Upload app.py with huggingface_hub

app.py

@@ -0,0 +1,333 @@
+import spaces
+import gradio as gr
+from transformers import LlavaForConditionalGeneration, TextIteratorStreamer, AutoProcessor
+import torch
+from PIL import Image
+from threading import Thread
+from typing import Generator
+from liger_kernel.transformers import apply_liger_kernel_to_llama
+
+LOGO_SRC = """data:image/svg+xml;base64,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"""
+
+MODEL_PATH = "fancyfeast/llama-joycaption-beta-one-hf-llava"
+
+# Enhanced title with ENHANCED badge
+TITLE = f"""<style>
+  .joy-header   {{display:flex; align-items:center; justify-content:center;
+                 gap:16px; margin:4px 0 12px;}}
+  .joy-header h1{{margin:0; font-size:1.9rem; line-height:1.2;}}
+  .joy-header p {{margin:2px 0 0; font-size:0.9rem; color:#666;}}
+  .joy-header img{{height:56px;}}
+  .enhanced-badge {{background: linear-gradient(45deg, #ff6b35, #f7931e); color: white; padding: 2px 8px; border-radius: 12px; font-size: 0.7rem; font-weight: bold;}}
+</style>
+
+<div class="joy-header">
+  <img src="{LOGO_SRC}" alt="JoyCaption logo">
+  <div>
+    <h1>JoyCaption <span style="font-weight:400">Beta&nbsp;One</span> <span class="enhanced-badge">ENHANCED</span></h1>
+    <p>Advanced Image-captioning with keyword injection &nbsp;|&nbsp; build enhanced</p>
+  </div>
+</div>
+<hr>"""
+
+# Enhanced description
+DESCRIPTION = """
+<div>
+<h2>🚀 Enhanced Features</h2>
+<ul>
+  <li><strong>🎯 Keyword Injection:</strong> Force specific words/phrases to appear in captions</li>
+  <li><strong>📋 Focus Areas:</strong> Emphasize particular aspects of the image</li>
+  <li><strong>🎨 Tone Control:</strong> Choose from professional, casual, technical styles</li>
+  <li><strong>⚙️ Advanced Prompting:</strong> Enhanced templates for better accuracy</li>
+  <li><strong>📊 Quality Metrics:</strong> Real-time feedback on keyword inclusion</li>
+</ul>
+
+<h2>Quick-start</h2>
+<ol>
+  <li><strong>Upload</strong> an image in the left panel</li>
+  <li><strong>Add Required Keywords</strong> that must appear in the caption (optional)</li>
+  <li><strong>Select Focus Areas</strong> to emphasize specific image aspects</li>
+  <li><strong>Choose Caption Type</strong> and adjust settings</li>
+  <li><strong>Generate</strong> your enhanced caption!</li>
+</ol>
+</div>
+"""
+
+# Original caption types (preserved)
+ORIGINAL_CAPTION_TYPES = {
+    "Descriptive": [
+        "Write a detailed description for this image.",
+        "Write a detailed description for this image in {word_count} words or less.",
+        "Write a {length} detailed description for this image.",
+    ],
+    "Art Critic": [
+        "Analyze this image like an art critic would with information about its composition, style, symbolism, the use of color, light, any artistic movement it might belong to, etc.",
+        "Analyze this image like an art critic would in {word_count} words.",
+        "Analyze this image like an art critic would. Keep it {length}.",
+    ],
+}
+
+# Enhanced caption types with keyword integration
+ENHANCED_CAPTION_TYPES = {
+    "Keyword-Focused Descriptive": [
+        "Write a detailed, accurate description of this image. {keyword_instruction} Integrate any specified terms naturally while maintaining descriptive accuracy. Focus on {focus_instruction}",
+        "Write a detailed description of this image in {word_count} words or less. {keyword_instruction} {focus_instruction}",
+        "Write a {length} detailed description of this image. {keyword_instruction} {focus_instruction}",
+    ],
+    "Enhanced Descriptive": [
+        "Provide a comprehensive, detailed description of this image. {keyword_instruction} Pay special attention to {focus_instruction} Use {tone} language throughout.",
+        "Describe this image comprehensively in {word_count} words. {keyword_instruction} Focus on {focus_instruction}",
+        "Write a {length} comprehensive description. {keyword_instruction} Emphasize {focus_instruction}",
+    ],
+}
+
+# Combine caption types
+ALL_CAPTION_TYPES = {**ENHANCED_CAPTION_TYPES, **ORIGINAL_CAPTION_TYPES}
+
+# Load model
+print("Loading JoyCaption model...")
