Dialog2Graph

Dialog2Graph allows you to effortlessly create chatsky flows and scripts from dialogs using Large Language Models.

Contents

./dialog2graph - source code
./examples - usage scenarios
./experiments - test field for conducting experiments
./prompt_cache - utils for LLM output caching
./scripts - scripts for `poethepoet` automation 

Current Progress

Supported graph types:

• chain
• single cycle
• multi-cycle graph
• complex graph with cycles

Currently unsupported graph types:

• single node cycle

How to Start

Install poetry v. 1.8.4 (detailed installation guide)

pipx install poetry==1.8.4

Clone this repo and install project dependencies

git clone https://github.com/deeppavlov/dialog2graph.git
cd dialog2graph
poetry install

Consider installing PyGraphviz from here, if you are planning to visualize your graphs. Then add it to the poetry environment.

poetry add pygraphviz

Ensure that dependencies were installed correctly by running any Python script

poetry run python <your_file_name>.py

Create .env file to store credentials

Note: never hardcode your personal tokens and other sensitive credentials. Use the .env file to store them.

How to Use

See dialog2graph usage examples:

1. Learn base classes usage

2. Learn dialog2graph pipelines and model configuration

3. Generate dialog graph using LLMs

4. Evaluate graph and dialogs

5. Learn CLI interface

6. Augment dialogs on dialog graph

7. Extend a dialog graph

How to Contribute

See contribution guideline CONTRIBUTING.md