Model comparison
Trinity-Mini vs gemma-4-E4B-it
At Q4_K_M, gemma-4-E4B-it is the smaller download — 4,977,171,584 bytes against 15,823,053,440. At long context the gap widens: gemma-4-E4B-it's KV cache at 32K is 1.2× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| Trinity-Mini | gemma-4-E4B-it | |
|---|---|---|
| Parameters | 26.1B | 8.0B |
| Architecture | afmoe | gemma4 |
| Layers | 32 | 42 |
| Native context | 131,072 | 131,072 |
| Mixture of experts | yes, 128 experts | no |
| Quantizations published | 57 | 36 |
| Smallest quantization | 6.11 GiB | 3.30 GiB |
| Q4_K_M | 14.74 GiB | 4.64 GiB |
| Licence | apache-2.0 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | Trinity-Mini | gemma-4-E4B-it | Ratio |
|---|---|---|---|
| 4,096 | 0.18 GiB | 0.12 GiB | 1.46× |
| 8,192 | 0.24 GiB | 0.18 GiB | 1.36× |
| 16,384 | 0.37 GiB | 0.29 GiB | 1.28× |
| 32,768 | 0.62 GiB | 0.51 GiB | 1.22× |
| 65,536 | 1.12 GiB | 0.94 GiB | 1.18× |
| 131,072 | 2.12 GiB | 1.82 GiB | 1.16× |