What this category of data looks like
This category covers the entire production process of hot rolling, cold rolling, section steel, and steel pipe. Daily documents include rolling plans, heating schedules, rolling specifications, plate shape and thickness records, surface quality reports, and quality certificates. Data sources include slab or billet batch information, heating furnace operating parameters, mill automatic control data, online thickness and width measurement results, surface inspection records, and MES system data. Data is updated in real time per coil or per product batch. Document fields include coil number, batch number, heating temperature, rolling reduction rate, thickness deviation, surface defect level, and other fields, with corresponding physical units attached.
What constraints these characteristics impose on the vector models and indexes link
Real-time batch-level updated data requires indexes to support incremental updates, avoiding performance loss caused by full reconstruction. Multi-source heterogeneous data formats require unified preprocessing to ensure consistency in vector generation. The data volume per batch is small, but total data volume grows with production scale. An index structure that supports dynamic expansion is required. Numeric fields such as plate shape and thickness must be processed separately from process specification text to avoid feature confusion. Roll-level traceability requirements require indexes to retain batch association information, ensuring recall results can bind to specific production units. High-frequency calls also require retrieval latency to meet real-time Q&A response requirements.
Configuration settings
This configuration recommendation is adjusted based on FastGPT v4.17.0. Specific parameters are as follows:
| Configuration Item | Recommended Value Range | Rationale |
|---|---|---|
| Chunk Length | 800–1200 characters | Adapt to the semantic integrity of long texts such as rolling specifications and quality certificates, and avoid splitting key process parameters |
| Overlap Length | 100–200 characters | Retain process-related information across chunks, such as the connection between heating schedules and subsequent rolling steps |
defaultToken | Set from representative sample documents | defaultToken is the default token count for direct text chunking; tune it on representative source documents. See the v4.17.2 model-field reference. It is not a vector dimension, embedding-model input cap, or retrieval/LLM context. |
maxToken | Set to the selected embedding model/service maximum supported input tokens | maxToken is the input-token maximum supported by the selected embedding model/service; verify it in that provider configuration. See the v4.17.2 model-field reference. It is not retrieval/LLM context or a FastGPT-wide fixed cap. |
| Similarity Threshold | 0.75–0.85 | Filter low-match non-process-related documents to ensure recall results align with technical standard requirements |
| Single Search Limit | Top 10 entries | Control the scale of returned results to avoid overload while covering major process reference materials |
Three common errors
- Image attachments in surface quality reports cannot be indexed automatically in local deployment environments. The cause is that the corresponding image parsing dependency package is not installed, or the image index switch of the knowledge base is not enabled.
- Embedding model calls fail, returning OneAPI-related errors. The cause is that the interface address and key of the embedding model are not correctly filled in the FastGPT configuration, or the used embedding model does not support hardware acceleration of the current deployment environment.
- Online recall test results do not change after enabling the reranking model. The cause is that the reranking switch is not enabled synchronously during the knowledge base indexing stage, or the calling parameters of the reranking model do not adapt to the length requirements of category documents.
How to confirm the configuration is correctly set
- Import a typical rolling specification document, perform a chunking test, and confirm that key process parameters such as heating temperature and rolling reduction rate are not truncated.
- Call the embedding model interface to verify that the returned vector data format meets FastGPT requirements, with no errors returned.
- Initiate a retrieval targeting a specific defect determination standard, and confirm that the recall results include the corresponding surface quality report or technical standard document.
- After enabling the reranking function, compare retrieval results before and after enabling, and confirm that the result ranking has changed.
The values provided are common starting points and should be measured against the reader's own samples.