This document is compatible with FastGPT v4.17.0.
What the Data for This Category Looks Like
Business data for photovoltaic cell manufacturing is multi-source and has clear structured characteristics. Data sources include silicon wafer incoming material records, production MES systems, operating parameters of diffusion furnaces, coating equipment, and printing presses, EL and IV test reports, and laboratory quality inspection data. Data is updated by production batch and single-chip test results. Daily documents include process cards, diffusion and coating parameters, printing and sintering records, EL inspection images, IV grading data, and yield reports. Core fields include diffusion temperature, coating thickness, EL defect coordinates, conversion efficiency, and others. Corresponding units include degrees Celsius, micrometers, millimeters, watts per square centimeter, and more.
Constraints for Multi-turn Dialogue and Prompt Engineering
Since data is updated by batch and single chip, multi-turn dialogue must bind context for the corresponding production batch to avoid mixing data across batches. Multimodal EL inspection images and IV test data require prompts that clearly specify parsing rules. This ensures output formats match business fields. Core process parameter fields and units are strictly defined, so prompts must limit the range of queryable fields to prevent parameter confusion. Frequently updated batch data means dialogue context cannot be overly redundant. Limit the number of historical dialogue entries to avoid overload that affects response speed.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
messages historical entry count | First 6–10 entries | Matches the update rhythm of single-batch data to avoid context overload |
chatId binding | Bind by production batch | Ensures dialogue is associated with process and inspection data for the same batch, avoiding cross-batch confusion |
| Multimodal input support | Enabled | Supports parsing non-text business data such as EL inspection images and IV test data |
variables field restriction | Restrict exclusive fields including diffusion temperature, coating thickness, etc. | Aligns with process parameter specifications for photovoltaic cells, preventing parameter confusion |
| Maximum context length | 1000–1500 tokens | Adapts to the total character count of single-batch business data, balancing context completeness and response speed |
| Single request timeout | 600 seconds | Adapts to the time required for multimodal image parsing, avoiding premature timeout triggers |
Three Common Misconfigurations
- Phenomenon: Multi-turn dialogue in the workflow cannot continue batch data. Cause: Context for the corresponding production batch is not bound via
chatId. - Phenomenon: Multimodal image dialogue triggers format errors. Cause: The prompt does not clearly specify parsing fields and output format for EL inspection images.
- Phenomenon: Online dialogue has no response for an extended period. Cause: Chat history entry count is not limited, leading to context overload and timeout triggers.
How to Verify Proper Configuration
- Check if the dialogue context automatically associates process parameters and inspection data for the current production batch.
- Test uploading an EL inspection image and asking about defect locations, confirming that corresponding business fields can be correctly extracted.
- Check the chat history configuration of workflow nodes, confirming that the
chatIdfor the corresponding batch has been bound. - Initiate consecutive multi-turn questions, confirming that the context can correctly continue the batch information and parameter range from previous questions.
The values provided here are common starting points, and should be evaluated against the reader's own samples.