What data for this category looks like
Industrial metals financial reports and industry data primarily come from official trading platforms including the London Metal Exchange and the Shanghai Futures Exchange, as well as industry institutions such as the China Nonferrous Metals Industry Association. It also includes quarterly and annual public financial reports of domestic listed copper and aluminum smelting enterprises.
Data update schedules differ:
- Monthly spot prices and inventory data are released by the 5th day of the following month
- Quarterly financial reports are disclosed within 45 working days after the end of the quarter
- Annual financial reports must be published by the end of March of the following year
Document structures include core financial statements and detailed category data tables. Fields cover production volume, inventory volume, import and export volume of products such as cathode copper and aluminum ingots. Pricing units include tons and kilograms, with two pricing systems: RMB and USD.
Constraints on citation sources and traceability
Dispersed multi-source data requires clear labeling of data source entities and release times during traceability, to avoid confusion between statistical standards of different institutions.
Differentiated update frequencies mean a single query may match both historical financial reports and real-time spot data. Timestamp filtering is required to ensure recalled data timeliness matches business requirements.
The multi-unit and multi-pricing system of fields requires retaining original unit identifiers during traceability. Without this, data interpretation errors may occur.
Long documents and dense detailed fields require recalled fragments to be accurately positioned to specific data units. This avoids generalized recall causing contextual redundancy. Original professional expressions of fields must also be fully retained.
How to set the configuration
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
recall_top_k | Top 8-12 entries | The volume of industrial metals financial reports and industry data is large. Excessive recall will exceed the context capacity limit of the large model, while insufficient recall may miss key data units |
vector_similarity_threshold | 0.75-0.85 | Financial reports and industry data have strong professionalism. A higher similarity threshold can filter irrelevant content and ensure recalled fragments highly match query requirements |
source_tag_field | Data source name + data release time | This can clearly distinguish different sources such as London Metal Exchange inventory data and Shanghai Futures Exchange financial reports, meeting precise traceability labeling requirements |
parse_chunk_size | 800-1200 characters | Industrial metals financial report appendices have dense fields. A moderate segment length retains complete data units and avoids losing field association relationships after splitting |
rag_enable_source_cite | Enabled | Forcibly attach data source identifiers to responses, meeting traceability compliance requirements for industrial metal scenarios |
function_call_max_rows | 50 | Limit the number of database fragments returned by a single Function CALL to avoid excessive content occupying the large model context |
The parameter values provided on this page are common starting points for configuration. Actual values are affected by material form, data volume and business rules. Specific issues require specific analysis. It is recommended to test on your own samples before finalizing.
Three common configuration mistakes
- Phenomenon: After calling MySQL to retrieve industrial metal inventory data, the response does not display the original database fragment, and only vaguely mentions the data source. Cause: The
rag_enable_source_citeconfiguration is not enabled, and original fragments of database query results are not cached to the conversation context. - Phenomenon: The number of recalled entries from the locally deployed FastGPT knowledge base is fixed and cannot be adjusted. Cause: The
recall_top_kparameter in the deployment configuration file is not modified, and the default value is still used. - Phenomenon: The
finish_reasonfield is not included in the conversation context of subsequent components. Cause: This field is not added to the transfer parameter list of the conversation context, so subsequent components cannot obtain the field value.
How to confirm correct configuration
- Initiate a query containing industrial metal financial report keywords, and check if the response ends with labels for data source names and data release times.
- View the knowledge base recall log to confirm the number of recalled fragments matches the
recall_top_kconfiguration value. - Call Function CALL to retrieve industrial metal data from the MySQL database, and check if original database field content is displayed in the response.
- Modify the
vector_similarity_thresholdvalue, and verify if recall result relevance changes as expected.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-14.