Reference only; verify against the target FastGPT release and deployment documentation.
What the data for this category looks like
Non-ferrous metal smelting document and material data mainly comes from internal process manuals, daily production logs, monthly quality inspection reports, and safety compliance documents. Process documents are divided by chapters, and include fields such as process number, material name, process parameter range, and equipment identifier. Units include degrees Celsius, tons, hours, and others. Production logs are updated daily. Process manuals are updated irregularly with technological iterations. Compliance documents are upgraded in version per regulatory requirements. Most single documents are several thousand characters long.
Constraints on knowledge base retrieval and recall from these characteristics
Documents contain a large number of precisely labeled process texts. Retrieval must support both semantic matching and exact field matching to avoid missing key parameter content. Production data has a high update frequency, so incremental synchronization must be supported to ensure the timeliness of recalled data. Single documents are relatively long. Retrieving too much content will exceed the context window, while retrieving too little will fail to cover all process-related information. Industrial documents of different categories have field differences. Retrieval scope must be strictly isolated to avoid introducing incorrect data from unrelated categories.
Configuration settings
| Configuration Item | Recommended Range | Rationale |
|---|---|---|
similarity | 0.85–0.92 | Non-ferrous metal smelting documents contain a large number of precise process parameters. A threshold that is too low will introduce unrelated non-process text, while a threshold that is too high may miss compliance documents for similar process adjustments |
limit | Top 8–12 entries | Single smelting process documents have relatively long content. Retrieving too many entries will exceed the context window, while retrieving too few will fail to cover all process-related content |
searchMode | hybrid | Balances the semantic matching needs of process requirements and the exact matching needs of material and equipment numbers, adapting to the hybrid retrieval scenario of smelting scenarios |
usingReRank | Enabled | Documents in smelting scenarios contain a large number of similar process descriptions. Reranking can filter results with similar semantic scores but actual mismatches |
datasetSearchUsingExtensionQuery | Disabled | Document drafting for smelting scenarios mostly relies on precise process parameters. Extended queries may introduce unrelated general text |
collectionIds | Fill in according to the smelting-exclusive dataset ID | Isolates knowledge base data from other categories to avoid recalling unrelated documents such as steel and coal |
Three common mistakes
- Phenomenon: No synchronized data in the knowledge base, and the interface shows an empty dataset. Cause: The
collectionIdswas not configured to bind the non-ferrous metal smelting exclusive dataset, causing the retrieval scope to cover unrelated knowledge bases. - Phenomenon: The
similarityscore of retrieval results is lower than 0.85, and the number of exact match results is insufficient. Cause:searchModewas not set to hybrid, and only semantic retrieval was used, which cannot match exact process parameter text. - Phenomenon: The interface continuously displays "Retrieving" with no results returned. Cause: The
limitvalue exceeds 15 entries, orusingReRankwas not configured, leading to a reranking timeout that exceeds the default request time limit.
How to confirm the configuration is correct
- View the dataset details page, confirm that
collectionIdsis bound to the specified non-ferrous metal smelting knowledge base collection, and no unrelated collections are linked. - Initiate a test retrieval, enter a typical process query term, and check whether the
similarityscore of the returned results is within the 0.85–0.92 range. - Check the retrieval log, confirm that
searchModeis set to hybrid,usingReRankis enabled, and there are no timeout errors. - Initiate an incremental synchronization task, confirm that production log documents complete synchronization updates within 1 hour, and retrieval results include the latest production data.
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.