What this category of data looks like
Daily documentation for game client performance optimization includes performance test reports, device compatibility lists, optimization records, and crash analysis reports. Data sources include engine Profiler, crash platforms, real-device testing, and frame rate metrics. Update frequency follows a daily or per-version release cycle. Documents split fields by device, version, and optimization dimension, and contain frame rate values, resource usage, crash log identifiers, and performance comparison records from before and after optimization. Units include fps, MB, milliseconds, and others. The size of a single document typically ranges from hundreds to thousands of characters.
What constraints these characteristics impose on the knowledge base retrieval and recall link
Since data sources cover real-time real-device testing and crash platforms, with updates occurring daily or per version, the knowledge base must support high-frequency synchronized document sets. The structure of documents split by device and version requires the retrieval step to support filtering recall results by device and version, to avoid invalid cross-device matches. Performance optimization documents contain many professional fields, so recall results must have high relevance. Otherwise, operations duty personnel may misjudge the root cause of lag or crashes. Additionally, single documents have large size, so retrieval must control the number of returned results to avoid context overload that reduces inference efficiency.
How to set the configuration
This configuration is based on FastGPT v4.17.0
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches professional documents related to game client performance, avoids recalling irrelevant general operations ticket content. Adjust based on actual testing. |
limit | Top 10–15 entries | Adapts to the size of performance optimization documents, avoids overly long context that impacts inference. Adjust based on actual testing. |
searchMode | Normal Mode | Meets general retrieval needs for operations duty, avoids strict filtering in QA mode that prevents valid documents from being recalled. |
usingReRank | Enabled | Targets multi-dimensional features of performance data, improves relevance of recall results. |
datasetSearchUsingExtensionQuery | Enabled | Performs extended queries on long-text performance test reports, covers more relevant optimization records. |
collectionIds | Knowledge base collection IDs grouped by device/version | Precisely matches the current duty device and version range, avoids recalling invalid cross-version documents. |
Three common mistakes
- Symptom: A parameter error is returned when uploading files to the knowledge base via an external interface. Cause: Exclusive fields for game client performance optimization such as device and version are not included, which does not meet the parameter requirements of the FastGPT interface.
- Symptom: The knowledge base enables QA mode by default, which prevents long-text documents such as performance test reports from being recalled normally. Cause: The retrieval mode is not set to Normal Mode, and the default QA mode configuration is used by mistake.
- Symptom: Multiple files cannot be uploaded in a single batch, and subsets cannot be specified during retrieval. Cause: Files are not grouped into different knowledge base collections by device and version in advance. Global retrieval is relied on, leading to redundant results, and the
collectionIdsconfiguration is not used to precisely specify the retrieval scope.
How to confirm the configuration is complete
- Submit a retrieval request, enter keywords for the current duty device's performance issues, and check if the returned results include performance test reports and optimization records for the corresponding device.
- Check the knowledge base's retrieval mode settings to confirm it is set to Normal Mode.
- Call the external upload interface, upload a test performance log, and verify that the file is successfully stored in the specified knowledge base collection.
- Attempt to batch upload documents for the same device group, confirm the upload process works normally, and that the
collectionIdsconfiguration can be used to specify retrieval of this group of documents.
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.