* plain input/output prompt strategy w/o chat templates * disable duplicate code check * make sure to add an eos/eot token to the end of the output so it will stop * multi turn segement support and test
type: llama2_chat