llm inference is fine on rocm. llama.cpp and vllm both have very good rocm support. llm training is also mostly fine. I have not encountered any issues yet.

llm inference is fine on rocm. llama.cpp and vllm both have very good rocm support. llm training is also mostly fine. I have not encountered any issues yet. most of the cuda moat comes from people who are repeating what they heard 5-10 years ago.