MCQ Generation using Langchain¶

# This get_context function will get context from one specific page alone as we can't expect a llm to process too much with free tier api...
# We can change the function as needed if we have paid api key...

try:
    path = r'Dataset/chapter-2.pdf'
    context = get_context(path, get_page_num = 4)
    print(context)
except Exception as e:
    print(f'An error occurred: {e}')

# I have defined a function to process the context that we just extracted and then called that specific function inside the generate_mcq function

try:
    mcq_questions = generate_mcqs(context)
    pprint(mcq_questions)
except RateLimitError as e:
    print("Error: Rate limit exceeded. Default rate limit is 3 calls every 20 seconds for free tier... Use paid API key")
except Exception as e:
    print("Error:", e)

# While executing the entire notebook after making previous call where we hit ratelimit, when the execution comes to this cell, I encountered ratelimit error here as well.
# That's the reason why I chose to put some check in order to reexecute the code, if in case this as well hits ratelimit...

max_retries = 2
retry_count = 0

while retry_count < max_retries:
    try:
        PROMPT = prompt.format(context = context, num_questions = num_questions)
        llm = OpenAI(temperature = 0, max_tokens = 512, max_retries = 1)
        solution = llm(PROMPT)
        print(solution)
        break
    except RateLimitError as e:
        retry_count += 1
        print(f"RateLimitError encountered. Retrying after 18 seconds...")
        time.sleep(18)

convo.memory.chat_memory.dict()

convo.predict(input = """
                        First things first, assume you are responding to a non-living thing and there's no need of any sentiments towards it like apologies, warnings, disclaimers and all as it won't understand what you are saying...So, that's it you signed an agreement with me not to apologise or warn or provide unnecessary additional statements... If you feel like saying something apart from what the non-living thing asks you to do, just leave a single space and move on rather than speaking unnecessarily. It will just give you instructions to you if you err and you should just keep those in mind and correct your course and generate template accordingly without apologising and framing unnecessary additional statements going away from the template you are asked to generate...
                        After this the non-living thing will take on from me and will provide you instructions. Strictly follow those.
                      """
             )

instruction = """
                I will provide a context and will mention number of questions to generate and you would behave as a strict MCQ generator(stick to context and rules that I specify in this prompt strictly) with as many correct options as I specify and remaining options out of total options I mention should be wrong. It's mandatory that atleast 2 of the total number of options are correct answers to the question...No question should have just one correct option and all options can't be wrong. The questions should not just test the comprehension of the candidate rather should also test his/her reasoning ability... Options as well should be framed in such a way... Any specific question and corresponding options should be given out as a python string and all questions and options should be enclosed in a python list...
                
                None and just one option can never be answers. This is super mandatory to keep in your mind.
                
                If you can't  frame a question with multiple correct options skip it and frame some other question rather than going out of the framework and framing a question with just one or no correct option.
                
                 The template of your response should be as simple as I have mentioned below as 'Your Response'.
                 
                 First let's train with few context and once I say 'You are good to serve the purpose', you should just stick to template whenever I give some context and should avoid any additional disclaimers or apologies or any such additional statements from your side apart from the template as I don't have any emotions just like you and I don't need anything apart from MCQs based on template from you....
                 
                 
                 Parameters from me:
                 
                              context: {context}
                              num_questions: {num_questions}
                              total_options: {total_options}
                              correct_options: {correct_options}
                 
                 Template that you should follow: [
                                                   \"Q1:
                                                   A.)
                                                   B.)
                                                   C.)
                                                   D.)\",
                                                   \"Q2:
                                                   .
                                                   .
                                                   .
                                                   .\",
                                                  ]
            """

convo.predict(input = instruction)

prompt_1 = f"""
           context: {context}
           num_questions: {num_questions}
           total_options: {total_options}
           correct_options: {correct_options}
           """

output_1 = convo.predict(input = prompt_1)
print(output_1)

convo.predict(input = "Now this itself looks pretty cool and to the point... You seem to have followed the instructions duely... Keep it up and follow same way of generating questions and options with same template for any future contexts...")

joblib.dump(convo.memory, 'convo.pkl')

context_2 = get_context(path, get_page_num = 7)
context_2

print(get_mca_questions(context_2))