A free PDF document reader that supports converting PDF documents in various languages to Chinese and supports integration with large-scale models for question answering based on document content.
-
Windows:
-
GitHub: https://github.com/zstar1003/FreePDF/releases/download/v5.1.2/FreePDF_v5.1.2.exe
-
Quark Drive: https://pan.quark.cn/s/ee59aa67b65d
-
Mac (arm64):
-
GitHub: https://github.com/zstar1003/FreePDF/releases/download/v5.1.2/FreePDF_v5.1.2_macOS.dmg
-
Quark Drive: https://pan.quark.cn/s/e96b0c3efc3a
-
HomeBrew: Run
brew installfreepdf
The translated PDF file will generate -mono.pdf (translation file) in its corresponding directory.
Configuration Environment:
uv sync
Start the Application:
python main.py
Configuration File (pdf2zh_config.json) Example:
{
"models": {
"doclayout_path": "./models/doclayout_yolo_docstructbench_imgsz1024.onnx"
},
"fonts": {
"zh": "./fonts/SourceHanSerifCN-Regular.ttf",
"ja": "./fonts/SourceHanSerifJP-Regular.ttf",
"ko": "./fonts/SourceHanSerifKR-Regular.ttf",
"zh-TW": "./fonts/SourceHanSerifTW-Regular.ttf",
"default": "./fonts/GoNotoKurrent-Regular.ttf"
},
"translation": {
"service": "bing", // Translation engine, options: bing, google, silicon, ollama, custom
"lang_in": "en", // Source language, options: en, zh, ja, ko, zh-TW
"lang_out": "zh", // Target language, same as above
"envs": {
// bing/google no configuration required
// silicon example:
// "SILICON_API_KEY": "Your API Key",
// "SILICON_MODEL": "Qwen/Qwen2.5-7B-Instruct"
// ollama example:
// "OLLAMA_HOST": "http://127.0.0.1:11434",
// "OLLAMA_MODEL": "deepseek-r1:1.5b"
// Custom example:
// "CUSTOM_HOST": "https://api.xxx.com",
// "CUSTOM_KEY": "Your Key",
// "CUSTOM_MODEL": "Model Name"
}
},
"qa_engine": {
"service": "Off", // Question answering engine, options: Off, Silicon, Ollama, Custom
"envs": {
// Configuration method is the same as above
}
},
"qa_settings": {
"pages": "", // Limit the range of PDF pages for question answering analysis, format such as "1-5,8,10-15", leave blank for all pages
"system_prompt": "You are a professional PDF document analysis assistant. A user has uploaded a PDF document, and you need to answer the user's questions based on the document content.\n\nThe PDF document content is as follows:\n{pdf_content}\n\nPlease note:\n1. Please answer the questions only based on the above PDF document content\n2. If the question is unrelated to the document content, please clearly state this\n3. Answers should be accurate, detailed, and cite relevant page information\n4. Answer in Chinese\n"
},
"translation_enabled": true, // Whether to enable translation
"NOTO_FONT_PATH": "./fonts/SourceHanSerifCN-Regular.ttf", // Global font path (
"pages": "" // Global page range
}-
models.doclayout_path: DocLayout-YOLO ONNX model path. -
fonts: Font path for PDF rendering in various languages. -
translation.service: Translation engine, supports Bing, Google, Silicon, Ollama, and custom. -
translation.lang_in/lang_out: Source/target language, supporting en (English), zh (Chinese), ja (Japanese), ko (Korean), and zh-TW (Traditional Chinese). -
translation.envs: API parameters for different translation engines. The configuration method is exactly the same asqa_engine.envsbelow; see the typical configuration example below for details. -
qa_engine.service: Question answering engine, supporting off, silicon, ollama, and custom. -
qa_engine.envs: API parameters for different question answering engines. The configuration method is exactly the same astranslation.envsabove; see the typical configuration example below for details. -
qa_settings.pages: The range of PDF pages analyzed during question answering, formatted as "1-5,8,10-15". -
qa_settings.system_prompt: Question answering system prompts; the{pdf_content}placeholder represents the specific document content. -
translation_enabled: Whether translation is enabled (true/false). -
NOTO_FONT_PATH: Global font path. -
pages: Global page scope.
- Silicon-based Flow Translation/Question Answering:
``json
"service": "silicon",
"envs": {
"SILICON_API_KEY": "Your API Key",
"SILICON_MODEL": "Qwen/Qwen2.5-7B-Instruct"
} ``
- Ollama Local Large Model:
``json
"service": "ollama",
"envs": {
"OLLAMA_HOST": "http://127.0.0.1:11434",
"OLLAMA_MODEL": "deepseek-r1:1.5b"
} ``
- Custom OpenAI Compatible API:
``json
"service": "Custom",
"envs": {
"CUSTOM_HOST": "https://api.xxx.com",
"CUSTOM_KEY": "Your Key",
"CUSTOM_MODEL": "Model Name"
}
> Whether for translation or Q&A, the envs field is filled in exactly the same way; only the corresponding parameters need to be filled in according to the selected engine.
> It is recommended to edit the configuration file with Notepad/VSCode, etc. Note that JSON format should not have comments; all comments are for reference only.
Optional Translation Engines:
- Bing Translation (Default)
Selecting Bing as the translation engine requires no additional parameters.
- Google Translate
Selecting Google as the translation engine requires no additional parameters.
- Silicon-based Translation
Selecting Silicon as the translation engine requires additional configuration of the [Silicon Flow](https://cloud.siliconflow.cn/i/bjDoFhPf) API Key and the specific chat model.
- Ollama Translation
Choose Ollama as the translation engine. First, deploy a local chat model using Ollama and configure the Ollama address and specific chat model.
- Custom Translation
Other custom model engines compliant with the `OpenAi API`, such as Volcano Engine.
Supports mutual translation between five languages: Chinese, English, Japanese, Korean, and Traditional Chinese.
The question-answering engine supports Silicon-based streaming (online), Ollama (local), and other custom methods compliant with the `OpenAi API`.
## ❓ Frequently Asked Questions
1. Does it support image-based PDFs, such as scanned documents?
**Answer:** No. Essentially, it uses `pdf2zh` to detect text block content and then translates and replaces it. Image-based PDFs cannot be directly replaced, which would result in overlapping content.
2. When using a large model for translation, some content is not translated?
**Answer:** Large models with low parameter counts often have poor instruction compliance, making them unresponsive to translation and causing this issue. Therefore, when translating large models locally, it's crucial to ensure they have a sufficient parameter size, ideally 7 bytes or more.
3. Why are the contents of the table not translated?
**Answer:** pdf2zh currently does not support translating table content. To translate tables, please check the `dev` branch of this repository and use `pdf2zh_next`. However, due to its slow speed, it hasn't been merged into the main branch.
For other questions, please submit an issue or contact me directly via WeChat (zstar1003) to report problems.
## 🛠️ How to Contribute
1. Fork this GitHub repository
2. Clone the fork to your local machine:
`git clone git@github.com:<your username>/FreePDF.git`
3. Create your local machine