Author: Alibaba Cloud DTS
Version: 0.0.2
Type: Dify Plugin Tool
Description
This tool leverages the Document Intelligence processing capabilities of Alibaba Qwen to enable conversion from PDF files to Markdown format. The processed content retains the original document structure (such as headings, paragraphs, lists, tables, etc.) and also support parsing images to attach them to corresponding positions. It is suitable for efficient processing of complex formats such as technical documents, data transmission protocols, API documents, etc.
Usage limit
Our plugin is available for free to Dify users. Simply provide AK/SK information for your Alibaba Cloud account according to the usage instructions, and each Alibaba Cloud account UID is limited to 50 files. At present ,the plugin only support uploading one file at a time.
Features
- High precision parsing and text extraction: Our service utilizes Qwen Document Intelligent proprietary model understanding capability, which can provide more accurate semantic understanding of technical documents compared to general OCR tools. The model can accurately recognize text, tables, image descriptions, and other content in PDFs, with higher resolution and restoration accuracy.
- Regarding image processing: Many PDF processing tools chose to skip image information. This plugin also supports image oriented processing, preserving the position of the image in the output markdown and using OSS caching to return it in URL format for download… Also before the expiration time, the image information supports restoration by converting Markdown back to other file formats and will be inserted in the corresponding position.
- Structured Markdown Output: The model can automatically recognize and preserve the hierarchical structure of the original document (such as # title, - list, | table, etc.)
- Complex format preservation: The model supports the conversion of special content such as code blocks, formulas, chart descriptions, etc. and accurately preserves their original positions.
- Multi language support: Supports documents in Chinese, English, and other mainstream languages.
Usage
- Install plugins: Install plugins on Dify Marketplace.
- Create sub account: Acess to Alibaba Cloud RAM Console through the following link Alibaba Cloud RAM Console.Select 'Create User' in the Users interface of the console to create a sub account (sub account is only used for plugin authentication to prevent AK/SK leakage of the main account).Customize account information and select the option to access Using a permanent AccessKey to access.
Copy the AccessKey ID and AccessKey Secret.
- Create permission policies:Click on the Policies on the left and select 'Create Policy'.Select Json and paste the code section below into the script box.Press OK and Set a policy name.
{
"Version": "1",
"Statement": [
{
"Effect": "Allow",
"Action": [
"dts:CreateDtsInstance",
"dts:ConfigureDtsJob",
"dts:StartDtsJob",
"dts:DescribeDtsJobDetail",
"dts:DescribeWorkflowJobResult",
"dts:DescribeUploadPolicy"
],
"Resource": "*"
}
]
}
- Grant Permission: Return to the users interface and select grant permission in the sub account details to authorize the policy just created to this account.
- Authorization: Create a workflow in Dify, click on authorization and paste the corresponding AK/SK information.
- Start to use: Upload the PDF document to be converted and start the task.Then the converted Markdown content can be downloaded directly or integrated into other systems.
Limitations
- Complex layout: Extremely complex tables and nested layouts may be partially lost.
- Longer required time: High precision processing requires multi-step flow and parsing, and the running time is longer than general OCR processing.
- Performance dependency: Processing large files may be affected by limited flow or network latency.
Safety Tips
Please keep your AK/SK information safely and only use it for plugin authentication in a trusted Dify environment. If you find that the environment is not trustworthy or there may be any abnormalities, please Do Not use your AK/SK information for authentication. Also please be careful not to disclose AK/SK information to others.
Contact us
In the future, we will expand and build more capabilities in the multimodal analysis section.If you have any questions and suggestions in use, such as roast or what multimodal parsing capabilities you expect us to provide, please feel free to contact us.
DingTalk Group: 60495056098