Reference only; verify against the target FastGPT release and deployment documentation.
What this category of data looks like
Data primarily comes from construction logs on job sites, material purchase ledgers, employee attendance records, safety training documents, project acceptance reports, and similar sources. Update cadence follows project milestones: construction logs are updated daily, material ledgers weekly, and employee service-related reports monthly. Documents include structured tables, long text passages, PDF drawings with complex formatting, and Word documents with embedded forms. Fields include construction duration (hours/days), material usage (tons/cubic meters), employee attendance days, and document IDs prefixed with project codes, among others.
What constraints do these characteristics impose on document parsing and chunking
Mixed multi-format document structures require parsing tools to support text, table, and complex-format PDF parsing to avoid losing critical information. Structured ledger tables require retaining cell relationships during chunking to prevent semantic breaks from split table content. Document IDs with project codes and long construction log passages require retaining contextual connections during chunking to avoid losing project identifiers across chunks. Bulk project documents and frequently updated logs require parsing services to support queue processing to avoid overloading single batches of tasks.
How to set configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
customPdfParse | Enabled | Infrastructure construction documents often include complexly formatted construction drawings and acceptance report PDFs. Enabling custom parsing preserves formatting and table structure |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Parsing large single construction drawing PDFs or bulk documents takes longer. This setting prevents task interruption from timeout |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Collections of large project CAD drawings and annual material ledger documents have large file sizes. This setting adapts to single-file upload limits |
chunkSize | 800–1200 characters | Construction logs and training documents are often long passages. This range preserves semantic integrity and avoids splitting technical terms |
chunkSplitter | Split by paragraph and heading | Infrastructure construction documents are organized by chapter and passage. Splitting this way ensures coherent chunk semantics |
chunkSettingMode | Custom configuration | Document structures vary significantly across projects. Custom configuration adapts to chunking needs for multiple document types including construction logs, ledgers, and drawings |
Three common mistakes
- Scenario: Uploading a large construction drawing PDF results in a parsing timeout error in FastGPT, but the task succeeds with a third-party parsing service. Cause: The
PARSE_FILE_TIMEOUT_SECONDSparameter is not adjusted. The default timeout duration is insufficient for large document parsing. - Scenario: Bulk uploading dozens of material ledger Excel files causes parsing tasks to interrupt mid-process, and the issue cannot be resolved by restarting Docker. Cause: A reasonable bulk parsing queue is not configured. Single-batch processing volume exceeds system limits.
- Scenario: Chunked documents cannot be associated and retrieved by the model, and responses do not reference technical construction terms from the documents. Cause: The
chunkSizevalue is too small, or thechunkSplitterconfiguration is incorrect. This splits logically connected content across passages.
How to confirm correct configuration
- When a single construction drawing PDF larger than 100 MB is uploaded, verify that the parsing task completes within the preset duration. Confirm the
PARSE_FILE_TIMEOUT_SECONDSvalue adapts to document size. - When a structured material ledger Excel file is imported, verify that the complete table structure is preserved after parsing. Confirm the
customPdfParseconfiguration is enabled. - When a long construction log passage containing technical terms is selected, review the chunking results to confirm full semantic retention. Confirm the
chunkSizeandchunkSplitterconfigurations match the document structure. - When five or more identical-type project documents are bulk uploaded, verify that the task queue proceeds normally. Confirm the system does not trigger interrupts due to single-batch processing volume exceeding limits.
The values provided are common starting points and should be measured against the reader's own samples.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-28.