about silence tokens during inference

[thivux]

i see that the default values for silence_tokens during inference are [1388,1898,131]. my questions:

1. why is there more than one silence token?
2. how do silence_tokens differ from the <SIL> phoneme in vocab.txt?
3. how can i find the silence tokens when training on my own dataset?

[royrs]

@thivux Have figured it out? At least how can one find the silence tokens?