Model comparison
Qwen3-Coder-30B-A3B-Instruct vs NSFW_13B_sft
At Q4_K_M, NSFW_13B_sft is the smaller download — 8,560,487,488 bytes against 18,556,689,568. At long context the gap widens: Qwen3-Coder-30B-A3B-Instruct's KV cache at 32K is 8.3× 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 | NSFW_13B_sft | |
|---|---|---|
| Parameters | 30.5B | 13.3B |
| Architecture | qwen3moe | baichuan |
| Layers | 48 | 40 |
| Native context | 262,144 | — |
| Mixture of experts | yes, 128 experts | no |
| Quantizations published | 46 | 13 |
| Smallest quantization | 7.46 GiB | 5.18 GiB |
| Q4_K_M | 17.28 GiB | 7.97 GiB |
| Licence | apache-2.0 | apache-2.0 |
KV cache by context
the term that decides long-context viability
| Context | Qwen3-Coder-30B-A3B-Instruct | NSFW_13B_sft | Ratio |
|---|---|---|---|
| 4,096 | 0.38 GiB | 3.13 GiB | 8.33× |
| 8,192 | 0.75 GiB | 6.25 GiB | 8.33× |
| 16,384 | 1.50 GiB | 12.50 GiB | 8.33× |
| 32,768 | 3.00 GiB | 25.00 GiB | 8.33× |
| 65,536 | 6.00 GiB | 50.00 GiB | 8.33× |
| 131,072 | 12.00 GiB | 100.00 GiB | 8.33× |