Model comparison

NSFW_13B_sft vs Qwen3-30B-A3B-Thinking-2507

At Q4_K_M, NSFW_13B_sft is the smaller download — 8,560,487,488 bytes against 18,556,685,824. At long context the gap widens: Qwen3-30B-A3B-Thinking-2507'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

NSFW_13B_sftQwen3-30B-A3B-Thinking-2507
Parameters13.3B30.5B
Architecturebaichuanqwen3moe
Layers4048
Native context—262,144
Mixture of expertsnoyes, 128 experts
Quantizations published1351
Smallest quantization5.18 GiB7.05 GiB
Q4_K_M7.97 GiB17.28 GiB
Licenceapache-2.0apache-2.0

KV cache by context

the term that decides long-context viability
ContextNSFW_13B_sftQwen3-30B-A3B-Thinking-2507Ratio
4,0963.13 GiB0.38 GiB8.33×
8,1926.25 GiB0.75 GiB8.33×
16,38412.50 GiB1.50 GiB8.33×
32,76825.00 GiB3.00 GiB8.33×
65,53650.00 GiB6.00 GiB8.33×
131,072100.00 GiB12.00 GiB8.33×
NSFW_13B_sft vs Qwen3-30B-A3B-Thinking-2507 — size, memory and hardware fit — ossmodeldb