“I have a PDF with 20 pages and need to generate at least 50 multiple-choice questions with 12 columns each (including questions, correct answer, and options a, b, c, d etc…). Using the OpenAI API with GPT models such as GPT-3.5 (model 1106), I can only generate 10 questions with their options, despite having a larger context length available. How can I efficiently generate the desired number of questions and options?”
the prompt is:
def generate_mcqs(self):
try:
text = self.load_documents()
if not text:
return None
parser = JsonOutputParser(pydantic_object=Mcq)
model = “gpt-4-turbo-2024-04-09” # Or
#model = “gpt-3.5-turbo-1106”
#model = ‘gpt-4-turbo’
model = ChatOpenAI(model=model, temperature=0)
system_message= '''Generate multiple-choice questions
Each MCQ should include four options, with one correct answer where “None of the above” and “all of the above” should not be an option.
Also, provide an explanation for why the chosen answer is correct.
Each question should be categorized according to the following criteria:
a) Difficulty Level: Easy, Medium, or Hard.
b) Complexity Level: P1 (single knowledge point), P2 (two knowledge points), or P3 (three or more knowledge points).
c) bloom's taxonomy level: Understand, Apply, Analyze, etc.
d) Divide and create three questions of P1, P2 and P3 each where P1 means only one learning outcome, P2 means > 1 and <=2 learning outcomes and Knowledge points clubbed together
and P3 means > 2 and <=3 learning outcomes and knowledge points clubbed together.
e) Create questions in the Indian context.
f) Include multiple distractors that reflect common mistakes, or misconceptions. Make sure that each option tests a different aspect of understanding, and all of them look plausible to increase the complexity.
g) Questions should test teacher's analytical thinking, computational thinking, and logical thinking skills along with the knowledge so add such components in each question
h) Be smart in asking questions and try not to repeat concepts in questions
f) Must integrate multiple knowledge points and learning outcomes into a single problem use outcomes from a given list to increase the complexity of the question for at least 60% of questions out of total questions
The output must be formatted in JSON, including fields for the question, options (a, b, c, d), correct answer, answer explanation, complexity level, bloom's taxonomy level, and difficulty level.
Provide the assessment in the following JSON format:
{
"question": "",
"option_a": "",
"option_b": "",
"option_c": "",
"option_d": "",
"correct_answer": "",
"answer_explanation": "",
"bloom_taxonomy_level": "",
"complexity_level": "",
"difficulty_level": ""
}
'''
chat_template = ChatPromptTemplate.from_messages(
[
SystemMessage(
content=(
system_message
)
),
HumanMessagePromptTemplate.from_template("You must Generate {num_questions} multiple-choice questions for grade: {grade} and subject: {subject} using {text}"),
]
)
print("......Model is processing.........")
chain = chat_template | model | parser
with get_openai_callback() as cb:
results = chain.invoke({"num_questions":self.num_questions,"grade":self.grade, "subject":self.subject, "text": text})
print(cb)
return results
except Exception as e:
print(f"Error generating MCQs: {e}")
return None
input is pdf
developement: Langchain>>
