local-deep-research 0.5.7__py3-none-any.whl → 0.6.0__py3-none-any.whl

local_deep_research/web/templates/pages/benchmark_simple.html

@@ -0,0 +1,453 @@
+{% extends "base.html" %}
+
+{% set active_page = 'benchmark' %}
+
+{% block title %}Benchmark - Deep Research System{% endblock %}
+
+{% block extra_head %}
+<meta name="csrf-token" content="{{ csrf_token() }}">
+<style>
+.benchmark-card {
+    max-width: 700px;
+    margin: 0 auto;
+}
+
+.current-config {
+    background: var(--bg-color);
+    border: 1px solid var(--border-color);
+    border-radius: 8px;
+    padding: 20px;
+    margin-bottom: 20px;
+}
+
+.config-item {
+    display: flex;
+    justify-content: space-between;
+    padding: 8px 0;
+    border-bottom: 1px solid var(--border-color);
+}
+
+.config-item:last-child {
+    border-bottom: none;
+}
+
+.config-label {
+    font-weight: 500;
+    color: var(--text-muted);
+}
+
+.config-value {
+    font-weight: 600;
+    color: var(--primary-color);
+}
+
+.dataset-selection {
+    margin: 20px 0;
+}
+
+.dataset-option {
+    display: flex;
+    align-items: center;
+    padding: 15px;
+    margin: 10px 0;
+    background: var(--card-bg);
+    border: 2px solid var(--border-color);
+    border-radius: 8px;
+    cursor: pointer;
+    transition: all 0.2s;
+}
+
+.dataset-option:hover {
+    border-color: var(--primary-color);
+}
+
+.dataset-option.selected {
+    border-color: var(--primary-color);
+    background: var(--primary-light);
+}
+
+.dataset-option input[type="number"] {
+    width: 80px;
+    margin-left: auto;
+}
+
+.start-button {
+    width: 100%;
+    padding: 15px;
+    font-size: 1.1rem;
+}
+
+.progress-section {
+    display: none;
+    margin-top: 30px;
+}
+
+.progress-bar {
+    width: 100%;
+    height: 30px;
+    background: var(--bg-color);
+    border-radius: 15px;
+    overflow: hidden;
+    margin: 20px 0;
+}
+
+.progress-fill {
+    height: 100%;
+    background: linear-gradient(90deg, var(--primary-color), var(--accent-color));
+    width: 0%;
+    transition: width 0.3s ease;
+}
+
+.stats-grid {
+    display: grid;
+    grid-template-columns: repeat(auto-fit, minmax(150px, 1fr));
+    gap: 15px;
+    margin-top: 20px;
+}
+
+.stat-card {
+    padding: 15px;
+    background: var(--bg-color);
+    border-radius: 8px;
+    text-align: center;
+}
+
+.stat-value {
+    font-size: 1.8rem;
+    font-weight: bold;
+    color: var(--primary-color);
+}
+
+.stat-label {
+    font-size: 0.9rem;
+    color: var(--text-muted);
+    margin-top: 5px;
+}
+</style>
+{% endblock %}
+
+{% block content %}
+<div class="page active" id="benchmark">
+    <div class="page-header">
+        <h1>Benchmark Your Configuration</h1>
+        <p class="page-subtitle">Test your current settings against standard datasets</p>
+    </div>
+
+    <div id="benchmark-alert" class="settings-alert-container" style="display:none"></div>
+
+    <div class="card benchmark-card">
+        <div class="card-content">
+            <!-- Current Configuration Display -->
+            <div class="current-config">
+                <h3>Current Configuration</h3>
+                <div id="config-display">
+                    <div class="config-item">
+                        <span class="config-label">Loading...</span>
+                        <span class="config-value"></span>
+                    </div>
+                </div>
+            </div>
+
+            <!-- Dataset Selection -->
+            <div class="dataset-selection">
+                <h3>Select Datasets to Test</h3>
+
+                <div class="dataset-option selected" data-dataset="simpleqa">
+                    <div>
+                        <h4>SimpleQA</h4>
+                        <p>Fact-based questions with clear answers</p>
+                    </div>
+                    <input type="number" id="simpleqa_count" value="20" min="1" max="500">
+                </div>
+
+                <div class="dataset-option selected" data-dataset="browsecomp">
+                    <div>
+                        <h4>BrowseComp</h4>
+                        <p>Complex browsing and comparison tasks</p>
+                    </div>
+                    <input type="number" id="browsecomp_count" value="10" min="1" max="200">
+                </div>
+            </div>
+
+            <!-- Start Button -->
+            <button type="button" id="start-benchmark" class="btn btn-primary start-button">
+                <i class="fas fa-play"></i> Start Benchmark
+            </button>
+
+            <!-- Progress Section -->
+            <div class="progress-section" id="progress-section">
+                <h3>Benchmark Progress</h3>
+
+                <div class="progress-bar">
+                    <div class="progress-fill" id="progress-fill"></div>
+                </div>
+
+                <div class="stats-grid">
+                    <div class="stat-card">
+                        <div class="stat-value" id="stat-accuracy">--%</div>
+                        <div class="stat-label">Accuracy</div>
+                    </div>
+                    <div class="stat-card">
+                        <div class="stat-value" id="stat-completed">0</div>
