Model comparison
translategemma-4b-it vs Qwen3-VL-8B-Instruct-abliterated-v1
At Q4_K_M, translategemma-4b-it is the smaller download — 2,489,909,312 bytes against 5,027,785,792.
From the file· summed bytes, KV per layer
Side by side
| translategemma-4b-it | Qwen3-VL-8B-Instruct-abliterated-v1 | |
|---|---|---|
| Parameters | 5.0B | 8.8B |
| Architecture | gemma3 | qwen3vl |
| Layers | — | — |
| Native context | — | — |
| Mixture of experts | no | no |
| Quantizations published | 19 | 48 |
| Smallest quantization | 1.61 GiB | 1.97 GiB |
| Q4_K_M | 2.32 GiB | 4.68 GiB |
| Licence | gemma | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | translategemma-4b-it | Qwen3-VL-8B-Instruct-abliterated-v1 | Ratio |
|---|---|---|---|
| 4,096 | — | — | — |
| 8,192 | — | — | — |
| 16,384 | — | — | — |
| 32,768 | — | — | — |
| 65,536 | — | — | — |
| 131,072 | — | — | — |