project.yml

title: "Prodigy LLM Prompt Tournament on Generating Tweets: Inform or Virality"
description: |
  This project runs a prompt tournament con generating tweets using
  different prompts for given topics and scenarios.
spacy_version: ">=3.5.0,<4.0.0"

# Variables can be referenced across the project.yml using ${vars.var_name}
vars:
  config: "config.cfg"
  name: "tweet_tournament"
  version: "1.0.0"
  topics-prompt: | 
    Generate topics that social media managers create social media content about.
    Choose topics that have the significant economic, social or political impact.

# These are the directories that the project needs. The project CLI will make
# sure that they always exist.
directories: ["assets", "corpus", "scripts", "metrics"]

# Assets that should be downloaded or available in the directory. We're shipping
# them with the project, so they won't have to be downloaded. But the
# 'project assets' command still lets you verify that the checksums match.
assets:
  - dest: "assets/topics.jsonl"
    description: "Topics and scenarios to be generated"
  - dest: "assets/prompt_folder/prompt1.jinja2"
    description: "Prompt instructing to inform the reader. The goal is education."
  - dest: "assets/prompt_folder/prompt2.jinja2"
    description: "Prompt instructing to attract the reader. The goal is virality."

# Workflows are sequences of commands (see below) executed in order. You can
# run them via "spacy project run [workflow]". If a commands's inputs/outputs
# haven't changed, it won't be re-run.
workflows:
  setup:
    - install
  prodigy:
    - prodigy-tournament-tweets

# Project commands, specified in a style similar to CI config files (e.g. Azure
# pipelines). The name is the command name that lets you trigger the command
# via "spacy project run [command] [path]". The help message is optional and
# shown when executing "spacy project run [optional command] [path] --help".
commands:
  - name: "install"
    help: "Install packages"
    script:
      - "python3 -m pip install --upgrade pip"
      - "python3 -m pip install -r requirements.txt"
      - "dotenv run -- python3 -m pip install --pre prodigy -f \"https://xxxx@download.prodi.gy\""

  - name: "clean-files"
    script:
      - "rm -rf data/*"
    help: "Clean all files in data/ folder"

  - name: "clean-venv"
    script:
      - "rm -rf venv"
    help: "Remove the virtual environment"

  - name: "clean-prodigy-tweets"
    script:
      - "python3 -m prodigy drop tweets"
    help: "Drop the tweets dataset from Prodigy"

  - name: "generate-topics"
    script:
      - "dotenv run -- python3 -m prodigy terms.openai.fetch ${vars.topics-prompt} assets/topics.jsonl --n 10"
    help: "Generate topics for tweet generation"

  - name: "prodigy-tweets"
    script:
      - "dotenv run -- python3 -m prodigy ab.openai.prompts tweets assets/topics.jsonl assets/display-template.jinja2 assets/prompt_folder/prompt1.jinja2 assets/prompt_folder/prompt2.jinja2 --repeat 4"
    help: "A/B evaluation of prompts"
    deps:
      - "assets/topics.jsonl"
      - "assets/display-template.jinja2"
      - "assets/prompt_folder/prompt1.jinja2"
      - "assets/prompt_folder/prompt2.jinja2"

  - name: "prodigy-tournament-tweets"
    script:
      - "dotenv run -- python3 -m prodigy ab.openai.tournament tweets-tournament assets/topics.jsonl assets/display-template.jinja2 assets/prompt_folder"
    help: "Prodigy tournament for tweet generation based on topics"
    deps:
      - "assets/topics.jsonl"
      - "assets/display-template.jinja2"
      - "assets/prompt_folder"