Model comparison
Qwen2.5-VL-72B-Instruct vs Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTP
At Q4_K_M, Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTP is the smaller download — 36,545,830,912 bytes against 47,415,713,248. At long context the gap widens: Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTP's KV cache at 32K is 5.0× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| Qwen2.5-VL-72B-Instruct | Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTP | |
|---|---|---|
| Parameters | 73.4B | 27.8B |
| Architecture | qwen2vl | qwen35 |
| Layers | 80 | 64 |
| Native context | 128,000 | 262,144 |
| Mixture of experts | no | no |
| Quantizations published | 45 | 34 |
| Smallest quantization | 22.11 GiB | 10.12 GiB |
| Q4_K_M | 44.16 GiB | 34.04 GiB |
| Licence | other | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | Qwen2.5-VL-72B-Instruct | Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTP | Ratio |
|---|---|---|---|
| 4,096 | 1.25 GiB | 0.25 GiB | 5.00× |
| 8,192 | 2.50 GiB | 0.50 GiB | 5.00× |
| 16,384 | 5.00 GiB | 1.00 GiB | 5.00× |
| 32,768 | 10.00 GiB | 2.00 GiB | 5.00× |
| 65,536 | 20.00 GiB | 4.00 GiB | 5.00× |
| 131,072 | 40.00 GiB | 8.00 GiB | 5.00× |