launch leaderboard

.gitignore

@@ -0,0 +1 @@
+__pycache__

Dockerfile

@@ -0,0 +1,36 @@
+# --- Base image ---
+FROM python:3.11-slim
+
+ENV PYTHONDONTWRITEBYTECODE=1 \
+    PYTHONUNBUFFERED=1 \
+    PIP_NO_CACHE_DIR=on
+
+# OS deps (git useful for HF repos; curl for HEALTHCHECK)
+RUN apt-get update && apt-get install -y --no-install-recommends \
+    git curl && \
+    rm -rf /var/lib/apt/lists/*
+
+# Non-root user
+RUN useradd -m -u 1000 user
+USER user
+ENV PATH="/home/user/.local/bin:$PATH"
+
+WORKDIR /app
+
+# Python deps
+COPY --chown=user ./requirements.txt requirements.txt
+RUN pip install --upgrade pip && pip install -r requirements.txt
+
+# App source
+COPY --chown=user . /app
+
+EXPOSE 7860
+
+# Healthcheck pings your FastAPI health endpoint
+HEALTHCHECK --interval=30s --timeout=3s --start-period=20s --retries=3 \
+  CMD curl -fsS http://localhost:7860/api/health || exit 1
+
+# Production server (Gunicorn + Uvicorn worker)
+#CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "7860"]
+
+CMD ["gunicorn", "-k", "uvicorn.workers.UvicornWorker", "-w", "2", "-b", "0.0.0.0:7860", "server:app"]

benchmark.json

@@ -0,0 +1,202 @@
+[
+  {
+    "provider": "OpenAI",
+    "name": "GPT-OSS-120B",
+    "repo": "openai/gpt-oss-120b",
+    "updated_at": "2025-09-16T00:00:00Z",
+    "scores": [
+      {"dataset_name": "TeleQna", "metric_type": "raw", "score": 78.51, "energy_consumed": "", "co2_consumed": ""},
+      {"dataset_name": "TeleMath", "metric_type": "raw", "score": 60.40, "energy_consumed": "", "co2_consumed": ""},
+      {"dataset_name": "TeleLogs", "metric_type": "raw", "score": 44.70, "energy_consumed": "", "co2_consumed": ""}
+    ]
+  },
+  {
+    "provider": "Qwen",
+    "name": "Qwen3-32B",
+    "repo": "qwen/qwen3-32b",
+    "updated_at": "2025-09-16T00:00:00Z",
+    "scores": [
+      {"dataset_name": "TeleMath", "metric_type": "raw", "score": 69.51, "energy_consumed": "", "co2_consumed": ""},
+      {"dataset_name": "TeleLogs", "metric_type": "raw", "score": 33.77, "energy_consumed": "", "co2_consumed": ""}
+    ]
+  },
+  {
+    "provider": "OpenAI",
+    "name": "GPT-OSS-20B",
+    "repo": "openai/gpt-oss-20b",
+    "updated_at": "2025-09-16T00:00:00Z",
+    "scores": [
+      {"dataset_name": "TeleQna", "metric_type": "raw", "score": 75.79, "energy_consumed": "", "co2_consumed": ""},
+      {"dataset_name": "TeleMath", "metric_type": "raw", "score": 53.80, "energy_consumed": "", "co2_consumed": ""},
+      {"dataset_name": "TeleLogs", "metric_type": "raw", "score": 40.10, "energy_consumed": "", "co2_consumed": ""}
+    ]
+  },
+  {
+    "provider": "DeepSeek",
+    "name": "R1-Distill-Llama-70B",
+    "repo": "deepseek/r1-distill-llama-70b",
+    "updated_at": "2025-09-16T00:00:00Z",
+    "scores": [
+      {"dataset_name": "TeleMath", "metric_type": "raw", "score": 53.21, "energy_consumed": "", "co2_consumed": ""},
+      {"dataset_name": "TeleLogs", "metric_type": "raw", "score": 29.42, "energy_consumed": "", "co2_consumed": ""}
+    ]
+  },
+  {
+    "provider": "ByteDance",
+    "name": "Seed-OSS-36B",
+    "repo": "bytedance/seed-oss-36b",
+    "updated_at": "2025-09-16T00:00:00Z",
+    "scores": [
+      {"dataset_name": "TeleQna", "metric_type": "raw", "score": 75.67, "energy_consumed": "", "co2_consumed": ""},
