Initial commit

README.md

@@ -1,13 +1,37 @@
 ---
-title: Sui Demo
-emoji: 🦀
-colorFrom: red
-colorTo: green
+title: sui-1 Summarizer
+emoji: 📄
+colorFrom: indigo
+colorTo: purple
 sdk: gradio
-sdk_version: 6.2.0
+sdk_version: 5.34.2
 app_file: app.py
 pinned: false
-license: apache-2.0
+license: mit
+short_description: Grounded Summarization with Source Citations
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# sui-1 Summarizer
+
+Generate grounded summaries with verifiable source citations using the [ellamind/sui-1-24b](https://huggingface.co/ellamind/sui-1-24b) model.
+
+## Features
+
+- **PDF Upload**: Extract text using Mistral OCR
+- **Text Input**: Paste text directly for summarization
+- **Multi-language**: Support for English, German, Spanish, French, and Italian
+- **Source Citations**: Every claim is backed by numbered citations from the source document
+- **Streaming Output**: See the summary generate in real-time
+
+## Requirements
+
+This Space requires:
+- **ZeroGPU** hardware (uses `@spaces.GPU` decorator)
+- **MISTRAL_API_KEY** secret for PDF OCR functionality
+
+## Setup
+
+1. Create a new Space on HuggingFace
+2. Set the hardware to **ZeroGPU**
+3. Add `MISTRAL_API_KEY` as a secret in Space settings
+4. Upload `app.py`, `requirements.txt`, and this `README.md`

app.py

@@ -1,7 +1,800 @@
+"""
+sui-1 Summarizer - Grounded Summarization with Source Citations
+A Gradio app for HuggingFace ZeroGPU Spaces
+"""
+
+import base64
+import hashlib
+import json
+import os
+import re
+from threading import Thread
+from typing import Generator
+
 import gradio as gr
+import spaces
+import spacy
+import spacy.cli
+import torch
+from mistralai import Mistral
+from mistral_common.protocol.instruct.request import ChatCompletionRequest
+from mistral_common.tokens.tokenizers.mistral import MistralTokenizer
+from transformers import AutoTokenizer, Mistral3ForConditionalGeneration, TextIteratorStreamer
+
+
+# ============================================================================
+# Configuration
+# ============================================================================
+
+MODEL_ID = "ellamind/sui-1-24b"
+HF_TOKENIZER_ID = "mistralai/Mistral-Small-3.1-24B-Instruct-2503"
+
+LANGUAGES = {
+    "English": "en",
+    "German": "de",
+    "Spanish": "es",
+    "French": "fr",
+    "Italian": "it",
+}
+
+SPACY_MODELS = {
+    "en": "en_core_web_sm",
+    "de": "de_core_news_sm",
+    "es": "es_core_news_sm",
+    "fr": "fr_core_news_sm",
+    "it": "it_core_news_sm",
+}
+
+
+# ============================================================================
+# Initialization
+# ============================================================================
+
+def download_spacy_models():
+    """Download all required spaCy models."""
+    for model_name in SPACY_MODELS.values():
+        spacy.cli.download(model_name)
+
+
+def load_model():
+    """Load the sui-1-24b model and tokenizers."""
+    print(f"Loading model: {MODEL_ID}")
+    
+    tokenizer = MistralTokenizer.from_hf_hub(MODEL_ID)
+    model = Mistral3ForConditionalGeneration.from_pretrained(
+        MODEL_ID,
+        dtype=torch.bfloat16,
+        device_map="auto",
+    )
+    hf_tokenizer = AutoTokenizer.from_pretrained(HF_TOKENIZER_ID)
+    
+    print("Model loaded successfully")
+    return model, tokenizer, hf_tokenizer
+
+
+# Initialize on startup
+download_spacy_models()
+model, tokenizer, hf_tokenizer = load_model()
+nlp_cache: dict[str, spacy.Language] = {}
+
+
+# ============================================================================
+# NLP Utilities
+# ============================================================================
+
+def get_nlp(lang_code: str) -> spacy.Language:
+    """Get or load spaCy model for sentence segmentation."""
