In pervious code, the followup template in prompt variations is used to get the grade of AI answer. But this method returns incorrect answer many times. So I changed the followup template to summarize the AI answer and return one of the question options. This works much better.
added an cli tool to get session results locally.
cli usage:
$ python cli.py -h
usage: cli.py [-h] [-j JOBS] [-d] [-a] [-t TMP_DIR]
optional arguments:
-h, --help show this help message and exit
-j JOBS, --jobs JOBS Use how many cpu processes to run
-d, --debug Run in debug mode
-a, --append Append the session result to GSpreadsheet
-t TMP_DIR, --tmp_dir TMP_DIR
Dir to store the cached session results (default to ./output)
NOTE: In my testing the maximum processes we can use without going over OpenAI's rate limit is -j 2
Changes:
cli usage:
NOTE: In my testing the maximum processes we can use without going over OpenAI's rate limit is
-j 2