What data looks like for this category
Data for this category originates from internal enterprise procurement systems, supplier quotation documents, budget spreadsheets, and contract management systems. Daily workflows produce documents including purchase requisitions, price comparison records, formal contracts, and acceptance documents. Common fields include purchase order numbers, supplier qualifications, budget amounts, delivery timelines, and clause details. Each purchase order triggers a data update cycle. Monthly cumulative update volume fluctuates with procurement scale. Individual documents have a standardized structure but high inter-field correlation. Cross-document associated information concentrates in areas such as supplier performance and budget execution.
Constraints on context and token workflows from these characteristics
Procurement-related documents have high inter-field correlation, and valid information in a single document is relatively scattered. Default chunking granularity risks breaking the association between purchase order numbers, supplier information, and clauses, leading to incomplete recalled information. Each purchase order triggers a data update cycle. New data uses a highly consistent format with existing data, so frequent adjustments to chunking rules are not required. A single recall must cover multiple associated documents under the same order, which easily exceeds the token limit. Supplier quotations and contract clauses have large differences in text length. Long text chunks consume excessive tokens, while short text chunks increase recall redundancy and raise token consumption.
How to configure settings
This article is written for FastGPT v4.17.0. The following configurations adapt to the business characteristics of this category:
| Configuration Item | Recommended Value Range | Rationale |
|---|---|---|
| Chunk Length | 800–1200 characters | Matches the length of single field groups in procurement documents, avoiding breaking associated information such as purchase order numbers and supplier names |
| Overlap Length | 150–200 characters | Retains key associated fields across chunks, ensuring the completeness of recalled information |
maxContext | 8000–10000 tokens | Covers the total token consumption of 3 to 4 associated documents under a single purchase order, aligning with the standard information scope of enterprise procurement workflows |
quoteMaxToken | 1500–2000 tokens | Limits the length of a single recalled text block, preventing long contract clauses or large quotation sheets from occupying excessive context space |
| Single Recall Limit | Top 6 entries | Matches the average number of associated documents per monthly purchase order, reducing token consumption from invalid recalls |
Three common misconfigurations
- Issue: With default chunking parameters, supplier quotation sheets in Excel format are split into overly large data blocks, leading to redundant recall results and excessive token consumption. Cause: Chunk length and
quoteMaxTokenparameters are not adjusted for the field structure of procurement documents, and default chunking does not match the field group length of purchase orders. - Issue: The error "Reached the max retries per request limit" appears during calls. Cause: The number of documents recalled in a single request is not limited, leading to token overlimit for a single request and triggering the retry limit.
- Issue: Context information returned in conversations exceeds the actual needs of the current procurement workflow. Cause: A reasonable
maxContextparameter is not set, and irrelevant historical purchase order information is included in the current context.
How to confirm proper configuration
- Upload a single purchase requisition and its associated price comparison record, and check if the chunked text retains the association between purchase order numbers and supplier names.
- Initiate a query containing procurement requirements, verify the number of recalled documents and the token proportion of single text blocks, and adjust parameters to align with the information scope of the current workflow.
- Review token consumption logs, confirm that token consumption per single conversation does not exceed the preset business threshold, and there are no unnecessary long-text recalls.
- Test multi-order associated queries, confirm that recalled context only includes documents related to the current procurement workflow, and no irrelevant historical data is included.
The values provided are common starting points and should be evaluated against the reader's own samples.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-10-19.