Home

Awesome

tests PyPI PyPI - Python Version DOI:10.1101/2020.07.15.204701 DOI

<picture> <source media="(prefers-color-scheme: dark)" srcset="assets/chemcrow_dark_bold.png" width='100%'> <source media="(prefers-color-scheme: light)" srcset="assets/chemcrow_light_bold.png" width='100%'> <img alt="ChemCrow logo" src="/assets/" width="100%"> </picture>

<br></br>

ChemCrow is an open source package for the accurate solution of reasoning-intensive chemical tasks.

Built with Langchain, it uses a collection of chemical tools including RDKit, paper-qa, as well as some relevant databases in chemistry, like Pubchem and chem-space.

🤗 Try it out in HuggingFace!

ChemCrow Demo

⚠️ Note

This package does not contain all the tools described in the ChemCrow paper because of API usage restrictions. This repo will not give the same results as that paper.

All the experiments have been released under ChemCrow runs.

👩‍💻 Installation

pip install chemcrow

🔥 Usage

First set up your API keys in your environment.

export OPENAI_API_KEY=your-openai-api-key

You can optionally use Serp API:

export SERP_API_KEY=your-serpapi-api-key

In a Python session:

from chemcrow.agents import ChemCrow

chem_model = ChemCrow(model="gpt-4-0613", temp=0.1, streaming=False)
chem_model.run("What is the molecular weight of tylenol?")

🛠️ Self-hosting of some tools.

By default, ChemCrow relies on the RXN4Chem API for retrosynthetic planning and reaction product prediction. This can however be slow and depends on you having an API key.

Optionally, you can also self host these tools by running some pre-made docker images.

Run

docker run --gpus all -d -p 8051:5000 doncamilom/rxnpred:latest
docker run --gpus all -d -p 8052:5000 doncamilom/retrosynthesis:latest

Now ChemCrow can be used like this:

from chemcrow.agents import ChemCrow

chem_model = ChemCrow(model="gpt-4-0613", temp=0.1, streaming=False, local_rxn=True)
chem_model.run("What is the product of the reaction between styrene and dibromine?")

✅ Citation

Bran, Andres M., et al. "ChemCrow: Augmenting large-language models with chemistry tools." arXiv preprint arXiv:2304.05376 (2023).

@article{bran2023chemcrow,
      title={ChemCrow: Augmenting large-language models with chemistry tools},
      author={Andres M Bran and Sam Cox and Oliver Schilter and Carlo Baldassari and Andrew D White and Philippe Schwaller},
      year={2023},
      eprint={2304.05376},
      archivePrefix={arXiv},
      primaryClass={physics.chem-ph},
      publisher={arXiv}
}