What data looks like for this category
Data for this category comes from the Chinese Center for Disease Control and Prevention Information System, medical institution HIS systems, infectious disease reporting systems, and laboratory testing systems. The update frequency is daily. Daily documents include infectious disease report cards, epidemic analysis reports, case correction records, clustered epidemic reports, and epidemiological investigation reports. Fields include basic case information, diagnosis results, reporting time limits, correction markers, testing data, and more. The length of individual documents varies widely. Some epidemiological investigation reports can reach several thousand characters.
What constraints do these characteristics impose on vector models and indexes
Batch data updated daily requires indexes to support high-frequency incremental synchronization, to avoid excessive time spent on full reconstruction. Differences in formats across multiple document types require vector models to adapt to mixed input of structured fields and unstructured text. Legal reporting rules for case-related fields require indexes to retain field-level associations. This ensures that retrieval can match case correction records to original report cards. The wide variation in individual document length requires chunking strategies to balance vector extraction accuracy for short texts and long documents. This avoids information truncation or redundancy.
How to set configurations
| Configuration Item | Recommended Values | Rationale |
|---|---|---|
| Chunk Length | 800–1200 characters | Ensures complete information per chunk for epidemiological investigation reports and epidemic analysis reports, avoids truncation of long texts |
| Overlap Length | 100–200 characters | Retains contextual association between chunks, reduces information fragmentation for case correction records |
| Similarity Threshold | 0.75–0.85 | Differentiates similar case reports and correction records, avoids mismatching key information required for legal reporting |
maxToken | 4096 | Adapts to the upper limit for vector extraction from long documents, aligns with input limits of most public vector models |
defaultToken | Set based on actual measurements | Adapts to batch processing quotas for daily batch data, avoids exceeding interface call limits |
| Single Recall Limit | Top 20 entries | Covers newly added case reports and correction records for the day, meets retrieval needs for internal office work |
Three common configuration mistakes
- After upgrading a version, calling the index interface returns a 400 status code with no response body. The cause is failure to adapt to the new version’s vector model interface parameter validation rules. The
defaultTokenvalue from the old configuration exceeds the current interface’s allowed quota. - After local deployment, server memory is fully occupied during daily scheduled index synchronization. The cause is failure to set a reasonable chunk length limit. When processing ultra-long epidemiological investigation reports in batch, the single-time data load is too large, exceeding the server memory threshold.
- Knowledge base file uploads occasionally get stuck at 1 or 2 index progress groups. The cause is that the field format of some case correction records does not meet the input requirements of the vector model. Automatic format validation is not triggered, leading to interruption of index construction.
How to confirm configurations are set correctly
- Import a single infectious disease report card and epidemiological investigation report. Check that the vector extraction results cover core fields with no obvious information truncation.
- Simulate daily batch data synchronization. Review interface call logs to confirm that the
maxTokenanddefaultTokenconfigurations do not trigger quota exceeded errors. - Retrieve correction records for a specified case. Confirm that the recall results include the original report and correction documents, and that the matching logic aligns with legal reporting rules.
- Export the current knowledge base configuration in FastGPT v4.17.0. Verify that saved parameters such as chunk length and overlap length match the set values.
The values provided here are common starting points, and should be evaluated against samples relevant to the actual deployment scenario.
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.