What This Category of Data Looks Like
This category of data is generated by manufacturing enterprise change management teams. Sources include PLM, ERP, MES systems, supplier materials, validation test documents, and customer-specific requirement documents. Daily documents include ECN/ECR forms, change assessment reports, drawings and BOM version records, process validation reports, production switching plans, inventory disposal documents, and customer approval files.
Data updates have no fixed cycle. Updates are triggered by each process or engineering change. Documents include core fields such as change order number, version identifier, and effective node. There is no unified standardized format for these documents.
Constraints on Vector Models and Indexes
The data characteristics of this category impose multiple constraints on the vector models and indexes workflow. Multi-source heterogeneous document sources and non-uniform formats require the index module to support cross-system document parsing and format adaptation. Updates triggered without a fixed cycle require incremental index logic to avoid resource waste from full reindexing each time. Documents include non-plain-text content such as drawings and BOM lists, so multi-modal vector generation capabilities must be supported. The association between changes and multiple data links requires the index to accurately link core fields such as change order number and version identifier. This ensures all associated files are covered during retrieval.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Chunk Length | 800–1200 characters | Adapts to the paragraph length of documents such as ECN/ECR forms and process validation reports, avoiding information fragmentation within chunks |
| Overlap Length | 100–150 characters | Retains key information such as change order numbers and version identifiers across chunks, preventing associated content from being truncated |
defaultToken | Set from representative sample documents | defaultToken is the default token count for direct text chunking; tune it on representative source documents. See the v4.17.2 model-field reference. It is not a vector dimension, embedding-model input cap, or retrieval/LLM context. |
maxToken | Set to the selected embedding model/service maximum supported input tokens | maxToken is the input-token maximum supported by the selected embedding model/service; verify it in that provider configuration. See the v4.17.2 model-field reference. It is not retrieval/LLM context or a FastGPT-wide fixed cap. |
| Similarity Threshold | 0.72–0.80, calibrated via actual testing | Filters low-match irrelevant documents, preventing non-associated daily production files from being included in search results |
| Single Search Limit | Top 10 entries | Controls the number of search results, adapting to the volume of associated files in change management scenarios |
Three Common Misconfiguration Scenarios
- Phenomenon: After local deployment of FastGPT v4.17.0, uploading image documents such as ECN drawings shows no automatic indexing option. Knowledge base index results only include text content. Cause: Multimodal indexing-related deployment configurations are not enabled, or the deployment package does not include corresponding model dependencies.
- Phenomenon: After configuring an Embedding model to connect to OneAPI, a connection error occurs. Logs show model call failure. Cause: The OneAPI interface address and authentication key are not correctly configured, or the embedding model name does not match the model name configured in OneAPI.
- Phenomenon: When searching for BOM version files associated with changes, the number of results is far lower than expected. Some key documents are not retrieved. Cause: The similarity threshold is set too high, or the chunk overlap length is insufficient, leading to key change order number fields being truncated.
How to Confirm Successful Configuration
- Upload a standard ECN document and its corresponding drawing. Check if the generated vector chunks include core fields such as change order number and version identifier.
- Test calling the Embedding model interface. Confirm that FastGPT can normally retrieve vector return results without connection errors.
- Enter the test search term "change order number XXX". Verify that the search results include all documents associated with that order number, with no obvious omissions.
- View the knowledge base index logs. Confirm that incremental index tasks can be triggered normally, with no repeated full reindexing records.
The values provided are common starting points and should be evaluated against the reader's own samples.