lexoid 0.1.11.post1

No project description provided

 ___      _______  __   __  _______  ___   ______  
|   |    |       ||  |_|  ||       ||   | |      | 
|   |    |    ___||       ||   _   ||   | |  _    |
|   |    |   |___ |       ||  | |  ||   | | | |   |
|   |___ |    ___| |     | |  |_|  ||   | | |_|   |
|       ||   |___ |   _   ||       ||   | |       |
|_______||_______||__| |__||_______||___| |______| 
                                                                                                    

Lexoid is an efficient document parsing library that supports both LLM-based and non-LLM-based (static) PDF document parsing.

Documentation

Motivation:

• Use the multi-modal advancement of LLMs
• Enable convenience for users
• Collaborate with a permissive license

Installation

Installing with pip

pip install lexoid

To use LLM-based parsing, define the following environment variables or create a .env file with the following definitions

OPENAI_API_KEY=""
GOOGLE_API_KEY=""

Optionally, to use Playwright for retrieving web content (instead of the requests library):

playwright install --with-deps --only-shell chromium

Building .whl from source

make build

Creating a local installation

To install dependencies:

make install

or, to install with dev-dependencies:

make dev

To activate virtual environment:

source .venv/bin/activate

Usage

Example Notebook

Example Colab Notebook

Here's a quick example to parse documents using Lexoid:

from lexoid.api import parse
from lexoid.api import ParserType

parsed_md = parse("https://www.justice.gov/eoir/immigration-law-advisor", parser_type="LLM_PARSE")["raw"]
# or
pdf_path = "path/to/immigration-law-advisor.pdf"
parsed_md = parse(pdf_path, parser_type="LLM_PARSE")["raw"]

print(parsed_md)

Parameters

• path (str): The file path or URL.
• parser_type (str, optional): The type of parser to use ("LLM_PARSE" or "STATIC_PARSE"). Defaults to "AUTO".
• pages_per_split (int, optional): Number of pages per split for chunking. Defaults to 4.
• max_threads (int, optional): Maximum number of threads for parallel processing. Defaults to 4.
• **kwargs: Additional arguments for the parser.

Benchmark

Results aggregated across 5 iterations each for 5 documents.

Note: Benchmarks are currently done in the zero-shot setting.

Rank	Model	Mean Similarity	Std. Dev.	Time (s)
1	gemini-2.0-flash	0.829	0.102	7.41
2	gemini-2.0-flash-001	0.814	0.176	6.85
3	gemini-1.5-flash	0.797	0.143	9.54
4	gemini-2.0-pro-exp	0.764	0.227	11.95
5	gemini-2.0-flash-thinking-exp	0.746	0.266	10.46
6	gemini-1.5-pro	0.732	0.265	11.44
7	gpt-4o	0.687	0.247	10.16
8	gpt-4o-mini	0.642	0.213	9.71
9	gemini-1.5-flash-8b	0.551	0.223	3.91
10	Llama-Vision-Free (via Together AI)	0.531	0.198	6.93
11	Llama-3.2-11B-Vision-Instruct-Turbo (via Together AI)	0.524	0.192	3.68
12	Llama-3.2-90B-Vision-Instruct-Turbo (via Together AI)	0.461	0.306	19.26
13	Llama-3.2-11B-Vision-Instruct (via Hugging Face)	0.451	0.257	4.54