What Data for This Category Looks Like
Coke-related data is primarily sourced from domestic main producing area mine ledgers, coastal port spot trading systems, futures delivery warehouse records, and industry association monitoring databases. Update frequencies follow multiple dimensions: spot transaction prices are updated daily, futures settlement prices are updated per trading day, industry supply and demand monitoring weekly reports are released weekly, and monthly industry analysis reports are updated monthly. Common data formats are structured tables or CSV files, with fields including origin, coke grade, fixed carbon content, sulfur content, purchase price, and inventory. Units are respectively Chinese origin names, letter grades, mass fraction, mg/kg, yuan/ton, and tons.
Constraints Imposed on Multi-turn Dialogue and Prompt Engineering
Dispersed data sources and varying update frequencies require multi-turn dialogue to clearly define call priorities for different channel data, to avoid returning outdated information. Structured table and CSV document formats require specifying field matching rules in prompts, to ensure the AI accurately extracts coke data for specific grades and origins. Fields with multiple coexisting units require preset unit unification logic in prompts, to avoid output confusion. Processing large-volume data files requires the dialogue system to support long document parsing and batch data reading, to prevent dialogue interruptions or timeouts caused by excessive data volume.
Configuration Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
UPLOAD_FILE_MAX_SIZE | 1000 MB | Supports upload requirements for large-volume coke data files such as 15,000-row Excel files and 100,000-character Chinese documents |
maxContext | 8000–12000 characters | Accommodates multi-turn dialogue history and parsed multiple sets of coke data fragments to avoid context overflow |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Covers the complete parsing process for large-volume CSV or text documents to avoid mid-process timeout interruptions |
Recall count | Top 8–10 entries | Accurately covers multi-dimensional query requirements such as coke grade and origin mentioned by users, while controlling context load |
Similarity threshold | 0.75–0.85 | Accurately matches user-specified coke attributes, filters irrelevant data recalls, and improves dialogue efficiency |
Rerank result count | Top 5 entries | Focuses on core coke data indicators required for marketing, ensuring output content is concise and meets information density requirements |
The parameter values provided on this page are common recommended starting points for configuration. Actual values are affected by material form, data volume, and business rules. Specific issues require individual analysis, and it is recommended to test on relevant in-house samples before finalizing settings.
Three Common Misconfigurations
- Issue: A
413 Request Entity Too Largeerror is returned when uploading large-volume coke data files via the dialogue interface. Cause: TheUPLOAD_FILE_MAX_SIZEparameter has not been adjusted to a value suitable for large-volume files. - Issue: AI dialogue output takes more than 30 seconds after searching the knowledge base. Cause: The
maxContextvalue is too large or theRecall countconfiguration is set too high, leading to excessive computational load from context loading and data recall. - Issue: Images are incomplete when calling MCP tools to generate coke marketing content. Cause: The prompt does not clearly specify image size specifications adapted to the marketing scenario, or the MCP tool's image output parameters are not configured.
How to Verify Correct Configuration
- Upload test large-volume Excel or text documents, verify the interface returns a normal status code, and confirm the
UPLOAD_FILE_MAX_SIZEconfiguration aligns with the current data scale. - Initiate multi-turn dialogue queries for fields such as coke grade and price, check that the AI output uses unified units, and confirm the unit rules in the prompt have taken effect.
- Simulate a multi-turn interaction scenario, test the information call priority of different data sources, and confirm the output information is the latest updated content.
- Call MCP tools to generate marketing content, check the image display effect, and confirm the prompt matches the parameters configured for the MCP tool.
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.