repository·main·Indexed 19 days ago
https://github.com/niutrans/latextransAn end-to-end LaTeX document translation system that uses multi-agent collaboration to translate LaTeX source code into multilingual PDFs. It preserves formulas, layout, and cross-references by translating pre-processed source code rather than PDFs. The system includes a CLI for processing arXiv IDs and local projects, a Streamlit-based GUI, and specialized classes for parsing (LatexParser), compiling (LaTexCompiler), and reconstructing (LatexConstructor) LaTeX documents.
LaTeXTrans is a structured LaTeX document translation system based on multi-agent collaboration. Unlike traditional PDF translation methods that often break formulas and formatting, LaTeXTrans translates the pre-processed LaTeX source code directly.
The system uses a workflow consisting of six specialized agents:
This approach ensures that formulas, layout, and cross-references remain intact while maintaining terminology consistency and generating a high-quality translated PDF that matches the original layout.
For a graphical interface with progress bars, logs, and configurable parameters, you can use the Streamlit-based GUI.
Option 1: Run the dedicated command
latextrans-guiOption 2: Run via Streamlit directly
streamlit run src/gui/streamlit_app.pylatextrans-guiTo install LaTeXTrans, clone the repository and install it in editable mode using pip. It is recommended to use a Conda virtual environment.
To compile LaTeX files and generate PDF outputs, you must have a LaTeX distribution installed:
Using Conda (Recommended):
conda create -n latextrans python=3.10 -y
conda activate latextrans
git clone https://github.com/NiuTrans/LaTeXTrans.git
cd LaTeXTrans
pip install -e .Standard Installation:
git clone https://github.com/NiuTrans/LaTeXTrans.git
cd LaTeXTrans
pip install -e .pip install -e .For a browser-based interface that shows live workflow progress, logs, and runtime configuration, you can launch the Streamlit-based GUI.
Run either the dedicated command or the Streamlit command directly.
# Using the dedicated CLI command
latextrans-gui
# Or running Streamlit directly
streamlit run src/gui/streamlit_app.pyBefore running translations, you must configure your Large Language Model (LLM) credentials in the config/default.toml file. You need to specify the model name, your api_key, and the base_url for the API.
| Model | Recommended base_url |
|---|---|
deepseek-chat | https://api.deepseek.com/v1/chat/completions |
gpt-4o | https://api.openai.com/v1/chat/completions |
gemini-2.5-pro | https://generativelanguage.googleapis.com/v1beta/openai/chat/completions |
model = "deepseek-chat"
api_key = "your_api_key_here"
base_url = "https://api.deepseek.com/v1/chat/completions"Before running translations, you must configure your Large Language Model (LLM) settings in the config/default.toml file. You need to provide the model name, your api_key, and the base_url for the API.
Recommended base_url examples:
https://api.deepseek.com/v1/chat/completionshttps://api.openai.com/v1/chat/completionshttps://generativelanguage.googleapis.com/v1beta/openai/chat/completionsmodel = "deepseek-chat"
api_key = "your_api_key_here"
base_url = "https://api.deepseek.com/v1/chat/completions"The latextrans CLI (via main.py) allows you to automate the translation of LaTeX documents. You can provide input via arXiv IDs, local project directories, or compressed archives. The system uses a CoordinatorAgent to manage the translation workflow based on a TOML configuration file.
Key Input Methods:
--arxiv to provide one or more arXiv IDs.--project to provide paths to local directories or archive files (.zip, .tar, .tar.gz, .tgz).--all-existing to process all projects already present in the configured tex_sources_dir.# Translate a single arXiv paper
latextrans --arxiv 2508.18791
# Batch translate multiple arXiv papers
latextrans --arxiv 2508.18791v2, 2407.01648
# Translate a local project directory or archive
latextrans --project /path/to/project_dir
latextrans --project /path/to/archive.zip
# Process all existing projects in the source directory
latextrans --all-existingThe latextrans graphical user interface can be launched using the Python entrypoint in src/gui/launcher.py. This command starts a Streamlit application that provides a visual interface for the translation tools. It is equivalent to running python -m streamlit run src/gui/streamlit_app.py from the project root.
python src/gui/launcher.pyThe translation behavior is controlled by a TOML configuration file. The CLI can override several keys in this configuration.
Configurable sections and keys via CLI:
llm_config.base_url: Set via --url.llm_config.model: Set via --model.llm_config.api_key: Set via --key.paper_list: Extended by providing --arxiv IDs.tex_sources_dir: Set via --source.output_dir: Set via --output.If user_term is not set in the config, the system will automatically attempt to determine the arXiv category for downloaded papers.
LaTeXTrans supports translating local files or directories using the --project flag.
Provide a path to a .zip, .tar, .tar.gz, or .tgz file:
latextrans --project D:\path\to\paper_source.tar.gzProvide the path to the extracted project directory:
latextrans --project D:\path\to\paper_project_dirTo process all existing projects within the tex source directory, use:
latextrans --all-existingNote: When using --arxiv or --project, LaTeXTrans only processes the explicitly specified input and will not scan other projects in tex source.
latextrans --project D:\path\to\paper_project_dirYou can translate papers directly using their arXiv ID. The system will download the LaTeX source, run the translation workflow (parsing, translating, reconstructing, and compiling), and save the results in the outputs folder.
Use the --arxiv flag with the ID. You can also specify a version (e.g., v2).
latextrans --arxiv 2508.18791
# Or with version
latextrans --arxiv 2508.18791v2Provide multiple comma-separated arXiv IDs to translate them in bulk.
latextrans --arxiv 2508.18791v2, 2407.01648latextrans --arxiv 2508.18791