What Data for This Category Looks Like
Daily documents for this category include performance test reports, device compatibility lists, optimization records, and crash analysis reports. Data sources include engine Profiler, crash platform, real-device test results, and frame rate collection data. Documents update daily or per version. Document fields include frame rate (fps), package size (MB), device model, crash log ID, resource usage values before and after optimization, and more. Performance data for different devices and versions is archived independently. The structure adjusts quickly with version iterations.
Constraints These Characteristics Impose on Knowledge Base Retrieval and Recall
High-frequency updates require the knowledge base retrieval and recall link to support incremental synchronization. This ensures the latest performance data can quickly join the retrieval chain after each version update. The independent archiving structure for multiple devices and versions requires binding the dataset scope of the corresponding version and device during retrieval. This prevents irrelevant cross-version or cross-device data from being retrieved. The presence of professional performance fields requires a mixed mode of keyword and field matching. This accurately matches the device, frame rate, or resource usage indicators corresponding to the current issue. The long content of single performance documents requires limiting the number of single recall results. This prevents exceptions in the generation link caused by exceeding the context window.
Configuration Settings
The following configurations are based on FastGPT v4.17.0, adapted for knowledge base retrieval needs for game client performance optimization:
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.72–0.88 | Matches precise documents related to game performance, filters low-relevance general optimization content, and adapts the semantic similarity threshold for professional performance data |
limit | Top 8 entries | Adapts to the average length of single performance documents, controls total context usage, and avoids truncation triggered by excessive content in the generation link |
usingReRank | Enabled | Performs secondary sorting on retrieved performance data, prioritizes matching key indicators such as the device, frame rate, or crash log ID specified in the current issue |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands keywords for performance issues, for example, associating "stuttering" with "frame rate drop" and "excessive resource usage", to improve retrieval coverage |
collectionIds | Configured by version | Differentiates performance knowledge bases for different game versions, avoids interference from cross-version data, and ensures retrieval results only include optimization content for the current version |
Three Common Misconfigurations
- Phenomenon: The knowledge base retrieval link returns results, but the generation link outputs no content. Cause: After upgrading FastGPT version, daily updated performance test reports were not resynced, and the format of retrieved data is incompatible with new model input requirements.
- Phenomenon: Local testing can normally generate content with knowledge base references, but cannot associate knowledge base data when called through external release channels. Cause: The
collectionIdsconfiguration to bind the knowledge base collection for the corresponding game version was not set, and the external request did not carry the correct dataset identification parameter. - Phenomenon: The generated optimization suggestions include internal paragraph IDs of crash analysis reports or performance test reports. Cause: The output switch for paragraph IDs was not turned off, causing retrieved document fragments with IDs to be directly included in the generation context.
How to Confirm Proper Configuration
- Upload a single latest device compatibility list, initiate a query containing the specified device name, and check whether the retrieval results include relevant fragments of this document.
- Trigger the knowledge base synchronization task for version updates, and check whether the synchronization log shows that the update status of all performance documents is successful.
- Initiate a query containing "stuttering" and "frame rate drop", and check whether the sorting of retrieval results prioritizes matching the device parameters specified in the current issue.
- View the context preview of the generated content, and confirm that unnecessary internal paragraph identifiers are not included.
The values given 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.