What Data Looks Like for This Category
The primary entities in this category are manufacturers of power battery packs, energy storage cabinets, and BMS systems. Daily documents include grouping rules, BOMs, welding and torque records, BMS flashing records, insulation and voltage withstand reports, EOL test results, and system configuration lists. Data sources are cell capacity grading data, MES systems, welding equipment, torque tools, BMS software versions, and EOL test benches. Updates are created once per pack or cabinet, and sync whenever a BMS software version changes. Data includes structured numeric fields, semi-structured test reports, and long-text rule documents. The number of fields and format of a single data entry varies based on the business process.
Constraints Imposed on Vector Models and Indexes
Data sources for this category are scattered, and formats vary widely. It includes both short time-series device records and long-text grouping rules, so the system must support generating text vectors of different lengths. Update frequency is high, and per-batch data volume is large. Indexes must support incremental updates to avoid the time cost of full reconstruction. There is a high demand for exact matching of structured fields. Vector models must retain semantic features of numeric and rule-based text, and avoid losing key business information due to improper chunking. Errors in a single data entry will directly affect subsequent audit results. The recall accuracy of vector indexes must meet business verification standards.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Chunk Length | 800–1200 characters | Adapts to the common length of battery PACK grouping rules and EOL test reports, avoids single chunk text exceeding the maxToken limit |
| Overlap Length | 100–200 characters | Retains cross-chunk context information, ensures semantic coherence of welding records and torque data |
maxToken | 1024 | Matches the input length limit of most open-source embedding models, adapts to the average length of single business data entries for this category |
| Similarity Threshold | 0.75–0.85 | Balances matching accuracy and recall count, meets the audit and verification requirements for insulation tests and BMS configurations |
| Single Search Limit | Top 8 entries | Covers the number of associated records required for single-batch audits in this category, avoids redundant results interfering with judgment |
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 Misconfigurations
- Phenomenon: After uploading Excel-format BOMs or welding records, the number of chunks generated by the knowledge base is far higher than expected, consuming significant computing resources. Cause: Data is not split by single business records, and the entire table or row is used as the chunk unit, resulting in single chunk text exceeding the
maxTokenlimit. - Phenomenon: When initiating an inspection and audit query, the returned matching results only contain a small number of fields, and cannot be associated with complete cell grouping rules. Cause: The chunk overlap length is set too low, resulting in loss of cross-chunk associated information.
- Phenomenon: After locally deploying an ollama model, vector index generation fails with a connection timeout error. Cause: The access port for the vector model is not opened in docker-compose.yml, or
defaultTokenis not correctly configured to adapt to the context window of the local model.
How to Verify Proper Configuration
- Upload a single grouping rule document of 1000 characters, check the chunk preview in the FastGPT v4.17.0 backend, and confirm the number of chunks matches expectations.
- Initiate a test query by entering "Check the cell torque records for the corresponding batch", verify that the returned results include the corresponding business fields and associated information.
- Upload new BMS flashing records, check the vector index update log, and confirm that the new data has been correctly indexed.
- Adjust the similarity threshold, observe changes in the number of returned results, and confirm that it meets the matching accuracy requirements for business audits.
The values given are common starting points and should be measured against the reader's own samples.