Model comparison
Luna-7B-A4B vs Qwen3-VL-8B-Instruct-abliterated-v1
At Q4_K_M, Luna-7B-A4B is the smaller download — 4,126,489,312 bytes against 5,027,785,792.
From the file· summed bytes, KV per layer
Side by side
| Luna-7B-A4B | Qwen3-VL-8B-Instruct-abliterated-v1 | |
|---|---|---|
| Parameters | 6.7B | 8.8B |
| Architecture | qwen3moe | qwen3vl |
| Layers | 36 | — |
| Native context | 262,144 | — |
| Mixture of experts | yes, 2 experts | no |
| Quantizations published | 36 | 48 |
| Smallest quantization | 1.49 GiB | 1.97 GiB |
| Q4_K_M | 3.84 GiB | 4.68 GiB |
| Licence | mit | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | Luna-7B-A4B | Qwen3-VL-8B-Instruct-abliterated-v1 | Ratio |
|---|---|---|---|
| 4,096 | 0.56 GiB | — | — |
| 8,192 | 1.13 GiB | — | — |
| 16,384 | 2.25 GiB | — | — |
| 32,768 | 4.50 GiB | — | — |
| 65,536 | 9.00 GiB | — | — |
| 131,072 | 18.00 GiB | — | — |