This document is based on configuration instructions for FastGPT v4.17.0.
What Data for This Category Looks Like
The data for this category mainly comes from annotation platforms, image samples, category dictionaries, and inspection results. Daily circulating documents include annotation specifications, bounding box examples, quality inspection reports, and rework lists. Data updates follow a daily or single-batch annotation task schedule. Each document typically contains fields such as visual target coordinates, category labels, and rework requirements. There are no unified fixed units, and values adjust based on the category and requirements of the annotation project. Typical processing objects are image annotation data with location information, which differs from pure text-based annotation projects.
What Constraints Do These Characteristics Impose on Knowledge Base Retrieval and Recall
The data characteristics of this category impose multiple constraints on retrieval and recall. First, the data includes image annotation samples with coordinates and rule documents. Retrieval must match both text semantics and visual annotation scenarios, and cannot rely solely on pure text semantic matching. Second, the update frequency is daily or single-batch. The knowledge base must support incremental synchronization of the latest quality inspection reports and rework lists from the annotation platform. This avoids retrieval results lagging behind actual annotation tasks. In addition, the content structure of annotation specifications and bounding box examples is not uniform. During retrieval, priority must be given to matching high-frequency keywords related to boundary framing and occlusion judgment. It must also be compatible with custom fields from different batches of annotation tasks.
How to Configure the Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Semantic similarity range for matching high-frequency text such as annotation specifications and bounding box examples, adapted to the rule matching needs of image annotations |
limit | 6 | Covers typical rework scenarios and reference examples for single-batch annotation tasks, avoids overly long context that affects inference efficiency |
searchMode | hybrid | Matches both text semantics and structured features of annotation rules, adapted to the multimodal data scenario of image annotations |
usingReRank | Enabled | Ranks recall results by relevance, prioritizes highly matched content related to boundary framing and occlusion judgment |
datasetSearchUsingExtensionQuery | Enabled | Expands user questions for pre-sales Q&A, supplements general terminology related to annotations, improves recall accuracy |
collectionIds | Configured according to the exclusive annotation project collection | Isolates annotation data from different batches, avoids retrieval interference across projects |
Three Common Misconfigurations
- Symptom: Newly added annotation specifications or rework lists in the knowledge base cannot be retrieved. Cause: The new document is not synchronized to the exclusive collection configured in
collectionIds, or incremental index updates are not triggered. - Symptom: A large amount of non-image annotation document content appears in retrieval results. Cause:
collectionIdsis not configured for data isolation, resulting in retrieval of documents across categories or projects. - Symptom: Disk usage of locally deployed FastGPT continues to exceed expected levels. Cause: Expired original files, split file chunks, and embedding vectors are not cleaned up regularly, and the knowledge base storage upper limit for single-batch annotation tasks is not restricted.
How to Confirm the Configuration Is Correct
- Run a retrieval test for a single annotation rule, verify that the recall results include matching document content.
- View the knowledge base storage statistics panel, confirm that disk usage only includes three types of data: original files, split file chunks, and embedding vectors.
- Check the
collectionIdsconfiguration, confirm that it only associates the exclusive knowledge base collection for the current image annotation project. - Trigger the synchronization operation for new documents, verify that the new content can be retrieved within a reasonable time frame.
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.