- Python 3.10
python scripts/simulate_student.py \
--input_file data/raw/questions.json \
--output_file data/simulated/questions_stu_answers.json \
--model_name mistralai/Mistral-7B-Instruct-v0.3 \
--num_simulations 5 \
--removal_probability 0.3 \
--number_modification_range 0.2 \
--number_modification_chance 0.5python scripts/generate_llm_feedback.py \
--input_file data/simulated/questions_stu_answers.json \
--output_file data/simulated/questions_llm_feedbacks.json \
--model_names allenai/OLMo-2-1124-7B-Instruct meta-llama/Llama-3.1-8B-Instructpython scripts/generate_ta_feedback.py \
--input_file data/simulated/questions_llm_feedbacks.json \
--output_file data/simulated/questions_ta_feedbacks.json \
--model_name microsoft/Phi-4-mini-instruct \
--max_prompt_tokens 8000Convert the feedback data to DPO training format:
python scripts/convert_to_dpo.py \
--input_file data/simulated/questions_ta_feedbacks.json \
--output_file data/processed/dpo_samples.json