Reference only; verify against the target FastGPT release and deployment documentation.
What Data for This Category Looks Like
Data for this category primarily comes from the Chinese Center for Disease Control and Prevention Information System, medical institution HIS and infectious disease reporting systems, and laboratory testing systems. Daily documents include infectious disease report cards, epidemic analysis reports, case correction records, clustered epidemic reports, and epidemiological investigation reports. Data is updated daily. Each individual document contains fields such as case unique identifier, diagnosis time, reporting unit, correction status, and test results. Some documents require cross-source data association across multiple systems, and field formats must comply with regulatory notifiable infectious disease reporting standards.
What Constraints Do These Characteristics Impose on Vector Models and Indexes
Daily update frequency requires indexes to support incremental synchronization, avoiding the time cost of full reconstruction. Cross-source data association requirements require indexes to retain original field identifiers from each system, facilitating subsequent verification of case association relationships. Significant differences exist in document lengths across categories, from short report cards to thousands-character epidemiological investigation reports, requiring adaptation to different chunking strategies. Regulatory reporting rules require indexes to retain key fields such as case corrections and reporting time limits, ensuring that recall results can be traced for compliance. At the same time, the timeliness of epidemic early warning requires vector recall to be completed within a reasonable time frame, avoiding delays in identifying abnormal signals.
How to Configure Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Chunk Length | 800–1200 characters | Adapts to the lengths of multiple document types such as infectious disease report cards and epidemiological investigation reports, avoiding information breakage across chunks |
| Overlap Length | 100–200 characters | Retains key information such as case identifiers across chunks, preventing loss of associated fields after chunking |
| Similarity Threshold | 0.75–0.85 | Balances recall accuracy and recall volume, avoiding missed reports of critical case information |
| Single Search Limit | Top 10 entries | Controls the amount of information in recall results, adapting to the rapid response needs of service consultations |
maxToken | 4000 token | Limits the total length of a single round of context, avoiding exceeding the model's processing limits |
| Reranking Model Switch | Enabled | Improves the sorting priority of highly relevant results, optimizing the accuracy of service consultation responses |
Three Common Configuration Errors
- Connection errors occur when configuring the Embedding model to connect to OneAPI, with the interface displaying "model call failed". The cause is that the OneAPI model mapping identifier was not filled in correctly, resulting in vector requests failing to be forwarded properly.
- After enabling the reranking model configuration and checking the corresponding option during the indexing phase, the online recall test results do not display according to the reranking priority. The phenomenon is that the recall result sorting is consistent with the initial similarity sorting. The cause is that the reranking call switch was not enabled synchronously during the question-and-answer call phase; only configuring reranking during the indexing phase does not take effect.
- Interface incompatibility errors occur when attempting to connect to Tongyi Multimodal Vector Model, with the return message "specified call format not supported". The cause is that the dedicated call parameters adapted to Tongyi Multimodal Vector were not adjusted, and the configuration using the general format was retained.
How to Confirm Configuration Is Correct
- Manually upload a single infectious disease report card, perform vector indexing, and check the index details page to confirm that all preset fields are stored correctly.
- Launch a consultation test containing a case number or epidemic keyword, verify whether the matched document content is included in the recall results, and confirm that the similarity threshold setting is effective.
- After enabling the reranking switch, compare the recall result sorting before and after enabling to confirm that the reranking model has been called normally.
- Check the system operation logs to confirm that the
maxTokenparameter for vector model calls does not exceed the set limit, and there are no timeouts or connection errors.
The values provided above 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.