Upload folder using huggingface_hub

.claude/settings.local.json

@@ -2,7 +2,8 @@
   "permissions": {
     "allow": [
       "Bash(pip install:*)",
-      "Bash(huggingface-cli:*)"
+      "Bash(huggingface-cli:*)",
+      "WebFetch(domain:huggingface.co)"
     ],
     "deny": []
   }

README.md

@@ -1,232 +1,54 @@
-import gradio as gr
-from transformers import pipeline
-import torch
-import os
-
-# Initialize the text generation pipeline
-generator = None
-
-def initialize_model():
-    global generator
-    try:
-        # Use Qwen 2.5 - state-of-the-art 2024/2025 model for document generation
-        generator = pipeline(
-            "text-generation",
-            model="Qwen/Qwen2.5-1.5B-Instruct",  # Latest Qwen model, perfect size for Spaces
-            device=-1,  # CPU only for compatibility
-            torch_dtype=torch.float32,
-            trust_remote_code=True
-        )
-        return "Qwen 2.5-1.5B-Instruct loaded successfully!"
-    except Exception as e:
-        # Fallback to smaller Qwen model
-        try:
-            generator = pipeline(
-                "text-generation",
-                model="Qwen/Qwen2.5-0.5B-Instruct", 
-                device=-1,
-                torch_dtype=torch.float32,
-                trust_remote_code=True
-            )
-            return "Fallback model (Qwen 2.5-0.5B-Instruct) loaded successfully!"
-        except Exception as e2:
-            # Final fallback to Phi-3 mini
-            try:
-                generator = pipeline(
-                    "text-generation", 
-                    model="microsoft/Phi-3-mini-4k-instruct",
-                    device=-1,
-                    torch_dtype=torch.float32,
-                    trust_remote_code=True
-                )
-                return "Final fallback model (Phi-3-mini) loaded successfully!"
-            except Exception as e3:
-                return f"All modern models failed: Qwen: {str(e)}, Qwen-small: {str(e2)}, Phi-3: {str(e3)}"
-
-def generate_onepager(topic, target_audience, key_points, tone, length):
-    if generator is None:
-        return "Error: Model not initialized. Please wait for the model to load."
-    
-    # Create a structured prompt for one-pager generation
-    length_tokens = {"Short": 200, "Medium": 400, "Long": 600}
-    max_tokens = length_tokens.get(length, 400)
-    
-    # Create an optimized prompt for Qwen 2.5 instruction format
-    prompt = f"""<|im_start|>system
-You are a professional document writer specializing in creating concise, well-structured one-page business documents.
-<|im_end|>
-<|im_start|>user
-Create a professional one-page document about "{topic}" targeted at {target_audience}.
-
-Requirements:
-- Tone: {tone.lower()}
-- Key points to include: {key_points}
-- Length: {length}
-- Format: Use clear headers and bullet points
-- Structure: Title, Executive Summary, Key Points, Benefits, Recommendations, Conclusion
-
-Please write the complete one-page document now.
-<|im_end|>
-<|im_start|>assistant
-# {topic}
-
-## Executive Summary
-
-"""
-    
-    try:
-        # Generate the one-pager
-        result = generator(
-            prompt,
-            max_length=len(prompt.split()) + max_tokens,
-            num_return_sequences=1,
-            temperature=0.8,
-            do_sample=True,
-            pad_token_id=generator.tokenizer.eos_token_id,
-            eos_token_id=generator.tokenizer.eos_token_id,
-            repetition_penalty=1.1
-        )
-        
-        # Extract the generated text
-        generated_text = result[0]['generated_text']
-        
-        # Clean up the output
-        onepager = generated_text.replace(prompt, "").strip()
-        
-        # If output is too short, provide a structured fallback
-        if len(onepager) < 50:
-            onepager = create_structured_onepager(topic, target_audience, key_points, tone)
-        
-        return onepager
-        
-    except Exception as e:
-        # Fallback to structured template
-        return create_structured_onepager(topic, target_audience, key_points, tone)
-
-def create_structured_onepager(topic, target_audience, key_points, tone):
-    """Create a structured one-pager when AI generation fails"""
-    
-    tone_styles = {
-        "Professional": "formal and business-oriented",
-        "Casual": "friendly and approachable",
-        "Academic": "scholarly and research-focused", 
-        "Persuasive": "compelling and action-oriented",
-        "Informative": "clear and educational"
-    }
-    
-    style_desc = tone_styles.get(tone, "professional")
-    
-    template = f"""# {topic}
-
-## Executive Summary
-This document provides a comprehensive overview of {topic.lower()} tailored for {target_audience.lower()}. The content is presented in a {style_desc} manner to ensure maximum impact and understanding.
-
-## Key Points
-
-{chr(10).join([f"• {point.strip()}" for point in key_points.split(',') if point.strip()])}
-
-## Background
-{topic} represents an important area that requires careful consideration and strategic thinking. Understanding the core concepts and implications is essential for {target_audience.lower()}.
-
-## Main Content
-The fundamental aspects of {topic.lower()} encompass several critical areas that directly impact stakeholders. These elements work together to create a comprehensive framework for understanding and implementation.
-
-## Benefits & Opportunities
-- Enhanced understanding of core concepts
-- Improved decision-making capabilities  
-- Strategic advantages for implementation
-- Clear actionable insights
