Upload app(2).py

app(2).py

@@ -0,0 +1,352 @@
+import spaces
+import json
+import math
+import os
+import traceback
+from io import BytesIO
+from typing import Any, Dict, List, Optional, Tuple
+import re
+import time
+from threading import Thread
+from io import BytesIO
+import uuid
+import tempfile
+
+import gradio as gr
+import requests
+import torch
+from PIL import Image
+import fitz
+
+from transformers import (
+    Qwen2_5_VLForConditionalGeneration,
+    Qwen2VLForConditionalGeneration,
+    AutoModelForCausalLM,
+    AutoModelForVision2Seq,
+    AutoModelForImageTextToText,
+    AutoModel,
+    AutoProcessor,
+    TextIteratorStreamer,
+    AutoTokenizer,
+)
+
+from transformers.image_utils import load_image
+
+from reportlab.lib.pagesizes import A4
+from reportlab.lib.styles import getSampleStyleSheet
+from reportlab.platypus import SimpleDocTemplate, Image as RLImage, Paragraph, Spacer
+from reportlab.lib.units import inch
+
+# --- Constants and Model Setup ---
+MAX_INPUT_TOKEN_LENGTH = 4096
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+
+print("CUDA_VISIBLE_DEVICES=", os.environ.get("CUDA_VISIBLE_DEVICES"))
+print("torch.__version__ =", torch.__version__)
+print("torch.version.cuda =", torch.version.cuda)
+print("cuda available:", torch.cuda.is_available())
+print("cuda device count:", torch.cuda.device_count())
+if torch.cuda.is_available():
+    print("current device:", torch.cuda.current_device())
+    print("device name:", torch.cuda.get_device_name(torch.cuda.current_device()))
+
+print("Using device:", device)
+
+# --- Model Loading ---
+MODEL_ID_M = "LiquidAI/LFM2-VL-450M"
+processor_m = AutoProcessor.from_pretrained(MODEL_ID_M, trust_remote_code=True)
+model_m = AutoModelForImageTextToText.from_pretrained(
+    MODEL_ID_M, trust_remote_code=True, torch_dtype=torch.float16
+).to(device).eval()
+
+MODEL_ID_T = "LiquidAI/LFM2-VL-1.6B"
+processor_t = AutoProcessor.from_pretrained(MODEL_ID_T, trust_remote_code=True)
+model_t = AutoModelForImageTextToText.from_pretrained(
+    MODEL_ID_T, trust_remote_code=True, torch_dtype=torch.float16
+).to(device).eval()
+
+MODEL_ID_C = "HuggingFaceTB/SmolVLM-Instruct-250M"
+processor_c = AutoProcessor.from_pretrained(MODEL_ID_C, trust_remote_code=True)
+model_c = AutoModelForVision2Seq.from_pretrained(
+    MODEL_ID_C, trust_remote_code=True, torch_dtype=torch.float16, _attn_implementation="flash_attention_2"
+).to(device).eval()
+
+MODEL_ID_G = "echo840/MonkeyOCR-pro-1.2B"
+SUBFOLDER = "Recognition"
+processor_g = AutoProcessor.from_pretrained(
+    MODEL_ID_G, trust_remote_code=True, subfolder=SUBFOLDER
+)
+model_g = Qwen2_5_VLForConditionalGeneration.from_pretrained(
+    MODEL_ID_G, trust_remote_code=True, subfolder=SUBFOLDER, torch_dtype=torch.float16
+).to(device).eval()
+
+MODEL_ID_I = "UCSC-VLAA/VLAA-Thinker-Qwen2VL-2B"
+processor_i = AutoProcessor.from_pretrained(MODEL_ID_I, trust_remote_code=True)
+model_i = Qwen2VLForConditionalGeneration.from_pretrained(
+    MODEL_ID_I, trust_remote_code=True, torch_dtype=torch.float16
+).to(device).eval()
+
+MODEL_ID_A = "nanonets/Nanonets-OCR-s"
+processor_a = AutoProcessor.from_pretrained(MODEL_ID_A, trust_remote_code=True)
+model_a = Qwen2_5_VLForConditionalGeneration.from_pretrained(
+    MODEL_ID_A, trust_remote_code=True, torch_dtype=torch.float16
+).to(device).eval()
+
+MODEL_ID_X = "prithivMLmods/Megalodon-OCR-Sync-0713"
+processor_x = AutoProcessor.from_pretrained(MODEL_ID_X, trust_remote_code=True)
+model_x = Qwen2_5_VLForConditionalGeneration.from_pretrained(
+    MODEL_ID_X, trust_remote_code=True, torch_dtype=torch.float16