+      {"dataset_name": "TeleMath", "metric_type": "raw", "score": 56.05, "energy_consumed": "", "co2_consumed": ""},
+      {"dataset_name": "TeleLogs", "metric_type": "raw", "score": 57.00, "energy_consumed": "", "co2_consumed": ""}
+    ]
+  },
+  {
+    "provider": "Qwen",
+    "name": "QwQ-32B",
+    "repo": "qwen/qwq-32b",
+    "updated_at": "2025-09-16T00:00:00Z",
+    "scores": [
+      {"dataset_name": "TeleLogs", "metric_type": "raw", "score": 33.62, "energy_consumed": "", "co2_consumed": ""}
+    ]
+  },
+  {
+    "provider": "Qwen",
+    "name": "Qwen2.5-72B",
+    "repo": "qwen/qwen2.5-72b",
+    "updated_at": "2025-09-16T00:00:00Z",
+    "scores": [
+      {"dataset_name": "TeleQna", "metric_type": "raw", "score": 76.50, "energy_consumed": "", "co2_consumed": ""}
+    ]
+  },
+  {
+    "provider": "Meta",
+    "name": "Llama-3.3-70B-Instruct",
+    "repo": "meta/llama-3.3-70b-instruct",
+    "updated_at": "2025-09-16T00:00:00Z",
+    "scores": [
+      {"dataset_name": "TeleQna", "metric_type": "raw", "score": 74.98, "energy_consumed": "", "co2_consumed": ""},
+      {"dataset_name": "TeleMath", "metric_type": "raw", "score": 36.23, "energy_consumed": "", "co2_consumed": ""}
+    ]
+  },
+  {
+    "provider": "OpenAI",
+    "name": "GPT-4",
+    "repo": "openai/gpt-4",
+    "updated_at": "2025-09-16T00:00:00Z",
+    "scores": [
+      {"dataset_name": "TeleQna", "metric_type": "raw", "score": 74.91, "energy_consumed": "", "co2_consumed": ""}
+    ]
+  },
+  {
+    "provider": "Qwen",
+    "name": "Qwen3-8B",
+    "repo": "qwen/qwen3-8b",
+    "updated_at": "2025-09-16T00:00:00Z",
+    "scores": [
+      {"dataset_name": "TeleQna", "metric_type": "raw", "score": 73.21, "energy_consumed": "", "co2_consumed": ""}
+    ]
+  },
+  {
+    "provider": "Qwen",
+    "name": "Qwen2.5-7B-Instruct",
+    "repo": "qwen/qwen2.5-7b-instruct",
+    "updated_at": "2025-09-16T00:00:00Z",
+    "scores": [
+      {"dataset_name": "TeleQna", "metric_type": "raw", "score": 69.31, "energy_consumed": "", "co2_consumed": ""},
+      {"dataset_name": "TeleLogs", "metric_type": "raw", "score": 12.05, "energy_consumed": "", "co2_consumed": ""}
+    ]
+  },
+  {
+    "provider": "Meta",
+    "name": "Llama-3.1-8B-Instruct",
+    "repo": "meta/llama-3.1-8b-instruct",
+    "updated_at": "2025-09-16T00:00:00Z",
+    "scores": [
+      {"dataset_name": "TeleQna", "metric_type": "raw", "score": 68.03, "energy_consumed": "", "co2_consumed": ""},
+      {"dataset_name": "TeleMath", "metric_type": "raw", "score": 13.56, "energy_consumed": "", "co2_consumed": ""}
+    ]
+  },
+  {
+    "provider": "OpenAI",
+    "name": "GPT-3.5",
+    "repo": "openai/gpt-3.5",
+    "updated_at": "2025-09-16T00:00:00Z",
+    "scores": [
+      {"dataset_name": "TeleQna", "metric_type": "raw", "score": 67.29, "energy_consumed": "", "co2_consumed": ""}
+    ]
+  },
+  {
+    "provider": "Mistral",
+    "name": "Mixtral",
+    "repo": "mistral/mixtral",
+    "updated_at": "2025-09-16T00:00:00Z",
+    "scores": [
+      {"dataset_name": "TeleQna", "metric_type": "raw", "score": 67.74, "energy_consumed": "", "co2_consumed": ""}
+    ]
+  },
+  {
+    "provider": "Mistral",
+    "name": "Mistral-7B",
+    "repo": "mistral/mistral-7b",
