Data Profile for This Category
Traditional Chinese Medicine (TCM) enterprise financial report data is sourced primarily from compliant channels including official securities exchange disclosure platforms and third-party disclosure portals. Disclosure follows a fixed schedule: annual reports once per year, semi-annual reports once every six months, and quarterly reports once per quarter. A single financial report document typically ranges from tens to hundreds of pages in length, with a structure that includes consolidated financial statements, management's discussion and analysis, and core business data sections. Fields related to TCM raw materials include raw material purchase volume (unit: tons or kilograms), book balance of TCM raw material inventory, and TCM decoction pieces production capacity data. R&D expenditure also includes expenses related to TCM new drug clinical trials.
Constraints on Model Access and Configuration
The long-form documents, specialized terminology, and fixed disclosure schedule of TCM financial reports impose clear constraints on model access and configuration. First, the considerable length of individual financial reports requires connected models to have a sufficiently large context window to prevent truncation of core sections such as consolidated financial statements and management's discussion and analysis. Second, specialized terminology related to TCM raw materials and specific units such as tons and kilograms require domain-specific tokenization adaptation to avoid incorrect term splitting. Third, the fixed quarterly, semi-annual, and annual disclosure schedule can be mapped to scheduled data synchronization trigger rules. Fourth, unique fields including core raw material purchases and TCM production capacity require assigning priority recall weights during index configuration.
Configuration Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 1000–1500 characters | TCM financial reports contain a large volume of specialized terminology and table content. This segment length preserves contextual association and avoids splitting core professional expressions |
maxContext | 12000–20000 characters | The core content of a single TCM financial report is lengthy. A sufficiently large context window can fully load report data for analysis |
similarityThreshold | 0.75–0.85 | TCM financial reports have a high proportion of specialized terminology. This threshold filters low-relevance recall results and avoids interference from invalid information during analysis |
RECALL_NUMBER | 5–8 entries | Core financial indicators and business data in TCM financial reports are concentrated in distribution. 5–8 recall entries can cover key analysis dimensions |
PARSE_FILE_TIMEOUT_SECONDS | 300 seconds | TCM financial report documents have large file sizes and contain multi-page tables. This duration ensures complete parsing without interruption |
SCHEDULE_SYNC_INTERVAL | 86400 seconds | Matches the quarterly, semi-annual, and annual disclosure schedule of TCM financial reports. Daily synchronization enables timely access to newly disclosed data |
The parameter values provided on this page are common recommended starting points for configuration. Actual values are affected by material format, 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 Mistakes
- Scenario: After upgrading to v4.8.19 or later, calls to a locally deployed m3e model via openapi fail, returning the
Model not founderror. Cause: The model configuration validation logic was updated in the new version. Failure to correctly match the model name with the identification field returned by the interface prevents normal model invocation. - Scenario: Configuring oneapi as an embedding model access point results in persistent errors, and financial report data index construction cannot be completed. Cause: A dedicated embedding model was not separately configured for index tasks. Using a general-purpose conversation model as the embedding source leads to a mismatch between model task types.
- Scenario: File parsing times out and interrupts during local deployment when the model interface is not properly configured. Cause: The
PARSE_FILE_TIMEOUT_SECONDSparameter was not adjusted for the large size of TCM financial report documents. Using the default short duration configuration fails to meet the requirements of long document parsing.
How to Verify Successful Configuration
- Upload a single TCM financial report test document, check whether specialized terminology in the parsing results is fully retained with no abnormal splitting, and adjust the segment length parameter based on actual performance.
- Initiate a financial report analysis query, verify whether the recall results include content related to core fields such as TCM raw material purchases and R&D expenditure, and adjust the similarity threshold and number of recall entries to meet analysis requirements.
- Configure a scheduled synchronization task, wait for the task to complete execution, check whether financial report data for the corresponding disclosure period has been added to the knowledge base, and confirm that the synchronization interval and trigger logic are functioning correctly.
- Call the model interface to initiate a test request, check whether the returned results include structured data consistent with TCM financial report analysis logic, and confirm that the context window configuration can cover core analysis content.
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.