Advice on dual MI50 32gb build
I just installed Ubuntu 24.04 and rock 6.3.3 after using windows and vulkan for a long time.
Now I wanna install Hermes and llama cpp
But llama cpp has a lot of forks so I don’t know which one to use. I know there are some that are specifically for gfx 906 which is the vega 20 gpu
I’m interested in vllm since I have two of my mi50s and tensor parallelism but I hear horror stories in how hard it is to set up
I’m planning on primarily using the llm to power a Hermes agent for general use, basic research, and if possible a dnd game.
I greatly welcome any and all suggestions no matter how basic since I think I have a lot of knowledge gaps.
Thanks everyone!