Async HTML loader and HTML2Text transformer

docs/extras/modules/data_connection/document_loaders/integrations/async_html.ipynb

@@ -0,0 +1,107 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "id": "e229e34c",
+   "metadata": {},
+   "source": [
+    "# AsyncHtmlLoader\n",
+    "\n",
+    "AsyncHtmlLoader loads raw HTML from a list of urls concurrently."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 4,
+   "id": "4c8e4dab",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from langchain.document_loaders import AsyncHtmlLoader"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 5,
+   "id": "e76b5ddc",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Fetching pages: 100%|############| 2/2 [00:00<00:00,  9.96it/s]\n"
+     ]
+    }
+   ],
+   "source": [
+    "urls = [\"https://www.espn.com\", \"https://lilianweng.github.io/posts/2023-06-23-agent/\"]\n",
+    "loader = AsyncHtmlLoader(urls)\n",
+    "docs = loader.load()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 6,
+   "id": "5dca1c0c",
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "' news. Stream exclusive games on ESPN+ and play fantasy sports.\" />\\n<meta property=\"og:image\" content=\"https://a1.espncdn.com/combiner/i?img=%2Fi%2Fespn%2Fespn_logos%2Fespn_red.png\"/>\\n<meta property=\"og:image:width\" content=\"1200\" />\\n<meta property=\"og:image:height\" content=\"630\" />\\n<meta property=\"og:type\" content=\"website\" />\\n<meta name=\"twitter:site\" content=\"espn\" />\\n<meta name=\"twitter:url\" content=\"https://www.espn.com\" />\\n<meta name=\"twitter:title\" content=\"ESPN - Serving Sports Fans. Anytime. Anywhere.\"/>\\n<meta name=\"twitter:description\" content=\"Visit ESPN for live scores, highlights and sports news. Stream exclusive games on ESPN+ and play fantasy sports.\" />\\n<meta name=\"twitter:card\" content=\"summary\">\\n<meta name=\"twitter:app:name:iphone\" content=\"ESPN\"/>\\n<meta name=\"twitter:app:id:iphone\" content=\"317469184\"/>\\n<meta name=\"twitter:app:name:googleplay\" content=\"ESPN\"/>\\n<meta name=\"twitter:app:id:googleplay\" content=\"com.espn.score_center\"/>\\n<meta name=\"title\" content=\"ESPN - '"
+      ]
+     },
+     "execution_count": 6,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "docs[0].page_content[1000:2000]"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 7,
+   "id": "4d024f0f",
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "'al\" href=\"https://lilianweng.github.io/posts/2023-06-23-agent/\" />\\n<link crossorigin=\"anonymous\" href=\"/assets/css/stylesheet.min.67a6fb6e33089cb29e856bcc95d7aa39f70049a42b123105531265a0d9f1258b.css\" integrity=\"sha256-Z6b7bjMInLKehWvMldeqOfcASaQrEjEFUxJloNnxJYs=\" rel=\"preload stylesheet\" as=\"style\">\\n<script defer crossorigin=\"anonymous\" src=\"/assets/js/highlight.min.7680afc38aa6b15ddf158a4f3780b7b1f7dde7e91d26f073e6229bb7a0793c92.js\" integrity=\"sha256-doCvw4qmsV3fFYpPN4C3sffd5&#43;kdJvBz5iKbt6B5PJI=\"\\n    onload=\"hljs.initHighlightingOnLoad();\"></script>\\n<link rel=\"icon\" href=\"https://lilianweng.github.io/favicon_peach.ico\">\\n<link rel=\"icon\" type=\"image/png\" sizes=\"16x16\" href=\"https://lilianweng.github.io/favicon-16x16.png\">\\n<link rel=\"icon\" type=\"image/png\" sizes=\"32x32\" href=\"https://lilianweng.github.io/favicon-32x32.png\">\\n<link rel=\"apple-touch-icon\" href=\"https://lilianweng.github.io/apple-touch-icon.png\">\\n<link rel=\"mask-icon\" href=\"https://lilianweng.github.io/safari-pinned-tab.'"
+      ]
+     },
+     "execution_count": 7,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "docs[1].page_content[1000:2000]"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3 (ipykernel)",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.9.16"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}

docs/extras/modules/data_connection/document_transformers/integrations/html2text.ipynb

