What data looks like for this category
The data for this category primarily originates from ERP orders and inventory, bills of materials (BOMs) and process routes, equipment capacity, staff shifts, work-in-progress items, and supplier deliveries. Daily processed documents include sales orders, master production schedules, material requirements plans, weekly and daily scheduling tables, work orders, material shortage lists, and delivery change notices. Fields include order numbers, material codes, process working hours, equipment numbers, shift counts, and more. Units include pieces, hours, days, and others. Data update frequencies are daily, weekly, or triggered in real time alongside orders or abnormal events.
What constraints these characteristics impose on model access and configuration
Documents for production planning and shop scheduling vary widely in length. Short documents such as material shortage lists may only contain a few dozen characters, while long documents like weekly scheduling tables can span thousands of characters. This requires model context configuration to adapt to different input lengths. Multi-source heterogeneous data structures require the model to accurately retrieve relevant document fragments during calls. High data update frequencies and random trigger scenarios require configurations that support dynamic document synchronization and real-time calls. Additionally, scheduling reviews must maintain consistent results, avoid suggestions that violate production constraints due to model randomness, and support autonomous tool calls to handle dynamic production scenarios such as order insertion, material shortages, and rescheduling.
How to set configurations
The following configurations apply to FastGPT v4.17.0.
| Configuration Item | Recommended Value Range | Rationale |
|---|---|---|
AIPROXY_API_ENDPOINT | Fill in the official access endpoint provided by the third-party large model platform | Adapt to the request path of the target model to ensure that call requests are forwarded correctly |
AIPROXY_API_TOKEN | Fill in the valid API access key generated by the platform | Complete identity verification to obtain legal permissions for model calls |
maxContext | 8000–12000 token | Match the average length of production planning-related documents, avoid context truncation that loses core constraint information such as ERP orders and BOMs |
maxTokens | 2000–3000 token | Adapt to the output length of scheduling analysis results, ensure that content such as process planning and adjustment suggestions is fully returned |
quoteMaxToken | 1500–2000 token | Sufficient to retrieve relevant document fragments such as work orders and bills of materials, support basis matching for technical cross-review |
temperature | 0.2–0.4 | Reduce the randomness of model output, ensure that review results related to scheduling remain consistent and comply with production constraint requirements |
| Workflow/API tool selection | auto | Allow the model to independently identify tools that need to be called, adapt to dynamic production scenarios such as order insertion, material shortages, and rescheduling |
Three common configuration errors
- Model calls return a 401 status code, prompting that the key is invalid. This occurs when
AIPROXY_API_TOKENis not filled in correctly, or an expired third-party platform key is used. - Scheduling suggestions output by the model lack key material constraint information. This occurs when
maxContextis set too small, causing core documents such as ERP orders and BOMs to be truncated. - Tool calls fail, and knowledge base retrieval cannot be triggered. This occurs when
toolChoiceis incorrectly set to a fixed mode, limiting the model's ability to call tools autonomously.
How to confirm configurations are correctly set
- Initiate a test request, enter a query that includes an order number and material shortage situation, confirm that the interface returns a normal response with no error messages.
- Review the model output content, confirm that relevant work order and scheduling table fragments are included, verify that the retrieval range of
quoteMaxTokenmeets expectations. - Adjust the
temperaturevalue, compare the outputs of two identical queries, confirm that result consistency meets the stability requirements of production reviews. - Trigger a test request for an order insertion scenario, confirm that the model can autonomously call tools to generate an adjusted scheduling plan, verify that the
toolChoiceconfiguration takes effect.
The values provided are common starting points and should be measured against the reader's own samples.