+    if lang_code not in nlp_cache:
+        model_name = SPACY_MODELS.get(lang_code, "en_core_web_sm")
+        try:
+            nlp_cache[lang_code] = spacy.load(model_name)
+        except OSError:
+            print(f"spaCy model '{model_name}' not found, using English")
+            nlp_cache[lang_code] = spacy.load("en_core_web_sm")
+    return nlp_cache[lang_code]
+
+
+def tag_sentences(text: str, lang_code: str) -> tuple[str, dict[str, str], dict[str, int]]:
+    """
+    Tag each sentence with a unique XML identifier.
+    
+    Returns:
+        tagged: Text with XML tags around each sentence
+        tag_to_sentence: Mapping of tag to original sentence
+        tag_to_pos: Mapping of tag to character position
+    """
+    nlp = get_nlp(lang_code)
+    doc = nlp(text)
+    
+    tagged = ""
+    tag_to_sentence = {}
+    tag_to_pos = {}
+    
+    for i, sent in enumerate(doc.sents):
+        sentence = sent.text.strip()
+        if sentence:
+            tag = hashlib.md5(f"{i}_{sentence[:50]}".encode()).hexdigest()[:8]
+            tag_to_sentence[tag] = sentence
+            tag_to_pos[tag] = sent.start_char
+            tagged += f"<{tag}>{sentence}</{tag}>"
+    
+    return tagged, tag_to_sentence, tag_to_pos
+
+
+# ============================================================================
+# PDF Processing
+# ============================================================================
+
+def extract_pdf_text(pdf_path: str) -> tuple[str, dict[int, tuple[int, int]]]:
+    """
+    Extract text from PDF using Mistral OCR.
+    
+    Returns:
+        text: Concatenated markdown text from all pages
+        page_ranges: Dict mapping page number (1-indexed) to (start_char, end_char)
+    """
+    api_key = os.environ.get("MISTRAL_API_KEY")
+    if not api_key:
+        raise ValueError("MISTRAL_API_KEY not set. Please add it to your Space secrets.")
+    
+    with open(pdf_path, "rb") as f:
+        base64_pdf = base64.b64encode(f.read()).decode("utf-8")
+    
+    client = Mistral(api_key=api_key)
+    response = client.ocr.process(
+        model="mistral-ocr-latest",
+        document={
+            "type": "document_url",
+            "document_url": f"data:application/pdf;base64,{base64_pdf}",
+        },
+        include_image_base64=False,
+    )
+    
+    full_text = ""
+    page_ranges = {}
+    
+    for page in response.pages:
+        page_num = page.index + 1
+        start = len(full_text)
+        page_text = page.markdown or ""
+        full_text += page_text + "\n\n"
+        page_ranges[page_num] = (start, len(full_text))
+    
+    return full_text.strip(), page_ranges
+
+
+def find_page(char_pos: int, page_ranges: dict[int, tuple[int, int]]) -> int:
+    """Find which page a character position belongs to."""
+    for page_num, (start, end) in page_ranges.items():
+        if start <= char_pos < end:
+            return page_num
+    return 1
+
+
+# ============================================================================
+# Output Processing
+# ============================================================================
+
+def process_output(
+    text: str,
+    tag_to_sentence: dict[str, str],
+    tag_to_pos: dict[str, int],
+    page_ranges: dict[int, tuple[int, int]] | None,
+) -> tuple[str, list[dict]]:
+    """
+    Process model output to replace XML tags with numbered citations.
+    
+    Returns:
+        processed_text: Summary with [1], [2], etc.