+    "updated_at": "2025-09-16T00:00:00Z",
+    "scores": [
+      {"dataset_name": "TeleQna", "metric_type": "raw", "score": 47.07, "energy_consumed": "", "co2_consumed": ""}
+    ]
+  },
+  {
+    "provider": "Microsoft",
+    "name": "Phi-4-Reasoning+",
+    "repo": "microsoft/phi-4-reasoning+",
+    "updated_at": "2025-09-16T00:00:00Z",
+    "scores": [
+      {"dataset_name": "TeleMath", "metric_type": "raw", "score": 53.56, "energy_consumed": "", "co2_consumed": ""}
+    ]
+  },
+  {
+    "provider": "Qwen",
+    "name": "Qwen3-4B-Instruct",
+    "repo": "qwen/qwen3-4b-instruct",
+    "updated_at": "2025-09-16T00:00:00Z",
+    "scores": [
+      {"dataset_name": "TeleMath", "metric_type": "raw", "score": 45.62, "energy_consumed": "", "co2_consumed": ""}
+    ]
+  },
+  {
+    "provider": "Qwen",
+    "name": "Qwen2.5-Math-72B-Instruct",
+    "repo": "qwen/qwen2.5-math-72b-instruct",
+    "updated_at": "2025-09-16T00:00:00Z",
+    "scores": [
+      {"dataset_name": "TeleMath", "metric_type": "raw", "score": 39.99, "energy_consumed": "", "co2_consumed": ""}
+    ]
+  },
+  {
+    "provider": "Qwen",
+    "name": "Qwen2.5-Math-7B-Instruct",
+    "repo": "qwen/qwen2.5-math-7b-instruct",
+    "updated_at": "2025-09-16T00:00:00Z",
+    "scores": [
+      {"dataset_name": "TeleMath", "metric_type": "raw", "score": 22.38, "energy_consumed": "", "co2_consumed": ""}
+    ]
+  },
+  {
+    "provider": "Qwen",
+    "name": "Qwen2.5-32B-Instruct",
+    "repo": "qwen/qwen2.5-32b-instruct",
+    "updated_at": "2025-09-16T00:00:00Z",
+    "scores": [
+      {"dataset_name": "TeleLogs", "metric_type": "raw", "score": 18.85, "energy_consumed": "", "co2_consumed": ""}
+    ]
+  },
+  {
+    "provider": "Qwen",
+    "name": "Qwen2.5-1.5B-Instruct",
+    "repo": "qwen/qwen2.5-1.5b-instruct",
+    "updated_at": "2025-09-16T00:00:00Z",
+    "scores": [
+      {"dataset_name": "TeleLogs", "metric_type": "raw", "score": 11.25, "energy_consumed": "", "co2_consumed": ""}
+    ]
+  }
+]

index.html

@@ -0,0 +1,598 @@
+<!DOCTYPE html>
+<html lang="en">
+<head>
+  <meta charset="UTF-8" />
+  <meta name="viewport" content="width=device-width, initial-scale=1.0"/>
+  <title>GSMA Open-Telco LLM Benchmarks</title>
+
+  <!-- Tailwind + Chart.js -->
+  <script src="https://cdn.tailwindcss.com"></script>
+  <script src="https://cdn.jsdelivr.net/npm/chart.js"></script>
+
+  <!-- Fonts -->
+  <link rel="preconnect" href="https://fonts.googleapis.com">
+  <link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
+  <link href="https://fonts.googleapis.com/css2?family=Inter:wght@300;400;500;600;700;800&display=swap" rel="stylesheet">
+
+  <style>
+    :root{
+      color-scheme: light dark;
+      --text: #0f172a;
+      --bg: #f8fafc;
+      --card-bg: rgba(255,255,255,.75);
+      --border: #e2e8f0;
+      --header-bg: rgba(241,245,249,.85);
+      --sticky-bg: rgba(248,250,252,.92);
+      --chip-bg:#f1f5f9;
+    }
+    .dark{
+      --text: #e2e8f0;
+      --bg: #020617;
+      --card-bg: rgba(2,6,23,.6);
+      --border:#334155;
+      --header-bg: rgba(30,41,59,.75);
+      --sticky-bg: rgba(2,6,23,.92);
+      --chip-bg: rgba(15,23,42,.6);
+    }
+    html { font-family: Inter, system-ui, -apple-system, Segoe UI, Roboto, "Helvetica Neue", Arial, "Noto Sans"; }
