yanfu 0.1.1

PDF/CAJ to Markdown translator with layout-preserving PDF generation using local LLMs

YanFu

PDF/CAJ document translator with layout-preserving PDF generation using local LLMs.

Zero-configuration: Install once, run immediately. Models auto-download on first use.

Features / 功能特性

• Zero-Config Setup: pip install yanfu then run. No Ollama, no API keys, no external services.

• Auto-Download Models: GGUF models download automatically from ModelScope/HuggingFace on first run.

• Multi-format Support: Parse PDF and CAJ (Chinese Academic Journal) files.

• Local LLM Translation: Translate using GGUF models (gemma3:1b, qwen3:0.6b) via llama-cpp-python.

• Layout Preservation: Generate PDF output with preserved layout, images, and formulas.

• OCR Support: Handle scanned documents with OCR.

• Batch Processing: Process multiple files or entire directories.

• CLI & API: Command-line interface and Python API with ToolResult pattern.

• Agent Integration: OpenAI function-calling tools for LLM agents.

• 零配置：pip install yanfu 后即可运行。无需 Ollama、无需 API 密钥、无需外部服务。

• 自动下载模型：GGUF 模型在首次运行时自动从 ModelScope/HuggingFace 下载。

• 多格式支持：解析 PDF 和 CAJ（中国学术期刊）文件。

• 本地大模型翻译：使用 GGUF 模型（gemma3:1b、qwen3:0.6b）通过 llama-cpp-python 翻译。

• 排版保留：生成保留排版、图片和公式的 PDF 输出。

• OCR 支持：处理扫描文档。

• 批量处理：处理多个文件或整个目录。

• CLI 和 API：命令行界面和带有 ToolResult 模式的 Python API。

• 智能体集成：用于 LLM 智能体的 OpenAI 函数调用工具。

Requirements / 系统要求

• Python 3.10+
• macOS / Linux / Windows
• CPU only (GGUF models run on CPU, no GPU required)
• ~1GB disk space for model

Installation / 安装

# Basic installation (includes all dependencies)
pip install yanfu

# With OCR support
pip install yanfu[ocr]

# Full installation
pip install yanfu[all]

That's it! No additional setup needed. Models will auto-download on first run.

Quick Start / 快速开始

GUI Application / 图形界面

# Launch GUI
yanfu --gui

# Or with PySide6 installed
pip install yanfu[gui]
yanfu --gui

CLI / 命令行

# Translate a PDF to English (model downloads automatically on first run)
yanfu paper.pdf

# Translate to Chinese
yanfu paper.pdf -l zh

# Translate to Japanese with faster model
yanfu paper.pdf -l ja --model qwen3:0.6b

# Translate multiple files
yanfu paper1.pdf paper2.pdf -l fr

# Batch process a directory
yanfu ./papers --batch -l es

# Verbose output
yanfu paper.pdf -v

# JSON output
yanfu paper.pdf --json

Usage / 使用方法

CLI Flags / 命令行参数

Flag	Description
-V, --version	Show version
-v, --verbose	Enable verbose output
-o, --output	Output directory
--json	JSON output format
-q, --quiet	Suppress non-essential output
-l, --lang	Target language (default: en)
--source-lang	Source language (default: auto)
--model	Translation model (default: gemma3:1b)
--model-path	Direct path to GGUF file
--cache-dir	Model cache directory
--use-ocr	Enable OCR for scanned docs
--engine	PDF parser (auto/pymupdf/marker/pdfplumber)
--temperature	Translation temperature (0.0-1.0)
--batch	Batch process directory
--list-langs	List supported languages
--list-models	List available models
--download-model	Download a model without translating
--list-downloaded	List downloaded models
--cleanup-models	Remove all downloaded models

Translation Models / 翻译模型

Model	Size	Quality	Speed
gemma3:1b	~780MB	Good	Medium
qwen3:0.6b	~420MB	Basic	Fast
qwen3:1.8b	~1.1GB	Best	Slow

Models are stored in ~/.cache/yanfu/models/ after download.

