AMD

[bennmann]

If you can give me the step by step timing commands, I can help update your table with AMD 6900 XT data soonish; I already have some RWKV running on AmD.

AMD has come a long way

[ArEnSc]

We did a very course grain calculation we put the timer at the start of the context generation, at the end of it, and then calculated the number of tokens generated per second. This is through the interface too, we didn't even get to the root of the code. Ideally I should write a tiny wrapper to do timing but that's what we did!