What the data for this category looks like
Daily documents for this category include territorial spatial master plans, detailed plans, special plans, planning implementation evaluation reports, public opinion summaries, and more. Data sources include the territorial spatial basic information platform, natural resource survey and monitoring data, population and economic statistics, and remote sensing images. The data contains structured spatial coordinates, planning indicator fields, and unstructured text content. The update frequency is annual. The overall data volume grows as the scope of the corresponding planning area expands.
What constraints these characteristics impose on the vector models and indexes link
The multi-source and heterogeneous data structure requires vector models and indexes to support both structured planning indicator text and long-paragraph planning reports, to avoid semantic coding deviation. The annual update frequency means high-frequency incremental indexing is unnecessary. A periodic full refresh strategy can be adopted to reduce operation and maintenance costs. Planning businesses have high requirements for approval accuracy. The accuracy of vector recall directly affects the effectiveness of business process guidance. The ratio of false recalls and missed recalls must be strictly controlled to avoid impacting subsequent planning approval processes.
Reference only; verify against the target FastGPT release and deployment documentation.
How to set the configurations
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
| Chunk length | 800–1200 characters | Adapts to the standard paragraph length of territorial spatial planning reports, avoids semantic fragmentation |
| Overlap length | 100–150 characters | Connects the spatial correlation and text semantics of adjacent chunks, avoids loss of boundary information |
| Similarity threshold | 0.72–0.78 | Balances recall accuracy and coverage for planning-related content, avoids false recalls of planning content from unrelated areas |
| Single search limit | Top 8 entries | Planning businesses need to balance multi-dimensional reference materials, and a small number of entries can cover core planning content |
| maxToken | 8192 | Adapts to the input length limit of mainstream embedding models, ensures that the complete semantics of long-text planning reports are encoded |
| defaultToken | Calibrated via actual testing | Adapts to the token quota of the deployment environment, avoids call limit exceeded errors |
Three common configuration mistakes
- The interface returns the error "This token does not have permission to use the model". Cause: The corresponding embedding model channel has not been configured in FastGPT, or the configured channel has not obtained the usage permission for the corresponding model.
- An unavailable channel prompt still appears after configuring the bce-embedding channel. Cause: The channel's service address, key and other parameters are filled incorrectly, or FastGPT has not completed channel refresh and synchronization.
- Knowledge base search takes longer than expected. Cause: The chunk length is set too large, causing the single-batch vector calculation load to exceed the hardware processing capacity, or the index is not built partitioned by planning document type.
How to confirm the configuration is complete
- Upload a territorial spatial master plan document to trigger the vector encoding process, and check that there are no errors related to embedding model calls in the task logs.
- Enter planning-related query terms, such as "three zones and three lines layout requirements", and verify that the recall results include corresponding content from the planning implementation evaluation report.
- Test multiple rounds of query operations, confirm that the search response time meets business expectations, and there are no continuous timeouts.
- View the index management page, confirm that the index refresh cycle matches the annual data update frequency, and there are no abnormal incremental refresh tasks.
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.