Upload 3 files

app.py

@@ -0,0 +1,417 @@
+"""
+Main application for Dynamic Highscores system.
+
+This file integrates all components into a unified application.
+"""
+
+import os
+import gradio as gr
+import threading
+import time
+from database_schema import DynamicHighscoresDB
+from auth import HuggingFaceAuth
+from benchmark_selection import BenchmarkSelector, create_benchmark_selection_ui
+from evaluation_queue import EvaluationQueue, create_model_submission_ui
+from leaderboard import Leaderboard, create_leaderboard_ui
+from sample_benchmarks import add_sample_benchmarks
+
+# Initialize components in main thread
+db = DynamicHighscoresDB()
+auth_manager = HuggingFaceAuth(db)
+benchmark_selector = BenchmarkSelector(db, auth_manager)
+evaluation_queue = EvaluationQueue(db, auth_manager)
+leaderboard = Leaderboard(db)
+
+# Initialize sample benchmarks if none exist
+print("Checking for existing benchmarks...")
+benchmarks = db.get_benchmarks()
+if not benchmarks or len(benchmarks) == 0:
+    print("No benchmarks found. Adding sample benchmarks...")
+    try:
+        # Make sure the database path is clear
+        print(f"Database path: {db.db_path}")
+        
+        # Import and call the function directly 
+        num_added = add_sample_benchmarks()
+        print(f"Added {num_added} sample benchmarks.")
+    except Exception as e:
+        print(f"Error adding sample benchmarks: {str(e)}")
+        # Try direct DB insertion as fallback
+        try:
+            print("Attempting direct benchmark insertion...")
+            db.add_benchmark(
+                name="MMLU (Massive Multitask Language Understanding)",
+                dataset_id="cais/mmlu",
+                description="Tests knowledge across 57 subjects"
+            )
+            print("Added fallback benchmark.")
+        except Exception as inner_e:
+            print(f"Fallback insertion failed: {str(inner_e)}")
+else:
+    print(f"Found {len(benchmarks)} existing benchmarks.")
+
+# Custom CSS with theme awareness
+css = """
+/* Theme-adaptive colored info box */
+.info-text {
+    background-color: rgba(53, 130, 220, 0.1);
+    padding: 12px;
+    border-radius: 8px;
+    border-left: 4px solid #3498db;
+    margin: 12px 0;
+}
+
+/* High-contrast text for elements - works in light and dark themes */
+.info-text, .header, .footer, .tab-content, 
+button, input, textarea, select, option, 
+.gradio-container *, .markdown-text {
+    color: var(--text-color, inherit) !important;
+}
+
+/* Container styling */
+.container {
+    max-width: 1200px;
+    margin: 0 auto;
+}
+
+/* Header styling */
+.header {
+    text-align: center;
+    margin-bottom: 20px;
+    font-weight: bold;
+    font-size: 24px;
+}
+
+/* Footer styling */
+.footer {
+    text-align: center;
+    margin-top: 40px;
+    padding: 20px;
+    border-top: 1px solid var(--border-color-primary, #eee);
+}
+
+/* Login section styling */
+.login-section {
+    padding: 10px;
+    margin-bottom: 15px;
+    border-radius: 8px;
+    background-color: rgba(250, 250, 250, 0.1);
+    text-align: center;
+}
+
+/* Token input styling */
+.token-input {
+    margin: 10px 0;
+    padding: 8px;
+    border-radius: 4px;
+    border: 1px solid #ccc;
+    width: 100%;
+}
+
+/* Force high contrast on specific input areas */
+input[type="text"], input[type="password"], textarea {
+    background-color: var(--background-fill-primary) !important;
+    color: var(--body-text-color) !important;
+}
+
+/* Force text visibility in multiple contexts */
+.gradio-markdown p, .gradio-markdown h1, .gradio-markdown h2, 
+.gradio-markdown h3, .gradio-markdown h4, .gradio-markdown li {
+    color: var(--body-text-color) !important;
+}
+
+/* Fix dark mode text visibility */
+@media (prefers-color-scheme: dark) {
+    input, textarea, select {
+        color: #ffffff !important;
+    }
+    
+    ::placeholder {
+        color: rgba(255, 255, 255, 0.5) !important;
+    }
+}
+"""
+
+# Create token input UI
+def create_token_input_ui():
+    with gr.Row():
+        with gr.Column():
+            gr.Markdown("### HuggingFace Token Authentication")
+            gr.Markdown("""
+            Enter your HuggingFace tokens to use this application. 
+            You can find your tokens in your [HuggingFace settings](https://huggingface.co/settings/tokens).
+            
+            - **Read Token**: Required for accessing models and datasets
+            - **Write Token**: Required for submitting evaluation results
+            
+            Your tokens are stored only in your browser's local storage and are not saved on the server.