-
-## Recommendations
-1. Begin with thorough analysis of current situation
-2. Develop comprehensive implementation strategy
-3. Monitor progress and adjust approach as needed
-4. Measure results and iterate for continuous improvement
-
-## Conclusion
-{topic} offers significant opportunities for {target_audience.lower()} when approached strategically. The key points outlined above provide a solid foundation for moving forward with confidence and clarity.
-
----
-*This one-pager was generated to provide quick, actionable insights on {topic.lower()}.*"""
-    
-    return template
-
-# Create the Gradio interface
-def create_interface():
-    with gr.Blocks(title="One-Pager Generator", theme=gr.themes.Soft()) as demo:
-        gr.Markdown("# 📄 AI One-Pager Generator")
-        gr.Markdown("Generate professional one-page documents on any topic using AI!")
-        
-        with gr.Row():
-            with gr.Column(scale=1):
-                topic_input = gr.Textbox(
-                    label="Topic",
-                    placeholder="e.g., Digital Marketing Strategy, Climate Change Solutions, etc.",
-                    lines=2,
-                    value="Artificial Intelligence in Healthcare"
-                )
-                
-                audience_input = gr.Textbox(
-                    label="Target Audience",
-                    placeholder="e.g., Business executives, Students, General public, etc.",
-                    lines=1,
-                    value="Healthcare professionals"
-                )
-                
-                keypoints_input = gr.Textbox(
-                    label="Key Points to Cover",
-                    placeholder="Enter main points separated by commas",
-                    lines=4,
-                    value="Machine learning applications, Data privacy, Cost-effectiveness, Implementation challenges"
-                )
-                
-                tone_dropdown = gr.Dropdown(
-                    choices=["Professional", "Casual", "Academic", "Persuasive", "Informative"],
-                    label="Tone",
-                    value="Professional"
-                )
-                
-                length_dropdown = gr.Dropdown(
-                    choices=["Short", "Medium", "Long"],
-                    label="Length",
-                    value="Medium"
-                )
-                
-                generate_btn = gr.Button("🚀 Generate One-Pager", variant="primary")
-                
-            with gr.Column(scale=2):
-                output_text = gr.Textbox(
-                    label="Generated One-Pager",
-                    lines=25,
-                    max_lines=35,
-                    show_copy_button=True,
-                    placeholder="Your generated one-pager will appear here..."
-                )
-        
-        with gr.Row():
-            gr.Markdown("""
-            ### 💡 Tips for Best Results:
-            - **Be specific** with your topic for more targeted content
-            - **Include 3-5 key points** separated by commas
-            - **Choose the right tone** for your intended audience  
-            - **Use descriptive audience** details (e.g., "C-level executives" vs "executives")
-            """)
-        
-        # Connect the generate button to the function
-        generate_btn.click(
-            fn=generate_onepager,
-            inputs=[topic_input, audience_input, keypoints_input, tone_dropdown, length_dropdown],
-            outputs=output_text
-        )
-        
-    return demo
-
-# Initialize model and launch
-if __name__ == "__main__":
-    print("🚀 Starting One-Pager Generator...")
-    print("📥 Loading AI model...")
-    initialize_model()
-    print("✅ Model loaded! Launching interface...")
-    
-    demo = create_interface()
-    demo.launch()
+# 📄 AI One-Pager Generator
+
+An intelligent Gradio application that generates professional one-page documents on any topic using Hugging Face transformers.
+
+## Features
+
+- **Topic-based Generation**: Create one-pagers on any subject
+- **Customizable Parameters**: 
+  - Target audience specification
+  - Tone selection (Professional, Casual, Academic, Persuasive, Informative)
+  - Length control (Short, Medium, Long)
+  - Key points input
+- **Professional Formatting**: Structured output with title, executive summary, key points, and conclusion
+- **Easy to Use**: Simple web interface powered by Gradio
+
+## Installation
+
+1. Clone this repository
+2. Install dependencies:
+```bash
+pip install -r requirements.txt
+```
+
+3. Run the application:
+```bash
+python app.py
+```
+
+## Usage
+
+1. Enter your topic in the "Topic" field
+2. Specify your target audience
+3. List key points you want covered
+4. Select the appropriate tone and length
+5. Click "Generate One-Pager"
+6. Copy and use your generated document!
+
+## Deployment to Hugging Face Spaces
+
+This app is ready to be deployed to Hugging Face Spaces:
+
+1. Create a new Space on Hugging Face
+2. Upload these files to your Space
+3. Your app will be automatically deployed!
+
+## Model Information
+
+This application uses Qwen 2.5-Omni-7B, a state-of-the-art 2024/2025 multimodal language model with 7B parameters, specifically optimized for instruction following and high-quality document generation. Fallback models include Qwen 2.5-7B-Instruct and Qwen 2.5-1.5B-Instruct for compatibility across different hardware configurations.
+
+**Note**: This model performs best on GPU-enabled Spaces for optimal speed and quality.
+
+## License
+
+MIT License

