6.50 GiB | 12.58 GiB |
| Q4_K_M | — | 18.27 GiB |
| Licence | apache-2.0 | other |
KV cache by context
the term that decides long-context viability
| Context | gemma-4-12B-it-qat-q4_0-unquantized | WizardLM-Uncensored-SuperCOT-StoryTelling-30b | Ratio |
|---|---|---|---|
| 4,096 | 0.72 GiB | 6.09 GiB | 8.48× |
| 8,192 | 0.97 GiB | 12.19 GiB | 12.58× |
| 16,384 | 1.47 GiB | 24.38 GiB | 16.60× |
| 32,768 | 2.47 GiB | 48.75 GiB | 19.75× |
| 65,536 | 4.47 GiB | 97.50 GiB | 21.82× |
| 131,072 | 8.47 GiB | 195.00 GiB | 23.03× |