This entity does not have a full profile yet. Below are related knowledge claims.
RTX 5090上Qwen3.6 35B-A3B推理速度达220 token/s，显存占用约23GB
Qwen3.6 35B-A3B是混合专家架构（MoE），总参数量35B，每次推理实际激活参数量约3B