Update app.py

app.py

@@ -7,6 +7,7 @@ import asyncio
 from threading import Thread
 from pathlib import Path
 from io import BytesIO
+from typing import Optional, Tuple, Dict, Any
 
 import gradio as gr
 import spaces
@@ -91,6 +92,55 @@ def convert_pdf_to_images(file_path: str, dpi: int = 200):
     pdf_document.close()
     return images
 
+def get_initial_pdf_state() -> Dict[str, Any]:
+    """Returns the default initial state for the PDF viewer."""
+    return {"pages": [], "total_pages": 0, "current_page_index": 0}
+
+def load_and_preview_pdf(file_path: Optional[str]) -> Tuple[Optional[Image.Image], Dict[str, Any], str]:
+    """
+    Loads a PDF, converts pages to images, and prepares the state for preview.
+    """
+    state = get_initial_pdf_state()
+    if not file_path:
+        return None, state, '<div style="text-align:center;">No file loaded</div>'
+
+    try:
+        pages = convert_pdf_to_images(file_path)
+        if not pages:
+            return None, state, '<div style="text-align:center;">Could not load file</div>'
+        
+        state["pages"] = pages
+        state["total_pages"] = len(pages)
+        page_info_html = f'<div style="text-align:center;">Page 1 / {state["total_pages"]}</div>'
+        return pages[0], state, page_info_html
+    except Exception as e:
+        return None, state, f'<div style="text-align:center;">Failed to load preview: {e}</div>'
+
+def navigate_pdf_page(direction: str, state: Dict[str, Any]):
+    """
+    Navigates to the previous or next page in the PDF preview.
+    """
+    if not state or not state["pages"]:
+        return None, state, '<div style="text-align:center;">No file loaded</div>'
+
+    current_index = state["current_page_index"]
+    total_pages = state["total_pages"]
+    
+    if direction == "prev":
+        new_index = max(0, current_index - 1)
+    elif direction == "next":
+        new_index = min(total_pages - 1, current_index + 1)
+    else:
+        new_index = current_index
+        
+    state["current_page_index"] = new_index
+    
+    image_preview = state["pages"][new_index]
+    page_info_html = f'<div style="text-align:center;">Page {new_index + 1} / {total_pages}</div>'
+    
+    return image_preview, state, page_info_html
+
+
 @spaces.GPU
 def generate_image(text: str, image: Image.Image,
                    max_new_tokens: int = 1024,
@@ -162,32 +212,24 @@ def generate_video(text: str, video_path: str,
         yield buffer, buffer
 
 @spaces.GPU
-def generate_pdf(text: str, pdf_path: str,
+def generate_pdf(text: str, state: Dict[str, Any],
                  max_new_tokens: int = 2048,
                  temperature: float = 0.6,
                  top_p: float = 0.9,
                  top_k: int = 50,
                  repetition_penalty: float = 1.2):
     """
-    Processes a PDF file page by page and generates a combined textual output.
+    Processes a PDF file page by page using the pre-loaded images from the state.
     """
-    if not pdf_path:
-        yield "Please upload a PDF file.", "Please upload a PDF file."
-        return
-
-    try:
-        page_images = convert_pdf_to_images(pdf_path)
-        if not page_images:
-            yield "Could not extract pages from the PDF.", "Could not extract pages from the PDF."
-            return
-    except Exception as e:
-        yield f"Error processing PDF: {e}", f"Error processing PDF: {e}"
+    if not state or not state["pages"]:
+        yield "Please upload a PDF file first.", "Please upload a PDF file first."
         return
 
+    page_images = state["pages"]
     full_response = ""
     for i, image in enumerate(page_images):
         page_header = f"--- Page {i+1}/{len(page_images)} ---\n"
-        yield page_header, page_header
+        yield full_response + page_header, full_response + page_header
         
         messages = [{"role": "user", "content": [{"type": "image"}, {"type": "text", "text": text}]}]
         prompt_full = processor_q3vl.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
@@ -205,24 +247,15 @@ def generate_pdf(text: str, pdf_path: str,
             yield full_response + page_header + page_buffer, full_response + page_header + page_buffer
             time.sleep(0.01)
         
-        full_response += page_header + page_buffer + "\n"
+        full_response += page_header + page_buffer + "\n\n"
 
