📄 Marker PDF Converter

with PP-DocLayout V3 & Qwen3.5

Upload a PDF and choose your processing options.

• PP-DocLayout V3: Replaces Surya layout detection with PaddlePaddle's RT-DETR instance segmentation model for higher bbox accuracy.
• Qwen3.5 LLM: Enables Marker's use_llm pass for table cleanup, equation rendering, and reading-order correction via the HF inference router.

The first page bbox overlay colours body regions blue and margin/aside regions orange, derived from font-geometric x-regime analysis.

Notes

• LLM calls use your Space's HF_TOKEN free inference credit (~$2). Once exhausted, the LLM option is automatically disabled for the session.
• Processing time: ~15-30s for a 10-page PDF without LLM; ~60s with LLM.
• Overlay shows page 1 only. Aside detection uses PyMuPDF font-geometry (no extra GPU cost).