What data for this category looks like
Core data and documentation for this category include grouping rules for battery packs, energy storage cabinets, and BMS systems. It also covers BOM lists, welding and torque records, BMS flashing logs, insulation and voltage withstand reports, EOL test data, and system configuration lists. Data originates from cell capacity grading systems, MES systems, welding and torque equipment, BMS software version repositories, and EOL test benches. Updates are made per individual pack or cabinet, and synchronized with each software version change. Document fields include cell parameters, electrical connection specifications, thermal management thresholds, insulation ratings, test results, and additional details. Some documents require matching to unique device serial numbers and version numbers.
Constraints on model access and configuration workflows
This category has data that updates frequently and in small per-update volumes. Model access must support high-frequency, small-batch document indexing and updates. This avoids resource consumption from full re-scans. Documents include multiple types of structured and semi-structured data. Examples include BOM hierarchical structures and numeric fields in test data. Configurations must support parsing multi-modal documents, and retain original units and precision for all fields. Errors can lead to serious issues such as thermal management failure and BMS protection malfunction. Strict cross-check logic must be configured to ensure parameter matching across every document. Data sources are dispersed. A unified data source access path must be configured to integrate outputs from different devices and systems.
How to set configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
maxContext | 8000–12000 token | Documents include multiple sections of test data and grouping rules. Sufficient context is required for cross-document cross-checks, based on FastGPT v4.17.0 context configuration specifications |
maxTokens | 2000–3000 token | Output for a single cross-review task must cover verification results across multiple documents, to avoid parsing errors from overly long outputs |
quoteMaxToken | 1500–2000 token | Key fields from original documents such as insulation ratings and torque values must be referenced. Sufficient original text length is needed to support verification activities |
| Workflow/API tool selection | auto | Adapts to models that do not support function call. Automatically triggers document retrieval and process orchestration to complete context-bound cross-review |
temperature | 0.1–0.3 | Cross-review requires rigor. Reduces random generation probability to ensure consistency of review results |
AIPROXY_API_ENDPOINT | Fill in the address of the deployed proxy service | Based on FastGPT v4.17.0 proxy configuration specifications, unifies model call entry to adapt to multi-data-source access requirements |
Three Common Misconfigurations
- Symptom: The OneApi page fails to load normally, and returns a 502 or 404 status code when accessed. Cause: After deploying FastGPT v4.17.0, the OneApi reverse proxy address was not configured correctly, or the image version does not match the current FastGPT version.
- Symptom: The review result returned by model call does not link to original document fragments, and the reference field is empty. Cause: The
quoteMaxTokenparameter was not configured correctly, or the context reference binding logic was not enabled, so key content from original documents cannot be extracted. - Symptom: Models that do not support function call cannot automatically trigger the document retrieval process, and cannot complete cross-review using context. Cause: The
toolChoiceparameter was not set toauto, or tool call trigger rules were not configured, so automatic retrieval and review process orchestration cannot be completed.
How to Confirm Configurations Are Correct
- Upload a single grouping rule document and its corresponding BOM list, initiate a cross-review task, and verify if the returned result includes parameter matching items between documents.
- View the model call log to confirm that the
AIPROXY_API_ENDPOINTandAIPROXY_API_TOKENconfigurations were read correctly, with no call failure records. - Test models that do not support function call, confirm whether the document retrieval process is automatically triggered, and check if tool call logs are generated normally.
- Adjust the
temperatureparameter to different ranges, observe changes in the rigor of review results, and confirm that the parameter takes effect.
The values provided here are common starting points, and should be evaluated against the reader's own samples.