What This Category of Data Looks Like
Coal transportation and sales business data primarily comes from internal transportation and sales management systems, contract ledgers, and on-site scheduling records. Data updates trigger per individual business process node. Sync immediately after a single business completes. Bulk data updates run daily at midnight. Each individual business document includes fields such as contract number, cargo type, departure location, arrival location, carried tonnage, settlement amount, and process nodes. Most fields are structured text, with a small amount of long text descriptions. Units include tons, yuan, train trips, batches, and others.
Constraints Imposed on Vector Models and Indexing
Coal transportation and sales business data includes structured fields and long text descriptions. Vector models must support text vectorization after concatenating multiple fields, and adapt to industrial transportation and sales terminology. Data updates follow two rhythms: immediate and bulk. Indexes must support incremental synchronization to avoid excessive time spent on full reconstruction. Single business documents link to multiple process nodes. Indexes must establish association mappings between nodes and document content. Indexes must also support semantic recognition of multi-unit fields such as tons, yuan, and train trips, to avoid losing unit association information during vector recall.
Configuration Settings
These configurations apply to FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
| Chunk Length | 800–1200 characters | Adapts to the average length of long text descriptions for coal transportation and sales business. Avoids over-splitting to lose process semantic association, and matches the vector model's maxToken limit. |
| Overlap Length | 100–150 characters | Retains cross-block process node context, and avoids breaks in business process steps after chunking. |
| Similarity Threshold | 0.72–0.80 | Adapts to the term specificity of coal transportation and sales business. Filters low-match irrelevant documents, and retains associated results for same-category business. |
| Single Search Limit | Top 8 entries | The number of linked process nodes and documents for a single coal transportation and sales business is limited. Too many results increase context processing burden. |
| Reranking Model | dengcao/Qwen3-Rerank | This model adapts to semantic understanding of industrial transportation and sales terminology, and improves the accuracy of recall results. |
defaultToken | 4096 | Matches the maximum input token count of the Qwen3-Embedding-8B model, and avoids truncation of ultra-long business documents. |
Three Common Misconfigurations
- Phenomenon: Milvus fails to start. Logs indicate missing PostgreSQL dependency or incorrect configuration. Cause: Accidentally used a Milvus Compose file with redundant PostgreSQL container configuration, and did not select the pure vector storage version adapted to lightweight indexing requirements.
- Phenomenon: Number of vector recall results does not match expectations. A large number of documents outside the coal category appear. Cause: Similarity threshold set too low, not adapted to the term specificity of coal transportation and sales business, leading to recall of low-match non-target documents.
- Phenomenon: Timeout errors or empty vector results when calling the Qwen3-Embedding-8B model. Cause: Did not configure
defaultTokento match the model's maximum input limit, leading to ultra-long business documents being truncated and unable to generate vectors normally, or a locally deployed ollama port not correctly mapped to the FastGPT service.
How to Confirm Configuration Is Correct
- Upload a single coal transportation and sales business document, view the chunked text fragments, and confirm that chunk length and overlap length configurations meet preset requirements.
- Enter query terms related to business processes, view the number of recall results, and confirm that the single search limit configuration is effective.
- Call the vector model to generate document vectors, check that the returned vector dimensions match the model's preset dimensions, and confirm that
defaultTokenandmaxTokenconfigurations do not exceed model limits. - Start the Milvus service, view the index synchronization logs, and confirm that incremental update tasks execute normally according to the rhythm triggered by business nodes.
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.