+        sources: List of dicts with citation info
+    """
+    counter = [0]
+    sources = []
+    
+    def replace(match):
+        tag = match.group(1)
+        counter[0] += 1
+        num = counter[0]
+        
+        source_info = {
+            "num": num,
+            "tag": tag,
+            "sentence": tag_to_sentence.get(tag, ""),
+        }
+        
+        if page_ranges and tag in tag_to_pos:
+            source_info["page"] = find_page(tag_to_pos[tag], page_ranges)
+        
+        sources.append(source_info)
+        return f"[{num}]"
+    
+    processed = re.sub(r'\[<([a-f0-9]{8})>\]', replace, text)
+    return processed, sources
+
+
+def extract_json_field(raw_output: str, field: str) -> str:
+    """Extract a field from JSON output (handles partial JSON during streaming)."""
+    # Try complete JSON first
+    json_match = re.search(r'\{[\s\S]*\}', raw_output)
+    if json_match:
+        try:
+            data = json.loads(json_match.group())
+            return data.get(field, "")
+        except json.JSONDecodeError:
+            pass
+    
+    # Fallback: extract partial field string
+    match = re.search(rf'"{field}"\s*:\s*"((?:[^"\\]|\\.)*)', raw_output)
+    if match:
+        partial = match.group(1)
+        return partial.replace('\\"', '"').replace('\\n', '\n')
+    
+    return ""
+
+
+def extract_summary_and_structure(raw_output: str) -> tuple[str, str]:
+    """
+    Extract both summary and structure from JSON output.
+    
+    Returns:
+        summary: The main summary text
+        structure: The reasoning/planning text
+    """
+    summary = extract_json_field(raw_output, "summary")
+    structure = extract_json_field(raw_output, "structure")
+    return summary, structure
+
+
+def format_sources(sources: list[dict], has_pages: bool) -> str:
+    """Format sources as markdown."""
+    lines = []
+    for src in sources:
+        sentence = src["sentence"]
+        if len(sentence) > 120:
+            sentence = sentence[:117] + "..."
+        
+        if has_pages and "page" in src:
+            lines.append(f"**[{src['num']}]** (p.{src['page']}) {sentence}")
+        else:
+            lines.append(f"**[{src['num']}]** {sentence}")
+    
+    return "\n\n".join(lines)
+
+
+# ============================================================================
+# Prompt Building
+# ============================================================================
+
+def build_prompt(
+    tagged_text: str,
+    words: int,
+    language: str,
+    custom_instruction: str = "",
+) -> str:
+    """Build the summarization prompt."""
+    num_tags = max(3, min(15, words // 40))
+    
+    custom_section = ""
+    if custom_instruction.strip():
+        custom_section = f"\n# Custom Instruction\n{custom_instruction.strip()}\n"
+    
+    return f"""You are a professional summarizer, following all given instructions with the utmost care.
+<text>
+{tagged_text}
+</text>
+# Output Format
+The output must be in JSON format with the following structure:
+1. A "structure" string containing your thoughts about the content and structure of the summary
+2. An "xml_tags" list containing the XML tag identifiers from the tagged text (e.g., "<a1b2c3d4>")
+3. A "summary" string containing the actual summary with inline XML tag references
+# Instructions
+1. Start by thinking about and explaining the structure and content of your summary. Select {num_tags} XML tags from the tagged text that capture the most significant data and facts.
+2. Begin with an executive summary introducing the title, author (if available), and key findings.
+3. Structure the summary in coherent paragraphs. Every paragraph should contain at least one XML tag reference.
+4. Reference XML tags inline in square brackets (e.g., [<a1b2c3d4>]) immediately after the statement they support.
+5. Each XML tag must appear exactly once in the summary.
+6. Avoid a concluding paragraph that merely restates points.
+7. Do not use bullet points or headings unless explicitly requested.
