Training-free hack to reduce MoE reasoning tokens by 8.5% without losing accuracy but increasing Active parameters per token in MOE (Qwen 35B A4B+) | 入行 365 深度情报

正在加载入行工作台

Training-free hack to reduce MoE reasoning tokens by 8.5% without losing accuracy but increasing Active parameters per token in MOE (Qwen 35B A4B+) | 入行 365 深度情报 | 入行 365 旗下的「入行之路」