@@ -0,0 +1,133 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "id": "fe6e5c82",
+   "metadata": {},
+   "source": [
+    "# html2text\n",
+    "\n",
+    "[html2text](https://github.com/Alir3z4/html2text/) is a Python script that converts a page of HTML into clean, easy-to-read plain ASCII text. \n",
+    "\n",
+    "The ASCII also happens to be valid Markdown (a text-to-HTML format)."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "ce77e0cb",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "! pip install html2text"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 4,
+   "id": "8ca0974b",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Fetching pages: 100%|############| 2/2 [00:00<00:00, 10.75it/s]\n"
+     ]
+    }
+   ],
+   "source": [
+    "from langchain.document_loaders import AsyncHtmlLoader\n",
+    "\n",
+    "urls = [\"https://www.espn.com\", \"https://lilianweng.github.io/posts/2023-06-23-agent/\"]\n",
+    "loader = AsyncHtmlLoader(urls)\n",
+    "docs = loader.load()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "id": "ddf2be97",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from langchain.document_transformers import Html2TextTransformer"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 5,
+   "id": "a95a928c",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "urls = [\"https://www.espn.com\", \"https://lilianweng.github.io/posts/2023-06-23-agent/\"]\n",
+    "html2text = Html2TextTransformer()\n",
+    "docs_transformed = html2text.transform_documents(docs)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 6,
+   "id": "18ef9fe9",
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "\"  * ESPNFC\\n\\n  * X Games\\n\\n  * SEC Network\\n\\n## ESPN Apps\\n\\n  * ESPN\\n\\n  * ESPN Fantasy\\n\\n## Follow ESPN\\n\\n  * Facebook\\n\\n  * Twitter\\n\\n  * Instagram\\n\\n  * Snapchat\\n\\n  * YouTube\\n\\n  * The ESPN Daily Podcast\\n\\n2023 FIFA Women's World Cup\\n\\n## Follow live: Canada takes on Nigeria in group stage of Women's World Cup\\n\\n2m\\n\\nEPA/Morgan Hancock\\n\\n## TOP HEADLINES\\n\\n  * Snyder fined $60M over findings in investigation\\n  * NFL owners approve $6.05B sale of Commanders\\n  * Jags assistant comes out as gay in NFL milestone\\n  * O's alone atop East after topping slumping Rays\\n  * ACC's Phillips: Never condoned hazing at NU\\n\\n  * Vikings WR Addison cited for driving 140 mph\\n  * 'Taking his time': Patient QB Rodgers wows Jets\\n  * Reyna got U.S. assurances after Berhalter rehire\\n  * NFL Future Power Rankings\\n\\n## USWNT AT THE WORLD CUP\\n\\n### USA VS. VIETNAM: 9 P.M. ET FRIDAY\\n\\n## How do you defend against Alex Morgan? Former opponents sound off\\n\\nThe U.S. forward is unstoppable at this level, scoring 121 goals and adding 49\""
+      ]
+     },
+     "execution_count": 6,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "docs_transformed[0].page_content[1000:2000]"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 7,
+   "id": "6045d660",
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "\"t's brain,\\ncomplemented by several key components:\\n\\n  * **Planning**\\n    * Subgoal and decomposition: The agent breaks down large tasks into smaller, manageable subgoals, enabling efficient handling of complex tasks.\\n    * Reflection and refinement: The agent can do self-criticism and self-reflection over past actions, learn from mistakes and refine them for future steps, thereby improving the quality of final results.\\n  * **Memory**\\n    * Short-term memory: I would consider all the in-context learning (See Prompt Engineering) as utilizing short-term memory of the model to learn.\\n    * Long-term memory: This provides the agent with the capability to retain and recall (infinite) information over extended periods, often by leveraging an external vector store and fast retrieval.\\n  * **Tool use**\\n    * The agent learns to call external APIs for extra information that is missing from the model weights (often hard to change after pre-training), including current information, code execution c\""
+      ]
+     },
+     "execution_count": 7,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "docs_transformed[1].page_content[1000:2000]"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3 (ipykernel)",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.9.16"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}

langchain/document_loaders/__init__.py

@@ -5,6 +5,7 @@
 from langchain.document_loaders.airtable import AirtableLoader
 from langchain.document_loaders.apify_dataset import ApifyDatasetLoader
 from langchain.document_loaders.arxiv import ArxivLoader
+from langchain.document_loaders.async_html import AsyncHtmlLoader
 from langchain.document_loaders.azlyrics import AZLyricsLoader
 from langchain.document_loaders.azure_blob_storage_container import (
     AzureBlobStorageContainerLoader,
@@ -161,6 +162,7 @@
 