+    body { color: var(--text); background: var(--bg); }
+
+    .card{ border-radius:1rem; box-shadow:0 10px 25px rgba(2,6,23,.08); background:var(--card-bg); backdrop-filter: blur(8px); border:1px solid var(--border); }
+    .btn{ display:inline-flex; align-items:center; justify-content:center; gap:.5rem; border-radius:.8rem; padding:.625rem 1rem; font-weight:700; transition: transform .05s ease; }
+    .btn:active{ transform: scale(.98); }
+    .btn-primary{ background:#4f46e5; color:#fff; } .btn-primary:hover{ background:#6366f1; }
+    .btn-ghost{ background:transparent; border:1px solid var(--border); } .btn-ghost:hover{ background:var(--chip-bg); }
+    .btn-outline{ border:1px solid #4f46e5; color:#4338ca; } .dark .btn-outline{ color:#a5b4fc; }
+    .btn-outline:hover{ background:#eef2ff; } .dark .btn-outline:hover{ background: rgba(30,27,75,.5); }
+    .input{ width:100%; border:1px solid var(--border); border-radius:.8rem; padding:.6rem .8rem; background:#fff; color:#0f172a; }
+    .dark .input{ background:#0b1220; color:#e2e8f0; }
+    .input:focus{ outline:none; box-shadow:0 0 0 2px rgba(99,102,241,.6); }
+    .label{ font-size:.875rem; font-weight:600; color:#334155; } .dark .label{ color:#cbd5e1; }
+    .tab{ padding:.6rem 1rem; border-radius:.8rem; cursor:pointer; font-weight:700; }
+    .tab-active{ background:#fff; border:1px solid var(--border); box-shadow:0 1px 2px rgba(0,0,0,.04); } .dark .tab-active{ background:#0b1220; }
+    .pill{ display:inline-flex; align-items:center; padding:.1rem .5rem; border-radius:999px; font-size:.53rem; font-weight:700; }
+    .metric-badge{ background:#ecfdf5; color:#065f46; } .dark .metric-badge{ background:rgba(16,185,129,.18); color:#d1fae5; }
+    .metric-badge-judge{ background:#e0f2fe; color:#075985; } .dark .metric-badge-judge{ background:rgba(56,189,248,.18); color:#bae6fd; }
+    .kpi{ color:#475569; font-size:.875rem; } .dark .kpi{ color:#94a3b8; }
+
+    /* Column widths */
+    /* .rank-col{ width:1.5rem; }
+    .provider-col{ width:4rem;  }
+    .model-col{ width:10rem; white-space:nowrap; }
+    .mean-col{ width:3rem; white-space:nowrap; }
+    .ds-col{ min-width:9.5rem; white-space:nowrap; } */
+
+    .rank-col{ }
+    .provider-col{   }
+    .model-col{ white-space:nowrap; }
+    .mean-col{ white-space:nowrap;  }
+    .ds-col{ white-space:nowrap; }
+
+    /* Sticky columns: use calc so offsets update easily */
+    .sticky-rank{ position: sticky; left: 0; z-index: 3; background: var(--sticky-bg); }
+    .sticky-provider{ position: sticky; left: 2.3rem; z-index: 3; background: var(--sticky-bg); }
+    .sticky-model{ position: sticky; left: calc(1.85rem + 6rem); z-index: 3; background: var(--sticky-bg); }
+    .sticky-mean{ position: sticky; left: calc(2.3rem + 4rem + 10rem + 0.5rem); z-index: 3; background: var(--sticky-bg); }
+
+    /* Mobile: hide rank & dataset columns; update sticky offsets */
+    @media (max-width: 768px){
+      .rank-col, .ds-col, .ds-head { display:none !important; }
+      .sticky-provider{ left: 0; }
+      .sticky-model{ left: 14rem; }
+      .sticky-mean{ left: calc(14rem + 18rem); }
+    }
+
+    /* Header background */
+    thead th{ background: var(--header-bg); }
+
+    /* Global gradient */
+    .gradient-bg{
