Upload 3 files

README.md

@@ -1,15 +1,52 @@
 ---
-title: Docling  Layout Demo
-emoji: 📚
-colorFrom: pink
-colorTo: gray
+title: PP-DocLayoutV3 Empirical Parser
+emoji: 📄
+colorFrom: blue
+colorTo: indigo
 sdk: gradio
 sdk_version: 6.9.0
-python_version: '3.12'
 app_file: app.py
 pinned: false
 license: mit
-short_description: docling-pp-doc-layout based document conversion demo
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# PP-DocLayoutV3 Pipeline: Empirical Iteration Guide
+
+This application provides an extraction pipeline using `docling-pp-doc-layout`
+running on Hugging Face's ZeroGPU infrastructure (70 GB VRAM NVIDIA H200).
+Because instance-segmentation-based layout parsing exhibits high variance in
+memory utilisation based on polygon density and image resolution, this Space is
+engineered for iterative, data-driven optimisation.
+
+## Architecture
+
+| Component | Value |
+|---|---|
+| Hardware | Hugging Face ZeroGPU (`@spaces.GPU`, large tier — half H200) |
+| SDK | Gradio 6.9.0 |
+| Python | 3.12 (ZeroGPU supports 3.12.12 and 3.10.13; 3.13 is **not** supported) |
+| Layout model | `PaddlePaddle/PP-DocLayoutV3_safetensors` |
+| GPU timeout | 120 s (`duration=120`) |
+
+## Iterative Deployment Protocol
+
+### 1. Memory Profiling and Batch Optimisation
+
+`PPDocLayoutV3Options` is initialised with `batch_size=2` as a conservative
+baseline. Monitor ZeroGPU hardware logs for OOM evictions. The large tier
+provides 70 GB VRAM, so `batch_size` can be incremented sequentially until
+utilisation approaches the ceiling.
+
+### 2. Confidence Threshold Calibration
+
+`confidence_threshold=0.5` is the default decision boundary. Evaluate output
+classifications against a validation set:
+
+- **Higher threshold** → higher precision, fewer false positives
+- **Lower threshold** → higher recall, fewer missed bounding boxes
+
+### 3. Queue Latency and Hardware Timeouts
+
+ZeroGPU enforces a 60 s default GPU lease. The `@spaces.GPU(duration=120)`
+annotation extends this to 120 s. If empirical data shows consistent sub-60 s
+inference, reduce `duration` to improve queue priority for Space visitors.

app.py

@@ -0,0 +1,55 @@
+import gradio as gr
+import spaces
+from docling.datamodel.base_models import InputFormat
+from docling.document_converter import DocumentConverter, PdfFormatOption
+from docling.datamodel.pipeline_options import PdfPipelineOptions
+from docling_pp_doc_layout.options import PPDocLayoutV3Options
+
+# Global initialisation — pipeline is constructed lazily on the first
+# convert() call, which happens inside @spaces.GPU, so decide_device()
+# correctly resolves "cuda:0" when the H200 is allocated.
+pipeline_options = PdfPipelineOptions(
+    layout_options=PPDocLayoutV3Options(
+        batch_size=2,
+        confidence_threshold=0.5,
+    )
+)
+
+converter = DocumentConverter(
+    format_options={
+        InputFormat.PDF: PdfFormatOption(pipeline_options=pipeline_options)
+    }
+)
+
+
+@spaces.GPU(duration=120)
+def infer_layout(file_path: str | None):
+    if not file_path:
+        return {"error": "No file uploaded"}
+    try:
+        result = converter.convert(file_path)
+        structured_data = []
+        for item, _level in result.document.iterate_items():
+            structured_data.append({
+                "type": type(item).__name__,
+                "content": getattr(item, "text", "No text mapping"),
+            })
+        return structured_data
+    except Exception as e:
+        return {"runtime_exception": str(e)}
+
+
+with gr.Blocks(title="PP-DocLayoutV3 Empirical Parser") as interface:
+    gr.Markdown(
+        "## Layout Detection Inference\n"
+        "Upload a PDF to parse structural components through the "
+        "PaddlePaddle PP-DocLayoutV3 model."
+    )
+    with gr.Row():
+        pdf_input = gr.File(label="Source Document", file_types=[".pdf"])
+        json_output = gr.JSON(label="Structured Extraction Matrix")
+    execute_btn = gr.Button("Initialize Inference")
+    execute_btn.click(fn=infer_layout, inputs=pdf_input, outputs=json_output)
+
+if __name__ == "__main__":
+    interface.launch()

requirements.txt

@@ -0,0 +1,2 @@
+docling-pp-doc-layout
+spaces