+processor = AutoProcessor.from_pretrained(MODEL_PATH)
+model = LlavaForConditionalGeneration.from_pretrained(MODEL_PATH, torch_dtype="bfloat16", device_map=0)
+model.eval()
+apply_liger_kernel_to_llama(model=model.language_model)
+print("Model loaded successfully!")
+
+def build_enhanced_prompt(
+    caption_type: str, 
+    caption_length: str, 
+    required_keywords: str, 
+    focus_areas: list, 
+    tone_style: str
+) -> str:
+    """Build an enhanced prompt with keyword injection and focus areas"""
+    
+    # Choose template index
+    if caption_length == "any":
+        template_idx = 0
+    elif isinstance(caption_length, str) and caption_length.isdigit():
+        template_idx = 1
+    else:
+        template_idx = 2
+    
+    # Get base prompt template
+    base_prompt = ALL_CAPTION_TYPES[caption_type][template_idx]
+    
+    # Build keyword instruction
+    keyword_instruction = ""
+    if required_keywords.strip():
+        keywords = [k.strip() for k in required_keywords.split(',') if k.strip()]
+        if keywords:
+            keyword_instruction = f"REQUIRED KEYWORDS - You MUST naturally include these words/phrases: {', '.join(keywords)}."
+    
+    # Build focus instruction
+    focus_instruction = ""
+    if focus_areas:
+        focus_instruction = f"Pay special attention to: {', '.join(focus_areas)}"
+    else:
+        focus_instruction = "all visible elements"
+    
+    # Add tone
+    tone = tone_style.lower() if tone_style != "Auto-detect" else "appropriate"
+    
+    # Format the prompt
+    try:
+        formatted_prompt = base_prompt.format(
+            keyword_instruction=keyword_instruction,
+            focus_instruction=focus_instruction,
+            tone=tone,
+            length=caption_length,
+            word_count=caption_length
+        )
+    except KeyError:
+        # Fallback for original templates that don't have these placeholders
+        formatted_prompt = base_prompt.format(
+            length=caption_length,
+            word_count=caption_length
+        )
+    
+    return formatted_prompt
+
+def check_keyword_inclusion(caption: str, required_keywords: str) -> str:
+    """Check if required keywords are included in the caption"""
+    if not required_keywords.strip():
+        return "✅ No required keywords specified"
+    
+    keywords = [k.strip().lower() for k in required_keywords.split(',') if k.strip()]
+    caption_lower = caption.lower()
+    
+    included = []
+    missing = []
+    
+    for keyword in keywords:
+        if keyword in caption_lower:
+            included.append(keyword)
+        else:
+            missing.append(keyword)
+    
+    result = f"✅ Included ({len(included)}/{len(keywords)}): {', '.join(included) if included else 'None'}"
+    if missing:
+        result += f"\n❌ Missing: {', '.join(missing)}"
+    
+    return result
+
+@spaces.GPU()
+@torch.no_grad()
+def chat_joycaption(input_image: Image.Image, prompt: str, temperature: float, top_p: float, max_new_tokens: int) -> Generator[str, None, None]:
+    torch.cuda.empty_cache()
+
+    if input_image is None:
+        yield "No image provided. Please upload an image."
+        return
+
+    system_content = "You are a helpful assistant specialized in providing detailed, accurate image descriptions. Focus on being precise, comprehensive, and following instructions exactly."
