dataset.py

# /// script
# dependencies = [
#     "aiohttp",
#     "datasets",
#     "beautifulsoup4",
#     "llms-txt-rs",
#     "tqdm"
# ]
# ///

from dataclasses import dataclass
from urllib.parse import urlparse

import asyncio
import aiohttp
from bs4 import BeautifulSoup
from llms_txt_rs import parse_llms_txt
from datasets import Dataset
from tqdm.asyncio import tqdm_asyncio
import json

# Type alias
ParsedTxt = dict[str, str | None | dict[str, list[dict[str, str | None]]]]


@dataclass
class Page:
    name: str
    url: str | None
    url_full: str | None
    # Data added to the Page object once downloaded
    llms_txt: str | None = None
    llms_txt_parsed: ParsedTxt | None = None
    llms_txt_full: str | None = None

    @property
    def base_url(self) -> str:
        parsed = urlparse(self.url)
        return f"{parsed.scheme}://{parsed.netloc}"

    def add_text(self, txt: str, parsed_txt: ParsedTxt, is_full: bool = False, url: str | None = None) -> None:
        if url and not url.startswith(self.base_url):
            return
        if not is_full:
            self.llms_txt = txt
            self.llms_txt_parsed = parsed_txt
        else:
            self.llms_txt_full = txt

    def record(self) -> dict:
        # This is necessary to ensure the 'llms_txt_parsed' field is saved appropriately,
        # otherwise datasets will mix the fields to make the field homogeneus across the different
        # pages.
        return {
            "name": self.name,
            "url": self.url,
            "url_full": self.url_full,
            "llms_txt": self.llms_txt,
            "llms_txt_parsed": json.dumps(self.llms_txt_parsed),
            "llms_txt_full": self.llms_txt_full,
        }


async def download_directory_llmstxt(url: str = "https://directory.llmstxt.cloud/") -> list[Page]:
    async with aiohttp.ClientSession() as session:
        async with session.get(url) as response:
            content = await response.read()
            soup = BeautifulSoup(content, 'html.parser')

            pages = [parse_row(row) for row in soup.find_all('li')]

    return pages


# Extracts the pages from the html
def parse_row(row: list[str]) -> Page:
    a_elems = row.find_all("a")
    name = a_elems[0].find("h3").text
    url_full = None
    if len(a_elems) > 2:  # contains both url llms.txt and llms-full.txt
        for elem in a_elems[1:]:
            url = elem.get("href")
            if "llms-full.txt" in url:
                url_full = url
    else:
        url = a_elems[1].get("href")

    return Page(
        name=name,
        url=url,
        url_full=url_full
    )


async def download_url(session: aiohttp.ClientSession, page: Page, url: str) -> Page:
    try:
        async with session.get(url) as response:
            if response.status == 200:
                txt = await response.text()
                try:
                    parsed = parse_llms_txt(txt)
                    if "llms.txt" in url:
                        page.add_text(txt, parsed, is_full=False, url=url)
                    elif "llms-full.txt" in url:
                        page.add_text(txt, None, is_full=True, url=url)
                    else:
                        print(f"Unexpected url downloaded from the directory: {url}")
                        return page

                except Exception as e:
                    # No need to do anything, the fields will remain None
                    print(f"Error parsing '{url}': {e}")

    except Exception as e:
        print(f"Error downloading url '{url}': {e}")

    return page


async def process_page(session: aiohttp.ClientSession, page: Page) -> list[Page]:
    if url := page.url:
        page = await download_url(session, page, url)
    if url := page.url_full:
        page = await download_url(session, page, url)

    return page


async def download_pages(pages: list[Page]) -> list[Page]:
    async with aiohttp.ClientSession() as session:
        tasks = [process_page(session, page) for page in pages]
        return await tqdm_asyncio.gather(*tasks, total=len(pages))


async def main():
    print("Downloading directory...")
    pages = await download_directory_llmstxt()

    print("Downloading pages...")
    pages = await download_pages(pages)

    print("Preparing dataset...")
    ds = Dataset.from_list([page.record() for page in pages])
    
    ds.push_to_hub("plaguss/llms-txt", commit_message="Initial commit")


if __name__ == "__main__":
    asyncio.run(main())