Reference only; verify against the target FastGPT release and deployment documentation.
What the data for this category looks like
Core data for this category comes from national public resource trading platforms, local public resource trading platforms, project approval systems, and bidding and tendering supervision systems. The update frequency is once per day. Commonly processed documents include bidding announcements, qualification pre-examination documents, bidding documents, tender documents, bid evaluation reports, successful bid candidate announcements, and successful bid result announcements. Document structures include clear section titles, section divisions, qualification requirements, bill of quantities, bid evaluation records and other fields. Some documents include images, tables, or multi-format attachments.
Constraints Imposed on Document Parsing and Chunking
Document formats from multiple sources are inconsistent. Parsing must support multiple formats including PDF, Word, and images. Daily updated batch documents require high efficiency in parsing and chunking to avoid delays. Documents contain highly business-related content such as sections and qualification requirements. Chunking must retain semantic integrity and avoid splitting critical business units. Some documents have large file sizes. Resource limits for large file upload and parsing must be accommodated. The party building learning Q&A scenario requires parsed content to accurately match bidding and tendering related policy requirements, and avoid missing key compliance information.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Bidding and tendering documents include long paragraphs of bills of quantities and bid evaluation records. This length retains business semantic integrity while avoiding overly fragmented chunking |
chunkSplitter | Split by paragraph + title | Documents have clear section titles and section divisions. Splitting by this rule retains the coherence of business logic |
PARSE_FILE_TIMEOUT_SECONDS | 300–600 seconds | Single large bidding documents include multiple attachments. This duration covers the parsing time of most documents and avoids mid-process timeout interrupts |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Accommodates large document packages containing multiple tender document attachments, meeting batch upload requirements |
customPdfParse | Enabled | Most bidding documents are official standardized PDF formats. Enabling custom parsing allows more accurate extraction of structured content such as tables and titles |
chunkSettingMode | Manual configuration | Chunking parameters need to be adjusted for the specific structure of bidding documents. Manual configuration allows flexible adaptation to business needs |
Common Mistakes
- Phenomenon: Parsing result is empty or contains no text after providing an image URL. Cause: Access permissions for remote file parsing are not configured, or custom parsing rules for image formats are not enabled.
- Phenomenon: Parameter error is triggered when calling the workflow API to upload files. Cause: Image or document parameters are not passed in the multipart/form-data format required by the interface, and JSON format is incorrectly used for parameter passing.
- Phenomenon: Section information is truncated in chunking results. Cause:
chunkSplitteris configured to split by fixed characters, instead of splitting by title or paragraph boundaries, causing section content to be split into different chunks.
How to Verify Correct Configuration
- Upload a standard bidding document sample, check whether the parsed text completely extracts core business content such as sections and qualification requirements.
- Call the parsing interface, check whether the length of each chunk in the returned chunk list matches the
chunkSizesetting range. - Check system operation logs to confirm that the time taken to parse a single document does not exceed the
PARSE_FILE_TIMEOUT_SECONDSconfiguration value. - Test uploading a document with image attachments, confirm that the parsed result includes text information from the images.
The values provided here 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 2026-09-28.