LLMWhisperer Modes
API Mode Mapping
Each mode below corresponds to a specific value to be passed as the mode parameter in the API:
- Native Text →
mode=native_text - Low Cost →
mode=low_cost - High Quality →
mode=high_quality - Form →
mode=form - Table →
mode=table
💡 Use the
modeparameter in your API call to select one of the modes above.
Feature matrix for LLMWhisperer modes
| Native Text | Low Cost | High Quality | Form | Table | |
|---|---|---|---|---|---|
| PDF (not scanned) | ✓ Yes | ✓ Yes | ✓ Yes | ✓ Yes | ✓ Yes |
| PDF (scanned) | ✗ No | ✓ Yes | ✓ Yes | ✓ Yes | ✓ Yes |
| PDF (with forms) | ✗ No | ✗ No | ✗ No | ✓ Yes | ✓ Yes |
| Images | ✗ No | ✓ Yes | ✓ Yes | ✓ Yes | ✓ Yes |
| MS Office Document | ✗ No | ✓ Yes | ✓ Yes | ✓ Yes | ✓ Yes |
| MS Office Excel | ✗ No | ✗ No | ✗ No | ✓ Yes | ✗ No |
| MS Office Powerpoint | ✗ No | ✓ Yes | ✓ Yes | ✓ Yes | ✓ Yes |
| LibreOffice Writer | ✗ No | ✓ Yes | ✓ Yes | ✓ Yes | ✓ Yes |
| LibreOffice Calc | ✗ No | ✓ Yes | ✓ Yes | ✓ Yes | ✓ Yes |
| LibreOffice Impress | ✗ No | ✓ Yes | ✓ Yes | ✓ Yes | ✓ Yes |
| Checkbox and Radio button detection | ✗ No | ✗ No | ✗ No | ✓ Yes | ✓ Yes |
| Lines reproduction in output | ✗ No | ✓ Yes | ✓ Yes | ✓ Yes | ✓ Yes |
| Extraction performance | Very fast | Medium | Fast | Fast | Fast |
| Image preprocessing (median filter and gaussian blur) | ✗ No | ✓ Yes | ✗ No | ✗ No | ✗ No |
| Line splitting strategy choice | ✓ Yes | ✓ Yes | ✓ Yes | ✓ Yes | ✓ Yes |
| Supported languages | All (unicode) | 120+ | 300+ | 300+ | 300+ |
| Handwriting recognition | ✗ No | Basic support | ✓ Yes | ✓ Yes | ✓ Yes |
| Layout preserving output | ✓ Yes | ✓ Yes | ✓ Yes | ✓ Yes | ✓ Yes |
| AI/ML based enhancement | ✗ No | ✗ No | ✓ Yes | ✓ Yes | ✓ Yes |
| Rotation and skew compensation | NA | ✗ No | ✓ Yes | ✓ Yes | ✓ Yes |
| Auto repair PDFs | ✓ Yes | ✓ Yes | ✓ Yes | ✓ Yes | ✓ Yes |
| Dense text content | Best performance | Very good | Very good | Very good | Very good |
| High entropy content (each page contains large variety of text sizes) | Best performance | Very good | Very good | Very good | Very good |
Recommended use cases
| Native Text | Low Cost | High Quality | Form | Table | |
|---|---|---|---|---|---|
| Recommended use cases | • Low latency requirement • All documents are PDFs • PDFs are native text PDFs • Cost sensitive application | • High quality scanned PDFs • High quality scanned images • No handwritten documents | • Medium/low quality scanned PDFs • Medium/low quality scanned images • Handwritten documents | • Checkbox and radio button detection • Medium/low quality scanned PDFs • Medium/low quality scanned images • Handwritten documents | • Documents containing structured tables • Financial statements and reports • Invoices, bills, spreadsheets • Table-heavy scanned documents |