+).to(device).eval()
+
+MODEL_ID_Z = "Vchitect/ShotVL-3B"
+processor_z = AutoProcessor.from_pretrained(MODEL_ID_Z, trust_remote_code=True)
+model_z = Qwen2_5_VLForConditionalGeneration.from_pretrained(
+    MODEL_ID_Z, trust_remote_code=True, torch_dtype=torch.float16
+).to(device).eval()
+
+# --- Moondream2 Model Loading ---
+MODEL_ID_MD = "vikhyatk/moondream2"
+REVISION_MD = "2025-06-21"
+moondream = AutoModelForCausalLM.from_pretrained(
+  MODEL_ID_MD,
+  revision=REVISION_MD,
+  trust_remote_code=True,
+  torch_dtype=torch.float16,
+  device_map={"": "cuda"},
+)
+tokenizer_md = AutoTokenizer.from_pretrained(MODEL_ID_MD, revision=REVISION_MD)
+
+
+# --- PDF Generation and Preview Utility Function ---
+def generate_and_preview_pdf(image: Image.Image, text_content: str, font_size: int, line_spacing: float, alignment: str, image_size: str):
+    """
+    Generates a PDF, saves it, and then creates image previews of its pages.
+    Returns the path to the PDF and a list of paths to the preview images.
+    """
+    if image is None or not text_content or not text_content.strip():
+        raise gr.Error("Cannot generate PDF. Image or text content is missing.")
+
+    # --- 1. Generate the PDF ---
+    temp_dir = tempfile.gettempdir()
+    pdf_filename = os.path.join(temp_dir, f"output_{uuid.uuid4()}.pdf")
+    doc = SimpleDocTemplate(
+        pdf_filename,
+        pagesize=A4,
+        rightMargin=inch, leftMargin=inch,
+        topMargin=inch, bottomMargin=inch
+    )
+    styles = getSampleStyleSheet()
+    style_normal = styles["Normal"]
+    style_normal.fontSize = int(font_size)
+    style_normal.leading = int(font_size) * line_spacing
+    style_normal.alignment = {"Left": 0, "Center": 1, "Right": 2, "Justified": 4}[alignment]
+
+    story = []
+
+    img_buffer = BytesIO()
+    image.save(img_buffer, format='PNG')
+    img_buffer.seek(0)
+    
+    page_width, _ = A4
+    available_width = page_width - 2 * inch
+    image_widths = {
+        "Small": available_width * 0.3,
+        "Medium": available_width * 0.6,
+        "Large": available_width * 0.9,
+    }
+    img_width = image_widths[image_size]
+    img = RLImage(img_buffer, width=img_width, height=image.height * (img_width / image.width))
+    story.append(img)
+    story.append(Spacer(1, 12))
+
+    cleaned_text = re.sub(r'#+\s*', '', text_content).replace("*", "")
+    text_paragraphs = cleaned_text.split('\n')
+    
+    for para in text_paragraphs:
+        if para.strip():
+            story.append(Paragraph(para, style_normal))
+
+    doc.build(story)
+
+    # --- 2. Render PDF pages as images for preview ---
+    preview_images = []
+    try:
+        pdf_doc = fitz.open(pdf_filename)
+        for page_num in range(len(pdf_doc)):
+            page = pdf_doc.load_page(page_num)
+            pix = page.get_pixmap(dpi=150)
+            preview_img_path = os.path.join(temp_dir, f"preview_{uuid.uuid4()}_p{page_num}.png")
+            pix.save(preview_img_path)
+            preview_images.append(preview_img_path)
+        pdf_doc.close()
+    except Exception as e:
+        print(f"Error generating PDF preview: {e}")
+        
+    return pdf_filename, preview_images
+
+
+# --- Core Application Logic ---
+@spaces.GPU
+def process_document_stream(
+    model_name: str, 
+    image: Image.Image, 
+    prompt_input: str, 
+    max_new_tokens: int,
+    temperature: float,
+    top_p: float,
+    top_k: int,
+    repetition_penalty: float
+):
+    """
+    Main generator function that handles model inference tasks with advanced generation parameters.
