This document is written for FastGPT v4.17.0.
What This Category’s Data Looks Like
Data sources for this category include wafer test maps, MES systems, bonding and molding equipment logs, ATE tester data, X-ray/SAM inspection reports, and customer specification documents. Daily documents cover packaging process cards, bonding and molding parameter sheets, incoming wafer maps, test program version files, yield reports, and failure analysis reports. Data updates occur per production batch, per tested chip, and daily aggregation. Fields include wafer ID, batch number, bonding pressure, test voltage, bin code, and others, with corresponding units such as millimeters, newtons, volts, and more.
Constraints on Model Access and Configuration
The high-frequency, fine-grained update rhythm of this category requires model access configuration to match incremental data sync frequency, preventing context overflow. Diverse data types include structured parameters, image detection results, and text reports. Configure multimodal processing parameters to support different input formats. Fields with dedicated units require accurate matching between fields and units during model invocation to avoid parsing errors. Decentralized data sources need unified access configuration to ensure logs from all devices and systems can be pulled and parsed correctly.
Configuration Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
maxContext | 8000–12000 token | Adapts to the ledger data summarization needs of multiple batches and multiple fields for this category, avoiding context overflow |
maxTokens | 4000 token | Meets the full generation requirements for a single yield report and failure analysis report |
vision | Enabled | Used to parse multimodal data such as incoming wafer maps and X-ray/SAM inspection images |
temperature | 0.2–0.4 | Ensures the accuracy of ledger summarization, avoiding deviations in generated results |
AIPROXY_API_ENDPOINT | Fill in according to the actual deployed proxy address | Adapts to decentralized data sources and cross-device access needs |
AIPROXY_API_TOKEN | Fill in according to the assigned exclusive token | Ensures access security and permission control |
Three Common Misconfigurations
- Phenomenon: An error stating that multimodal input is not supported is returned when calling the model. Cause: The
visionconfiguration item is not enabled, or the connected model does not support multimodal capabilities. - Phenomenon: A 404 body not found error is returned after configuring the proxy. Cause: The
AIPROXY_API_ENDPOINTis filled incorrectly, or there is a configuration conflict with other proxy services. - Phenomenon: Field units in the generated ledger do not match actual values. Cause: No unit matching logic is configured for dedicated fields of this category (such as bonding pressure, test voltage), leading to model parsing deviations.
How to Confirm Proper Configuration
- Upload one incoming wafer map and the corresponding batch’s MES data, verify that the model can properly parse the image content and extract relevant fields.
- Initiate a ledger summarization request, check that the returned context length does not exceed the configured limit.
- View the proxy service’s running logs, confirm that calls to
AIPROXY_API_ENDPOINTandAIPROXY_API_TOKENhave no abnormalities. - Test multiple sets of different types of input data, verify that the report fields generated by the model match actual business rules accurately.
The values provided are common starting points and should be measured against the reader's own samples.