LocalForge
返回指南
MODEL / ALIBABA QWEN

Qwen3 30B-A3B

All 30B parameters must be stored even though roughly 3B are active per token.

一手來源
總參數30B3B active / token
估算記憶體~22.5 GBGGUF Q4 MoE
上下文32K+Text
LicenseApache-2.0100+ languages and dialects
01 / 適合
  • reasoning
  • coding
  • efficient MoE
02 / 估算適配
Apple silicon · 32 GBTightNVIDIA · 24 GBTightApple silicon · 64 GBComfortableWorkstation · 48 GBComfortableDatacenter · 80 GBComfortable
03 / 部署方式
OllamaEasyollama run qwen3:30b-a3b
llama.cppMediumllama-server -m /path/to/verified-model.gguf --port 8080
LM StudioEasySearch “Qwen/Qwen3-30B-A3B” → Load a Q4/INT4 build → Start Server
MLX LMMediummlx_lm.chat --model Qwen/Qwen3-30B-A3B
注意

All 30B parameters must be stored even though roughly 3B are active per token.