+    """
+    if image is None:
+        yield "Please upload an image.", ""
+        return
+    if not prompt_input or not prompt_input.strip():
+        yield "Please enter a prompt.", ""
+        return
+
+    if model_name == "Moondream2(vision)":
+        image_embeds = moondream.encode_image(image)
+        answer = moondream.answer_question(
+            image_embeds=image_embeds,
+            question=prompt_input,
+            tokenizer=tokenizer_md
+        )
+        yield answer, answer
+        return
+
+    if model_name == "LFM2-VL-450M(fast)": processor, model = processor_m, model_m
+    elif model_name == "LFM2-VL-1.6B(fast)": processor, model = processor_t, model_t
+    elif model_name == "ShotVL-3B(cinematic)": processor, model = processor_z, model_z
+    elif model_name == "SmolVLM-Instruct-250M(smol)": processor, model = processor_c, model_c
+    elif model_name == "MonkeyOCR-pro-1.2B(ocr)": processor, model = processor_g, model_g
+    elif model_name == "VLAA-Thinker-Qwen2VL-2B(reason)": processor, model = processor_i, model_i
+    elif model_name == "Nanonets-OCR-s(ocr)": processor, model = processor_a, model_a
+    elif model_name == "Megalodon-OCR-Sync-0713(ocr)": processor, model = processor_x, model_x
+    else:
+        yield "Invalid model selected.", ""
+        return
+
+    messages = [{"role": "user", "content": [{"type": "image", "image": image}, {"type": "text", "text": prompt_input}]}]
+    prompt_full = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
+    inputs = processor(text=[prompt_full], images=[image], return_tensors="pt", padding=True, truncation=True, max_length=MAX_INPUT_TOKEN_LENGTH).to(device)
+    streamer = TextIteratorStreamer(processor, skip_prompt=True, skip_special_tokens=True)
+    
+    generation_kwargs = {
+        **inputs, 
+        "streamer": streamer, 
+        "max_new_tokens": max_new_tokens,
+        "temperature": temperature,
+        "top_p": top_p,
+        "top_k": top_k,
+        "repetition_penalty": repetition_penalty,
+        "do_sample": True
+    }
+
+    thread = Thread(target=model.generate, kwargs=generation_kwargs)
+    thread.start()
+
+    buffer = ""
+    for new_text in streamer:
+        buffer += new_text
+        buffer = buffer.replace("<|im_end|>", "")
+        time.sleep(0.01)
+        yield buffer , buffer
+
+    yield buffer, buffer
+
+
+# --- Gradio UI Definition ---
+def create_gradio_interface():
+    """Builds and returns the Gradio web interface."""
+    css = """
+    .main-container { max-width: 1400px; margin: 0 auto; }
+    .process-button { border: none !important; color: white !important; font-weight: bold !important; background-color: blue !important;}
+    .process-button:hover { background-color: darkblue !important; transform: translateY(-2px) !important; box-shadow: 0 4px 8px rgba(0,0,0,0.2) !important; }
+    #gallery { min-height: 400px; }
+    """
+    with gr.Blocks(theme="bethecloud/storj_theme", css=css) as demo:
+        gr.HTML("""
+        <div class="title" style="text-align: center">
+            <h1>Tiny VLMs Lab🧪</h1>
+            <p style="font-size: 1.1em; color: #6b7280; margin-bottom: 0.6em;">
+                Advanced Vision-Language Model for Image Content and Layout Extraction
+            </p>
+        </div>
+        """)
+
+        with gr.Row():
+            # Left Column (Inputs)
+            with gr.Column(scale=1):
+                model_choice = gr.Dropdown(
+                    choices=["LFM2-VL-450M(fast)", "LFM2-VL-1.6B(fast)", "SmolVLM-Instruct-250M(smol)", "Moondream2(vision)", "ShotVL-3B(cinematic)", "Megalodon-OCR-Sync-0713(ocr)", 
+                             "VLAA-Thinker-Qwen2VL-2B(reason)", "MonkeyOCR-pro-1.2B(ocr)", "Nanonets-OCR-s(ocr)"],
+                    label="Select Model", value= "LFM2-VL-450M(fast)"
+                )
+                prompt_input = gr.Textbox(label="Query Input", placeholder="✦︎ Enter your query", value="Describe the image!")