+                        <div class="stat-label">Completed</div>
+                    </div>
+                    <div class="stat-card">
+                        <div class="stat-value" id="stat-rate">--</div>
+                        <div class="stat-label">Per Minute</div>
+                    </div>
+                    <div class="stat-card">
+                        <div class="stat-value" id="stat-remaining">--</div>
+                        <div class="stat-label">Time Left</div>
+                    </div>
+                </div>
+
+                <div style="margin-top: 20px; text-align: center;">
+                    <button type="button" id="cancel-benchmark" class="btn btn-secondary">
+                        <i class="fas fa-stop"></i> Cancel
+                    </button>
+                </div>
+            </div>
+        </div>
+    </div>
+</div>
+
+<script>
+let currentBenchmarkId = null;
+let progressInterval = null;
+
+document.addEventListener('DOMContentLoaded', function() {
+    loadCurrentConfig();
+    setupEventListeners();
+});
+
+async function loadCurrentConfig() {
+    try {
+        // Get current settings from the database
+        const response = await fetch('/research/settings/api');
+        const data = await response.json();
+
+        if (data.settings) {
+            const config = {
+                provider: data.settings['llm.provider']?.value || 'Not configured',
+                model: data.settings['llm.model']?.value || 'Not configured',
+                search_tool: data.settings['search.tool']?.value || 'searxng',
+                iterations: data.settings['search.iterations']?.value || '8',
+                questions_per_iteration: data.settings['search.questions_per_iteration']?.value || '5',
+                search_strategy: data.settings['search.search_strategy']?.value || 'focused_iteration'
+            };
+            displayConfig(config);
+        }
+    } catch (error) {
+        console.error('Error loading config:', error);
+        // Use fallback display
+        displayConfig({
+            provider: 'Loading...',
+            model: 'Loading...',
+            search_tool: 'Loading...',
+            iterations: '-'
+        });
+    }
+}
+
+function displayConfig(config) {
+    const configDisplay = document.getElementById('config-display');
+    configDisplay.innerHTML = `
+        <div class="config-item">
+            <span class="config-label">LLM Provider</span>
+            <span class="config-value">${config.provider || 'Not set'}</span>
+        </div>
+        <div class="config-item">
+            <span class="config-label">Model</span>
+            <span class="config-value">${config.model || 'Not set'}</span>
+        </div>
+        <div class="config-item">
+            <span class="config-label">Search Engine</span>
+            <span class="config-value">${config.search_tool || 'searxng'}</span>
+        </div>
+        <div class="config-item">
+            <span class="config-label">Search Iterations</span>
+            <span class="config-value">${config.iterations || '8'}</span>
+        </div>
+    `;
+}
+
+function setupEventListeners() {
+    // Dataset selection
+    document.querySelectorAll('.dataset-option').forEach(option => {
+        option.addEventListener('click', function(e) {
+            if (!e.target.matches('input')) {
+                this.classList.toggle('selected');
+            }
+        });
+    });
+
+    // Start benchmark
+    document.getElementById('start-benchmark').addEventListener('click', startBenchmark);
+
+    // Cancel benchmark
+    document.getElementById('cancel-benchmark').addEventListener('click', cancelBenchmark);
+}
+
+async function startBenchmark() {
+    // Get dataset configuration
+    const datasets = {};
+    document.querySelectorAll('.dataset-option.selected').forEach(option => {
+        const dataset = option.getAttribute('data-dataset');
+        const count = parseInt(option.querySelector('input').value) || 0;
+        if (count > 0) {
+            datasets[dataset] = { count: count };
+        }
+    });
+
+    if (Object.keys(datasets).length === 0) {
+        showAlert('Please select at least one dataset', 'error');
+        return;
+    }
+
+    // Disable start button
+    document.getElementById('start-benchmark').disabled = true;
+
+    try {
+        // Get CSRF token
+        const csrfToken = document.querySelector('meta[name="csrf-token"]').getAttribute('content');
+
+        const response = await fetch('/benchmark/api/start-simple', {
+            method: 'POST',
+            headers: {
+                'Content-Type': 'application/json',
+                'X-CSRFToken': csrfToken
+            },
+            body: JSON.stringify({ datasets_config: datasets })
+        });
+
+        const data = await response.json();
+
+        if (data.success) {
+            currentBenchmarkId = data.benchmark_run_id;
+            showProgress();
+            startProgressTracking();
+            showAlert('Benchmark started!', 'success');
+        } else {
+            showAlert('Error: ' + data.error, 'error');