+{custom_section}
+Parameters:
+- Word count (excl. XML tags): {words}
+- Number of XML tags: {num_tags}
+- Language: {language}
+"""
+
+
+# ============================================================================
+# Summary Generation
+# ============================================================================
+
+@spaces.GPU(duration=100)
+def generate_summary(
+    text: str,
+    lang_code: str,
+    language_name: str,
+    words: int,
+    custom_instruction: str,
+    page_ranges: dict[int, tuple[int, int]] | None,
+) -> Generator[tuple[str, str, list[dict], int], None, None]:
+    """Generate summary with streaming output."""
+    tagged_text, tag_to_sentence, tag_to_pos = tag_sentences(text, lang_code)
+    num_sentences = len(tag_to_sentence)
+    
+    prompt = build_prompt(tagged_text, words, language_name, custom_instruction)
+    messages = [{"role": "user", "content": prompt}]
+    tokenized = tokenizer.encode_chat_completion(ChatCompletionRequest(messages=messages))
+    inputs = torch.tensor([tokenized.tokens]).to(model.device)
+
+    streamer = TextIteratorStreamer(hf_tokenizer, skip_prompt=True, skip_special_tokens=True)
+
+    gen_kwargs = {
+        "input_ids": inputs,
+        "streamer": streamer,
+        "max_new_tokens": 4096,
+        "do_sample": False,
+        "pad_token_id": hf_tokenizer.eos_token_id,
+    }
+    
+    thread = Thread(target=model.generate, kwargs=gen_kwargs)
+    thread.start()
+    
+    full_output = ""
+    for new_text in streamer:
+        full_output += new_text
+        summary, structure = extract_summary_and_structure(full_output)
+        processed, sources = process_output(summary, tag_to_sentence, tag_to_pos, page_ranges)
+        yield processed, structure, sources, num_sentences
+    
+    thread.join()
+    
+    summary, structure = extract_summary_and_structure(full_output)
+    processed, sources = process_output(summary, tag_to_sentence, tag_to_pos, page_ranges)
+    yield processed, structure, sources, num_sentences
+
+
+# ============================================================================
+# Main Summarization Function
+# ============================================================================
+
+def format_output(summary: str, structure: str) -> str:
+    """Format the summary with thinking section if available."""
+    if not summary and not structure:
+        return "*Your summary will appear here...*"
+    
+    parts = []
+    
+    if structure:
+        parts.append(f'<div class="thinking-section">\n\n💭 **Thinking**\n\n*{structure}*\n\n</div>')
+    
+    if summary:
+        if structure:
+            parts.append('<div class="summary-divider"></div>')
+        parts.append(f'<div class="summary-section">\n\n{summary}\n\n</div>')
+    
+    return "\n".join(parts)
+
+
+def summarize(
+    pdf_file: str | None,
+    text_input: str | None,
+    language: str,
+    words: int,
+    custom_instruction: str,
+    progress=gr.Progress(),
+):
+    """Main entry point for summarization."""
+    page_ranges = None
+    lang_code = LANGUAGES.get(language, "en")
+    
+    # Determine input source
+    if pdf_file:
+        progress(0.1, desc="📄 Extracting text from PDF...")
+        try:
+            text, page_ranges = extract_pdf_text(pdf_file)
+        except Exception as e:
+            yield (
+                f"❌ **Error extracting PDF:** {str(e)}",
+                "Please check that MISTRAL_API_KEY is set correctly.",
+                gr.update(visible=False),
+            )
+            return
+    elif text_input and text_input.strip():
+        text = text_input.strip()
+    else:
+        yield (
+            "⚠️ Please provide a PDF file or paste some text to summarize.",
+            "",
+            gr.update(visible=False),
+        )
+        return
+    
+    if not text.strip():
+        yield (
+            "⚠️ No text content found in the document.",
+            "",
+            gr.update(visible=False),
+        )
+        return
+    
+    progress(0.2, desc="🤖 Generating summary...")
