What data for this category looks like
Data for this category originates from silicon wafer incoming material records, workshop MES systems, operation logs of diffusion furnaces, coating equipment, and printing presses, EL and IV test reports, and laboratory test data. Update frequencies cover each production batch, single-piece product inspection, and daily summary workflows. Daily documents include process cards, diffusion and coating parameter tables, printing and sintering records, EL inspection images, IV grading data tables, and yield analysis reports. Fields include equipment operating temperature, process duration, electrical performance parameters, and product grading results, with corresponding units of degrees Celsius, seconds, watts, and grade identifiers.
What constraints these characteristics impose on the model access and configuration link
Data sources for this category are scattered, including structured equipment parameters, unstructured EL inspection images, and multi-system summary reports. Update frequencies cover single batches, single-piece inspections, and daily summaries. This requires the model access link to support both structured text and visual inputs. Configuration must enable unified parsing of multi-format data. Single-batch data volume fluctuates with production rhythm, so context and token length limits must be set to avoid excessive resource usage. Cross-system data has strong field correlations, so model calls must associate process and inspection information from different sources. The configuration must also support stable interface calls for batch data.
How to set the configuration
This configuration is based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
maxContext | 8000–12000 token | Photovoltaic cell manufacturing process data includes multi-process parameters and batch inspection records, requiring sufficient historical context for correlation analysis |
maxTokens | 4000–6000 token | Exception report drafting must fully cover EL inspection results and process deviation details to avoid truncating key analysis content |
vision | Enabled | Visual data such as EL inspection images and printing appearance images must be processed to identify abnormalities including microcracks and color differences that are hard to locate quickly with the naked eye |
temperature | 0.1–0.3 | Process documents and exception reports must maintain rigor, preventing generated content from deviating from actual production parameters |
AIPROXY_API_ENDPOINT | Fill in the proxy address deployed within the enterprise | Ensure transmission security of sensitive production data such as silicon wafer incoming materials and equipment operating status |
AIPROXY_API_TOKEN | Exclusive token generated per enterprise permission rules | Restrict unauthorized interface calls to avoid production data leakage |
Three common errors
- Symptom: Interface returns a 503 status code. Cause: Using the default group to call the embedding model without configuring a dedicated model group for photovoltaic production scenarios, leading to resource competition failure.
- Symptom: Interface calls return a
maxTokenslimit exceeded error. Cause: Failing to adjust the token limit based on single-batch process data volume, using general configuration directly and resulting in truncation of key parameters. - Symptom: EL inspection images cannot be parsed normally. Cause: The
visionconfiguration is not enabled, so the model cannot process visual input data.
How to confirm the configuration is complete
- Submit a test request containing single-batch process parameters and EL inspection images, check if the interface returns a normal status, and confirm that the visual parsing function triggers correctly.
- Generate a simulated exception report, verify that the generated content covers core information such as equipment operating parameters and inspection results, and confirm that the context and token configuration do not truncate key content.
- Review the interface call log to confirm that the proxy address and token configuration are correct, with no records of unauthorized access or parameter errors.
- Submit a test request after adjusting the
temperatureparameter, compare the rigor of the generated content, and confirm that the parameter value meets scenario requirements.
The values provided are common starting points and should be measured against the reader's own samples.