pub const PAD_TOKEN: i32 = 50256;
Token 50256 is used as the padding token, which corresponds to the <|endoftext|> token in the OpenAI GPT-2 encoder.
50256
<|endoftext|>