What Data for This Category Looks Like
This category’s data comes primarily from engine Profiler, crash platforms, real-device testing, and frame rate collection tools. Updates are released daily or with each new version. Documents used in daily operations and tickets include performance test reports, device adaptation lists, historical optimization records, and crash analysis reports. Data fields include device model, frame rate (unit: fps), memory usage (unit: MB), crash timestamp, and performance comparison parameters from before and after optimization.
Constraints Imposed by These Characteristics on Re-ranking and Hybrid Retrieval
Performance data updates frequently and mostly consists of real-time collected real-device test results. Hybrid retrieval must prioritize recalling the latest, most time-sensitive data. Hybrid retrieval for multiple document types must support both structured test reports and unstructured crash logs. The re-ranking link must adjust sorting based on field priority. For example, frame rate anomaly entries should display first. Search results must accurately match the device and performance problem type of the current ticket. This prevents irrelevant data from interfering with operations and maintenance judgments, and reduces troubleshooting time for stuttering, crash, and freeze issues.
How to Set Configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
usingReRank | Enabled | Game client performance data has high relevance requirements. Re-ranking filters low-relevance historical optimization records and improves troubleshooting efficiency |
similarity | 0.75-0.85 | The semantic matching degree between performance problem descriptions and historical data must fall within this range to avoid recalling irrelevant device adaptation documents |
limit | Top 10 entries | Troubleshooting for operations and maintenance tickets only requires a small number of highly relevant core data. Too many entries increase reading burden |
searchMode | Hybrid retrieval mode | Must recall both structured performance test reports and unstructured crash logs to cover all business data |
Three Common Mistakes
- Empty re-ranking results in conversation details. Cause: The
usingReRankconfiguration is not enabled, or theAIPROXY_API_ENDPOINTandAIPROXY_API_TOKENsettings for the App/Admin server have not completed startup verification. - The number of search results does not match the set
limitvalue. Cause: Thelimitparameter value exceeds the total number of performance-related documents in the current knowledge base, or the hybrid retrieval recall logic does not correctly apply this parameter. - Deployed re-ranking models do not appear in the FastGPT interface. Cause: Information for the corresponding re-ranking service was not added via model channel configuration. Only running the re-ranking model locally without associating it with the FastGPT model channel.
How to Confirm Configuration Is Complete
- Check the environment variables of the App/Admin server. Confirm that
AIPROXY_API_ENDPOINTandAIPROXY_API_TOKENare correctly configured, and thatAIPROXY_API_ENDPOINTdoes not have a/v1or/chat/completionssuffix attached. - Submit a simulated internal ticket for a performance problem. Check whether the number of search recall results in the conversation details matches the
limitsetting. - Trigger a re-ranking test. Confirm that the sorting priority of search results aligns with the importance of performance data. For example, frame rate anomaly entries appear before regular logs.
- Confirm that the current FastGPT version is v4.17.0 or higher. This ensures compatibility with re-ranking and hybrid retrieval functions.
The values provided are common starting points and should be measured 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 2026-10-19.