What This Category of Data Looks Like
Daily documents for this category include cotton blending or ingredient lists, process setup sheets, equipment operation records, yarn inspection sheets, grey fabric inspection sheets, and defect records. Data sources cover raw fiber inspection reports, ERP systems, real-time data from spinning and weaving machines, evenness tester test results, fabric inspection machine collected data, and order specification files. Data updates align with production batches, team shifts, and individual grey fabric production progress. Corresponding documents and data entries are generated for each batch of raw materials, each shift’s operations, and each finished fabric roll. Document fields include fiber indicators, spinning twist, weaving structure parameters, defect types and locations, and more. Some fields use industry-specific units such as tex and m.
Constraints on Citation and Traceability
Data for this category comes from multiple sources and has a high update frequency. Independent data entries are generated for each batch of raw materials, each shift’s operations, and each finished fabric roll. This requires the citation and traceability process to accurately match data generation times and production batches, to avoid citing expired cotton blending or process parameters. Documents contain industry-specific fields and units. The knowledge base must fully retain details such as fiber indicators, twist, and defect locations during chunking. Otherwise, original records for specific production links cannot be matched during traceability. Additionally, multi-source data has format differences. Offline documents and real-time collected data have different structures. Structured processing rules for different data sources must be clearly defined in the configuration, to prevent confusion in citation logic across different document types.
Configuration Guidelines
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
quoteMaxToken | 2000–3000 token | This category of documents has dense fields and long single-item content. Sufficient tokens are needed to carry complete citation content, avoiding truncation of key parameters |
responseData | Enable structured parsing | Daily documents for this category are mostly structured process sheets and inspection records. Enabling this allows accurate extraction of industry-specific fields such as fiber indicators and defect types, facilitating traceability matching |
detail | Enable | Original document fields and industry-specific units must be retained, to ensure original records of specific production parameters can be verified during traceability |
| Single search citation limit | Top 8–12 entries | A single production batch corresponds to multiple data entries. Too many citations will cause context overload, while too few will fail to cover core process reference content |
| Data source time range | Last 7 days | Data for this category is updated in real time with production batches. A 7-day range can cover core reference documents for the current production cycle, avoiding citing expired data |
Three Common Configuration Errors
- Phenomenon: After setting
quoteMaxTokento 3000 tokens, the interface shows configuration successful, but the large language model does not receive any citation context. Reason: In FastGPT v4.17.0, if the value ofquoteMaxTokenexceeds the system's default total context limit, a content discard logic is triggered, causing citation content to fail to be passed to the large language model. - Phenomenon: Directly uploading equipment operation data returned by HTTP interfaces as knowledge base documents, and no relevant results are matched after configuring citations. Reason: Structured organization of HTTP response data was not performed. The original data does not meet the indexing format requirements of FastGPT, so it cannot be correctly identified.
- Phenomenon: The interface shows that knowledge base content has been cited, but the large language model output does not include professional parameters such as yarn twist and grey fabric defects. Reason: The
detailconfiguration was not enabled, causing the cited documents to only retain summary content and lose industry-specific key field information.
How to Verify Proper Configuration
- View the knowledge base indexing logs to confirm that uploaded documents such as cotton blending lists and process setup sheets have been correctly parsed, and industry-specific fields have been extracted.
- Launch a test query, enter a question containing yarn fineness and weaving structure parameters, and check whether the large language model output attaches the document name and generation time of the citation source.
- Adjust the value of
quoteMaxToken, test the context loading status under different values, and confirm that no context loss occurs. - Verify the indexing results after uploading HTTP response data, and confirm that structured fields can be normally retrieved and cited.
The values provided are common starting points and should be measured against the reader's own samples.