This document is written per the configuration specifications for FastGPT v4.17.0.
What this category of data looks like
Data for this category primarily comes from national maritime administration bodies, shipbuilding industry standardization organizations, and internal corporate compliance archives. Updates are released irregularly alongside policy announcements, standard revisions, and business adjustments. Documents include multi-level directory policy compilations, technical specifications with parameter tables, and process requirements with embedded diagrams. Supported formats include official PDF files and internal DOC notifications. Fields include clause numbers, effective dates, applicable ship types, and units of measurement such as gross tonnage and ship length. Some documents include nested subclauses and linked reference files.
What constraints do these characteristics impose on document parsing and chunking?
Multi-level directory policy compilations require retaining hierarchical connections during parsing, to avoid splitting clauses and breaking context. Technical specifications with measurement units and ship type parameters require retaining the binding relationship between parameters and clauses during chunking, and must not split continuous content that spans units. Process documents with embedded diagrams require extracting annotated text synchronously, to ensure chunks include complete technical explanations. Different formats of official documents and internal notifications require adapting respective parsing logic, to avoid content loss caused by format differences. Long policy compilations require controlling chunk granularity, to avoid excessively long single chunks that reduce retrieval accuracy.
How to set configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Single clauses in shipbuilding policy documents are mostly 500–1000 characters long. This range preserves clause integrity and avoids splitting across multiple clauses |
chunkSplitter | Split by title + paragraph | Multi-level directory policy documents require retaining hierarchical structure. Splitting by title avoids breaking context connections |
customPdfParse | Enable table recognition + image alt text extraction | Shipbuilding documents contain numerous parameter tables and process diagrams. This configuration fully extracts structured content and annotated information |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Parsing long policy document compilations requires significant time. This setting avoids timeout interruptions |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Individual shipbuilding standard documents may exceed 500 MB. This upper limit covers most file sizes |
chunkSettingMode | Manual configuration | Shipbuilding documents have long clauses and structured characteristics. Chunk parameters must be adjusted for these specific scenarios. Automatic mode cannot adapt to dedicated use cases |
Three common mistakes
- Issue: Multi-level directory policy compilations only parse top-level directory content, and lower-level clauses cannot be retrieved. Cause: The chunking logic set to split by title is not enabled, and only top-level title text is extracted.
- Issue: After uploading internal DOC format notifications, the parsing result is empty or only contains titles. Cause: Parsing rules compatible with DOC format are not configured, and the default parsing logic does not adapt to nested styles in older DOC files.
- Issue: Split-related error logs appear during parsing. Cause: The
chunkSizesetting is too small, causing long paragraphs to fail to split normally and triggering chunking exceptions.
How to confirm configurations are correctly set
- Upload a shipbuilding policy PDF that includes multi-level directories, and verify the parsed result contains all hierarchical clause content.
- Upload a shipbuilding specification DOC that includes parameter tables, and verify the parsed result fully extracts measurement units and ship type parameters from the tables.
- Call the file parsing API, verify that the returned result fields include parsed chunk text, and confirm no empty content is returned.
- View parsing logs to confirm there are no timeout or chunking failure error messages. Adjust the values of corresponding parameters based on actual file length.
The values provided here are common starting points, and should be evaluated 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.