arxiv-translator 0.4.0

Translates arXiv LaTeX papers to Chinese using Google Gemini models.

arXiv LaTeX Translator

arXiv LaTeX Translator is a powerful tool configured to automatically translate arXiv papers from English to Chinese. It downloads the LaTeX source, translates the content using Google Gemini 3.0 (Flash or Pro), and recompiles the paper into a professional PDF, preserving the original layout, equations, and citations.

✨ Features

• Automated Workflow: Downloads source -> Extracts -> Translates -> Recompiles.
• Model Selection: Choose between Gemini 3.0 Flash (fast/cheap) or Gemini 3.0 Pro (higher quality).
• Academic Quality: Uses specialized prompts to ensure accurate translation of AI/ML terminology and academic tone.
• Robust Processing:
  • Handles Large Files: Automatically chunks large LaTeX files to avoid API limits.
  • Error Resilience: Retries on network failures.
  • LaTeX Preservation: Strictly preserves mathematical formulas, citations, and structural commands.
• Automated Workflow: Downloads source -> Extracts -> Translates -> Recompiles.
• Model Selection: Choose between Gemini 3.0 Flash (fast/cheap) or Gemini 3.0 Pro (higher quality).
• Academic Quality: Uses specialized prompts to ensure accurate translation of AI/ML terminology and academic tone.
• Robust Processing:
  • Handles Large Files: Automatically chunks large LaTeX files to avoid API limits.
  • Error Resilience: Retries on network failures.
  • LaTeX Preservation: Strictly preserves mathematical formulas, citations, and structural commands.
• DeepDive Analysis: AI-powered technical analysis that injects explanation boxes into the PDF for complex formulas and concepts.
• Concurrent Processing: Uses parallel workers (12 processes) for both translation and analysis to significantly reduce wait times.
• Chinese Support: Automatically injects ctex package for proper Chinese rendering.

🚀 Installation

Option 1: Install via PyPI (Recommended)

pip install arxiv-translator

Option 2: Install from Source (For Development)

git clone https://github.com/ZeyuChen/arxiv-translator.git
cd arxiv-translator
pip install .

Option 2: Development Setup (Recommended for Contributors)

git clone https://github.com/ZeyuChen/arxiv-translator.git
cd arxiv-translator

# Install micromamba environment (optional but recommended)
micromamba create -f environment.yml
micromamba activate arxiv-translator

# Install package in editable mode with dev dependencies
pip install -e .

3. Install Tectonic (TeX Engine)

The translator uses Tectonic for robust PDF compilation.

# If using micromamba environment (already installed via environment.yml)
micromamba install tectonic -c conda-forge

# Or install manually
curl --proto '=https' --tlsv1.2 -fsSL https://drop-sh.fullyjustified.net | sh

4. Get Gemini API Key

1. Visit Google AI Studio.
2. Sign in with your Google account.
3. Click "Get API key" in the sidebar.
4. Click "Create API key" (you can create it in a new or existing Google Cloud project).
5. Copy the key string (starts with AIza...).

5. Configuration

Quick Setup (v0.2+): Run the following command to save your API key globally (stored in ~/.arxiv-translator/config.json):

arxiv-translator --set-key YOUR_API_KEY

Alternative: Set the environment variable:

export GEMINI_API_KEY=your_api_key_here

📖 Usage

Basic Usage

arxiv-translator https://arxiv.org/abs/2602.04705

Advanced Usage

Select Model:

# Use Gemini 3.0 Pro (Better quality, slower)
arxiv-translator 2602.04705 --model pro

# Use Gemini 3.0 Flash (Default, faster)
arxiv-translator 2602.04705 --model flash

# Enable DeepDive Analysis (Technical Explanations)
arxiv-translator 2602.04705 --deepdive

Custom Output:

arxiv-translator 2602.04705 --output my_translated_paper.pdf

Full Help:

arxiv-translator --help

📂 Output

The translated PDF will be generated in the project root with the format:

• {arxiv_id}_zh_flash.pdf (for Flash model)
• {arxiv_id}_zh_pro.pdf (for Pro model)

🔧 Technical Details

• Parser: Extracts the main LaTeX file automatically.
• Translator: Uses google-genai SDK. Implements smart chunking for long sections.
• Compiler: Uses tectonic for hassle-free compilation, automatically downloading necessary LaTeX packages.

🤝 Contributing

Contributions are welcome! Please submit a Pull Request.

📊 Performance Benchmark

Test Case: arXiv:2602.04705 (ERNIE 5.0 Technical Report)

Model	Time	Translation Quality	Note
Gemini 3.0 Flash	~11 min	Good	Fast, reliable. Recommended for most papers.
Gemini 3.0 Pro	~20 min	Excellent	Slower, higher precision in detailed academic phrasing.

Comparison Preview

Original English	Gemini 3.0 Flash	Gemini 3.0 Pro

📄 License

This project is licensed under the Apache License 2.0.