What the data for this category looks like
Core data sources for thermal coal include domestic bulk commodity spot exchanges, industry associations and coastal port monitoring institutions. Data updates follow two rhythms: spot trading prices are updated daily, long-term contract prices are updated monthly. Port inventory and arriving vessel data is updated every half day. Standard document structure includes fields such as origin identifier, calorific value parameter (unit: kilocalories per kilogram), listed price (unit: yuan per ton), weekly trading volume, downstream industry demand share reference items. Some monthly reports include supplementary data on regional transportation scheduling.
Constraints Imposed on Tool Calling and Plugins by These Characteristics
First, daily updated spot price data requires tool calling to adapt to high-frequency APIs. Polling intervals must match data update rhythms. Avoid triggering API rate limits from overly frequent calls, or delayed marketing content from overly long intervals. Second, calorific value uses kilocalories per kilogram as the unit, which differs from other bulk commodity unit conventions. The plugin's field mapping link needs separate unit conversion rules to prevent unit confusion in displayed data. Third, long-term contract prices and spot prices have different update cycles. When calling tools, distinguish API types and adapt to monthly caching and real-time caching strategies respectively. This ensures accurate display of both price types in marketing content. Fourth, port arriving vessel data updates every half day. The tool calling cache expiration time must not exceed 6 hours. This ensures the timeliness of marketing materials.
Configuration Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
stream_response_mode | Enabled | Adapts to large model calls that only support stream mode, ensuring normal synchronization of large model output to the marketing content generation workflow |
plugin_timeout | 300 seconds | Third-party thermal coal data interfaces typically have high response delays. 300 seconds covers most normal call durations, avoiding premature timeout errors |
pgvector_plugin_version | v1.2.0 | This version fixes overflow issues with multi-field vector storage, adapting to the vector retrieval requirements of thermal coal data with multiple fields (calorific value, price, origin) |
text_extract_max_length | 8000 characters | Single texts from thermal coal industry research reports and spot weekly reports typically range from 5000 to 10000 characters. 8000 characters covers conventional extraction scenarios |
api_request_retry_times | 2 retries | Bulk commodity data interfaces occasionally experience network fluctuations. 2 retries reduce call failure rates without excessively consuming interface quotas |
vector_similarity_threshold | 0.75 | Thermal coal marketing content needs to match industry terminology and data characteristics. This threshold filters low-correlation recall results, improving content accuracy |
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 case-by-case analysis. It is recommended to test on your own samples before finalizing.
Three Common Configuration Errors
- Setting
stream_response_modeto disabled will trigger error logs when calling large models that only support stream mode. The error message includes "Failed to obtain data". The cause is a mismatch between the large model's output mode and the configured mode, preventing FastGPT from normally receiving streaming data returned by the large model. - Failing to back up vector indexes before upgrading the PgVector plugin will result in loss of stored thermal coal data vectors. The cause is that no index migration operation was performed during the plugin upgrade, causing local indexes to be incompatible with the new plugin version.
- The text extraction tool returns a
400 Invalid JSON payload received. Unknown nameerror, with the symptom of abnormal tool return fields. The cause is that the submitted JSON request contains fields not defined by the tool, and no field whitelist or validation rules were configured in advance.
How to Confirm Configuration Is Complete
- Enter the FastGPT model configuration page, check the switch status of
stream_response_mode, confirm it matches the output mode of the target large model. - Call the test interface, enter keywords related to thermal coal, check if the tool return fields include core data such as calorific value and price, and that the units conform to industry standards.
- View the PgVector plugin version information, confirm it has been upgraded to the target version, and that the vector index storage status is normal.
- Run a text extraction tool test, upload a thermal coal industry report, check if the returned JSON data only includes preset fields, with no unknown field errors.
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.