Questions about example training scripts

[arman-kazemi]

Hi,
I have some questions about parameters in the agentevol training script:

1. Where can we get the training data? iter_data_path="./iter_data/iter_data_all_task"
2. What's the difference between the test step and the inference step?
3. Where can we get the inference data?
   inference_file_list=("webshop.json" "alfworld.json" "textcraft.json" "sciworld.json" "sqlgym.json" "wordle.json" "maze.json" "babyai.json" "weather.json" "movie.json" "todo.json" "academia.json" "sheet.json" "webarena.json")
cur_inference_file=./small_exploration_data/${inference_file_list[$index]}
4. What do I need to change to train a model that does NOT use the Llama 2 Chat template?

Thanks,