+    
+    for summary, structure, sources, _ in generate_summary(
+        text, lang_code, language, words, custom_instruction, page_ranges
+    ):
+        formatted_output = format_output(summary, structure)
+        sources_md = format_sources(sources, page_ranges is not None) if sources else "*Identifying sources...*"
+        yield formatted_output, sources_md, gr.update(visible=True)
+    
+    progress(1.0, desc="✅ Complete!")
+
+
+# ============================================================================
+# Theme & Styling
+# ============================================================================
+
+THEME = gr.themes.Soft(
+    primary_hue="indigo",
+    secondary_hue="blue",
+    neutral_hue="slate",
+    font=gr.themes.GoogleFont("Inter"),
+).set(
+    body_background_fill="linear-gradient(135deg, #667eea 0%, #764ba2 100%)",
+    body_background_fill_dark="linear-gradient(135deg, #1a1a2e 0%, #16213e 100%)",
+    block_background_fill="white",
+    block_background_fill_dark="#1e1e2e",
+    block_radius="12px",
+    block_shadow="0 4px 6px -1px rgb(0 0 0 / 0.1), 0 2px 4px -2px rgb(0 0 0 / 0.1)",
+)
+
+CSS = """
+/* Full width layout */
+.gradio-container {
+    max-width: 100% !important;
+    padding: 0 3rem !important;
+}
+
+/* Header styling */
+.main-header {
+    text-align: center;
+    padding: 2.5rem 2rem;
+    background: rgba(30, 30, 46, 0.6);
+    backdrop-filter: blur(10px);
+    border-radius: 16px;
+    margin-bottom: 2rem;
+    border: 1px solid rgba(255, 255, 255, 0.1);
+}
+
+.main-header h1 {
+    font-size: 2.75rem;
+    font-weight: 700;
+    background: linear-gradient(135deg, #a5b4fc 0%, #c4b5fd 100%);
+    -webkit-background-clip: text;
+    -webkit-text-fill-color: transparent;
+    background-clip: text;
+    margin-bottom: 0.75rem;
+}
+
+.main-header p {
+    color: #94a3b8;
+    font-size: 1.15rem;
+    margin: 0;
+}
+
+/* Card styling */
+.input-card, .output-card {
+    background: rgba(30, 30, 46, 0.8) !important;
+    border: 1px solid rgba(255, 255, 255, 0.08) !important;
+    border-radius: 16px !important;
+    padding: 1.5rem !important;
+}
+
+/* Section headers */
+.section-title {
+    color: #e2e8f0 !important;
+    font-size: 1.1rem !important;
+    font-weight: 600 !important;
+    margin-bottom: 1rem !important;
+    display: flex;
+    align-items: center;
+    gap: 0.5rem;
+}
+
+/* Summary output */
+.summary-text {
+    font-size: 1.05rem !important;
+    line-height: 1.85 !important;
+    color: #e2e8f0 !important;
+    background: transparent !important;
+}
+
+/* Thinking section */
+.thinking-section {
+    background: rgba(99, 102, 241, 0.08);
+    border-left: 3px solid #6366f1;
+    border-radius: 0 8px 8px 0;
+    padding: 1rem 1.25rem;
+    margin-bottom: 0;
+}
+
+.thinking-section p {
+    color: #a5b4fc !important;
+    font-size: 0.95rem !important;
+    line-height: 1.7 !important;
+    margin: 0 !important;
+}
+
+/* Summary divider */
+.summary-divider {
+    height: 1px;
+    background: linear-gradient(90deg, transparent, rgba(99, 102, 241, 0.3), transparent);
+    margin: 1.5rem 0;
+}
+
+/* Summary section */
+.summary-section {
+    padding-top: 0.5rem;
+}
+
+/* Sources panel - fixed styling */
+.sources-panel {
+    background: #252536 !important;
+    border-radius: 12px !important;
+    padding: 1.25rem !important;
+    font-size: 0.95rem !important;
+    max-height: 400px !important;
+    overflow-y: auto !important;
