Home

Awesome

⛓️🛠️ ChainForge

An open-source visual programming environment for battle-testing prompts to LLMs.

<img width="1517" alt="banner" src="https://github.com/ianarawjo/ChainForge/assets/5251713/570879ef-ef8a-4e00-b37c-b49bc3c1a370">

ChainForge is a data flow prompt engineering environment for analyzing and evaluating LLM responses. It is geared towards early-stage, quick-and-dirty exploration of prompts, chat responses, and response quality that goes beyond ad-hoc chatting with individual LLMs. With ChainForge, you can:

Read the docs to learn more. ChainForge comes with a number of example evaluation flows to give you a sense of what's possible, including 188 example flows generated from benchmarks in OpenAI evals.

This is an open beta of Chainforge. We support model providers OpenAI, HuggingFace, Anthropic, Google PaLM2, Azure OpenAI endpoints, and Dalai-hosted models Alpaca and Llama. You can change the exact model and individual model settings. Visualization nodes support numeric and boolean evaluation metrics. Try it and let us know what you think! :)

ChainForge is built on ReactFlow and Flask.

Table of Contents

Installation

You can install ChainForge locally, or try it out on the web at https://chainforge.ai/play/. The web version of ChainForge has a limited feature set. In a locally installed version you can load API keys automatically from environment variables, write Python code to evaluate LLM responses, or query locally-run Alpaca/Llama models hosted via Dalai.

To install Chainforge on your machine, make sure you have Python 3.8 or higher, then run

pip install chainforge

Once installed, do

chainforge serve

Open localhost:8000 in a Google Chrome, Firefox, Microsoft Edge, or Brave browser.

You can set your API keys by clicking the Settings icon in the top-right corner. If you prefer to not worry about this everytime you open ChainForge, we recommend that save your OpenAI, Anthropic, Google PaLM API keys and/or Amazon AWS credentials to your local environment. For more details, see the How to Install.

Run using Docker

You can use our Dockerfile to run ChainForge locally using Docker Desktop:

Now you can open the browser of your choice and open http://127.0.0.1:8000.

Supported providers

Example experiments

We've prepared a couple example flows to give you a sense of what's possible with Chainforge. Click the "Example Flows" button on the top-right corner and select one. Here is a basic comparison example, plotting the length of responses across different models and arguments for the prompt parameter {game}:

<img width="1593" alt="basic-compare" src="https://github.com/ianarawjo/ChainForge/assets/5251713/43c87ab7-aabd-41ba-8d9b-e7e9ebe25c75">

You can also conduct ground truth evaluations using Tabular Data nodes. For instance, we can compare each LLM's ability to answer math problems by comparing each response to the expected answer:

<img width="1775" alt="Screen Shot 2023-07-04 at 9 21 50 AM" src="https://github.com/ianarawjo/ChainForge/assets/5251713/6d842f7a-f747-44f9-b317-95bec73653c5">

Compare responses across models and prompts

Compare across models and prompt variables with an interactive response inspector, including a formatted table and exportable data:

<img width="1460" alt="Screen Shot 2023-07-19 at 5 03 55 PM" src="https://github.com/ianarawjo/ChainForge/assets/5251713/6aca2bd7-7820-4256-9e8b-3a87795f3e50">

Here's a tutorial to get started comparing across prompt templates.

Share with others

The web version of ChainForge (https://chainforge.ai/play/) includes a Share button.

Simply click Share to generate a unique link for your flow and copy it to your clipboard:

ezgif-2-a4d8048bba

For instance, here's a experiment I made that tries to get an LLM to reveal a secret key: https://chainforge.ai/play/?f=28puvwc788bog

Note To prevent abuse, you can only share up to 10 flows at a time, and each flow must be <5MB after compression. If you share more than 10 flows, the oldest link will break, so make sure to always Export important flows to cforge files, and use Share to only pass data ephemerally.

For finer details about the features of specific nodes, check out the List of Nodes.

Features

A key goal of ChainForge is facilitating comparison and evaluation of prompts and models. Basic features are:

Taken together, these features let you easily:

We've also found that some users simply want to use ChainForge to make tons of parametrized queries to LLMs (e.g., chaining prompt templates into prompt templates), possibly score them, and then output the results to a spreadsheet (Excel xlsx). To do this, attach an Inspect node to the output of a Prompt node and click Export Data.

For more specific details, see our documentation.


Development

ChainForge was created by Ian Arawjo, a postdoctoral scholar in Harvard HCI's Glassman Lab with support from the Harvard HCI community. Collaborators include PhD students Priyan Vaithilingam and Chelse Swoopes, Harvard undergraduate Sean Yang, and faculty members Elena Glassman and Martin Wattenberg.

This work was partially funded by the NSF grants IIS-2107391, IIS-2040880, and IIS-1955699. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.

We provide ongoing releases of this tool in the hopes that others find it useful for their projects.

Inspiration and Links

ChainForge is meant to be general-purpose, and is not developed for a specific API or LLM back-end. Our ultimate goal is integration into other tools for the systematic evaluation and auditing of LLMs. We hope to help others who are developing prompt-analysis flows in LLMs, or otherwise auditing LLM outputs. This project was inspired by own our use case, but also shares some comraderie with two related (closed-source) research projects, both led by Sherry Wu:

Unlike these projects, we are focusing on supporting evaluation across prompts, prompt parameters, and models.

How to collaborate?

We welcome open-source collaborators. If you want to report a bug or request a feature, open an Issue. We also encourage users to implement the requested feature / bug fix and submit a Pull Request.


Cite Us

If you use ChainForge for research purposes, or build upon the source code, we ask that you cite our arXiv pre-print in any related publications. The BibTeX you can use is:

@misc{arawjo2023chainforge,
      title={ChainForge: A Visual Toolkit for Prompt Engineering and LLM Hypothesis Testing},
      author={Ian Arawjo and Chelse Swoopes and Priyan Vaithilingam and Martin Wattenberg and Elena Glassman},
      year={2023},
      eprint={2309.09128},
      archivePrefix={arXiv},
      primaryClass={cs.HC}
}

License

ChainForge is released under the MIT License.