What the data for this category looks like
Data for this category comes primarily from art asset libraries, project management systems, engine resource packages, and outsourced delivery packages. Update frequency is daily or weekly. Commonly processed documents include art requirement sheets, asset specification sheets, outsourced feedback forms, and acceptance records. Most documents have hierarchical entries, numerical parameters, and associated attachments. Fields include resource ID, delivery deadline, polygon count, texture resolution, and more. Some documents contain embedded images or table data.
Constraints on document parsing and chunking
Documents in this category often include structured parameters and embedded resources. Parsing must retain the full integrity of parameters and their associated units to prevent errors during subsequent compliance checks. Mixed data formats from multiple sources require support for Excel files with embedded images, PDF documents with mixed text and images, and other formats. Failure to support these formats will result in loss of key supporting content from outsourced feedback or acceptance records. Frequently updated documents require stable parsing efficiency. Timeouts or parsing failures will delay asset delivery schedules. Differences in hierarchical structures across documents require chunking rules that match entry logic, to avoid splitting a single requirement into multiple incomplete chunks.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Fit the field length of art requirement sheets and asset specification sheets, avoid splitting complete parameter entries with units |
chunkSplitter | Split by heading + paragraph | Match the hierarchical structure of art documents, merge text content under the same requirement entry into one chunk |
customPdfParse | Enable text-image linked parsing | Handle screenshots and text descriptions in PDF acceptance records, retain their corresponding relationship |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Adapt to batch document and compressed package upload requirements for outsourced delivery packages |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Address parsing time for large-volume asset specification documents, avoid triggering default timeout errors |
chunkSettingMode | Customize by document type | Adjust parsing parameters separately for Excel files with embedded images and PDF documents with mixed text and images |
Three Common Misconfigurations
- Phenomenon: When parsing Excel asset specification sheets with embedded images, only table text content is returned, and images are not extracted. Cause: Parsing rules for embedded resources are not configured, only plain text content is extracted.
- Phenomenon: Text and images are separated when parsing PDF documents, and the corresponding relationship between screenshots and descriptions cannot be retained. Cause: The text-image linked parsing mode of
customPdfParseis not enabled. - Phenomenon: Parsing tasks only use CPU resources and do not call idle GPUs. Cause: GPU acceleration parameters are not configured, resulting in excessively long parsing time for large-volume documents.
How to Verify Correct Configuration
- Log in to the FastGPT v4.17.0 backend, enter the parsing configuration page, confirm that parameters such as
chunkSize,chunkSplitter,customPdfParseare configured according to preset values. - Upload a single Excel asset specification sheet with embedded images, check whether the parsing result includes image-related descriptions and complete table content.
- Upload an outsourced delivery package document of about 100 MB, check whether the parsing task is completed within 600 seconds without timeout errors.
- Split a long art requirement sheet, check that the chunking result does not split parameter entries with units, such as retaining complete content like "10000 polygons".
The values provided are common starting points and should be measured 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.