+            document.getElementById('start-benchmark').disabled = false;
+        }
+    } catch (error) {
+        console.error('Error starting benchmark:', error);
+        showAlert('Failed to start benchmark', 'error');
+        document.getElementById('start-benchmark').disabled = false;
+    }
+}
+
+function showProgress() {
+    document.querySelector('.dataset-selection').style.display = 'none';
+    document.getElementById('start-benchmark').style.display = 'none';
+    document.getElementById('progress-section').style.display = 'block';
+}
+
+function hideProgress() {
+    document.querySelector('.dataset-selection').style.display = 'block';
+    document.getElementById('start-benchmark').style.display = 'block';
+    document.getElementById('progress-section').style.display = 'none';
+    document.getElementById('start-benchmark').disabled = false;
+}
+
+function startProgressTracking() {
+    updateProgress(); // Initial update
+    progressInterval = setInterval(updateProgress, 2000);
+}
+
+async function updateProgress() {
+    if (!currentBenchmarkId) return;
+
+    try {
+        const response = await fetch(`/benchmark/api/status/${currentBenchmarkId}`);
+        const data = await response.json();
+
+        if (data.success) {
+            const status = data.status;
+
+            // Update progress bar
+            const percentage = status.total_examples > 0 ?
+                (status.completed_examples / status.total_examples * 100) : 0;
+            document.getElementById('progress-fill').style.width = percentage + '%';
+
+            // Update stats
+            document.getElementById('stat-accuracy').textContent =
+                status.overall_accuracy ? status.overall_accuracy.toFixed(1) + '%' : '--%';
+            document.getElementById('stat-completed').textContent =
+                `${status.completed_examples}/${status.total_examples}`;
+            document.getElementById('stat-rate').textContent =
+                status.processing_rate ? status.processing_rate.toFixed(1) : '--';
+
+            // Estimate remaining time
+            if (status.processing_rate > 0 && status.completed_examples < status.total_examples) {
+                const remaining = status.total_examples - status.completed_examples;
+                const minutes = Math.ceil(remaining / status.processing_rate);
+                document.getElementById('stat-remaining').textContent = `${minutes}m`;
+            } else {
+                document.getElementById('stat-remaining').textContent = '--';
+            }
+
+            // Check if completed
+            if (status.status === 'completed' || status.status === 'failed' || status.status === 'cancelled') {
+                clearInterval(progressInterval);
+                progressInterval = null;
+
+                if (status.status === 'completed') {
+                    showAlert(`Benchmark completed! Final accuracy: ${status.overall_accuracy.toFixed(1)}%`, 'success');
+                } else {
+                    showAlert(`Benchmark ${status.status}`, 'error');
+                }
+
+                setTimeout(() => {
+                    hideProgress();
+                    currentBenchmarkId = null;
+                }, 3000);
+            }
+        }
+    } catch (error) {
+        console.error('Error updating progress:', error);
+    }
+}
+
+async function cancelBenchmark() {
+    if (!currentBenchmarkId) return;
+
+    try {
+        const csrfToken = document.querySelector('meta[name="csrf-token"]').getAttribute('content');
+        const response = await fetch(`/benchmark/api/cancel/${currentBenchmarkId}`, {
+            method: 'POST',
+            headers: {
+                'X-CSRFToken': csrfToken
+            }
+        });
+
+        const data = await response.json();
+        if (data.success) {
+            showAlert('Benchmark cancelled', 'info');
+        }
+    } catch (error) {
+        console.error('Error cancelling benchmark:', error);
+    }
+}
+
+function showAlert(message, type) {
+    const alertContainer = document.getElementById('benchmark-alert');
+    alertContainer.innerHTML = `
+        <div class="settings-alert alert-${type}">
+            <span>${message}</span>
+            <button type="button" class="close-alert" onclick="this.parentElement.parentElement.style.display='none'">
+                <i class="fas fa-times"></i>
+            </button>
+        </div>
+    `;
+    alertContainer.style.display = 'block';
+
+    if (type === 'success') {
+        setTimeout(() => {
+            alertContainer.style.display = 'none';
+        }, 5000);
+    }
+}
+</script>
+{% endblock %}

local_deep_research/web/templates/pages/cost_analytics.html

@@ -1094,7 +1094,7 @@
             div.innerHTML = `
                 <div class="cost-item-info">
                     <div class="cost-item-name">
-                        <a href="/research/results/${item.research_id}" class="research-link">
+                        <a href="/results/${item.research_id}" class="research-link">
                             Research Session #${item.research_id}
                         </a>
                     </div>