refactor(gradio): focus app on structured extraction

Remove the content task and Markdown outputs so the Space exposes one structured extraction pipeline with one JSON result. Update examples and documentation to match the simplified API.

README.md

@@ -10,14 +10,14 @@ app_file: app.py
 pinned: false
 models:
   - numind/NuExtract3
-short_description: Extract Markdown or JSON from documents with NuExtract3.
+short_description: Extract structured JSON from documents with NuExtract3.
 ---
 
 # NuMark
 
-NuMark is a Gradio app for extracting Markdown or structured JSON from document images with [NuExtract3](https://huggingface.co/numind/NuExtract3).
+NuMark is a Gradio app for extracting structured JSON from document images with [NuExtract3](https://huggingface.co/numind/NuExtract3).
 
-Upload a receipt, invoice, or other document image, optionally add accompanying text, and choose either content extraction or schema-guided structured extraction.
+Upload a receipt, invoice, or other document image, optionally add accompanying text, and choose a JSON template for schema-guided structured extraction.
 
 The inference function uses Hugging Face ZeroGPU dynamic GPU allocation. Select ZeroGPU in the Space hardware settings before launching the app.
 

app/gradio.py

@@ -262,7 +262,7 @@ def run_nuextract(messages, **chat_template_kwargs):
     )[0].strip()
 
 
-def extract(image, text, mode, template, enable_thinking):
+def extract(image, text, template, enable_thinking):
     result = run_nuextract(
         [
             {
@@ -273,21 +273,18 @@ def extract(image, text, mode, template, enable_thinking):
                 ],
             }
         ],
-        mode=mode,
+        mode="structured",
         template=template,
         enable_thinking=enable_thinking,
     )
 
-    if mode == "content":
-        return result, result, gr.skip()
-
-    return gr.skip(), gr.skip(), json.loads(result)
+    return json.loads(result)
 
 
 with gr.Blocks(title="NuMarkApp") as demo:
     gr.Markdown(
         "# NuMarkApp\n"
-        "Extract Markdown or structured JSON from an image and optional text using "
+        "Extract structured JSON from an image and optional text using "
         "[numind/NuExtract3](https://huggingface.co/numind/NuExtract3)."
     )
     with gr.Row():
@@ -298,11 +295,6 @@ with gr.Blocks(title="NuMarkApp") as demo:
                 placeholder="Add text to process alongside the image",
                 lines=3,
             )
-            mode = gr.Radio(
-                ["content", "structured"],
-                value="content",
-                label="Task",
-            )
             enable_thinking = gr.Checkbox(label="Enable thinking")
             template_preset = gr.Dropdown(
                 choices=list(structured_json_templates),
@@ -317,22 +309,13 @@ with gr.Blocks(title="NuMarkApp") as demo:
                     lines=18,
                     info="Used for structured extraction. Edit the example to match your document.",
                 )
-            run = gr.Button("Extract", variant="primary")
-        with gr.Column(), gr.Tabs():
-            with gr.Tab("Content"):
-                markdown_text = gr.Textbox(
-                    label="Markdown source",
-                    lines=28,
-                    buttons=["copy"],
-                )
-                with gr.Accordion("Markdown preview (optional)", open=False):
-                    markdown_preview = gr.Markdown()
-            with gr.Tab("Structured"):
-                structured_output = gr.JSON(
-                    label="Structured JSON",
-                    open=True,
-                    show_indices=True,
-                )
+            run = gr.Button("Extract structured JSON", variant="primary")
+        with gr.Column():
+            structured_output = gr.JSON(
+                label="Structured JSON",
+                open=True,
+                show_indices=True,
+            )
 
     gr.Examples(
         examples=[
@@ -345,7 +328,6 @@ with gr.Blocks(title="NuMarkApp") as demo:
                     / "receipt-ocr-original.webp"
                 ),
                 "",
-                "structured",
                 False,
             ],
             [
@@ -357,7 +339,6 @@ with gr.Blocks(title="NuMarkApp") as demo:
                     / "invoice-with-items.png"
                 ),
                 "",
-                "structured",
                 False,
             ],
             [
@@ -369,7 +350,6 @@ with gr.Blocks(title="NuMarkApp") as demo:
                     / "BankStatementChequing.png"
                 ),
                 "",
-                "structured",
                 False,
             ],
             [
@@ -381,7 +361,6 @@ with gr.Blocks(title="NuMarkApp") as demo:
                     / "bank statement blog image.webp"
                 ),
                 "",
-                "structured",
                 False,
             ],
             *[
@@ -389,7 +368,6 @@ with gr.Blocks(title="NuMarkApp") as demo:
                     default_structured_json_template,
                     str(path),
                     "",
-                    "content",
                     False,
                 ]
                 for path in sorted(
@@ -402,7 +380,7 @@ with gr.Blocks(title="NuMarkApp") as demo:
                 )
             ],
         ],
-        inputs=[template_preset, image, text, mode, enable_thinking],
+        inputs=[template_preset, image, text, enable_thinking],
     )
 
     template_preset.change(
@@ -413,8 +391,8 @@ with gr.Blocks(title="NuMarkApp") as demo:
 
     run.click(
         extract,
-        inputs=[image, text, mode, template, enable_thinking],
-        outputs=[markdown_preview, markdown_text, structured_output],
+        inputs=[image, text, template, enable_thinking],
+        outputs=structured_output,
         api_name="extract",
     )