* [llama4] fix the mm yaml, add scout single gpu yaml * add README for llama4 * rename to specify fsdp