What data for this category looks like
Daily documents for this category include purchase orders, delivery acceptance slips, serial number lists, and invoice materials. Data sources cover ERP systems, supplier shipping documents, logistics tracking information, and equipment manufacturer backends. Data updates follow per-order or weekly rhythms. Single-task documents may include structured Excel serial number lists, plus unstructured PDF acceptance slips and invoices. Core document fields include equipment model, serial number, purchase order number, delivery date, and acceptance result. Some documents include unit information such as units or pieces.
Constraints imposed on document parsing and chunking
Multi-source document format differences require the parsing step to support multiple file types. This prevents loss of key fields due to format incompatibility. Per-order data updates require chunking rules to match the document volume of a single task. This avoids disrupting subsequent model configuration checks and serial number registration due to overly fine or coarse chunking. Structured fields such as device serial numbers and models must not be split. Splitting these fields will cause errors in subsequent checks. High document update frequency means parsing timeout settings must align with business processing rhythms. This prevents process delays from timeout interruptions.
How to set configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Fits the field density of purchase orders and delivery acceptance slips, avoids splitting key fields such as device serial numbers and models |
chunkSplitter | Line break + custom delimiter | Fits Excel and PDF documents pre-segmented with line breaks, preserves integrity of structured fields |
customPdfParse | Enabled | Processes PDF formats of delivery acceptance slips and invoice materials, retains field information from original layout |
UPLOAD_FILE_MAX_SIZE | 3000 MB | Accommodates large-volume document packages for single-batch multiple devices, prevents upload errors |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Fits parsing duration of multi-page PDFs and batch Excel files, prevents timeout interruptions |
chunkSettingMode | Manual mode | Allows adjustment of chunking rules for highly structured documents such as serial number lists |
Three common misconfigurations
- Uploading a 3 MB PDF document triggers an error. The cause is failure to adjust the
UPLOAD_FILE_MAX_SIZEconfiguration. The default setting does not cover this document size. - After setting a custom delimiter, chunking results still merge multiple paragraphs or split single paragraphs. The cause is failure to set
chunkSettingModeto manual mode. The system automatically overrides the custom chunking rules. - Processing PDFs throws an error after deploying Marker and configuring environment variables. The cause is failure to correctly enable the
customPdfParsefunction, or the deployed Marker version is incompatible with FastGPT v4.17.0.
How to confirm configurations are properly set
- Upload a single 3 MB PDF delivery acceptance slip. Check that the upload and parsing process has no errors. Confirm that the
UPLOAD_FILE_MAX_SIZEconfiguration takes effect. - Upload an Excel serial number list pre-segmented with line breaks. Check that the chunking results do not split serial number and model information for a single device. Confirm that the
chunkSplitterandchunkSettingModeconfigurations match. - Submit a batch parsing task and wait no more than 600 seconds. Check the task completion status. Confirm that the
PARSE_FILE_TIMEOUT_SECONDSconfiguration is appropriate. - After enabling
customPdfParse, upload an invoice PDF. Check that the parsing result retains the layout integrity of original fields such as invoice number and invoice date.
The values provided here are common starting points, and should be evaluated against the deploying party’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.