Model comparison
Qwen3.6-27B vs ERNIE-4.5-21B-A3B-PT
At Q4_K_M, ERNIE-4.5-21B-A3B-PT is the smaller download — 13,331,015,456 bytes against 16,547,397,888.
From the file· summed bytes, KV per layer
Side by side
| Qwen3.6-27B | ERNIE-4.5-21B-A3B-PT | |
|---|---|---|
| Parameters | 27.8B | 21.9B |
| Architecture | qwen35 | ernie4_5-moe |
| Layers | 64 | 28 |
| Native context | 262,144 | 131,072 |
| Mixture of experts | no | no |
| Quantizations published | 40 | 50 |
| Smallest quantization | 8.74 GiB | 5.91 GiB |
| Q4_K_M | 15.41 GiB | 12.42 GiB |
| Licence | apache-2.0 | — |
KV cache by context
the term that decides long-context viability
| Context | Qwen3.6-27B | ERNIE-4.5-21B-A3B-PT | Ratio |
|---|---|---|---|
| 4,096 | 0.25 GiB | 0.22 GiB | 1.14× |
| 8,192 | 0.50 GiB | 0.44 GiB | 1.14× |
| 16,384 | 1.00 GiB | 0.88 GiB | 1.14× |
| 32,768 | 2.00 GiB | 1.75 GiB | 1.14× |
| 65,536 | 4.00 GiB | 3.50 GiB | 1.14× |
| 131,072 | 8.00 GiB | 7.00 GiB | 1.14× |