Creating a Whisper pipeline from local checkpoint

[DiwakarBasnet]

While creating a whisper huggingface pipeline from local checkpoint, how to save the trainer? Usually we would save model by,
trainer.save_model("path/to/local/dir") but how to save model along with feature extractor and tokenizer for a whisper model, so that a pipeline can be created by just uploading the local checkpoint.

[Atefeh197]

Hi

Did you find any way for that?

[DiwakarBasnet]

Hello, I first saved the WhisperProcessor before training by using processor.save_pretrained("path/to/local/dir") then after the training is completed I saved the trained model by using trainer.save_model("path/to/local/dir"). Then for using it as a pipeline I used pipeline("automatic-speech-recognition", model="path/to/local/dir")