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