What the data for this category looks like
Data for special steel yield and market daily reports comes primarily from official websites of domestic special steel industry associations and daily closing quotes from spot trading platforms. All daily data is published before 17:00 on each trading day. Documents are typically in CSV or Excel format. Each row corresponds to one independent special steel quote entry, with the following fields: special steel grade, delivery condition, nominal diameter/thickness, manufacturer, daily listed price, previous day's listed price, price change amount. Units are uniformly yuan per ton. Price change amounts are measured in actual currency units. No percentage-based statistical fields are included.
What constraints do these characteristics impose on deployment and upgrade
Multiple scattered data sources require configuring multi-task pull rules. Set independent pull times and verification logic for each data source. The fixed daily update schedule requires avoiding upgrade operations during the 16:00-18:00 update window to prevent data pull conflicts. Entries uniquely identified by grade, specification and manufacturer require configuring precise deduplication rules to avoid duplicate data entering the knowledge base. Fields include non-standard specification parameters. Retain the original field mapping logic when upgrading parsing templates to prevent parsing failures.
How to set the configurations
These configurations apply to FastGPT 4.8.10 and later versions:
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
PARSE_FILE_TIMEOUT_SECONDS | 300-600 seconds | Special steel daily reports typically contain 50-100 quote entries. A reasonable timeout prevents parsing interruptions |
UPLOAD_FILE_MAX_SIZE | 200-500 MB | Special steel daily reports may include historical comparison attachments, with individual file sizes usually not exceeding 500 MB |
rag_parse_splitter_chunk_size | 800-1200 characters | The text length of a single special steel quote entry is approximately 100 characters. The chunk size can cover complete business units |
recall_top_k | Top 8-12 entries | There are many types and specifications of special steel. Excessive recall increases context pressure, while insufficient recall may miss relevant quotes |
schedule_interval | 16:30 daily | Most spot trading platforms complete daily quote updates before 17:00. Pulling data one hour in advance ensures data timeliness |
duplicate_removal_threshold | Exact match | Special steel quotes are uniquely identified by grade, specification and manufacturer. Deduplication is only performed for exact matches |
The parameter values provided on this page are common recommended starting points for configuration. Actual values are affected by material forms, data volume and business rules. Specific issues require targeted analysis, and it is recommended to test on your own samples before finalizing settings.
Three common configuration mistakes
- A 500 error occurs when importing a special steel daily report Excel file. This happens when a reasonable value for
PARSE_FILE_TIMEOUT_SECONDSis not configured. Parsing timeouts for documents with many rows trigger server errors. - The deployed vLLM inference service cannot be added in the connected platform, with a connection refused prompt. This occurs when FastGPT's
LLM_API_BASEis not configured to the internal network IP and port of the vLLM service, or when firewall permissions for the service port are not opened. - The container runs normally but the FastGPT frontend page cannot be accessed. This happens when the host port and the container's
3000port are not correctly mapped, or when the host firewall blocks the mapped port.
How to confirm the configuration is correct
- Upload a single special steel daily report document, check that the parsed fields include the preset special steel business fields, and that field matching meets business requirements.
- Run a manual data pull task once, confirm that the pulled data sources cover the preset special steel quote channels, and that the data update time matches expectations.
- Test the scheduled task trigger time, confirm that the task completes before the daily update window, with no missing or duplicate data.
- Call the knowledge base retrieval interface, enter a specified special steel grade, check that the returned quote data complies with the preset recall rules.
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.