What data for this category looks like
Reference only; verify against the target FastGPT release and deployment documentation.
Constraints Imposed on Vector Models and Indexes
Daily updated data sources require indexes to support incremental synchronization, avoiding resource consumption from full reconstruction. Heterogeneous data from multiple sources requires unified field mapping rules, ensuring consistent processing of documents exported from different systems. Diverse document types and large length differences require adaptive chunking strategies to preserve key information. Content related to hazard inspections has extremely high accuracy requirements. Vector retrieval recall results must strictly match keywords from the business scenario, and irrelevant content must not be returned. Additionally, retrieval results in the business workflow guidance scenario must be strongly associated with the currently handled hazard matters, and irrelevant historical inspection records must not be returned.
How to Set Configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Chunk Length | 600–800 characters | Adapts to the length of common documents such as hazard inspection checklists and rectification notices, preserving complete context for individual hazard entries |
| Overlap Length | 100–200 characters | Prevents key information from being broken during chunking, ensuring that associated content in adjacent chunks is fully retained |
| Similarity Threshold | 0.75–0.85 | Matches the precise retrieval needs of hazard inspection work, filtering low-relevance historical records |
| Single Search Limit | Top 5 results | Aligns with the single-call scenario of business workflow guidance, avoiding returning excessive redundant information |
defaultToken | Corresponding vector model access key | Used to verify model call permissions, ensuring that embedding requests are initiated normally |
maxToken | 8192 token | Adapts to the input length limits of mainstream vector models, preventing embedding requests from failing due to exceeding limits |
Three Common Misconfigurations
- A 401 unauthorized error occurs when calling the vector model. The cause is incorrect configuration of
defaultToken, or insufficient key permissions to initiate embedding requests. - The knowledge base automatically deletes duplicate document chunks, leading to disordered index order from custom chunking. The cause is that the document deduplication configuration is enabled by default, and original order markers from chunking are not retained.
- When adding a vector model via a custom channel, requests are always routed to large language model calls. The cause is that the vector model type is not explicitly marked in the configuration, leading to incorrect routing rule matching.
How to Confirm Correct Configuration
- Upload a single hazard inspection checklist document, wait for embedding to complete, then check the embedding status log to confirm there are no failure prompts.
- Enter the knowledge base settings page, check the switch status of the document deduplication configuration item to confirm it meets business requirements.
- Initiate a test query for the business workflow guidance scenario, verify that the number of returned retrieval results matches the single search limit setting.
- Check the vector model call log to confirm that
defaultTokenis correctly carried, with no permission error messages.
The values provided here are common starting points, and should be evaluated against the reader's own samples.
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.