Using Padding for ASR models

Hi! I have recently started using ASR models and found huggingface to be very helpful. However, I have a query related to the Wav2Vec2 processor. I found that the generated output is padded but I can not find the position of eos_token in the generated text ids. Is the token not used if we use padding?

I have used the WavLM and executed the demo code provided in the link: WavLM