Model comparison
v6-Finch-14B-HF 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 12.4× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| v6-Finch-14B-HF | gemma-4-12B-it-qat-q4_0-unquantized | |
|---|---|---|
| Parameters | 14.1B | 12.0B |
| Architecture | rwkv6 | gemma4 |
| Layers | 61 | 48 |
| Native context | — | 262,144 |
| Mixture of experts | no | no |
| Quantizations published | 20 | 2 |
| Smallest quantization | 5.10 GiB | 6.50 GiB |
| Q4_K_M | 8.17 GiB | — |
| Licence | apache-2.0 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | v6-Finch-14B-HF | gemma-4-12B-it-qat-q4_0-unquantized | Ratio |
|---|---|---|---|
| 4,096 | 3.81 GiB | 0.72 GiB | 5.30× |
| 8,192 | 7.63 GiB | 0.97 GiB | 7.87× |
| 16,384 | 15.25 GiB | 1.47 GiB | 10.38× |
| 32,768 | 30.50 GiB | 2.47 GiB | 12.35× |
| 65,536 | 61.00 GiB | 4.47 GiB | 13.65× |
| 131,072 | 122.00 GiB | 8.47 GiB | 14.41× |