Data Characteristics in this Category
Policy and SOP documents in health management typically originate from medical institutions, insurance companies, or internal corporate health departments. Content includes disease prevention guidelines, medical examination report interpretation standards, chronic disease management processes, and health consultation SOPs. These documents are updated infrequently, usually quarterly or annually, but may undergo temporary updates due to policy changes or medical breakthroughs. Document structures are primarily hierarchical, containing chapters, sections, charts, and attachments, often in PDF, Word, or internal knowledge base page formats. Core fields include disease codes (e.g., ICD-10), drug dosage units (e.g., mg/kg), test indicator units (e.g., mmol/L), and risk assessment levels. The data often contains extensive specialized terminology, abbreviations, and cross-references.
Constraints from these Characteristics on Deployment and Upgrade
The data characteristics of health management policy documents impose specific requirements on FastGPT's deployment and upgrade. Infrequent document updates mean knowledge base reconstruction or incremental updates do not need to be frequent. However, each update must ensure content completeness and consistency. The hierarchical structure and high density of specialized terminology require a chunking strategy that balances semantic integrity and retrieval efficiency, preventing loss of context after splitting. The presence of numerous specialized fields and units necessitates that the model possess high sensitivity to professional knowledge when understanding and generating responses. Tool calling capabilities become crucial for handling specific calculation or query scenarios. Furthermore, in intranet deployment scenarios, compatibility with external model dependencies, support for tool calling in local models (e.g., Qwen2.5 on Ollama), and decoupling from existing OneAPI services are key considerations during upgrades.
Configuration Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
Chunk size (Chunk Length) | 800–1200 characters | Ensures semantic units in health management policy documents remain complete, preventing truncation of important information. |
Recall count (Recall Count) | Top 8–12 | Considering the professional nature and cross-referencing characteristics of the documents, increasing recall count captures more comprehensive contextual information. |
Similarity threshold (Similarity Threshold) | 0.75–0.85 | Accuracy requirements are high in health management Q&A. A higher threshold filters out irrelevant recall results, improving answer quality. |
Rerank result count (Rerank Return Count) | Top 5 | When recall count is high, reranking selects the most relevant key information to the user's question, reducing the model's processing burden. |
maxContext | 3000–4000 tokens | Health management policy documents contain extensive details and specialized terminology, requiring sufficient context for model understanding. |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Allows ample time for parsing large PDF or Word policy documents, preventing processing failures due to timeouts. |
Three Common Mistakes
- The model fails to execute predefined tools, such as asking for time or performing unit conversions. This usually occurs because locally deployed models (e.g., Qwen2.5 based on Ollama) lack native support for FastGPT's tool calling protocol or are misconfigured.
- Knowledge base retrieval results do not match expectations. Even with reasonable chunk sizes and citation limits, important information may not be cited. This can happen if document chunking breaks critical semantic units or if the retrieval model misunderstands specific professional terms.
- After upgrading the FastGPT main application, some functions are abnormal or fail to start, while the OneAPI service runs normally. This typically results from only updating the FastGPT container image in a Docker environment without simultaneously addressing its dependencies, leading to version compatibility issues.
How to Verify Configuration
- Upload a health management policy document with complex hierarchies and specialized terminology. Check the knowledge base chunk preview to ensure each chunk maintains semantic integrity, especially for sections involving process steps and diagnostic standards.
- Ask questions about specific disease codes, drug dosages, or medical examination indicator units from the document. Verify if the model can accurately identify and provide answers with correct units, or trigger tools for calculations when necessary.
- In a simulated production environment, perform retrieval tests on the knowledge base using the FastGPT client or API. Check if the
recall_chunksfield contains highly relevant and complete text snippets, and compare thesimilarityvalues against the expected threshold. - Perform a FastGPT version upgrade. Then, test core Q&A, knowledge base management, and tool calling functionalities. Confirm all components work correctly, and no significant errors appear in the logs.
The values provided are common starting points and should be measured against the reader's own samples.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-21.