What This Category’s Data Looks Like
Data typically comes from performance logs exported via engine Profiler, frame rate and resource usage records from real-device testing, exception reports from crash platforms, and optimization records from version iterations. Regularly produced documents include performance test reports, device adaptation lists, optimization iteration records, and crash analysis ledgers. Data is updated daily or per version iteration cycle. Document fields usually include device model, frame rate value, memory usage percentage, resource loading duration, and crash stack fragments. There is no fixed mandatory format, but content must align with the corresponding test scenario.
Constraints on Citation Sources and Traceability
Scattered multi-source data requires support for multi-format data source access and incremental synchronization, to align with daily or per-version update rhythms. Unfixed document structures require custom extraction rules for citation fields, to accommodate format differences across test reports. Single-round analysis must process multiple device test data sets, so the content length of individual citations must be limited to avoid exceeding context processing thresholds. The accuracy of performance data directly impacts optimization direction, so strict traceability to specific test batches and device models is required to ensure traceability for compliance review.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
quoteMaxToken | 800–1200 characters | Aligns with the typical length of a single test report for game client performance optimization, avoiding exceeding context processing limits |
responseData | Return only core performance fields | Filters non-essential document metadata, focusing on content directly related to optimization such as device model, frame rate, and resource usage |
detail | Enable full traceability | Retains key information from citation sources including document ID, test batch, and device model, to meet compliance review requirements |
| Maximum context | 4000–6000 tokens | Accommodates citation content from multiple performance test reports, adapting to multi-device comparative analysis scenarios |
| Maximum knowledge base citations | Top 6 entries | Matches the typical reference sample size for daily performance analysis, avoiding redundant information interfering with model output |
| Maximum response tokens | 1500–2000 tokens | Ensures the output length of optimization suggestions covers specific adjustment directions and corresponding sources |
This page's configurations are based on FastGPT v4.17.0
Three Common Misconfigurations
- Phenomenon: Redundant citation markers and source links are included in returned results, interfering with the display of game client performance optimization suggestions. Cause: The
responseDataparameter is not configured correctly, and all full traceability fields are retained by default, with non-essential document metadata not filtered out. - Phenomenon: Preserved line break escape characters are output as-is in cited performance test report content, without proper text line break formatting. Cause: Text formatting configuration for the
detailparameter is not enabled, resulting in escape characters from original logs not being parsed correctly. - Phenomenon: Only a small amount of citation data is returned in a single round of performance analysis, failing to cover test results across multiple devices. Cause: The value range for
quoteMaxTokenis set too small, or the maximum knowledge base citation count limit is too low, truncating valid performance data.
How to Verify Proper Configuration
- Initiate a query that includes multi-device performance data, and check whether citation source fields in returned results include key information such as test batch and device model.
- Review the formatting of cited content, confirming that preset line break escape characters have been correctly parsed into text line break formats.
- After adjusting configuration items, verify that the length of returned citation content matches the preset value range.
- Check returned results to confirm only core fields directly related to performance optimization are included, with no redundant non-essential metadata.
Note: The configuration values listed are common starting points, and should be validated against specific 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.