Product Recommender Model Based on LLM

Overview

This repository contains work for a research project exploring product recommender models using fine-tuned LLMs.

Directory Structure

• data/: Directory storing formatted data used by the model.

  • video/: Movie and TV review datasets.
    • preprocessed_movies.json:
    • embed_movies.py: Python script for embedding movies.
    • test_query.ipynb: Jupyter notebook for testing queries against the movie database.
    • chroma_db/: Contains the sqlite3 database for Chroma.

• src/: Python files and notebooks for conducting experiments and building models.

  • preprocess_data.ipynb: Jupyter notebook for pre-processing the data. It combines metadata and reviews into a single JSON file.
  • local_call.py: Python script for evaluating local models.
  • openai_call.py: Python script for evaluating OpenAI models.
  • request_test.ipynb: Jupyter notebook for testing LLM queries and chains
  • tools/: Directory containing utility scripts.
    • local_llm_chains.py: Python script defining chains for local models.
    • openai_chains.py: Python script defining chains for OpenAI models
    • utils.py: Python script with various utility functions.

• results/: Directory where experiment results are saved.

  • Mixtral-8x7B-Instruct-v0.1/: Contains outputs and evaluations for the Mixtral-8x7B-Instruct-v0.1 model.
  • gpt-3.5-turbo/: Contains outputs and evaluations for the gpt-3.5-turbo model.
  • gpt-4-1106-preview/: Contains outputs for the gpt-4-1106-preview model.

• figures/: Directory for generating and saving figures and plots.

  • plot_evals.py: Script for creating bar plots of eval files