Supported Languages / 支持的语言

Code	Language	Code	Language
en	English	zh	Chinese (Simplified)
zh-Hant	Chinese (Traditional)	ja	Japanese
ko	Korean	fr	French
de	German	es	Spanish
ru	Russian	it	Italian
pt	Portuguese	ar	Arabic
hi	Hindi	th	Thai
vi	Vietnamese

Standalone Installer / 独立安装包

Windows MSI

Download the MSI installer for a complete offline experience:

# Build MSI with bundled model
./scripts/build_msi.sh

# Or on Windows
scripts\build_msi.bat gemma3:1b

The MSI installer includes:

• All dependencies
• PySide6 GUI
• Pre-downloaded GGUF model (~780MB)
• No internet required after installation

macOS DMG / Linux AppImage

briefcase create macOS dmg && briefcase build macOS dmg && briefcase package macOS dmg
briefcase create linux appimage && briefcase build linux appimage && briefcase package linux appimage

See PACKAGING.md for detailed instructions.

Python API

from yanfu import yanfu_translate_file, ToolResult

# Translate a single file (model auto-downloads on first use)
result = yanfu_translate_file(
    input_path="paper.pdf",
    target_lang="zh",
    model_name="gemma3:1b",
)

print(result.success)    # True / False
print(result.data)       # Output paths and metadata
print(result.metadata)   # Version and timing info

Batch Processing / 批量处理

from yanfu import yanfu_translate_files

result = yanfu_translate_files(
    input_paths=["paper1.pdf", "paper2.caj"],
    target_lang="ja",
    use_ocr=True,
)

for r in result.data["results"]:
    print(f"{r['file']}: {'OK' if r['success'] else 'Failed'}")

Model Management / 模型管理

from yanfu.translator import ModelManager

mm = ModelManager()

# List downloaded models
print(mm.list_downloaded_models())

# Download a specific model
mm.download_model("qwen3:0.6b")

# Check if model exists
print(mm.is_model_downloaded("gemma3:1b"))

# Cleanup to free disk space
mm.cleanup()  # Remove all models
mm.cleanup("qwen3:0.6b")  # Remove specific model

Agent Integration / 智能体集成

YanFu provides OpenAI function-calling tools for LLM agent integration:

from yanfu.tools import TOOLS, dispatch

# TOOLS contains the function schema for OpenAI API
# dispatch() routes tool calls to the appropriate function

# Example with OpenAI API
response = client.chat.completions.create(
    model="gpt-4",
    messages=[{"role": "user", "content": "Translate this PDF to Chinese"}],
    tools=TOOLS,
)

# Dispatch the tool call
tool_call = response.choices[0].message.tool_calls[0]
result = dispatch(tool_call.function.name, tool_call.function.arguments)

CLI Help / 命令行帮助

$ yanfu --help
usage: yanfu [-h] [-V] [-v] [-o OUTPUT] [--json] [-q] [-l LANG]
             [--source-lang SOURCE_LANG] [--model MODEL] [--model-path MODEL_PATH]
             [--cache-dir CACHE_DIR] [--use-ocr] [--engine ENGINE]
             [--temperature TEMPERATURE] [--page-size PAGE_SIZE]
             [--font FONT] [--font-size FONT_SIZE] [--margin MARGIN] [--batch]
             [--list-langs] [--list-models] [--download-model MODEL]
             [--list-downloaded] [--cleanup-models]
             [input ...]

YanFu - Translate PDF/CAJ documents using local LLMs (zero-configuration)

Development / 开发

# Clone and install for development
git clone https://github.com/CodeOfMe/YanFu.git
cd YanFu
pip install -e ".[dev]"

# Run tests
pytest tests/ -v

# Lint and format
ruff check .
ruff format .

License / 许可证

GPL-3.0-or-later

See Also / 参见

• NuoYi - PDF/DOCX to Markdown converter
• TransPaste - Local LLM clipboard translator