Spaces:
Running
What about add the `temperature` option to the WebUI?
Motivation
I'm a translator and I often rely on AI models to transcribe audio footages. And I often need to transcribe a single segment of audio for multiple times, in order to get the most correct result.
What about add the temperature parameter on the WebUI? Setting the temperature to non-0 will make the whisper model select token randomly from the logit instead of choose the token with highest probability. Sometimes the highest one is not the correct one, so that by running the model for multiple times, it can give me the correct result by chance.
PoC
I've made a pull request for this feature: #6.
I've added the "temperature" options to the WebUI now, along with every other option in the "Full" version of the interface. You can also specify an initial prompt in that version.
Thank you so much for your amazing work! Now I can tune almost every single parameter of the Whisper model.