data split

[xiaoyuxin1002]

Hi, after reading your code, can i check with you the following regarding how you split your dataset?

1. 5 samples are generated from prompt_gen_data to induce the instruction from the open-source LLM
2. 20 samples are generated from eval_data to evaluate the quality of the induced instruction during BO iterations
3. 100 samples are generated from test_data to evaluate the quality of the proposed instruction after BO iterations
   Thanks!

[xiaoyuxin1002]

Hi, it also seems that your data/instruction_induction_raw/induce folder contains 33 datasets while your data/instruction_induction_raw/execute folder contains only 31 datasets.
And Table 2 in your paper didn't include the tuned prompts for word_in_context, but had 2 rows both showing the prompts for negation.
Could you please fix them? Thanks!