+
+    convo = [
+        {
+            "role": "system",
+            "content": system_content,
+        },
+        {
+            "role": "user",
+            "content": prompt.strip(),
+        },
+    ]
+
+    convo_string = processor.apply_chat_template(convo, tokenize=False, add_generation_prompt=True)
+    inputs = processor(text=[convo_string], images=[input_image], return_tensors="pt").to('cuda')
+    inputs['pixel_values'] = inputs['pixel_values'].to(torch.bfloat16)
+
+    streamer = TextIteratorStreamer(processor.tokenizer, timeout=10.0, skip_prompt=True, skip_special_tokens=True)
+
+    generate_kwargs = dict(
+        **inputs,
+        max_new_tokens=max_new_tokens,
+        do_sample=True if temperature > 0 else False,
+        suppress_tokens=None,
+        use_cache=True,
+        temperature=temperature if temperature > 0 else None,
+        top_k=None,
+        top_p=top_p if temperature > 0 else None,
+        streamer=streamer,
+    )
+
+    t = Thread(target=model.generate, kwargs=generate_kwargs)
+    t.start()
+
+    outputs = []
+    for text in streamer:
+        outputs.append(text)
+        yield "".join(outputs)
+
+def generate_with_quality_check(input_image, prompt, temperature, top_p, max_new_tokens, required_keywords):
+    """Generate caption and provide quality feedback"""
+    
+    caption_generator = chat_joycaption(input_image, prompt, temperature, top_p, max_new_tokens)
+    
+    final_caption = ""
+    for partial_caption in caption_generator:
+        final_caption = partial_caption
+        yield partial_caption, ""
+    
+    # Check keyword inclusion
+    quality_feedback = check_keyword_inclusion(final_caption, required_keywords)
+    yield final_caption, quality_feedback
+
+# Create Gradio interface
+with gr.Blocks(title="JoyCaption Enhanced") as demo:
+    gr.HTML(TITLE)
+
+    with gr.Row():
+        with gr.Column():
+            input_image = gr.Image(type="pil", label="Input Image", height=512, width=512)
+            
+            # Enhanced keyword injection section
+            with gr.Group():
+                gr.HTML("<h3>🎯 Keyword Control</h3>")
+                required_keywords = gr.Textbox(
+                    label="Required Keywords/Phrases",
+                    placeholder="Enter comma-separated words that MUST appear in the caption (e.g., 'sunset, peaceful, golden hour')",
+                    info="These keywords will be naturally integrated into the description"
+                )
+                
+                focus_areas = gr.CheckboxGroup(
+                    choices=[
+                        "Foreground objects", "Background elements", "Colors and lighting", 
+                        "Textures and materials", "Spatial relationships", "Actions and poses",
+                        "Facial expressions", "Clothing and accessories", "Artistic style",
+                        "Composition and framing", "Mood and atmosphere"
+                    ],
+                    label="Focus Areas (emphasize these aspects)",
+                    value=["Foreground objects", "Colors and lighting"]
+                )
+                
+                tone_style = gr.Dropdown(
+                    choices=["Auto-detect", "Professional", "Casual", "Technical", "Creative", "Academic", "Marketing"],
+                    value="Professional",
+                    label="Description Tone"
+                )
+
+            caption_type = gr.Dropdown(
+                choices=list(ALL_CAPTION_TYPES.keys()),
+                value="Enhanced Descriptive",
+                label="Caption Type",
+                info="Enhanced types support keyword injection and focus areas"
+            )
+
+            caption_length = gr.Dropdown(
+                choices=["any", "very short", "short", "medium-length", "long", "very long"] +
+                        [str(i) for i in range(20, 261, 10)],
+                label="Caption Length",
+                value="long",
+            )
+
+            with gr.Accordion("Generation Settings", open=False):
+                temperature_slider = gr.Slider(
+                    minimum=0.0, maximum=2.0, value=0.6, step=0.05,
+                    label="Temperature",
+                    info="Higher values make output more creative"
+                )
+                top_p_slider = gr.Slider(
+                    minimum=0.0, maximum=1.0, value=0.9, step=0.01,
+                    label="Top-p"
+                )
+                max_tokens_slider = gr.Slider(
+                    minimum=1, maximum=2048, value=512, step=1,
+                    label="Max New Tokens"
+                )
+
+        with gr.Column():
+            prompt_box = gr.Textbox(lines=6, label="Generated Prompt", interactive=True)
+            
+            run_button = gr.Button("🚀 Generate Enhanced Caption", variant="primary", size="lg")
+
+            output_caption = gr.Textbox(label="Generated Caption", lines=8)
+            
+            quality_feedback = gr.Textbox(
+                label="📊 Keyword Quality Check",
+                lines=3
+            )
+
+    # Event handlers
+    for ctrl in [caption_type, caption_length, required_keywords, focus_areas, tone_style]:
+        ctrl.change(
+            build_enhanced_prompt,
+            inputs=[caption_type, caption_length, required_keywords, focus_areas, tone_style],
+            outputs=prompt_box,
+        )
+
+    run_button.click(
+        generate_with_quality_check,
+        inputs=[input_image, prompt_box, temperature_slider, top_p_slider, max_tokens_slider, required_keywords],
+        outputs=[output_caption, quality_feedback],
+    )
+
+    # Initialize prompt
+    demo.load(
+        lambda: build_enhanced_prompt("Enhanced Descriptive", "long", "", ["Foreground objects", "Colors and lighting"], "Professional"),
+        outputs=prompt_box
+    )
+
+    gr.Markdown(DESCRIPTION)
+
+if __name__ == "__main__":
+    demo.launch()