initial_prompt #7
Description
Firstly, thank you for the tremendous work on the Whisper TRT project.
I have been exploring the Whisper model and had a couple of questions:
initial_prompt Usage: Is there a way to use initial_prompt similar to the original Whisper model? This feature is incredibly useful for improving decoding in certain contexts, and I am keen to know if it can be leveraged in the Whisper TRT version as well.
Improving Decoding for Unique Spellings: If initial_prompt are not supported, do you have any advice on which models or algorithms could be implemented to enhance decoding, particularly for unique or uncommon spellings? Any guidance or recommendations would be greatly appreciated.
Thank you for your time and assistance.