+      background:
+        radial-gradient(1200px 600px at 20% -10%, rgba(99,102,241,.25), rgba(99,102,241,0) 60%),
+        radial-gradient(1200px 600px at 80% -10%, rgba(20,184,166,.2), rgba(20,184,166,0) 60%);
+    }
+
+    /* Scrollbar for horizontal overflow */
+    .scrollbar-thin::-webkit-scrollbar{ height:10px; }
+    .scrollbar-thin::-webkit-scrollbar-thumb{ background:#c7d2fe; border-radius:999px; }
+    .scrollbar-thin::-webkit-scrollbar-track{ background:transparent; }
+
+    /* Chart container: responsive height */
+    #chartWrap{ height: clamp(260px, 42vh, 460px); }
+  </style>
+</head>
+<body class="min-h-screen dark:bg-gray-800">
+  <!-- Header -->
+  <header class="sticky top-0 z-40 backdrop-blur dark:bg-white-800 border-b" style="border-color: var(--border)">
+    <div class="mx-auto max-w-7xl px-4 md:px-6 py-3 md:py-4 flex items-center justify-between">
+      <div class="flex items-center gap-3 md:gap-4">
+        <div class="h-10 w-10 rounded-xl bg-indigo-600 text-white grid place-items-center shadow-lg">📡</div>
+        <div>
+          <h1 class="text-lg md:text-xl font-extrabold tracking-tight" style="color: var(--text)">GSMA Open-Telco LLM Benchmarks</h1>
+          <p class="text-xs md:text-sm">Benchmarking models across telecom datasets</p>
+        </div>
+      </div>
+      <div class="md:flex items-center gap-2 hidden">
+        <button id="refreshBtn" class="btn btn-ghost" title="Refresh results">⟲ Refresh</button>
+        <!--button id="exportCsvBtn" class="btn btn-outline" title="Export current view to CSV">⭳ Export CSV</button-->
+        <button id="themeToggle" class="btn btn-ghost" title="Toggle dark mode">🌙</button>
+      </div>
+    </div>
+  </header>
+
+  <!-- Main -->
+  <main class="mx-auto max-w-7xl px-4 md:px-6 py-6 md:py-8 space-y-6 md:space-y-8">
+    <!-- Tabs -->
+    <div class="flex gap-2 md:gap-3">
+      <button class="tab tab-active" data-tab="leaderboard">🏆 Leaderboard</button>
+      <button class="tab" data-tab="submit">📤 Submit Model</button>
+      <button class="tab" data-tab="about">ℹ️ About</button>
+    </div>
+
+    <!-- Leaderboard Tab -->
+    <section id="tab-leaderboard" class="space-y-6 md:space-y-8">
+      <!-- Controls -->
+      <div class="grid lg:grid-cols-4 gap-4 md:gap-6">
+        <div class="card p-4 md:p-6 lg:col-span-2">
+          <div class="flex items-center justify-between">
+            <h2 class="font-semibold text-base md:text-lg" style="color: var(--text)">Select Datasets</h2>
+            <div class="flex gap-2">
+              <button id="selectAllBtn" class="btn btn-ghost text-xs md:text-sm px-3">Select all</button>
+              <button id="clearAllBtn" class="btn btn-ghost text-xs md:text-sm px-3">Clear</button>
+            </div>
+          </div>
+          <div id="datasetFilters" class="mt-3 md:mt-4 flex flex-wrap  gap-2.5"></div> <!--grid-cols-1 sm:grid-cols-2 lg:grid-cols-3-->
+        </div>
+
+        <div class="card p-4 md:p-6">
+          <h2 class="font-semibold text-base md:text-lg" style="color: var(--text)">Search & Filter</h2>
+          <div class="mt-2 md:mt-3 space-y-2.5">
+            <input id="searchInput" class="input" placeholder="Search provider/model…"/>
+            <select id="providerSelect" class="input">
