What the data for this category looks like
This explanation is based on the configuration logic of FastGPT v4.17.0. Core documents for in-feed ad creative testing include creative scripts, material test sheets, delivery data reports, and iteration checklists. Data primarily comes from ad delivery platforms, material libraries, user interaction data, and conversion data. Documents are updated daily or weekly. Document structures contain both structured fields and unstructured text. Structured fields include material type, delivery time slot, conversion cost. Unstructured content includes creative copy, material optimization suggestions. Field units include common advertising industry metrics such as impressions, clicks, conversions. There is no unified fixed format, and values adjust dynamically based on delivery cycles and testing goals.
What constraints do these characteristics impose on document parsing and chunking
The mixed multi-type nature of category documents requires the parsing process to adapt to both structured tables and long-text scripts. This avoids splitting that breaks the contextual association of creative variables. Frequently updated documents need incremental parsing support. This stops repeated processing of full historical data. Data from multiple sources may have format differences. The parsing process must support document formats exported from different platforms, and retain associations between fields. Losing this association prevents subsequent search tests from matching complete creative combination logic.
How to set configurations
| Configuration Item | Recommended Value Range | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Category documents include long-text creative scripts and test sheets with multiple fields. This range retains the complete context of a single set of creative variables, avoiding splitting that breaks logical associations |
chunkSplitter | Split by paragraph + title | Creative scripts and iteration checklists have clear scene titles and paragraph divisions. Splitting by this rule retains the independent logic of each test unit |
customPdfParse | Enabled | Category PDF documents are mostly delivery data reports and material test sheets with tables. Enabling this retains table structures and field correspondence |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Parsing large delivery data reports takes significant time. This duration prevents parsing processes from being interrupted by timeouts |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Single multi-cycle delivery data reports may contain large volumes of historical data. This setting supports large file uploads and parsing |
chunkSettingMode | Manual mode | Category document structures vary significantly. Manually adjusting parameters adapts to different types of creative documents and test reports |
Three common mistakes
- After uploading a PDF, data processing returns empty, and search test results show no content. The cause is that
customPdfParseis not enabled. This prevents parsing of table-containing delivery data reports or material test sheets, resulting in empty extracted valid content. - After local deployment, training a knowledge base interrupts data processing. The cause is that
PARSE_FILE_TIMEOUT_SECONDSis set too short. Parsing of large data reports is not completed before a timeout interrupt triggers. - Search results show unrelated creative variable combinations. The cause is that
chunkSizeis set too small. This splits complete creative scripts from the same scene, causing contextual breaks. Searches then match unrelated fragments.
How to confirm correct configuration
- Upload a PDF material test sheet with tables. Check if the parsed text retains table fields and corresponding values. Confirm that table structures are not damaged.
- Upload a delivery data report of around 100 MB. Verify that the parsing process has no timeout errors. Confirm that
PARSE_FILE_TIMEOUT_SECONDSis set appropriately. - Adjust the
chunkSizeparameter. Test the contextual integrity of searched creative scripts. Confirm that creative variables from the same scene are not split into different chunks. - Upload a test document of around 500 MB. Confirm that
UPLOAD_FILE_MAX_SIZEmeets requirements, and no file upload failure prompts appear.
The values provided here are common starting points. They should be evaluated against samples relevant to the specific use case.
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.