What the Data for This Category Looks Like
Data related to refined oil sales business processes comes from multiple sources: offline gas station business ledgers, user processing records from the official online mini-program, inventory and sales data from the ERP inventory system, and transaction logs from the fuel card management system. Data updates follow two rhythms: real-time and scheduled. Offline transactions and online instant processing data update in real time. Daily business ledgers and monthly inventory and sales data sync on a natural day basis.
Document structures include processing node lists, required material descriptions, process steps, and form fields. Fields include license plate number, fueling volume, payment amount, processing duration, and some fields have clear attached units.
Constraints Imposed on Vector Models and Indexes
Refined oil sales business data mixes multiple sources, including structured transaction fields and unstructured process text. This requires vector models to support vectorization of mixed fields. Real-time updated transaction and processing data requires index systems to support incremental updates, to avoid performance loss from full index reconstruction.
Business processing flows use hierarchical node logic. When processing in chunks, context associations between nodes must be retained, to prevent breakage of process logic after splitting. Fields with attached units require retaining unit information during vectorization, to avoid retrieval matching deviations caused by ambiguity.
Bulk ledger documents have large file sizes, so configurations must support large file chunking and incremental indexing. All configurations on this page use FastGPT v4.17.0.
How to Configure Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Chunk Length | 800–1200 characters | Matches the length of single process steps in refined oil business processing documents, avoids splitting complete processing nodes |
| Overlap Length | 100–200 characters | Retains context associations between adjacent process steps, avoids losing processing logic after chunking |
| Similarity Threshold | 0.75–0.85 | Filters low-match irrelevant business data, accurately matches the processing links requested by user inquiries |
| Single Search Limit | Top 8 entries | Adapts to the number of nodes in business processing flows, returns sufficient step information without redundancy |
defaultToken | 4096 | Adapts to the default token limit of the large model, ensures complete display of process descriptions |
maxToken | 8192 | Adapts to the splicing needs of long documents, covers the complete business processing workflow |
Three Common Misconfigurations
- When uploading refined oil business ledger files larger than 10 MB, some chunks fail vectorization, and the interface displays "vector generation exception". The cause is that the chunk length is set too large, splitting paragraphs that contain complete processing nodes, leading to missing context associations during vectorization.
- After updating FastGPT from version 4.9 to 4.10, the original refined oil knowledge base cannot return search results, and the interface displays "no matching results". The cause is that the embedding dimension of the vector model changes after the version upgrade, and the original index was not regenerated.
- After the server crashes and restarts due to batch upload of multiple files, the knowledge base enters a "not ready" state with no automatic indexing action, and the interface displays "index not started". The cause is that incremental automatic indexing configuration is not enabled, and only manual full index triggering is supported.
How to Verify Correct Configuration
- Upload a single refined oil business processing document under 10 MB, check the chunk preview interface, confirm that complete processing flow nodes are not split.
- Submit a simulated user inquiry, such as "how to apply for fuel card invoicing", check the number of returned search results, confirm that it matches the single search limit configuration.
- Check the vector model operation logs, confirm that the generated vector dimensions match the current configured model dimensions.
- Upload updated business data, check whether the knowledge base automatically triggers incremental indexing without manual operation.
The values listed 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.