Model comparison
Qwen3.8-27B vs gemma-4-19B-A4B-it-The-DECKARD-Heretic-Uncensored-Thinking
At Q4_K_M, gemma-4-19B-A4B-it-The-DECKARD-Heretic-Uncensored-Thinking is the smaller download — 12,291,399,456 bytes against 16,810,714,336. At long context the gap widens: gemma-4-19B-A4B-it-The-DECKARD-Heretic-Uncensored-Thinking's KV cache at 32K is 1.3× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| Qwen3.8-27B | gemma-4-19B-A4B-it-The-DECKARD-Heretic-Uncensored-Thinking | |
|---|---|---|
| Parameters | 27.8B | 19.0B |
| Architecture | qwen35 | gemma4 |
| Layers | 64 | 30 |
| Native context | 262,144 | 262,144 |
| Mixture of experts | no | yes, 90 experts |
| Quantizations published | 22 | 35 |
| Smallest quantization | 8.39 GiB | 5.70 GiB |
| Q4_K_M | 15.66 GiB | 11.45 GiB |
| Licence | apache-2.0 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | Qwen3.8-27B | gemma-4-19B-A4B-it-The-DECKARD-Heretic-Uncensored-Thinking | Ratio |
|---|---|---|---|
| 4,096 | 0.25 GiB | 0.45 GiB | 1.80× |
| 8,192 | 0.50 GiB | 0.61 GiB | 1.21× |
| 16,384 | 1.00 GiB | 0.92 GiB | 1.09× |
| 32,768 | 2.00 GiB | 1.54 GiB | 1.30× |
| 65,536 | 4.00 GiB | 2.79 GiB | 1.43× |
| 131,072 | 8.00 GiB | 5.29 GiB | 1.51× |