Data Profile for This Category
This category’s data primarily comes from animal quarantine information, supplier batch records, production traceability ledgers, cold chain temperature collection, metal detection data, and laboratory microbial test reports. Daily documents include quarantine certificates, slaughter and cutting records, ingredient marinating records, thermal processing records, cold chain temperature logs, and microbial test sheets. Data is updated per production batch and per shift. Cold chain temperature data must be collected and stored continuously. Data fields include batch number, quarantine number, temperature value (℃), detection time, microbial indicator values, and more. Some documents are a mix of structured tables and unstructured text.
Constraints on Vector Models and Indexes
This category’s data has high update frequency and is tied to batch dimensions. Vector indexes must support incremental updates to accommodate high-frequency writes per batch and per shift. The data includes structured batch and temperature fields, as well as unstructured test records. Vector models must support matching both structured semantics and natural language semantics. Continuous cold chain temperature collection generates large volumes of time-series data. Indexes must support quick filtering by time range. Additionally, the data is tied to compliance-related content such as quarantine and microbial testing. Retrieval result accuracy has high requirements, and false recalls of irrelevant data must be avoided.
Configuration Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
| Chunk Length | 800–1200 characters | Matches the document length of individual slaughter records and quarantine certificates, avoids association breaks caused by cross-batch chunking |
| Overlap Length | 100–150 characters | Retains key context such as batch numbers and temperature nodes, ensures semantic continuity between chunks |
| Similarity Threshold | 0.75–0.85 | Matches semantic consistency for microbial test and quarantine records, reduces false recall probability |
| Single Search Limit | Top 10 entries | Covers the volume of associated data for a single production batch, avoids redundant retrieval results |
maxToken | Set to the selected embedding model/service input limit | This is the configured embedding model's maximum token count, not retrieval or LLM context. Check the selected model or provider documentation. See the v4.17.2 model-field reference |
| defaultToken | Set from representative sample documents | defaultToken is the default token count for direct text chunking; tune it on representative source documents. See the v4.17.2 model-field reference. It is not a vector dimension, embedding-model input cap, or retrieval/LLM context. |
Three Common Configuration Errors
- A 503 error appears after starting the docker-compose container, prompting that no available channel for the text-embedding-v3 model exists under the default group. The cause is incorrect configuration of the vector model’s API channel binding, with text-embedding-v3 not associated with the default group.
- Retrieval results fail to link to the corresponding production batch, with empty batch number fields. The cause is failure to retain associated fields such as batch numbers and supplier codes during chunking, preventing the vector index from matching batch dimensions.
- The edit button cannot be found during retrieval debugging, making it impossible to directly adjust index parameters. The cause is that in FastGPT v4.17.0, the retrieval debugging entry has been moved to the retrieval configuration module in the knowledge base content management page.
How to Confirm Proper Configuration
- After starting the docker-compose container, check that the CHATAPIKEY environment variable has been correctly written to the configuration file, with no null values or format errors.
- Upload a quarantine certificate document for retrieval testing, confirm that the returned results include associated fields such as batch numbers and detection times, and that the number of results matches the single search limit setting.
- View the vector model’s call logs, confirm that requests for the text-embedding-v3 model have been successfully routed to the default group, with no 503 error records.
- Adjust the similarity threshold, verify that the semantic matching degree of retrieval results meets business expectations, with no excessive recalls or insufficient recalls.
The values listed are common starting points and should be validated against local operational samples.