What this category’s data looks like
TCM financing daily report data primarily comes from public financing announcement disclosures for TCM enterprises on the Shanghai, Shenzhen, and Beijing Stock Exchanges, enterprise financing filing information from local financial regulatory departments, and vertical financing monitoring platforms for the pharmaceutical and biotechnology industry. The data updates once daily. Each single data entry includes fields such as enterprise short name, affiliated TCM sub-segment (e.g., proprietary Chinese medicine, Chinese herbal pieces), financing round, financing amount (unit: ten thousand yuan or hundred million yuan), investor entity, financing completion time, and disclosure channel. Most document formats are structured tables or text files with clear field labels.
What constraints these characteristics impose on the model access and configuration phase
The daily update requirement requires configuring scheduled incremental sync tasks during the model access phase, to avoid compute resource consumption from full data pulls. Financing amounts use both ten thousand yuan and hundred million yuan as units, and TCM sub-segments have multiple expression forms. Configure field standardization mapping rules to ensure the model can uniformly identify and process units and segment classifications. Each single data entry contains multiple entity-type information items. Configure entity extraction trigger conditions during model access. Additionally, due to the timeliness requirements of financing daily reports, limit the maximum context window length to avoid interference from redundant old data during real-time inference.
How to set the configurations
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
incremental sync interval | every 24 hours | Matches the daily update cadence of TCM financing daily reports, avoids repeated pulling of old data |
field unit mapping rule | multiply by 10000 when converting hundred million yuan to ten thousand yuan | Adapts to the field feature where financing amount uses both ten thousand yuan and hundred million yuan as units, unifies the numerical unit for model inference |
maxContext | previous 3 rounds of dialogue + current query | Financing daily reports belong to time-sensitive news scenarios, limiting context length avoids interference from redundant data |
chunk size | 800–1200 characters | The combined length of fields for a single TCM financing daily report entry is moderate, this interval ensures complete financing elements are retained after chunking |
entity extraction trigger threshold | 0.75 | Accurately identifies entities such as investors and TCM sub-segments, avoids misjudging non-core fields |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Adapts to the parsing duration of structured financing documents, avoids parsing timeout caused by complex data formats |
The parameter values provided on this page are conventional recommendations used as a starting point 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 settings.
Three common configuration errors
- Phenomenon: After configuring a local Ollama model, the console returns a "504 Gateway Timeout" error. Cause: The FastGPT container does not use host network mode, and cannot establish a network connection with the locally deployed Ollama service.
- Phenomenon: After the knowledge base indexing is complete, a "No available indexing model detected" prompt pops up when creating a new conversation. Cause: The indexing model associated with the knowledge base is not configured as a type that supports vector database retrieval, or the automatic indexing switch for the knowledge base is not enabled.
- Phenomenon: The retrieved financing entries contain duplicates, and the model's response mentions the same financing multiple times. Cause: The
repeat_penaltyparameter is set to a value less than 1.1, which does not effectively suppress repeated content generation, or a unique financing identification field is not added during chunking, leading to duplicate matching.
How to confirm the configuration is correct
- Perform a manual incremental sync once, check if the sync log displays a prompt for successfully pulling new financing data, confirming that the sync interval configuration is effective.
- Enter a query involving financing amount unit conversion, check if the model correctly completes the numerical conversion, confirming that the field mapping rule is effective.
- Initiate a retrieval involving TCM sub-segments, check if the returned results accurately match financing entries for the corresponding segment, confirming that the entity extraction threshold configuration is reasonable.
- Create a new conversation and initiate a query, check if the interface loads the response normally without errors such as "no available indexing model", confirming that the model association configuration is correct.
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.