About raw prompt display option in OntoGPT command line

[yy20716]

Hello @caufieldjh and @cmungall,

I'm reaching out to inquire whether ontoGPT could potentially offer additional command line options to display the raw prompt that's being passed to the language models. It appears that certain portions of the prompts are stored in the output yaml file, as evident in this example. However, it's uncertain if this represents the complete prompt.

I also noticed the existence of the "--show-prompt" option in the cli.py file, though it seems to be specifically related to SPINDOCTOR, if I've interpreted it correctly. Could you kindly clarify whether such an option already exists? Your assistance is greatly appreciated. Thank you for your valuable support.

[caufieldjh]

Hi @yy20716 - you're right, the output generally truncates the text of the prompt or omits the text input entirely, working under the assumption that the output contains the text in its own field. The recursion in the extraction approach also means that while this is the initial prompt in the above spaghetti example:

From the text below, extract the following entities in the following format:

label: <the name of the recipe>
description: <a brief textual description of the recipe>
categories: <a semicolon separated list of the categories to which this recipe belongs>
ingredients: <a semicolon separated list of the ingredients plus quantities of the recipe>
steps: <a semicolon separated list of the individual steps involved in this recipe>


Text:
SIMPLE SPAGHETTI

DIRECTIONS
On medium heat melt the butter and sautee the onion and bell peppers.
Add the hamburger meat and cook until meat is well done.
Add the tomato sauce, salt, pepper and garlic powder.
Salt, pepper and garlic powder can be adjusted to your own tastes.
Cook noodles as directed.
Mix the sauce and noodles if you like, I keep them separated.

INGREDIENTS
UNITS: US
1
small onion (chopped)
1
bell pepper (chopped)
2
tablespoons garlic powder
3
tablespoons butter
1
teaspoon salt
1
teaspoon pepper
2
(15 ounce) cans tomato sauce
1
(16 ounce) box spaghetti noodles
1 - 1 1⁄2
lb hamburger meat.

This prompt is then also created and queried:

Split the following piece of text into fields in the following format:

food_item: <the food item>
amount: <the quantity of the ingredient, e.g. 2 lbs>


Text:
1 small onion (chopped)

then this:

Split the following piece of text into fields in the following format:

food: <the food item>
state: <the state of the food item (e.g. chopped, diced)>


Text:
small onion

and this

Split the following piece of text into fields in the following format:

value: <the value of the quantity>
unit: <the unit of the quantity, e.g. grams, cups, etc.>


Text:
1

and so on.

That being said, it would certainly be useful to see all those prompts, so I'll add the --show-prompt option to other CLI commands.

[caufieldjh]

OK, the --show-prompt can now be used with most CLI commands, though it writes the prompt out to the logger, so it will need to be used with the verbosity setting like -vvv.