understanding input construction and special tokens:

[VikasRajashekar]

I see that you define 5 new tokens. bos eos persona_token speaker1 speaker2.

And the input is constructed like:

input_ids :
<persona_token> (persona sentence1) <persona_token> (persona sentence2) ... <speaker1> (history sentence1) <speaker2> <history sentence2> <bos> response <eos>

token_type_ids:
<persona_token> for all personal sentence +<speaker1> and <speaker2> for respective sentences + <bos> for response

lm_labels:
-1 for tokens except for response

My questions are:

1. Is my understanding correct?
2. in token_type_ids why is bos tag used as a type for all the response tokens?

Attaching the file I used for this analysis.
xyz_new.txt