 __all__ = [
     "AcreomLoader",
+    "AsyncHtmlLoader",
     "AZLyricsLoader",
     "AirbyteJSONLoader",
     "AirtableLoader",

langchain/document_loaders/async_html.py

@@ -0,0 +1,138 @@
+"""Web base loader class."""
+import asyncio
+import logging
+import warnings
+from typing import Any, Dict, Iterator, List, Optional, Union
+
+import aiohttp
+import requests
+
+from langchain.docstore.document import Document
+from langchain.document_loaders.base import BaseLoader
+
+logger = logging.getLogger(__name__)
+
+default_header_template = {
+    "User-Agent": "",
+    "Accept": "text/html,application/xhtml+xml,application/xml;q=0.9,image/webp,*/*"
+    ";q=0.8",
+    "Accept-Language": "en-US,en;q=0.5",
+    "Referer": "https://www.google.com/",
+    "DNT": "1",
+    "Connection": "keep-alive",
+    "Upgrade-Insecure-Requests": "1",
+}
+
+
+class AsyncHtmlLoader(BaseLoader):
+    """Loads HTML asynchronously."""
+
+    web_paths: List[str]
+
+    requests_per_second: int = 2
+    """Max number of concurrent requests to make."""
+
+    requests_kwargs: Dict[str, Any] = {}
+    """kwargs for requests"""
+
+    raise_for_status: bool = False
+    """Raise an exception if http status code denotes an error."""
+
+    def __init__(
+        self,
+        web_path: Union[str, List[str]],
+        header_template: Optional[dict] = None,
+        verify_ssl: Optional[bool] = True,
+        proxies: Optional[dict] = None,
+    ):
+        """Initialize with webpage path."""
+
+        # TODO: Deprecate web_path in favor of web_paths, and remove this
+        # left like this because there are a number of loaders that expect single
+        # urls
+        if isinstance(web_path, str):
+            self.web_paths = [web_path]
+        elif isinstance(web_path, List):
+            self.web_paths = web_path
+
+        headers = header_template or default_header_template
+        if not headers.get("User-Agent"):
+            try:
+                from fake_useragent import UserAgent
+
+                headers["User-Agent"] = UserAgent().random
+            except ImportError:
+                logger.info(
+                    "fake_useragent not found, using default user agent."
+                    "To get a realistic header for requests, "
+                    "`pip install fake_useragent`."
+                )
+
+        self.session = requests.Session()
+        self.session.headers = dict(headers)
+        self.session.verify = verify_ssl
+
+        if proxies:
+            self.session.proxies.update(proxies)
+
+    async def _fetch(
+        self, url: str, retries: int = 3, cooldown: int = 2, backoff: float = 1.5
+    ) -> str:
+        async with aiohttp.ClientSession() as session:
+            for i in range(retries):
+                try:
+                    async with session.get(
+                        url,
+                        headers=self.session.headers,
+                        ssl=None if self.session.verify else False,
+                    ) as response:
+                        return await response.text()
+                except aiohttp.ClientConnectionError as e:
+                    if i == retries - 1:
+                        raise
+                    else:
+                        logger.warning(
+                            f"Error fetching {url} with attempt "
+                            f"{i + 1}/{retries}: {e}. Retrying..."
+                        )
+                        await asyncio.sleep(cooldown * backoff**i)
+        raise ValueError("retry count exceeded")
+
+    async def _fetch_with_rate_limit(
+        self, url: str, semaphore: asyncio.Semaphore
+    ) -> str:
+        async with semaphore:
+            return await self._fetch(url)
+
+    async def fetch_all(self, urls: List[str]) -> Any:
+        """Fetch all urls concurrently with rate limiting."""
+        semaphore = asyncio.Semaphore(self.requests_per_second)
+        tasks = []
+        for url in urls:
+            task = asyncio.ensure_future(self._fetch_with_rate_limit(url, semaphore))
+            tasks.append(task)
+        try:
+            from tqdm.asyncio import tqdm_asyncio
+
+            return await tqdm_asyncio.gather(
+                *tasks, desc="Fetching pages", ascii=True, mininterval=1
+            )
+        except ImportError:
+            warnings.warn("For better logging of progress, `pip install tqdm`")
+            return await asyncio.gather(*tasks)
+
+    def lazy_load(self) -> Iterator[Document]:
+        """Lazy load text from the url(s) in web_path."""
+        for doc in self.load():
+            yield doc
+
+    def load(self) -> List[Document]:
+        """Load text from the url(s) in web_path."""
+
+        results = asyncio.run(self.fetch_all(self.web_paths))
+        docs = []
+        for i, text in enumerate(results):
+            metadata = {"source": self.web_paths[i]}
+            docs.append(Document(page_content=text, metadata=metadata))
+
+        return docs