What data for this category looks like
TCM investment research data originates primarily from official standards issued by the National Pharmacopoeia Committee, local TCM processing specifications, clinical diagnosis and treatment guidelines, component test reports, and medicinal material traceability archives. Data updates align with revisions to official standards, approval of new medicinal materials, and publication of clinical research results. There is no fixed single update cycle. A single standard document usually includes sections such as medicinal material origin, nature and flavor meridian tropism, functions and indications, usage and dosage, chemical components, pharmacological research, and processing methods. Documents for prepared slices additionally include specifications and content determination items. Most field units use general medical measurement standards such as grams, milligrams, and milliliters.
Constraints on context and token workflows
Long text features of TCM investment research documents, such as complete pharmacopoeia entries and multi-component test reports, increase the length of individual context segments and consume more token quotas. Documents contain a large number of professional terms and field associations, such as the binding relationship between nature and flavor meridian tropism and functions and indications. Improper segmentation will damage semantic integrity. Newly updated data must be synchronized to the knowledge base in a timely manner. If the context recall range is not adjusted, the latest research conclusions may be missed. Document lengths vary widely across different medicinal materials. Some pharmacological research documents can reach several thousand characters. If the single recall length is not limited, the context token limit will be exceeded.
Configuration recommendations
| Configuration Item | Recommended Range | Rationale |
|---|---|---|
maxContext | 8000–12000 characters | Adapts to the long text characteristics of TCM documents, prevents individual context segments from exceeding the model's token limit |
chunkSize | 1000–1500 characters | Preserves field integrity when splitting long documents, avoids breaking associated content such as components and pharmacological data |
topK | 6–10 entries | Balances recall coverage and token consumption, meets the precise recall requirements for multiple fields |
similarityThreshold | 0.75–0.85 | Filters low-match irrelevant documents, reduces invalid token usage |
rerankTopN | 3–5 entries | Retains highly relevant entries after reranking recall results, optimizes context token usage efficiency |
tokenLimitPerChat | 120000–150000 characters | Adapts to the context accumulation requirements of multi-round investment research conversations |
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. It is recommended to test on your own samples before finalizing settings.
Three common misconfigurations
- Phenomenon: The displayed token consumption does not match the configured model version. The actual billing logic has not switched to the target v2 version. Cause: The associated configuration items for the v2 version were not updated synchronously on the model binding page. The system still calls the token billing rules of the old version.
- Phenomenon: When initiating a call using a custom OpenAPI token, platform default billing records are still generated. Cause: The associated permissions for the custom token were not bound in the knowledge base call configuration. The system still calls the platform's default billing logic.
- Phenomenon: A single-round investment research conversation triggers a token limit exceeded error, returning status code 413. Cause: The token limit for single-round conversations was not configured, and excessively long TCM document fragments were not split. This causes the total context token to exceed the model's supported range.
How to confirm proper configuration
- Access the knowledge base configuration page, and verify that values for parameters including
maxContextandchunkSizematch the preset configuration. - Initiate a test recall for a single TCM document, and check whether returned context fragment length and number of recalled entries align with expected configuration settings.
- Initiate a multi-round investment research conversation, and observe whether token consumption statistics match the configured
tokenLimitPerChat. - Review billing-associated configuration on the model binding page, and confirm that parameters have been synchronously updated to the target v2 version.
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.