What Data for This Category Looks Like
Reference only; verify against the target FastGPT release and deployment documentation.
Constraints Imposed by These Characteristics on Vector Models and Indexing
A high proportion of long texts requires chunking strategies tailored to reasonable per-chunk content length, to avoid semantic breaks that reduce retrieval accuracy. Structured fields and unstructured text coexist, requiring simultaneous support for linked structured metadata indexing and unstructured content vector indexing. Scanned attachment files require enabled multimodal parsing and corresponding vector indexing capabilities. Updates follow individual declaration batch or monthly archiving cadences, requiring adapted incremental indexing trigger timing to avoid full index recreation that wastes computing resources.
How to Set Configurations
| Configuration Item | Recommended Setting Range | Rationale |
|---|---|---|
| Chunk Length | 800–1200 characters | Ensures semantic integrity for long declaration materials, avoids overly fragmented single chunks |
| Overlap Length | 100–150 characters | Compensates for semantic gaps after long text chunking, improves coherence across chunk retrieval |
maxToken | 4096 token | Aligns with input length limits of mainstream vector models, prevents encoding failures |
| Similarity Threshold | 0.72–0.85 | Filters low-relevance case and declaration material retrieval results, meets the rigor requirements of government review |
| Reranking | Enable reranking of top 3 results | Raises ranking priority of highly relevant content, fits the precise retrieval needs of case review |
defaultToken | Set based on actual measurements | Adapts to quota limits for locally deployed vector model calls, avoids exceeding resource caps |
Three Common Misconfiguration Issues
- Issue: After locally deploying FastGPT, no image indexing model option appears when creating a knowledge base and uploading documents. Cause: The multimodal parsing function switch is not enabled, or the deployment package does not include corresponding multimodal dependency components.
- Issue: An error is returned when calling the OneAPI service for an embedding model, while the chat model runs normally. Cause: The vector model interface configuration for OneAPI does not match the input format required by FastGPT, or rate limiting is triggered due to quota limits.
- Issue: The number of retrieval results is far lower than expected after index generation. Cause: The chunk length is set too large, leading to failure to correctly split some long texts, or the similarity threshold is set too high, filtering valid matching results.
How to Confirm Proper Configuration
- Navigate to the knowledge base settings page, verify that the chunk length and overlap length configurations match the document characteristics of this category.
- Upload a test case file that includes scanned attachments, check if the parsed results contain image text content.
- Call the vector model test interface, verify that the encoding result for input text has no errors, and the returned vector dimensions meet the configured requirements.
- Simulate an incremental indexing task for a single declaration batch, confirm that the index update only targets newly added data, and no full reindexing is triggered.
The values provided are common starting points, and should be evaluated against samples associated with the intended deployment.
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.