Max

以DeepSeek-V3-Chat-multi-gpu-4.yaml为例 打开官方注释文档即可(后面的配置都不要改,官方文档里提到先配置的优先), 大概每层要6个G(比如 layers 3–4 就是12个G左右),看自己情况加层级即可 另外,仔细阅读一下,官方文档 https://kvcache-ai.github.io/ktransformers/en/injection_tutorial.html ``` # === MLP Experts Replacement === # replace with marlin expert. Open and modify layer-num as needed. # Each layer of...