What this category of data looks like
Iron and steel smelting-related document and material drafting data mainly comes from internal enterprise sources: smelting process specifications, quality inspection reports, safety operation guidelines, production logs, and bidding and contract templates. Data update triggers include process iterations, new national standard releases, or monthly production data archiving. Most documents are long texts with chapter numbers, with individual documents reaching tens of thousands of characters. They contain clear process parameter fields such as smelting temperature, furnace internal pressure, and alloy component proportion, and use units like degrees Celsius, megapascals, and percentage.
What constraints these characteristics impose on the knowledge base retrieval and recall link
The high proportion of long texts requires retaining contextual connections during retrieval and recall. This avoids separating process parameters from corresponding operation steps, and prevents mismatches between parameters and operation logic. Fields with precise parameters require retrieval to not rely solely on keyword matching. It is necessary to consider the binding relationship between parameters and context, to ensure retrieved content can be directly used for document drafting. The lack of a fixed update cycle requires the knowledge base to support flexible incremental synchronization and full reindex configuration. This prevents data lag from affecting document compliance and accuracy. The fixed chapter structure requires retrieval to prioritize retrieving associated content within the same chapter. This improves the coherence and standardization of document drafting.
How to set the configurations
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.72–0.80 | Matches the precision requirements for iron and steel smelting process parameters, avoids retrieving irrelevant general documents |
limit | Top 8 entries | Single smelting documents have many chunks after splitting, limit the number of retrieved entries to avoid context overflow |
searchMode | embedding | For text with structured parameters, vector retrieval is more suitable for process parameter matching than keyword matching |
usingReRank | Enabled | Rerank retrieved results, filter out results that match parameters but have irrelevant context |
datasetSearchUsingExtensionQuery | Enabled | For document drafting scenarios, automatically expand query terms to match synonyms of process terminology |
collectionIds | Bind the IDs of three categories of directories: smelting processes, quality inspection, safety | Limit retrieval scope, avoid mixing knowledge base data from other categories |
Three common mistakes
- Phenomenon: The model prompts that the knowledge base is empty during a conversation, but all uploaded documents are visible on the knowledge base management page. Cause:
collectionIdsfor limiting retrieval directories are not correctly configured, or the association between directories and datasets is not rebound after database migration. - Phenomenon: The number of retrieved results is far lower than expected. Cause: The
similaritythreshold is set too high, filtering out process parameter-related text that meets matching requirements. - Phenomenon: Retrieval triggers an exception when the number of chunks after splitting a single document exceeds 3000. Cause: Single documents are not split by chapter, leading to exceeding the chunk limit under a single directory, and the
limitparameter is not adjusted to adapt to the chunk scale.
How to confirm the configuration is complete
- Initiate a query containing specific smelting process parameters, check whether the retrieved results include contextual text corresponding to the parameters.
- Check the retrieval log, confirm that only document chunks under the bound smelting-related directories are returned.
- Test the reranking function, check whether the retrieved results are sorted to prioritize matching core process requirements.
- Upload a new process document and complete synchronization, confirm that the retrieval results can cover the newly uploaded content.
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 2026-09-28.