+    color: #cbd5e1 !important;
+    border: 1px solid rgba(255, 255, 255, 0.06) !important;
+}
+
+.sources-panel p {
+    color: #cbd5e1 !important;
+    margin-bottom: 0.75rem !important;
+    padding: 0.5rem 0 !important;
+    border-bottom: 1px solid rgba(255, 255, 255, 0.05) !important;
+}
+
+.sources-panel p:last-child {
+    border-bottom: none !important;
+    margin-bottom: 0 !important;
+}
+
+.sources-panel strong {
+    color: #a5b4fc !important;
+}
+
+/* Accordion styling */
+.accordion {
+    background: transparent !important;
+    border: none !important;
+}
+
+.accordion > .label-wrap {
+    background: rgba(37, 37, 54, 0.6) !important;
+    border-radius: 10px !important;
+    padding: 0.75rem 1rem !important;
+    border: 1px solid rgba(255, 255, 255, 0.06) !important;
+}
+
+/* Generate button */
+.generate-btn {
+    background: linear-gradient(135deg, #6366f1 0%, #8b5cf6 100%) !important;
+    border: none !important;
+    font-weight: 600 !important;
+    font-size: 1.05rem !important;
+    padding: 0.875rem 2rem !important;
+    border-radius: 10px !important;
+    transition: all 0.2s ease !important;
+    box-shadow: 0 4px 14px rgba(99, 102, 241, 0.4) !important;
+}
+
+.generate-btn:hover {
+    transform: translateY(-2px) !important;
+    box-shadow: 0 6px 20px rgba(99, 102, 241, 0.5) !important;
+}
+
+/* Input fields */
+.dark input, .dark textarea {
+    background: #252536 !important;
+    border: 1px solid rgba(255, 255, 255, 0.1) !important;
+    color: #e2e8f0 !important;
+    border-radius: 8px !important;
+}
+
+.dark input:focus, .dark textarea:focus {
+    border-color: #6366f1 !important;
+    box-shadow: 0 0 0 2px rgba(99, 102, 241, 0.2) !important;
+}
+
+/* Dropdown */
+.dark select {
+    background: #252536 !important;
+    border: 1px solid rgba(255, 255, 255, 0.1) !important;
+    color: #e2e8f0 !important;
+}
+
+/* Slider */
+.dark .slider {
+    background: #252536 !important;
+}
+
+/* File upload */
+.dark .upload-button {
+    background: #252536 !important;
+    border: 2px dashed rgba(255, 255, 255, 0.15) !important;
+    border-radius: 12px !important;
+}
+
+.dark .upload-button:hover {
+    border-color: #6366f1 !important;
+    background: rgba(99, 102, 241, 0.1) !important;
+}
+
+/* Divider */
+.divider {
+    color: #64748b !important;
+    text-align: center;
+    margin: 1rem 0;
+}
+
+/* Footer */
+.footer {
+    text-align: center;
+    padding: 1.5rem;
+    margin-top: 2rem;
+    color: #94a3b8;
+    font-size: 0.9rem;
+}
+
+.footer a {
+    color: #a5b4fc;
+    text-decoration: none;
+    transition: color 0.2s;
+}
+
+.footer a:hover {
+    color: #c4b5fd;
+    text-decoration: underline;
+}
+
+.footer-divider {
+    color: #475569;
+    margin: 0 0.5rem;
+}
+
+/* Hide default footer */
+footer {
+    display: none !important;
+}
+
+/* Responsive adjustments */
+@media (max-width: 768px) {
+    .gradio-container {
+        padding: 0 1rem !important;
+    }
+    
+    .main-header h1 {
+        font-size: 2rem;
+    }
+}
+"""
+
+
+# ============================================================================
+# Gradio Interface
+# ============================================================================
+
+FORCE_DARK_MODE_JS = """
+() => {
+    document.body.classList.remove('light');
+    document.body.classList.add('dark');
+}
+"""
+
+
+def create_app() -> gr.Blocks:
+    """Create the Gradio application."""