+            """)
+            
+            read_token = gr.Textbox(
+                label="Read Token", 
+                placeholder="Enter your HuggingFace read token", 
+                type="password"
+            )
+            write_token = gr.Textbox(
+                label="Write Token", 
+                placeholder="Enter your HuggingFace write token", 
+                type="password"
+            )
+            save_button = gr.Button("Save Tokens")
+            clear_button = gr.Button("Clear Tokens")
+            token_status = gr.Markdown("Not authenticated")
+            
+            # Hidden field to store the token status
+            token_state = gr.State(None)
+    
+    # JavaScript to handle token storage
+    token_js = """
+    <script>
+    // Function to save tokens to localStorage
+    function saveTokens() {
+        const readToken = document.querySelector('input[placeholder="Enter your HuggingFace read token"]').value;
+        const writeToken = document.querySelector('input[placeholder="Enter your HuggingFace write token"]').value;
+        
+        if (readToken && writeToken) {
+            localStorage.setItem("hf_read_token", readToken);
+            localStorage.setItem("hf_write_token", writeToken);
+            
+            // Set token in cookie for server-side access
+            document.cookie = "hf_token=" + readToken + "; path=/; SameSite=Strict";
+            
+            // Update status
+            const statusElement = document.querySelector('div[data-testid="markdown"] p');
+            if (statusElement) {
+                statusElement.textContent = "Authenticated with tokens";
+                statusElement.style.color = "green";
+            }
+            
+            // Reload page to apply tokens
+            setTimeout(() => window.location.reload(), 1000);
+        } else {
+            alert("Please enter both read and write tokens");
+        }
+    }
+    
+    // Function to clear tokens from localStorage
+    function clearTokens() {
+        localStorage.removeItem("hf_read_token");
+        localStorage.removeItem("hf_write_token");
+        
+        // Clear token cookie
+        document.cookie = "hf_token=; path=/; max-age=0; SameSite=Strict";
+        
+        // Update status
+        const statusElement = document.querySelector('div[data-testid="markdown"] p');
+        if (statusElement) {
+            statusElement.textContent = "Not authenticated";
+            statusElement.style.color = "red";
+        }
+        
+        // Clear input fields
+        document.querySelector('input[placeholder="Enter your HuggingFace read token"]').value = "";
+        document.querySelector('input[placeholder="Enter your HuggingFace write token"]').value = "";
+        
+        // Reload page to apply changes
+        setTimeout(() => window.location.reload(), 1000);
+    }
+    
+    // Function to load tokens from localStorage
+    function loadTokens() {
+        const readToken = localStorage.getItem("hf_read_token");
+        const writeToken = localStorage.getItem("hf_write_token");
+        
+        if (readToken && writeToken) {
+            document.querySelector('input[placeholder="Enter your HuggingFace read token"]').value = readToken;
+            document.querySelector('input[placeholder="Enter your HuggingFace write token"]').value = writeToken;
+            
+            // Update status
+            const statusElement = document.querySelector('div[data-testid="markdown"] p');
+            if (statusElement) {
+                statusElement.textContent = "Authenticated with tokens";
+                statusElement.style.color = "green";
+            }
+            
+            // Set token in cookie for server-side access if not already set
+            if (!document.cookie.includes("hf_token=")) {
+                document.cookie = "hf_token=" + readToken + "; path=/; SameSite=Strict";
+            }
+        }
+    }
+    
+    // Add event listeners once DOM is loaded
+    document.addEventListener("DOMContentLoaded", function() {
+        // Load tokens from localStorage
+        loadTokens();
+        
+        // Add event listeners to buttons
+        const saveButton = document.querySelector('button:nth-of-type(1)');
+        const clearButton = document.querySelector('button:nth-of-type(2)');
+        
+        if (saveButton) {
+            saveButton.addEventListener("click", saveTokens);
+        }
+        
+        if (clearButton) {
+            clearButton.addEventListener("click", clearTokens);
+        }
+    });
+    </script>
+    """
+    
+    return read_token, write_token, save_button, clear_button, token_status, token_state, token_js
+
+# Simple manual authentication check
+def check_user(request: gr.Request):
+    if request:
+        # Check for token in cookies
+        token = request.cookies.get("hf_token")
+        
+        if token:
+            try:
+                # Validate token with HuggingFace
+                user_info = auth_manager.hf_api.whoami(token=token)
+                
+                if user_info:
+                    username = user_info.get("name", "")
+                    print(f"User authenticated via token: {username}")
+                    
+                    # Check if user exists in our database, create if not
+                    user = db.get_user_by_username(username)
+                    if not user:
+                        # Create user if they don't exist
+                        print(f"Creating new user: {username}")
+                        is_admin = (username == "Quazim0t0")
+                        db.add_user(username, username, is_admin)
+                        user = db.get_user_by_username(username)
+                    
+                    return username
+            except Exception as e:
+                print(f"Token validation error: {e}")
+    
+    return None
+
+# Start evaluation queue worker
+def start_queue_worker():
+    # Wait a moment to ensure app is initialized
+    time.sleep(2)
+    try:
+        print("Starting evaluation queue worker...")
