utils/summarizer.py

import ast
import openai
from transformers import GPT2Tokenizer

# Initialize tokenizer
tokenizer = GPT2Tokenizer.from_pretrained("gpt2")

# Prompt engineering
def get_prompt(text):
    # prompt_prefix = """Generate exactly 3 different and thought provoking discussion questions about given article below, and return the answers of these questions with the evidence.
    
    # Desired output format: [{"Q":<question>,"A":<answer>},{"Q":<question>,"A":<answer>},{"Q":<question>,"A":<answer>}].
    # """ 
    prompt_prefix = """Generate exactly 3 different and thought provoking discussion questions about given article below, and return the answers of these questions with the evidence. 
    
    Desired output should be a markdown format like this: 

    ## Q1: <question>

   <answer>

    ## Q2: <question>

   <answer>   

    ## Q3: <question>

   <answer> 

    """     
    prompt_postfix ="""
    Given article content: \"""{}.\"""
    """
    prompt = prompt_prefix + prompt_postfix.format(text)
    return prompt

def limit_tokens(text, n=3000):
    # Get the first n tokens from the input text
    input_ids = tokenizer.encode(text, return_tensors="pt")
    first_n_tokens = input_ids[:, :n]
    # Convert the first n tokens back to text format
    processed_text = tokenizer.decode(first_n_tokens[0], skip_special_tokens=True)    
    return processed_text


# Chat completion
def get_openai_chatcompletion(text):
    """Get OpenAI Chat Completion result.
    """
    messages = []
    processed_text = limit_tokens(text)
    augmented_prompt = get_prompt(processed_text)
    messages.append({"role":"user","content": augmented_prompt})
    
    try:
        result = openai.ChatCompletion.create(
            model="gpt-3.5-turbo",
            messages=messages,
            temperature=0.7
        )
    except:
        raise
    return result


def get_analyze(result):
    try:
        # analyze = ast.literal_eval(result["choices"][0]['text'])
        # analyze = eval(result["choices"][0]['text'])
        # analyze = result["choices"][0]['text']
        analyze = result["choices"][0]["message"]["content"]
    except:
        raise    
    return analyze


def get_analyze_result(text):
    result = get_openai_chatcompletion(text)
    analyze = get_analyze(result)
    return analyze