What the data for this category looks like
The data used for declaration materials and file review at state-owned capital operation companies mainly comes from internal state-owned assets supervision systems, internal enterprise declaration and approval systems, and file archive storage repositories. Data updates follow annual declaration batches and quarterly special review cycles. A single declaration material uses a fixed template structure, covering declaration subject qualification information, project investment details, supporting material attachment lists, multi-level approval node records, and other content. Fields include standardized entries such as a unified file number, declaration amount (unit: ten thousand yuan), registered capital (unit: ten thousand yuan), and approval document number. Some materials include multi-page scanned documents and structured tables.
What constraints these characteristics impose on model access and configuration
Data sources are scattered across multiple intranet systems. This requires model access to adapt to intranet environments, and reliance on public network third-party services is not allowed. The length of single file materials varies significantly. Some special review files can reach dozens of pages, which creates clear requirements for the model’s context carrying capacity. Fixed field formats and structured content demand that review results output by the model strictly match preset fields to prevent format deviations. Intranet deployment constraints also require all model services and dependent components to run within the enterprise intranet network segment. Issues related to service connectivity and permission verification in intranet environments must be resolved.
How to set the configurations
The following configurations follow the parameter setting logic of FastGPT v4.17.0:
| Configuration Item | Recommended Value Range | Rationale |
|---|---|---|
maxContext | 8000–12000 token | Single declaration files have a relatively long average length. Full context coverage is required to complete a full review |
maxTokens | 2000–3000 token | Review results must include detailed field checks and compliance judgments. Output length must match the display requirements of review key points |
quoteMaxToken | 1500 token | Supporting material fragments from the file must be cited as review basis. Limiting the length of cited fragments avoids context overflow |
toolChoice | auto | Internal field check tools must be automatically triggered based on declaration material content, to ensure review results match fixed field formats |
temperature | 0.1–0.3 | Reviews must strictly follow fixed state-owned assets supervision rules. Reducing the randomness of model output ensures result consistency |
AIPROXY_API_ENDPOINT | Intranet-deployed ollama service address, such as http://192.168.1.100:11434/v1 | Adapts to the intranet deployment requirements of state-owned capital operation companies, avoiding public network data leakage risks |
AIPROXY_API_TOKEN | Exclusive access key generated by the intranet model service | Verifies model access permissions, ensuring only authorized FastGPT services can call the intranet model |
Three common configuration mistakes
- A database connection refused error occurs during intranet deployment testing, with logs showing the database started normally. The cause is that the FastGPT service and database container were not added to the same intranet network segment, resulting in network connectivity failure.
- Normal calls to models deployed via ollama fail. The cause is an incorrect
AIPROXY_API_ENDPOINTconfiguration, failing to point to the correct intranet model service port or omitting the correct API path suffix. - Review results returned by the model do not accurately match the supporting content of the declaration materials. The cause is an unreasonable
quoteMaxTokenparameter setting, leading to incomplete cited supporting material fragments or fragments that exceed the allowed range.
How to confirm the configuration is properly set up
- In the intranet of the FastGPT deployment environment, use the curl command to test the connectivity of
AIPROXY_API_ENDPOINT, confirming that the model service response can be returned normally. - Upload a standard-format declaration file, and check whether the review results returned by the model include all preset field check items and compliance judgments.
- Check the model call logs to confirm that the actual usage values of
maxContextandmaxTokensdo not exceed the configured parameter ranges. - Test the intranet database connection, confirming that the FastGPT service can normally read data sources for declaration materials and files, with no permission or connection errors.
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 2026-09-28.