app.py

@@ -9,38 +9,42 @@ generator = None
 def initialize_model():
     global generator
     try:
-        # Use a modern instruction-tuned model for better document generation
+        # Use Qwen 2.5-Omni-7B - state-of-the-art multimodal model
+        device = 0 if torch.cuda.is_available() else -1
         generator = pipeline(
             "text-generation",
-            model="microsoft/phi-2",  # Small but very capable instruction model
-            device=-1,  # CPU only for compatibility
-            torch_dtype=torch.float32,
-            trust_remote_code=True
+            model="Qwen/Qwen2.5-Omni-7B",
+            device=device,
+            torch_dtype=torch.bfloat16 if torch.cuda.is_available() else torch.float32,
+            trust_remote_code=True,
+            model_kwargs={"attn_implementation": "flash_attention_2"} if torch.cuda.is_available() else {}
         )
-        return "Model loaded successfully!"
+        return f"Qwen 2.5-Omni-7B loaded successfully on {'GPU' if device == 0 else 'CPU'}!"
     except Exception as e:
-        # Fallback to Flan-T5 for instruction following
+        # Fallback to regular Qwen 7B instruct
         try:
-            from transformers import T5ForConditionalGeneration, T5Tokenizer
+            device = 0 if torch.cuda.is_available() else -1
             generator = pipeline(
-                "text2text-generation",
-                model="google/flan-t5-base",
-                device=-1,
-                torch_dtype=torch.float32
+                "text-generation",
+                model="Qwen/Qwen2.5-7B-Instruct",
+                device=device,
+                torch_dtype=torch.bfloat16 if torch.cuda.is_available() else torch.float32,
+                trust_remote_code=True
             )
-            return "Fallback model (Flan-T5-base) loaded successfully!"
+            return f"Fallback model (Qwen 2.5-7B-Instruct) loaded on {'GPU' if device == 0 else 'CPU'}!"
         except Exception as e2:
-            # Final fallback to GPT2-medium
+            # Final fallback to 1.5B model
             try:
                 generator = pipeline(
                     "text-generation", 
-                    model="gpt2-medium",
+                    model="Qwen/Qwen2.5-1.5B-Instruct",
                     device=-1,
-                    torch_dtype=torch.float32
+                    torch_dtype=torch.float32,
+                    trust_remote_code=True
                 )
-                return "Final fallback model (GPT2-medium) loaded successfully!"
+                return "Final fallback model (Qwen 2.5-1.5B-Instruct) loaded on CPU!"
             except Exception as e3:
-                return f"All models failed: Primary: {str(e)}, Secondary: {str(e2)}, Tertiary: {str(e3)}"
+                return f"All models failed: Omni: {str(e)}, 7B: {str(e2)}, 1.5B: {str(e3)}"
 
 def generate_onepager(topic, target_audience, key_points, tone, length):
     if generator is None:
@@ -50,23 +54,27 @@ def generate_onepager(topic, target_audience, key_points, tone, length):
     length_tokens = {"Short": 200, "Medium": 400, "Long": 600}
     max_tokens = length_tokens.get(length, 400)
     
-    # Create a better prompt for modern instruction-following models
-    prompt = f"""Write a professional one-page document about "{topic}" for {target_audience}. 
-
-Use a {tone.lower()} tone and include these key points: {key_points}
-
-Format the document with:
-- Clear title
-- Executive summary
-- Main sections with headers
-- Key benefits/recommendations
-- Conclusion
-
-Document:
-
+    # Create an optimized prompt for Qwen 2.5 instruction format
+    prompt = f"""<|im_start|>system
+You are a professional document writer specializing in creating concise, well-structured one-page business documents.
+<|im_end|>
+<|im_start|>user
+Create a professional one-page document about "{topic}" targeted at {target_audience}.
+
+Requirements:
+- Tone: {tone.lower()}
+- Key points to include: {key_points}
+- Length: {length}
+- Format: Use clear headers and bullet points
+- Structure: Title, Executive Summary, Key Points, Benefits, Recommendations, Conclusion
+
+Please write the complete one-page document now.
+<|im_end|>
+<|im_start|>assistant
 # {topic}
 
 ## Executive Summary
+
 """
     
     try:

config.yaml

@@ -7,4 +7,6 @@ sdk_version: 4.44.0
 app_file: app.py
 pinned: false
 license: mit
-short_description: Generate professional one-page documents on any topic using AI
+suggested_hardware: t4-small
+suggested_storage: small
+short_description: Generate professional one-page documents using Qwen 2.5-Omni-7B

requirements.txt

@@ -2,4 +2,7 @@ gradio
 transformers
 torch
 huggingface_hub
-tokenizers
+tokenizers
+flash-attn
+accelerate
+bitsandbytes