Model comparison
DeepSeek-V2-Lite-Chat-Uncensored-Unbiased-Reasoner 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: DeepSeek-V2-Lite-Chat-Uncensored-Unbiased-Reasoner's KV cache at 32K is 2.6× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| DeepSeek-V2-Lite-Chat-Uncensored-Unbiased-Reasoner | gemma-4-12B-it-qat-q4_0-unquantized | |
|---|---|---|
| Parameters | 15.7B | 12.0B |
| Architecture | deepseek2 | gemma4 |
| Layers | 27 | 48 |
| Native context | 4,096 | 262,144 |
| Mixture of experts | yes, 64 experts | no |
| Quantizations published | 12 | 2 |
| Smallest quantization | 5.99 GiB | 6.50 GiB |
| Q4_K_M | 9.65 GiB | — |
| Licence | llama3.3 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | DeepSeek-V2-Lite-Chat-Uncensored-Unbiased-Reasoner | gemma-4-12B-it-qat-q4_0-unquantized | Ratio |
|---|---|---|---|
| 4,096 | 0.12 GiB | 0.72 GiB | 6.06× |
| 8,192 | 0.24 GiB | 0.97 GiB | 4.08× |
| 16,384 | 0.47 GiB | 1.47 GiB | 3.09× |
| 32,768 | 0.95 GiB | 2.47 GiB | 2.60× |
| 65,536 | 1.90 GiB | 4.47 GiB | 2.35× |
| 131,072 | 3.80 GiB | 8.47 GiB | 2.23× |