from shiny import ui, reactive, render, App from chemcrow.agents import ChemCrow import asyncio import shinyswatch chem_model = ChemCrow(model="gpt-4-0613", temp=0.1, verbose=True) app_ui = ui.page_fluid( shinyswatch.theme.slate(), ui.panel_title("ChemCrow UI"), ui.p("An experiment with Shiny for Python and ChemCrow"), ui.br(), ui.row( ui.column(9, ui.input_text("prompt", label=None, placeholder="E.g., What is the molecular weight of tylenol?", width="100%")), ui.column(3, ui.input_action_button("chat", "Chat", class_="btn btn-primary btn-lg btn-block", width="100%"),) ), ui.output_text("txt"), ui.output_ui("prompt_ui"), ui.output_ui("result"), ui.hr(), ui.div( {"style": "align-items: center; display: flex; flex-direction: column; justify-content: center;"}, ui.img(src="https://github.com/ur-whitelab/chemcrow-public/raw/main/assets/chemcrow_dark_thin.png", width="400px") ), ui.br(), ui.markdown(f'ChemCrow was [introduced](https://arxiv.org/abs/2304.05376) by Bran, Andres M., et al. "ChemCrow: Augmenting large-language models with chemistry tools." arXiv preprint arXiv:2304.05376 (2023). This tool is an extension of that work that puts the code into an interactive web app created by [James Wade](https://jameshwade.com) using [Shiny for Python](https://shiny.posit.co/py/). Find the code for the app [here](https://github.com/jameshwade/chemcrow) and the original code [here](https://github.com/ur-whitelab/chemcrow-public).') ) def server(input, output, session): @reactive.Effect() def _(): if input.chat(): ui.update_text("prompt", value="") @output @render.ui @reactive.event(input.chat) def prompt_ui(): list_ui = [ui.strong("Prompt"), ui.markdown(input.prompt())] return list_ui @output @render.ui @reactive.event(input.chat) # triggered when the "Chat" button is clicked async def result(): ui.notification_show("Chatting with ChemCrow...", type="message") try: response = await asyncio.to_thread(chem_model.run, input.prompt()) list_ui = [ui.strong("Thoughts"), ui.markdown(response[0]), ui.strong("Reasoning"), ui.markdown(response[1]), ui.strong("Answer"), ui.markdown(response[2])] return list_ui except TypeError: async def error_coro(): return "An error occurred while processing your request." return await error_coro() app = App(app_ui, server, debug=True)