anyparse-python 1.1.3

Parse any file to Markdown and JSON

English | 简体中文

AnyParse

AnyParse is a powerful multimodal document parsing and understanding engine designed to seamlessly convert complex files into structured Markdown and JSON formats. Whether it's basic text processing, professional document conversion, or advanced Vision-Language Models (VLM) and OCR recognition, AnyParse provides a comprehensive, one-stop solution.

Core Capabilities

• Multimodal Document Understanding: Supports cross-modal parsing of images and documents. By combining OCR and VLM technologies, it accurately extracts unstructured data.
• Comprehensive Format Coverage: Easily parses office documents, web pages, spreadsheets, e-books, and emails with a single tool.
• Structured Output: Transforms complex files into standardized Markdown and JSON, streamlining downstream data processing and Large Language Model (LLM) applications.

Key Features

• Documents & Layouts: PDF, DOCX, PPTX, XLSX, EPUB, IPYNB
• Text & Markup: TXT, MD, RST, HTML/XHTML/HTM/SHTML
• Spreadsheets & Data: CSV, TSV
• Images & Multimedia: PNG, JPEG/JPG
• Others: EML (Emails)
• Built-in CLI, FastAPI
• Supports running in a pure CPU environment, and also supports GPU
• Output text in human reading order, suitable for single-column, multi-column and complex layouts
• Retain the original document structure, including titles, paragraphs, lists, etc.
• Extract images, image descriptions, tables, table titles and footnotes
• Automatically identify and convert formulas in documents to LaTeX format
• Automatically identify and convert tables in documents to HTML format

Insduction

Installation

pip install anyparse-python

# or

pip install -e .

Usage

please download config/config.yaml into your project directory.

Download Models

# use modelscope (default)
export ANYPARSE_MODEL_MIRROR="modelscope"

# use huggingface
export ANYPARSE_MODEL_MIRROR="huggingface"

# download models
anyparse-cli download --config config/config.yaml --model

Models Hub

• AnyParse Models Hub ModelScope
• AnyParse Models Hub HuggingFace

Python

# Sync
from anyparse import AnyParser

model = AnyParser(config="config/config.yaml")
res = model.invoke(file = "/path/to/your_file")



# or Async
from anyparse import AsyncAnyParser

model = AsyncAnyParser(config="config/config.yaml")
res = await model.ainvoke(file = "/path/to/your_file")

CLI

# help

anyparse-cli --help

# parse file
anyparse-cli parse --config config/config.yaml --file /path/to/your_file

# start api server
anyparse-cli api --config config/config.yaml

# see allowed file types
anyparse-cli allow --config config/config.yaml

# see commands help
anyparse-cli [COMMAND] --help

API

• start api server

# start fastapi server and openai proxy
## use restful api or openai client call
anyparse-cli api --config config/config.yaml --host 0.0.0.0 --port 18007 --seckey 'your_custom_secret_key'

• call api

# openai
from openai import OpenAI

client = OpenAI(
    base_url = "http://localhost:18007/anyparse/openai/v1",
    api_key = "your_custom_secret_key",
)
## get model id and allowed file types
print(client.models.list())



## parse file
import base64

with open("1.pdf", "r", encoding="utf-8") as f:
    text_content = f.read()

encoded_bytes = base64.b64encode(text_content.encode('utf-8'))
base64_str = encoded_bytes.decode('utf-8')

response = client.chat.completions.create(
    model="anyparse",
    messages=[
        {
            "role": "user",
            "content": [
                {
                    "type": "file",
                    "file": {
                        "file_data": f"data:application/pdf;base64,{base64_str}"
                    }
                }
            ]
        }
    ],  # data:application/pdf;base64 prefix follow: client.models.list().data[0].allow_mimetypes
    # extra_body={
    #     "runtimes_args": {
    #         "use_doc_layout": True
    #     }
    # }
)

print(response.choices[0].message.content)



# or restful
import requests as rq

headers = {
    "Authorization": "Bearer your_custom_secret_key"
}

url = "http://localhost:18007/anyparse/invoke/v1"

args = {
    "use_doc_cls": False,
    "use_doc_rectifier": False,
    "use_doc_layout": True
}

file = '/path/to/your_file'

files = {
    'file': open(file,'rb')
}

res = rq.post(url, files = files, data = args, headers = headers)
print(res.json())

Details and Documentation see docs

TODO LIST

• audio transcription
• video transcription

All Thanks To Our Contributors

License Information

This repository is licensed under the AnyParse Open Source License, based on Apache 2.0 with additional conditions.

Acknowledgments

• MinerU
• PaddleOCR
• GLM-OCR
• pypdfium2

Star History