> training smart LLMs like Qwen3-Next-80B-A3B-Instruct-NVFP4

Sounds interesting; can you suggest any good discussions of this (on the web)?