What the data for this category looks like
Daily technical documents for SMT mounting and electronic assembly include BOM and coordinate files, line change sheets, stencil and furnace temperature records, first article confirmation reports, SPI/AOI inspection reports, maintenance records and traceability labels. Data sources include BOM, Gerber and coordinate files provided by customers, as well as MES systems, solder paste and component batch ledgers, mounters, SPI/AOI equipment, and reflow oven temperature data. Data update frequency follows each work order, each line change, and real-time updates. Document fields include part number, reference designator, component specification, furnace temperature curve parameters, defect markers, and more. Field names and units vary across documents from different sources.
What constraints these characteristics impose on model access and configuration
The multi-source, heterogeneous document structure requires model access configuration to support parsing and adaptation of mixed text and image data. Real-time and high-frequency updated data requires model call links to have low latency characteristics. Context and output length parameters must be set appropriately. Differences in field names and units require enabling the model's format calibration capability in the configuration. This ensures accurate cross-document information comparison. Core verification goals for different documents, such as part number verification and defect analysis, require adjusting the model's temperature parameters. This balances accuracy and efficiency. Access configuration for different model channels must match the business needs of the SMT scenario. For example, enable the image parsing function to process image content in AOI reports.
How to set the configuration
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
maxContext | 8000–12000 token | Covers the context length after splicing multiple documents such as BOM, line change sheets, and furnace temperature records, avoiding truncation of key information |
maxTokens | 2000–3000 token | Meets output length requirements for tasks including part number verification, defect analysis, and parameter comparison |
temperature | 0.1–0.3 | Ensures result consistency for precision matching tasks such as part number and reference designator verification, reducing random deviation |
vision | Enabled | Supports parsing of defect images in SPI/AOI reports, enabling cross-review of image and text data |
AIPROXY_API_ENDPOINT | Exclusive address provided by the deployment service provider | Matches access paths for different model channels, preventing call failures caused by incorrect addresses |
AIPROXY_API_TOKEN | Exclusive key bound to the corresponding model | Ensures permission validity for model calls, preventing access failures caused by incorrect keys or insufficient permissions |
Three Common Misconfigurations
- Phenomenon: The model selection list in the text content extraction module only displays GPT series options. The accessed GLM model is not shown. Cause: Model access configuration was only completed in the AI chat module, and the adaptation rules for the corresponding model were not synchronously associated in the text extraction module.
- Phenomenon: A "channel not found" prompt appears when calling the AWS Claude3 model. Cause:
AIPROXY_API_ENDPOINTwas not correctly configured as the exclusive access address for the corresponding AWS region, or the boundAIPROXY_API_TOKENdoes not have call permissions for the Claude3 model. - Phenomenon: The cross-review results returned by the model omit component batch and furnace temperature parameters. Cause: The
maxContextvalue is set too low, failing to cover the full context content of BOM and furnace temperature records, leading to truncation of key information.
How to verify successful configuration
- Upload a test BOM file and SPI inspection report, trigger the cross-review task, and check if the model can correctly identify the corresponding relationship between part numbers, reference designators, and inspection defects.
- Enter the model management interface of FastGPT v4.17.0, confirm that all accessed models are displayed in the selection lists of modules such as text extraction and AI chat, and verify the validity of
AIPROXY_API_TOKEN. - Adjust the
temperatureparameter to 0.2, submit 3 sets of duplicate test samples, and check if the returned review results remain consistent. - Upload an AOI defect image, confirm that the model can normally parse the defect markers in the image, and associate the text information of the corresponding document to complete the review.
The values listed above are common starting points, and should be evaluated against applicable use cases and samples.