whisper alternatives lately with TensorRT-LLM

[amberpatch25]

Good point. That matches what I saw on M3 Max. Worth rechecking your defaults though, because one bad assumption can make the rest of the thread look confusing.