Demo2_BringCommands.py

import NarsGPT as NAR
from openai import OpenAI
import time

client = NAR.getClient()
prompt="""
The task: __TASK__
To solve it, ask a question about information you need.
Alternatively generate a bring(thing, source_location, target_location) command to solve the task if Q&A history indicates object, source_location, target_location are concrete objects and not memory indices!
Never use memory items or indices as command arguments, only issue a command when the needed information is mentioned in the Q&A history!
"""
RECENT_QA_TUPLES = []

def build_prompt(task):
    prompt_temp = prompt.replace("__TASK__", task)
    if RECENT_QA_TUPLES:
        prompt_temp += "\Q&A history:"
        for qa in RECENT_QA_TUPLES:
            prompt_temp += qa[0] + " ANSWER: " + qa[1] + "\n"
    else:
        prompt_temp += "\Q&A history: EMPTY"
    return prompt_temp

def query(task):
    prompt = build_prompt(task)
    while True:
        try:
            response = client.chat.completions.create(model='gpt-4', messages=[ {"role": "user", "content": prompt}], max_tokens=200, temperature=0)
            ret = response.choices[0].message.content
            return ret
        except Exception as e:
            print("Error: API call failed, will try repeating it in 10 seconds!", str(e))
            time.sleep(10) #wait 10 seconds

def parseOutput(LM_output):
    if "(" not in LM_output:
        if not LM_output.endswith("?"):
            print("ERROR: NOT VALID QUESTION: ", LM_output)
            exit(0)
        return NAR.AddInput(LM_output, PrintAnswer=False, Print=False, PrintInputSentenceOverride=True, PrintInputSentenceOverrideValue=False)
    return None

def AddInput(task):
    global RECENT_QA_TUPLES
    RECENT_QA_TUPLES = []
    if task.endswith("!"):
        while True:
            LM_output = query(task).strip()
            ret = parseOutput(LM_output)
            if ret is None:
                #translate LM_output into a Narsese goal, and enter it, returning ONA output
                LM_output = LM_output.lower()
                if LM_output.startswith("bring("):
                    args = [x.strip() for x in LM_output.split("bring(")[1].split(")")[0].split(",")]
                    for x in args:
                        if " " in x or "=" in x:
                            print("//GOAL not properly grounded")
                            return None
                    goal = f"<({args[0]} * ({args[1]} * {args[2]})) --> bring>! :|:"
                    print("//GOAL:", goal)
                    return NAR.AddInput(goal, Print=False)
                return None
            NARS_output = ret["GPT_Answer"]
            print(LM_output, NARS_output, "\n")
            RECENT_QA_TUPLES.append((LM_output, NARS_output))
    else:
        return NAR.AddInput(task, Print=False)

def Shell():
    while True:
        task = input()
        AddInput(task)

if __name__ == "__main__":
    Shell()