 # --- Gradio Interface ---
 image_examples = [
     ["Describe the safety measures in the image. Conclude (Safe / Unsafe)..", "images/5.jpg"],
     ["Convert this page to doc [markdown] precisely.", "images/3.png"],
-    ["Explain the creativity in the image.", "images/6.jpg"],
-]
-
-video_examples = [
-    ["Explain the video in detail.", "videos/2.mp4"],
-    ["Explain the ad in detail.", "videos/1.mp4"]
 ]
-
-#pdf_examples = [
-#    ["Summarize the key findings from this document.", "examples/sample-doc.pdf"],
-#    ["Extract the main points from each section.", "examples/research-paper.pdf"],
-#]
+video_examples = [["Explain the video in detail.", "videos/2.mp4"]]
+pdf_examples = [["examples/sample-doc.pdf"]]
 
 css = """
 .submit-btn { background-color: #2980b9 !important; color: white !important; }
@@ -231,9 +264,10 @@ css = """
 """
 
 with gr.Blocks(css=css) as demo:
+    pdf_state = gr.State(value=get_initial_pdf_state())
     gr.Markdown("# **Qwen3-VL-Processor**")
     with gr.Row():
-        with gr.Column():
+        with gr.Column(scale=2):
             with gr.Tabs():
                 with gr.TabItem("Image Inference"):
                     image_query = gr.Textbox(label="Query Input", placeholder="Enter your query here...")
@@ -248,10 +282,18 @@ with gr.Blocks(css=css) as demo:
                     gr.Examples(examples=video_examples, inputs=[video_query, video_upload])
 
                 with gr.TabItem("PDF Inference"):
-                    pdf_query = gr.Textbox(label="Query Input", placeholder="e.g., 'Summarize this document'")
-                    pdf_upload = gr.File(label="Upload PDF", file_types=[".pdf"])
-                    pdf_submit = gr.Button("Submit", elem_classes="submit-btn")
-                    #gr.Examples(examples=pdf_examples, inputs=[pdf_query, pdf_upload])
+                    with gr.Row():
+                        with gr.Column(scale=1):
+                            pdf_query = gr.Textbox(label="Query Input", placeholder="e.g., 'Summarize this document'")
+                            pdf_upload = gr.File(label="Upload PDF", file_types=[".pdf"])
+                            gr.Examples(examples=pdf_examples, inputs=[pdf_upload])
+                            pdf_submit = gr.Button("Submit", elem_classes="submit-btn")
+                        with gr.Column(scale=1):
+                            pdf_preview_img = gr.Image(label="PDF Preview", height=280)
+                            with gr.Row():
+                                prev_page_btn = gr.Button("◀ Previous")
+                                page_info = gr.HTML('<div style="text-align:center;">No file loaded</div>')
+                                next_page_btn = gr.Button("Next ▶")
 
             with gr.Accordion("Advanced options", open=False):
                 max_new_tokens = gr.Slider(label="Max new tokens", minimum=1, maximum=MAX_MAX_NEW_TOKENS, step=1, value=DEFAULT_MAX_NEW_TOKENS)
@@ -260,10 +302,10 @@ with gr.Blocks(css=css) as demo:
                 top_k = gr.Slider(label="Top-k", minimum=1, maximum=1000, step=1, value=50)
                 repetition_penalty = gr.Slider(label="Repetition penalty", minimum=1.0, maximum=2.0, step=0.05, value=1.2)
 
-        with gr.Column():
+        with gr.Column(scale=3):
             with gr.Column(elem_classes="canvas-output"):
                 gr.Markdown("## Output")
-                output = gr.Textbox(label="Raw Output Stream", interactive=False, lines=10, show_copy_button=True)
+                output = gr.Textbox(label="Raw Output Stream", interactive=False, lines=15, show_copy_button=True)
                 with gr.Accordion("(Result.md)", open=False):
                     markdown_output = gr.Markdown(label="(Result.Md)")
 
@@ -280,9 +322,24 @@ with gr.Blocks(css=css) as demo:
     )
     pdf_submit.click(
         fn=generate_pdf,
-        inputs=[pdf_query, pdf_upload, max_new_tokens, temperature, top_p, top_k, repetition_penalty],
+        inputs=[pdf_query, pdf_state, max_new_tokens, temperature, top_p, top_k, repetition_penalty],
         outputs=[output, markdown_output]
     )
+    pdf_upload.change(
+        fn=load_and_preview_pdf,
+        inputs=[pdf_upload],
+        outputs=[pdf_preview_img, pdf_state, page_info]
+    )
+    prev_page_btn.click(
+        fn=lambda s: navigate_pdf_page("prev", s),
+        inputs=[pdf_state],
+        outputs=[pdf_preview_img, pdf_state, page_info]
+    )
+    next_page_btn.click(
+        fn=lambda s: navigate_pdf_page("next", s),
+        inputs=[pdf_state],
+        outputs=[pdf_preview_img, pdf_state, page_info]
+    )
 
 if __name__ == "__main__":
     demo.queue(max_size=50).launch(mcp_server=True, ssr_mode=False, show_error=True)