What the data for this category looks like
Data sources for the photovoltaic module encapsulation process include the cell grading system, glass adhesive film and backplane batch ledgers, the workshop MES system, data collected from string welding and lamination equipment, and output reports from EL/IV testers. Data updates follow a schedule where a unique ledger record is generated for each module, and batch information is updated synchronously for each batch of raw materials. Daily documents include layout drawings, material batch records, lamination process parameter records, EL defect images, IV test reports, insulation and voltage withstand test records, and finished product packing lists. Fields covered include module number, cell current grade, lamination temperature and duration, crack location grade, power value, voltage withstand value, and more.
What constraints these characteristics impose on model access and configuration
This category has high data update frequency and small per-record size, so model access configuration must adapt to real-time small-batch data processing to avoid context overflow. Multiple types of documents coexist, including structured parameter records and image-based EL defect images, so model capabilities that support visual recognition must be configured. Professional fields such as cell current grade and lamination process parameters require the model to accurately identify and match them, so configuration must ensure the model’s understanding accuracy of electronic manufacturing terminology. Data sources are scattered and field standardization varies, so data mapping rules must be clearly defined during configuration to reduce subsequent summary errors.
How to set the configuration
This document is written based on the configuration logic of FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
maxContext | 8000–12000 token | Adapts to multi-document splicing for single-batch module ledgers, avoids context overflow |
maxTokens | 2000–3000 token | Matches the output length of a single report generation, covers all field summary requirements |
vision | Enabled | Supports content recognition of EL defect images and IV test report images, meets the processing needs of image-based documents in this category |
temperature | 0.1–0.3 | Reduces output randomness, ensures accuracy of professional parameter summary |
AIPROXY_API_ENDPOINT | http://{Intranet Deployment IP}:8001/v1 | Adapts to intranet deployment scenarios, ensures data transmission security |
AIPROXY_API_TOKEN | Access key generated during deployment | Verifies model call permissions, prevents unauthorized access |
Three Common Configuration Mistakes
- Phenomenon: After deploying a rerank model, the corresponding model option does not appear in the FastGPT platform. Cause: The rerank model access configuration was not correctly added to the system configuration file, or the deployment address was not added to the allowed access whitelist.
- Phenomenon: After configuring the model as the v2 version, the actual call still uses the v1 version. Cause: The model name selected on the platform does not match the model name registered in the configuration file, or the configuration file was not correctly reloaded to take effect.
- Phenomenon: After deploying the model intranet, the service process restarts repeatedly. Cause: Correct network access permissions were not configured in the intranet environment, causing external resources relied on by the model to fail to load, or insufficient memory allocation in the startup parameters.
How to Confirm Successful Configuration
- Enter the FastGPT model management page, view the list of connected models, and confirm that the target model is displayed normally.
- Initiate a simulated call, input test data containing EL image parameters and lamination records, and check whether the output result includes accurate summary of the corresponding fields.
- View the platform logs, confirm that the parameters of the model call request match the configured items, with no missing or incorrect parameters.
- Test model call connectivity in the intranet environment, confirm that the request can normally reach the deployment address and return valid results.
The values given are common starting points and should be measured against the reader's own samples.