What the Data for This Category Looks Like
Daily documents for this category include sales orders, master production plans, material requirement plans, weekly and daily scheduling tables, work orders, material shortage lists, and delivery change notices. Data sources cover ERP orders and inventory, BOMs, process routes, equipment capacity, staff shifts, work-in-progress, and supplier delivery information. Data updates are triggered by business events, including scheduled daily and weekly refreshes, as well as real-time updates when orders change or emergency insertions occur. Most documents contain structured fields and text descriptions, with fields covering specific identifiers and units such as order numbers, material codes, capacity units, and scheduling time periods.
Constraints Imposed on Vector Models and Indexes
Data sources for this category are scattered and update frequencies vary. This requires indexes to support incremental refreshes and multi-source data aggregation, avoiding full index rebuilds. Documents include both structured tables (such as scheduling tables and work orders) and unstructured text (such as delivery change explanations). Vector models must support parsing both text semantics and table structures. Data fields carry specific business units and identifiers, so vector indexes must retain field association relationships to prevent incorrect matches across business semantics. A single review must cover multiple related documents, so index recall capabilities must adapt to multi-document related query requirements. The defaultToken and maxToken parameters must be properly configured to meet the length requirements of documents in this category.
How to Configure Settings
The following configurations are formulated based on the interface and parameter logic of FastGPT v4.17.0:
| Configuration Item | Recommended Range | Rationale |
|---|---|---|
| Chunk Length | 800–1200 characters | Adapts to the paragraph and table cell lengths of documents in this category, avoiding splitting core business content across semantic chunks |
| Overlap Length | 100–150 characters | Compensates for semantic breaks at chunk boundaries, covering cross-row related content in scheduling tables |
| Similarity Threshold | 0.72–0.80 | Matches the semantic similarity range of order and scheduling data, reducing false matches for unrelated documents |
| Single Search Limit | Top 8 entries | Adapts to the number of multiple work orders and scheduling tables that need to be covered in a single review for this category |
| defaultToken | 8192 | Adapts to the total content length of master production plan documents, meeting the requirements for complete context input |
| maxToken | 16384 | Reserves expansion space for abnormal scenarios, addressing delivery change notice content added after emergency order insertions |
Three Common Misconfigurations
- Symptom: A 500 status code is returned when calling the vector model, and the interface displays "Model configuration invalid". Cause: The third-party vector model channel is not bound correctly, or the selected model cannot meet the structured document parsing requirements for this category.
- Symptom: After the index is built, search results only return a small amount of text content, with no table-related matching items. Cause: Multimodal vector support is not enabled, so structured table fields in scheduling tables and work orders cannot be parsed.
- Symptom: A timeout error occurs during cross-review, and the log shows context overflow. Cause: The
maxTokenparameter is not set reasonably, preventing the complete scheduling content of long documents from being loaded.
How to Verify Successful Configuration
- Upload a single test work order and weekly and daily scheduling table, wait for the index build to complete, and confirm that there are no abnormal prompts on the interface.
- Enter targeted queries to verify that the returned results cover the core fields and related business content of category documents.
- Check the vector model configuration items, confirm that the channel binding and parameter settings have been completed, and there are no error logs.
- Trigger an incremental update operation to verify that newly added material shortage lists can be indexed correctly.
The values provided are common starting points and should be measured against the reader's own samples.