What Data Looks Like for This Category
Data for this category originates primarily from ERP and WMS systems, barcode or RFID identification devices, purchase work orders, storage location management systems, AGV transport systems, and weighing equipment. Daily generated documents include receiving slips, putaway slips, material picking and return slips, inventory counts, lot ledgers, and expiration date records. Data update frequencies include real-time (such as scanned storage), daily (such as work order picking updates), and monthly (such as full inventory counts). Fields include material codes, lot numbers, storage location numbers, expiration dates, inspection statuses, and material quantities. Units include pieces, kilograms, days, and similar units.
Constraints on Vector Models and Indexes
Data from multiple sources with varied update frequencies requires indexes to support both incremental updates and full refresh modes. This supports fast synchronization of real-time scanned data and batch reconstruction for monthly inventory counts. Core fields such as lot numbers, expiration dates, and inspection statuses must serve as anchor points for vector association. This prevents loss of field information that occurs when only encoding text content for vectors. Numeric fields such as material quantities and storage locations must be converted to embeddable text formats. This ensures vector models capture business logic tied to quantities and locations. Document lengths vary widely. Short documents such as receiving slips have fewer than 100 characters. Long documents such as inventory counts can reach thousands of characters. Adjust chunking rules to avoid truncating critical information.
Configuration Settings
The following configurations are based on the interface and parameter logic of FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Chunk Length | 800–1200 characters | Matches the average length of documents for this category, avoids truncating critical fields such as lot numbers and expiration dates |
| Overlap Length | 50–100 characters | Retains associated information across adjacent chunks, ensures context continuity for lot flow tracking |
maxToken | 8192 tokens | Aligns with the input limit of most embedding models, avoids truncating long document content |
| Similarity Threshold | 0.75–0.85 | Filters low-match historical documents, reduces false positive risks for incorrect material audits |
| Single Search Limit | Top 10 results | Balances recall efficiency and coverage, supports multi-lot association queries for in-stock materials |
defaultToken | Calibrated via actual testing | Ensures consistent encoding of unstructured text by adapting to embedding conversion for unspecified fields |
Three Common Configuration Mistakes
- The interface displays "No available channels". This occurs when the embedding model API address and key are not correctly configured on the model vendor page, and the model is not bound to the default group.
- The number of vector recall results does not match the set single search limit. This occurs when the chunk length is set too small, causing a single document to be split into too many chunks and exceeding the system's default recall limit.
- Embedding fails when deploying the bge-m3 model via ollama. This occurs when the ollama embedding model interface path is not correctly configured in FastGPT, or the model's embedding service port is not enabled.
How to Verify Successful Configuration
- Upload a receiving slip document for this category to trigger a vector embedding task. Check the task logs to confirm embedding succeeded and no truncation prompts appear.
- Enter a query statement that includes lot numbers and inspection statuses on the retrieval page. Confirm that the similarity scores of returned results fall within the preset threshold range.
- Navigate to the model management page. Confirm the embedding model is bound to the default group, and the API key and interface address are configured correctly.
- Run a full index refresh. Check that system runtime aligns with the requirements of the business update rhythm.
All provided configuration values are common starting points. Adjust them based on your own specific samples and business requirements.