is it possible to start from a checkpoint?
ghosalya opened this issue · comments
Hello, is it possible to train this model starting with a checkpoint? e.g. starts with a whisper large-v2 model file?
If not, may I ask if there is a pre-trained model for WhisperBiasing available for public?
Would be a big help if we don't need to train from scratch.
Thank you for your attention.
Sure. The paper is to apply TCPGen to Whisper without fine-tuning any Whisper parameters. You can start with whisper large-v2 (though I only experimented with large) and apply TCPGen to only train the TCPGen parameters.