+                image_input = gr.Image(label="Upload Image", type="pil", sources=['upload'])
+                
+                with gr.Accordion("Advanced Settings", open=False):
+                    max_new_tokens = gr.Slider(minimum=512, maximum=8192, value=4096, step=256, label="Max New Tokens")
+                    temperature = gr.Slider(label="Temperature", minimum=0.1, maximum=4.0, step=0.1, value=0.6)
+                    top_p = gr.Slider(label="Top-p (nucleus sampling)", minimum=0.05, maximum=1.0, step=0.05, value=0.9)
+                    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)
+                    
+                    gr.Markdown("### PDF Export Settings")
+                    font_size = gr.Dropdown(choices=["8", "10", "12", "14", "16", "18"], value="12", label="Font Size")
+                    line_spacing = gr.Dropdown(choices=[1.0, 1.15, 1.5, 2.0], value=1.15, label="Line Spacing")
+                    alignment = gr.Dropdown(choices=["Left", "Center", "Right", "Justified"], value="Justified", label="Text Alignment")
+                    image_size = gr.Dropdown(choices=["Small", "Medium", "Large"], value="Medium", label="Image Size in PDF")
+
+                process_btn = gr.Button("🚀 Process Image", variant="primary", elem_classes=["process-button"], size="lg")
+                clear_btn = gr.Button("🗑️ Clear All", variant="secondary")
+
+            # Right Column (Outputs)
+            with gr.Column(scale=2):
+                with gr.Tabs() as tabs:
+                    with gr.Tab("📝 Extracted Content"):
+                        raw_output_stream = gr.Textbox(label="Raw Model Output Stream", interactive=False, lines=15, show_copy_button=True)
+                        with gr.Row():
+                            examples = gr.Examples(
+                                examples=["examples/1.png", "examples/2.png", "examples/3.png", "examples/4.png", "examples/5.png"],
+                                inputs=image_input, label="Examples"
+                            )
+                        gr.Markdown("[Report-Bug💻](https://huggingface.co/spaces/prithivMLmods/OCR-Comparator/discussions)")
+                    
+                    with gr.Tab("📰 README.md"):
+                        with gr.Accordion("(Result.md)", open=True): 
+                            markdown_output = gr.Markdown()
+
+                    with gr.Tab("📋 PDF Preview"):
+                        generate_pdf_btn = gr.Button("📄 Generate PDF & Render", variant="primary")
+                        pdf_output_file = gr.File(label="Download Generated PDF", interactive=False)
+                        pdf_preview_gallery = gr.Gallery(label="PDF Page Preview", show_label=True, elem_id="gallery", columns=2, object_fit="contain", height="auto")
+
+        # Event Handlers
+        def clear_all_outputs():
+            return None, "", "Raw output will appear here.", "", None, None
+
+        process_btn.click(
+            fn=process_document_stream,
+            inputs=[model_choice, image_input, prompt_input, max_new_tokens, temperature, top_p, top_k, repetition_penalty],
+            outputs=[raw_output_stream, markdown_output]
+        )
+        
+        generate_pdf_btn.click(
+            fn=generate_and_preview_pdf,
+            inputs=[image_input, raw_output_stream, font_size, line_spacing, alignment, image_size],
+            outputs=[pdf_output_file, pdf_preview_gallery]
+        )
+
+        clear_btn.click(
+            clear_all_outputs,
+            outputs=[image_input, prompt_input, raw_output_stream, markdown_output, pdf_output_file, pdf_preview_gallery]
+        )
+    return demo
+
+if __name__ == "__main__":
+    demo = create_gradio_interface()
+    demo.queue(max_size=50).launch(share=True, ssr_mode=False, show_error=True)