in tokenizers/src/decoders/wordpiece.rs [47:61]
fn decode_chain(&self, mut tokens: Vec<String>) -> Result<Vec<String>> {
for (i, token) in tokens.iter_mut().enumerate() {
if i != 0 {
if let Some(tk) = token.strip_prefix(&self.prefix) {
*token = tk.to_string();
} else {
*token = format!(" {token}");
}
}
if self.cleanup {
*token = cleanup(token);
}
}
Ok(tokens)
}