+        evaluation_queue.start_worker()
+    except Exception as e:
+        print(f"Error starting queue worker: {e}")
+
+# Create Gradio app
+with gr.Blocks(css=css, title="Dynamic Highscores") as app:
+    # State to track user
+    user_state = gr.State(None)
+    
+    # Token input UI
+    read_token, write_token, save_button, clear_button, token_status, token_state, token_js = create_token_input_ui()
+    
+    # Add the token handling JavaScript
+    gr.HTML(token_js)
+    
+    gr.Markdown("# 🏆 Dynamic Highscores", elem_classes=["header"])
+    gr.Markdown("""
+    Welcome to Dynamic Highscores - a community benchmark platform for evaluating and comparing language models.
+    
+    - **Add your own benchmarks** from HuggingFace datasets
+    - **Submit your models** for CPU-only evaluation
+    - **Compare performance** across different models and benchmarks
+    - **Filter results** by model type (Merge, Agent, Reasoning, Coding, etc.)
+    """, elem_classes=["info-text"])
+    
+    # Main tabs
+    with gr.Tabs() as tabs:
+        with gr.TabItem("📊 Leaderboard", id=0):
+            leaderboard_ui = create_leaderboard_ui(leaderboard, db)
+        
+        with gr.TabItem("🚀 Submit Model", id=1):
+            submission_ui = create_model_submission_ui(evaluation_queue, auth_manager, db)
+        
+        with gr.TabItem("🔍 Benchmarks", id=2):
+            benchmark_ui = create_benchmark_selection_ui(benchmark_selector, auth_manager)
+        
+        with gr.TabItem("🌐 Community Framework", id=3):
+            # Create a simple placeholder for the Community Framework tab
+            gr.Markdown("""
+            # 🌐 Dynamic Highscores Community Framework
+
+            ## About Dynamic Highscores
+
+            Dynamic Highscores is an open-source community benchmark system for evaluating language models on any dataset. This project was created to fill the gap left by the retirement of HuggingFace's "Open LLM Leaderboards" which were discontinued due to outdated benchmarks.
+
+            ### Key Features
+
+            - **Flexible Benchmarking**: Test models against any HuggingFace dataset, not just predefined benchmarks
+            - **Community-Driven**: Anyone can add new benchmarks and submit models for evaluation
+            - **Modern Evaluation**: Focus on contemporary benchmarks that better reflect current model capabilities
+            - **CPU-Only Evaluation**: Ensures fair comparisons across different models
+            - **Daily Submission Limits**: Prevents system abuse (one benchmark per day per user)
+            - **Model Tagging**: Categorize models as Merge, Agent, Reasoning, Coding, etc.
+            - **Unified Leaderboard**: View all models with filtering capabilities by tags
+
+            ### Why This Project Matters
+
+            When HuggingFace retired their "Open LLM Leaderboards," the community lost a valuable resource for comparing model performance. The benchmarks used had become outdated and didn't reflect the rapid advances in language model capabilities.
+
+            Dynamic Highscores addresses this issue by allowing users to select from any benchmark on HuggingFace, including the most recent and relevant datasets. This ensures that models are evaluated on tasks that matter for current applications.
+
+            ## Model Configuration System (Coming Soon)
+
+            We're working on a modular system for model configurations that will allow users to:
+
+            - Create and apply predefined configurations for different model types
+            - Define parameters such as Temperature, Top-K, Min-P, Top-P, and Repetition Penalty
+            - Share optimal configurations with the community
+
+            ### Example Configuration (Gemma)
+
+            ```
+            Temperature: 1.0
+            Top_K: 64
+            Min_P: 0.01
+            Top_P: 0.95
+            Repetition Penalty: 1.0
+            ```
+
+            ## Contributing to the Project
+
+            We welcome contributions from the community! If you'd like to improve Dynamic Highscores, here are some ways to get involved:
+
+            - **Add New Features**: Enhance the platform with additional functionality
+            - **Improve Evaluation Methods**: Help make model evaluations more accurate and efficient
+            - **Fix Bugs**: Address issues in the codebase
+            - **Enhance Documentation**: Make the project more accessible to new users
+            - **Add Model Configurations**: Contribute optimal configurations for different model types
+
+            To contribute, fork the repository, make your changes, and submit a pull request. We appreciate all contributions, big or small!
+            """)
+    
+    gr.Markdown("""
+    ### About Dynamic Highscores
+    
+    This platform allows users to select benchmarks from HuggingFace datasets and evaluate models against them.
+    Each user can submit one benchmark per day (admin users are exempt from this limit).
+    All evaluations run on CPU only to ensure fair comparisons.
+    
+    Created by Quazim0t0
+    """, elem_classes=["footer"])
+    
+    # Check login on page load
+    app.load(
+        fn=check_user,
+        inputs=[],
+        outputs=[user_state]
+    )
+
+# Launch the app
+if __name__ == "__main__":
+    # Start queue worker in a separate thread
+    queue_thread = threading.Thread(target=start_queue_worker)
+    queue_thread.daemon = True
+    queue_thread.start()
+    
+    # Launch the app
+    app.launch()