Model comparison
Qwen3-Coder-30B-A3B-Instruct vs Noromaid-20b-v0.1.1
At Q4_K_M, Noromaid-20b-v0.1.1 is the smaller download — 12,042,208,864 bytes against 18,556,689,568. At long context the gap widens: Qwen3-Coder-30B-A3B-Instruct's KV cache at 32K is 12.9× smaller, which usually matters more than the difference in weights.
From the file· summed bytes, KV per layer
Side by side
| Qwen3-Coder-30B-A3B-Instruct | Noromaid-20b-v0.1.1 | |
|---|---|---|
| Parameters | 30.5B | 20.0B |
| Architecture | qwen3moe | llama |
| Layers | 48 | 62 |
| Native context | 262,144 | 4,096 |
| Mixture of experts | yes, 128 experts | no |
| Quantizations published | 46 | 33 |
| Smallest quantization | 7.46 GiB | 4.09 GiB |
| Q4_K_M | 17.28 GiB | 11.22 GiB |
| Licence | apache-2.0 | cc-by-nc-4.0 |
KV cache by context
the term that decides long-context viability
| Context | Qwen3-Coder-30B-A3B-Instruct | Noromaid-20b-v0.1.1 | Ratio |
|---|---|---|---|
| 4,096 | 0.38 GiB | 4.84 GiB | 12.92× |
| 8,192 | 0.75 GiB | 9.69 GiB | 12.92× |
| 16,384 | 1.50 GiB | 19.38 GiB | 12.92× |
| 32,768 | 3.00 GiB | 38.75 GiB | 12.92× |
| 65,536 | 6.00 GiB | 77.50 GiB | 12.92× |
| 131,072 | 12.00 GiB | 155.00 GiB | 12.92× |