Speed up Whisper Models?

[usernameawesome]

There have been a few discussion posts about optimizing/speeding up Whispercpp models, however I wanted to collect general advice into one topic.

#1653

• Does not help
  • Increasing number of threads
• Helps
  • Compiling with acceleration like cuBLAS or CLBlast
  • Using tiny models
  • Quantizing the models to 4-bit

#611

• Helps
  • Using Windows port

#2662

• Helps
  • Using OpenVino: https://www.intel.com/content/www/us/en/support/articles/000094846/boards-and-kits.html

#1460

• Helps
  • Using appropriate model

Are there any posts I am missing? Are there any other ideas on performance improvement for larger models?