Gemini Paper Summarizer

Overview

A command-line tool that uses Gemini API to generate summaries of academic papers.

Note: Supports multiple languages: Chinese, English, French, German, Japanese, Korean, and Spanish.

Examples

See the examples directory for sample outputs, including summaries of:

• Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., … Polosukhin, I. (2017). Attention Is All You Need.

Other Summaries in Japanese

• (2018) What you can cram into a single vector: Probing sentence embeddings for linguistic properties
• (2018) Improving language understanding by generative pre-training
• (2018) BERT: Pre-training of deep bidirectional Transformers for language understanding
• (2019) BERT rediscovers the classical NLP pipeline
• (2019) Linguistic knowledge and transferability of contextual representations
• (2019) A Structural Probe for Finding Syntax in Word Representations
• (2019) What does BERT look at? An analysis of BERT’s attention
• (2023) Text Embeddings Reveal (Almost) As Much As Text
• (2024) Large Concept Models: Language Modeling in a Sentence Representation Space

DeepSeek

• (2024) DeepSeek LLM: Scaling open-source language models with longtermism.
• (2024) DeepSeek-V2: A strong, economical, and efficient Mixture-of-Experts language model.
• (2024) DeepSeek-Coder-V2: Breaking the barrier of closed-source models in code intelligence.
• (2024) DeepSeek-V3 Technical Report

Prerequisites

• Python 3.10+
• Gemini API Key (obtain from Google AI Studio)

Note: If you find it difficult to set up the environment locally, please refer to the following Google Colab Notebook by @shoei05 (explained in Japanese):

• 論文要約 gemini-paper-summarizer を Google Colab で使用する

Installation

1. Clone the repository
2. Install dependencies:

   uv sync
   

3. Create a .env file in the project directory with your Gemini API key:

   GEMINI_API_KEY=your_api_key_here

Usage

uv run gp-summarize path/to/paper.pdf

The tool will generate several markdown files with different types of summaries:

1. A translation of the abstract
2. A summary of the entire paper
3. A JSON structure of the paper's chapters and sections
4. Individual summaries (not translations) for each main section
5. Summaries of 1-4 above combined into a single file

The output files will be named based on the input PDF filename. Files 1-4 above will be saved in a directory and numbered continuously (e.g., paper/001.md, paper/002.md, etc.). The combined file will be named paper.md.

Note: If the process is interrupted (e.g. by Ctrl+C or by a 429 rate limit error, etc.), the process can be re-run smoothly, because any existing output files will be skipped.

Output Format

For each prompt, the tool generates a markdown file containing:

• Title (prompt number)
• Statistics information about tokens
• Prompt
• AI-generated response

The section structure will be displayed in both JSON format and as a hierarchical list.

Command-Line Options

python -m gp_summarize [-h] [-d OUTPUT_DIR] [-o OUTPUT] [-l {de,en,es,fr,ja,ko,zh}] [--version] pdf_paths [pdf_paths ...]

• pdf_paths: Required Path(s) to one or more PDF files to summarize

  • Multiple PDF files can be specified
  • Wildcards (*) are supported on Windows

• -d, --output-dir: Optional. Specify the output directory for intermediate files

  • Recommended when processing multiple PDF files

• -o, --output: Optional. Specify the output file for the summary

  • Can only be used with a single PDF file

• -l, --language: Optional. Specify the output language

  • Supported languages: de (German), en (English), es (Spanish), fr (French), ja (Japanese), ko (Korean), zh (Chinese)
  • Default: Based on system language settings

• --version: Display version information

Examples

Summarize a single PDF:

python -m gp_summarize paper.pdf

Summarize multiple PDFs:

python -m gp_summarize paper1.pdf paper2.pdf

Specify an output directory:

python -m gp_summarize -d ./outputs paper.pdf

Summarize in a specific language:

python -m gp_summarize -l ja paper.pdf

License

CC0 1.0 Universal