Delete app.py

app.py

@@ -1,225 +0,0 @@
-import os
-import time
-from threading import Thread
-import re
-from PIL import Image, ImageDraw
-
-import gradio as gr
-import spaces
-import torch
-
-from transformers import (
-    Qwen2_5_VLForConditionalGeneration,
-    AutoProcessor,
-    TextIteratorStreamer,
-)
-
-# Constants for text generation
-MAX_MAX_NEW_TOKENS = 2048
-DEFAULT_MAX_NEW_TOKENS = 1024
-MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "4096"))
-
-device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
-
-# Load Lumian2-VLR-7B-Thinking
-MODEL_ID_Y = "prithivMLmods/Lumian2-VLR-7B-Thinking"
-processor = AutoProcessor.from_pretrained(MODEL_ID_Y, trust_remote_code=True)
-model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
-    MODEL_ID_Y,
-    trust_remote_code=True,
-    torch_dtype=torch.float16
-).to(device).eval()
-
-def parse_model_output(text: str):
-    """
-    Parses the model output to extract the answer and bounding box coordinates.
-    """
-    # Extract coordinates from the <think> block
-    think_match = re.search(r"<think>(.*?)</think>", text, re.DOTALL)
-    coordinates = []
-    if think_match:
-        think_content = think_match.group(1)
-        # Find all occurrences of (x, y) coordinates
-        coords_raw = re.findall(r'\((\d+),\s*(\d+)\)', think_content)
-        coordinates = [(int(x), int(y)) for x, y in coords_raw]
-
-    # Extract the answer from the <answer> block
-    answer_match = re.search(r"<answer>(.*?)</answer>", text, re.DOTALL)
-    answer = answer_match.group(1).strip() if answer_match else text
-
-    return answer, coordinates
-
-def draw_bounding_boxes(image: Image.Image, coordinates: list, box_size: int = 60, use_dotted_style: bool = False):
-    """
-    Draws square bounding boxes on the image at the given coordinates.
-    """
-    if not coordinates:
-        return image
-
-    img_with_boxes = image.copy()
-    draw = ImageDraw.Draw(img_with_boxes, "RGBA")
-    
-    half_box = box_size // 2
-
-    for (x, y) in coordinates:
-        # Define the bounding box corners
-        x1 = x - half_box
-        y1 = y - half_box
-        x2 = x + half_box
-        y2 = y + half_box
-        
-        if use_dotted_style:
-            # "Dotted like seaborn" - a semi-transparent fill with a solid outline
-            fill_color = (0, 100, 255, 60)  # Light blue, semi-transparent
-            outline_color = (0, 0, 255) # Solid blue
-            draw.rectangle([x1, y1, x2, y2], fill=fill_color, outline=outline_color, width=2)
-        else:
-            # Default solid box
-            outline_color = (255, 0, 0) # Red
-            draw.rectangle([x1, y1, x2, y2], outline=outline_color, width=3)
-            
-    return img_with_boxes
-
-@spaces.GPU
-def generate_image(text: str, image: Image.Image,
-                   max_new_tokens: int,
-                   temperature: float,
-                   top_p: float,
-                   top_k: int,
-                   repetition_penalty: float,
-                   draw_boxes: bool,
-                   use_dotted_style: bool):
-    """
-    Generates responses and draws bounding boxes based on model output.
-    Yields raw text, markdown-formatted text, and the processed image.
-    """
-    if image is None:
-        yield "Please upload an image.", "Please upload an image.", None
-        return
-
-    # Yield the original image immediately for the output display
-    yield "", "", image
-
-    messages = [{
-        "role": "user",
-        "content": [
-            {"type": "image", "image": image},
-            {"type": "text", "text": text},
-        ]
-    }]
-    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=False,
-        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
-        time.sleep(0.01)
-        # During generation, yield text updates but keep the original image
-        yield buffer, buffer, image
-
-    # After generation is complete, parse the output and draw boxes
-    final_answer, coordinates = parse_model_output(buffer)
-    
-    output_image = image
-    if draw_boxes and coordinates:
-        output_image = draw_bounding_boxes(image, coordinates, use_dotted_style=use_dotted_style)
-    
-    # Yield the final result with the processed image
-    yield buffer, final_answer, output_image
-
-# Define examples for image inference
-image_examples = [
-    ["Explain the content in detail.", "images/D.jpg"],
-    ["Explain the content (ocr).", "images/O.jpg"],
-    ["What is the core meaning of the poem?", "images/S.jpg"],
-    ["Provide a detailed caption for the image.", "images/A.jpg"],
-    ["Explain the pie-chart in detail.", "images/2.jpg"],
-    ["Jsonify Data.", "images/1.jpg"],
-]
-
-css = """
-.submit-btn {
-    background-color: #2980b9 !important;
-    color: white !important;
-}
-.submit-btn:hover {
-    background-color: #3498db !important;
-}
-.canvas-output {
-    border: 2px solid #4682B4;
-    border-radius: 10px;
-    padding: 20px;
-}
-"""
-
-# Create the Gradio Interface
-with gr.Blocks(css=css, theme="bethecloud/storj_theme") as demo:
-    gr.Markdown("# **Lumian2-VLR-7B-Thinking Image Inference**")
-    with gr.Row():
-        with gr.Column(scale=1):
-            gr.Markdown("## Image Inference")
-            image_query = gr.Textbox(label="Query Input", placeholder="Enter your query here...")
-            image_upload = gr.Image(type="pil", label="Image")
-            image_submit = gr.Button("Submit", elem_classes="submit-btn")
-            
-            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)
-                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.Examples(
-                examples=image_examples,
-                inputs=[image_query, image_upload]
-            )
-
-        with gr.Column(scale=2):
-            gr.Markdown("## Output")
-            with gr.Tabs():
-                with gr.TabItem("Image with Bounding Box"):
-                    image_output = gr.Image(label="Processed Image")
-                with gr.TabItem("Raw Text"):
-                    output = gr.Textbox(label="Raw Model Output", interactive=False, lines=10)
-                with gr.TabItem("Parsed Answer"):
-                    markdown_output = gr.Markdown(label="Parsed Answer")
-
-            gr.Markdown("**Model Info 💻** | [Report Bug](https://huggingface.co/spaces/prithivMLmods/Qwen2.5-VL/discussions)")
-
-            gr.Markdown(
-                """> [Lumian2-VLR-7B-Thinking](https://huggingface.co/prithivMLmods/Lumian2-VLR-7B-Thinking): The Lumian2-VLR-7B-Thinking model is a high-fidelity vision-language reasoning (experimental model) system designed for fine-grained multimodal understanding. Built on Qwen2.5-VL-7B-Instruct, this model enhances image captioning, and document comprehension through explicit grounded reasoning. It produces structured reasoning traces aligned with visual coordinates, enabling explainable multimodal reasoning."""
-            )
-
-            with gr.Row():
-                draw_boxes_checkbox = gr.Checkbox(label="Draw Bounding Boxes", value=True)
-                dotted_style_checkbox = gr.Checkbox(label="Use Dotted Style for Boxes", value=False)
-
-
-    image_submit.click(
-        fn=generate_image,
-        inputs=[image_query, image_upload, max_new_tokens, temperature, top_p, top_k, repetition_penalty, draw_boxes_checkbox, dotted_style_checkbox],
-        outputs=[output, markdown_output, image_output]
-    )
-
-if __name__ == "__main__":
-    demo.queue(max_size=50).launch(share=True)