Data Profile for This Category
This category’s data comes from state-owned assets supervision systems, enterprise ERP and project management systems, industry market data, audit and financial statements. Daily processed documents cover annual investment plans, project proposals, feasibility study reports, due diligence reports, asset appraisal reports, and investment post-evaluation reports. The data update cycle is quarterly. Core users include central enterprise strategic investment departments, project companies, state-owned assets supervision departments, and external consulting and evaluation institutions. Most documents are formal business files combining structured and semi-structured content, containing business fields such as project numbers, investment amounts, and approval nodes.
Constraints on Vector Models and Indexing
This category’s data is multi-source and heterogeneous, covering structured reports exported from systems and formal business documents. Format unification and field alignment are required before indexing. The low-frequency quarterly update attribute means high-frequency incremental indexing is not needed. A scheduled batch update strategy can be used. Document length varies widely, from a few pages of project proposals to dozens of pages of post-evaluation reports. When splitting chunks, maintain a balance between business logic integrity and reasonable text granularity. Professional business fields such as decision-making nodes and post-evaluation indicators require vector models to accurately capture professional semantics, and avoid semantic drift. At the same time, the data involves content related to state-owned assets supervision. Indexing processes must ensure traceability and security.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Chunk Length | 800–1200 characters | Adapts to the business paragraph length of documents in this category, avoids splitting that disrupts business logic such as decision-making nodes and post-evaluation indicators |
| Overlap Length | 100–150 characters | Connects business context between adjacent chunks, prevents key approval process information from being truncated |
defaultToken | 4096 token | Matches the input limit of mainstream open-source vector models, adapts to long-text chunk content from post-evaluation reports |
| Similarity Threshold | 0.72–0.85 | Filters low-relevance non-business documents, ensures retrieval results focus on core content of investment approval and post-evaluation |
| Single Search Limit | Top 5 entries | Matches the number of core reference documents for investment decision-making meeting materials, avoids redundant information interfering with business judgment |
maxToken | 8192 token | Limits the total length of context spliced after retrieval, adapts to system context window limits |
Three Common Mistakes
- Symptom: When uploading post-evaluation reports or feasibility study reports over 10 MB, some chunks fail vectorization after splitting, and the interface shows "vector generation exception". Cause: No appropriate
maxTokenparameter configured for long texts, vectorization is interrupted due to exceeding the model input limit. - Symptom: After a server restart due to uploading a large number of approval and post-evaluation documents, the knowledge base status shows "not ready" and there is no automatic indexing progress. Cause: Breakpoint resume configuration for batch indexing is not enabled, and the system cannot resume unfinished indexing tasks.
Reference only; verify against the target FastGPT release and deployment documentation.
How to Confirm Configurations Are Set Correctly
- Upload a single post-evaluation report over 10 MB, check that the number of chunks matches expectations, and there are no abnormal chunk markers.
- Initiate a retrieval for "investment decision-making meeting approval process", verify that the similarity scores of returned results fall within the preset threshold range.
- Trigger a batch indexing task, restart the server, then check whether the task automatically resumes and completes.
- View indexing logs, confirm that all business fields are correctly vectorized, and there are no semantic loss prompts.
The values provided here 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 2026-09-28.