Model comparison
Apertus-8B-Instruct-2509 vs gemma-4-E4B-it
At Q4_K_M, gemma-4-E4B-it is the smaller download — 4,977,171,584 bytes against 5,057,885,440. At long context the gap widens: gemma-4-E4B-it's KV cache at 32K is 7.9× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| Apertus-8B-Instruct-2509 | gemma-4-E4B-it | |
|---|---|---|
| Parameters | 8.1B | 8.0B |
| Architecture | apertus | gemma4 |
| Layers | 32 | 42 |
| Native context | 65,536 | 131,072 |
| Mixture of experts | no | no |
| Quantizations published | 67 | 36 |
| Smallest quantization | 1.91 GiB | 3.30 GiB |
| Q4_K_M | 4.71 GiB | 4.64 GiB |
| Licence | apache-2.0 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | Apertus-8B-Instruct-2509 | gemma-4-E4B-it | Ratio |
|---|---|---|---|
| 4,096 | 0.50 GiB | 0.12 GiB | 4.06× |
| 8,192 | 1.00 GiB | 0.18 GiB | 5.63× |
| 16,384 | 2.00 GiB | 0.29 GiB | 6.97× |
| 32,768 | 4.00 GiB | 0.51 GiB | 7.91× |
| 65,536 | 8.00 GiB | 0.94 GiB | 8.48× |
| 131,072 | 16.00 GiB | 1.82 GiB | 8.80× |