What Data for This Category Looks Like
This category's data primarily comes from switch consoles, wireless controllers, monitoring systems, and user trouble reporting channels. The update frequency is daily or whenever a network change occurs. Commonly processed documents include network topology diagrams, change orders, fault records, and wireless coverage inspection reports. Document fields include professional content such as port numbers, VLAN IDs, signal strength, and inspection times. Some documents use semi-structured formats; for example, change orders have clear operation steps and impact scopes. Others are unstructured PDF topology diagrams and Excel inspection reports.
Constraints for Document Parsing and Chunking
The characteristics of this category's data create multiple constraints for document parsing and chunking. First, data sources are diverse and formats are inconsistent, covering PDF topology diagrams, DOCX change orders, CSV inspection tables, and more. Parsing logic must adapt to multiple file formats. Second, update frequency is high. Chunk size should not be too large to ensure timeliness of knowledge base synchronization updates. Third, documents contain strongly related professional fields such as port numbers and VLAN IDs. Contextual connections must be retained during chunking to avoid splitting critical information. Finally, the number of batch-imported files is large, and individual file sizes are large. Configuration must support parallel processing and large file parsing.
Configuration Settings
The following configurations are based on FastGPT v4.17.0, and are adapted to the business data characteristics of internal network and WiFi operations.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Ensures contextual integrity of professional fields in internal network documents, avoids splitting associated information such as port numbers and their affiliated VLANs |
chunkSplitter | By paragraph + heading hierarchy | Matches the structured heading hierarchy of network topology diagrams and change orders, retains the coherent logic of operation steps |
customPdfParse | Enabled | Adapts to PDF topology diagrams exported from switches and inspection reports exported from wireless controllers, retains annotation text next to charts |
PARSE_FILE_TIMEOUT_SECONDS | 300–600 seconds | Accommodates parsing time for large network topology PDF files, prevents parsing failures due to timeouts |
UPLOAD_FILE_MAX_SIZE | 500 MB | Supports batch import of multiple large change order and inspection report files, meets parallel processing requirements |
Three Common Mistakes
- Some chunks in the knowledge base chunk list only contain isolated port numbers, with no affiliated VLAN information displayed. The cause is that
chunkSizeis set too small, which forcibly splits professionally associated fields in the same paragraph. - When batch uploading multiple large files, some files fail to parse, and the interface displays a parsing timeout. The cause is that the
PARSE_FILE_TIMEOUT_SECONDSparameter is not adjusted, and the parsing time of large topology PDF files exceeds the default threshold. - The text stream returned by the API call is plain text, with no formatting such as heading hierarchies or lists. The cause is that the format retention logic of
chunkSplitteris not enabled, and direct character splitting leads to loss of formatting.
How to Confirm Configurations Are Correct
- Upload a typical network topology PDF file, verify that parsed chunks retain annotation text and heading hierarchy next to the chart.
- Upload a change order DOCX file larger than 200 MB, confirm normal upload progress and no file size limit error triggers.
- Call the API to test chunking parameters, verify returned chunks contain complete port and VLAN associated information.
- View the knowledge base chunk list, confirm chunk lengths match the preset chunk size range.
The values provided are common starting points, and should be evaluated against samples relevant to the specific use case.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-10-19.