+    with gr.Blocks(title="sui-1 Summarizer", theme=THEME, css=CSS, js=FORCE_DARK_MODE_JS) as app:
+        
+        # Header
+        gr.HTML("""
+        <div class="main-header">
+            <h1>📄 sui-1 Summarizer</h1>
+            <p>Generate grounded summaries with verifiable source citations</p>
+        </div>
+        """)
+        
+        with gr.Row(equal_height=False):
+            # Left column: Inputs
+            with gr.Column(scale=1):
+                gr.Markdown("### 📥 Input", elem_classes=["section-title"])
+                
+                pdf_input = gr.File(
+                    label="Upload PDF",
+                    file_types=[".pdf"],
+                    type="filepath",
+                )
+                
+                gr.Markdown("<div class='divider'>— or —</div>")
+                
+                text_input = gr.Textbox(
+                    label="Paste Text",
+                    placeholder="Paste your document text here...",
+                    lines=8,
+                    max_lines=15,
+                )
+                
+                gr.Markdown("### ⚙️ Settings", elem_classes=["section-title"])
+                
+                language = gr.Dropdown(
+                    choices=list(LANGUAGES.keys()),
+                    value="English",
+                    label="Output Language",
+                )
+                
+                words = gr.Slider(
+                    minimum=100,
+                    maximum=800,
+                    value=250,
+                    step=50,
+                    label="Target Word Count",
+                    info="Longer summaries include more citations",
+                )
+                
+                custom_instruction = gr.Textbox(
+                    label="Custom Instructions (Optional)",
+                    placeholder="e.g., 'Focus on methodology' or 'Use bullet points'",
+                    lines=2,
+                )
+                
+                generate_btn = gr.Button(
+                    "✨ Generate Summary",
+                    variant="primary",
+                    size="lg",
+                    elem_classes=["generate-btn"],
+                )
+            
+            # Right column: Output
+            with gr.Column(scale=1):
+                gr.Markdown("### 📋 Summary", elem_classes=["section-title"])
+
+                summary_output = gr.Markdown(
+                    value="*Your summary will appear here...*",
+                    elem_classes=["summary-text"],
+                )
+                
+                with gr.Accordion(
+                    "📚 Source Citations",
+                    open=False,
+                    visible=False,
+                ) as sources_accordion:
+                    sources_output = gr.Markdown(elem_classes=["sources-panel"])
+        
+        # Event handlers
+        generate_btn.click(
+            fn=summarize,
+            inputs=[pdf_input, text_input, language, words, custom_instruction],
+            outputs=[summary_output, sources_output, sources_accordion],
+        )
+        
+        # Footer
+        gr.HTML("""
+        <div class="footer">
+            <strong>Model:</strong> <a href="https://huggingface.co/ellamind/sui-1-24b" target="_blank">ellamind/sui-1-24b</a>
+            <span class="footer-divider">•</span>
+            <strong>PDF OCR:</strong> <a href="https://docs.mistral.ai/capabilities/document_ai" target="_blank">Mistral OCR</a>
+            <span class="footer-divider">•</span>
+            Made with 💜 by <a href="https://huggingface.co/ellamind" target="_blank">ellamind</a>
+        </div>
+        """)
+    
+    return app
+
 
-def greet(name):
-    return "Hello " + name + "!!"
+# ============================================================================
+# Entry Point
+# ============================================================================
 
-demo = gr.Interface(fn=greet, inputs="text", outputs="text")
-demo.launch()
+if __name__ == "__main__":
+    app = create_app()
+    app.launch()

requirements.txt

@@ -0,0 +1,8 @@
+gradio>=5.0.0
+spaces
+spacy>=3.7.0
+torch
+transformers>=4.45.0
+mistralai>=1.0.0
+mistral_common>=1.5.0
+accelerate