End to End App deployment using Huggingface

To move our analyses outside of our local development environment and create applications that are available to multiple users, we can deploy an application and host it on a public-facing site like Huggingface. Here we will demonstrate this workflow by deploying an app allowing users to upload a file and extract text-based content from it based on user-defined instructions. we will build this using the following packages:

• panel: Provides the interactive UI with a file dropper, text input, and extract button.
• marvin: Handles text extraction using an OpenAI language model (gpt-4o-mini).
• markitdown: Converts uploaded documents into text format.
• Huggingface: Hosts the deployed application.

The app supports text and image files but does not handle videos.

To deploy an app on HuggingFace (shareable & free) use the following steps:

1. first create a new space https://huggingface.co/new-space and select the following options to deploy a Panel app:

1. Then edit the app.py file under files

insert the following code:

import io
import os
import marvin
import io
import os
import marvin
import markitdown
import panel as pn

pn.extension("filedropper")

marvin.settings.openai.chat.completions.model = "gpt-4o-mini"

def convert(value: dict):
    for filename, content in value.items():
        file_extension = os.path.splitext(filename)[1]
        if file_extension in (".png", ".jpg", ".jpeg", ".gif"):
            return marvin.Image(content, format=file_extension.lstrip("."))
        elif file_extension in (".mp4", ".mov", ".avi"):
            raise NotImplementedError("Video files are not supported yet.")
        if isinstance(content, bytes):
            buf = io.BytesIO(content)
        else:
            buf = io.StringIO(content)
        content = mid.convert_stream(buf, file_extension=file_extension)
        return content.text_content


def extract(event):
    content = convert(file_dropper.value)
    instructions = text_input.value
    output = marvin.extract(content, instructions=instructions)
    results.object = output


mid = markitdown.MarkItDown()
file_dropper = pn.widgets.FileDropper(
    multiple=False, height=370, sizing_mode="stretch_width"
)
text_input = pn.widgets.TextInput(
    placeholder="Things to extract",
    visible=file_dropper.param.value.rx().rx.len() > 0,
    sizing_mode="stretch_width",
)
submit_button = pn.widgets.Button(
    name="Extract",
    on_click=extract,
    visible=text_input.param.value.rx().rx.len() > 0,
    sizing_mode="stretch_width",
)
results = pn.pane.Placeholder()
template = pn.template.FastListTemplate(
    main=[file_dropper, text_input, submit_button, results],
    main_max_width="800px",
    main_layout=None,
)
template.servable()

1. Now update requirements.txt which specifies the packages needed

panel
jupyter
markitdown
marvin
openai

1. Lastly, update the secrets to include your private info such as your API keys

---

A deployed version of this app can be found here https://huggingface.co/spaces/ahnaconda/pdf-extraction