+              <option value="">All providers</option>
+            </select>
+          </div>
+        </div>
+
+        <div class="card p-4 md:p-6">
+          <h2 class="font-semibold text-base md:text-lg" style="color: var(--text)">KPI</h2>
+          <div class="mt-2 grid grid-cols-2 gap-3">
+            <div>
+              <div class="text-2xl md:text-3xl font-extrabold" id="kpiModels">—</div>
+              <div class="kpi">Models</div>
+            </div>
+            <div>
+              <div class="text-2xl md:text-3xl font-extrabold" id="kpiDatasets">—</div>
+              <div class="kpi">Datasets</div>
+            </div>
+            <div class="col-span-2 text-xs text-slate-500 dark:text-slate-400" id="lastUpdated">Last updated —</div>
+          </div>
+        </div>
+      </div>
+
+      <!-- Visualization -->
+      <div class="card p-5 md:p-6 lg:p-8">
+        <div class="flex items-center justify-between gap-3">
+          <h2 class="font-semibold text-base md:text-lg" style="color: var(--text)">Model Comparison</h2>
+          <div class="text-xs md:text-sm text-slate-500">Click ⭐ on rows to compare (max 3)</div>
+        </div>
+        <div id="chartWrap" class="mt-3 md:mt-4">
+          <canvas id="scoresChart"></canvas>
+        </div>
+      </div>
+
+      <!-- Table (always horizontally scrollable) -->
+      <div class="card overflow-hidden">
+        <div class="px-4 md:px-6 py-3 md:py-4 flex items-center justify-between">
+          <div class="font-semibold" style="color: var(--text)">Leaderboard</div>
+          <div class="text-xs md:text-sm text-slate-500">Sorted by mean score across selected datasets</div>
+        </div>
+
+        <div class="overflow-x-auto overscroll-x-contain scrollbar-thin pb-2">
+          <table class="min-w-[64rem] w-full text-sm">
+            <thead class="border-t border-b" style="border-color: var(--border)">
+              <tr id="tableHeaderRow">
+                <th class="p-3 text-left rank-col sticky-rank">#</th>
+                <th class="p-3 text-left provider-col sticky-provider border-l">Provider</th>
+                <th class="p-3 text-left model-col sticky-model border-l">Model</th>
+                <th class="p-3 text-left mean-col sticky-mean border-l">
+                    Mean
+                    <div class="text-[11px] text-slate-500">on selected</div>
+                </th>
+                <!-- Dataset columns injected here -->
+              </tr>
+            </thead>
+            <tbody id="tableBody"></tbody>
+          </table>
+        </div>
+
+        <div class="flex items-center gap-2 justify-end p-3">
+          <button id="refreshBtn" class="btn btn-ghost md:hidden" title="Refresh results">⟲ Refresh</button>
+          <button id="exportCsvBtn" class="btn btn-outline" title="Export current view to CSV">⭳ Export CSV</button>
+          <button id="themeToggle" class="btn btn-ghost md:hidden" title="Toggle dark mode">🌙</button>
+        </div>
+      </div>
+    </section>
+
+    <!-- Submit Tab -->
+    <section id="tab-submit" class="hidden">
+      <div class="card p-5 md:p-6 lg:p-8">
+        <h2 class="text-lg md:text-xl font-bold" style="color: var(--text)">Submit a Model for Evaluation</h2>
+        <p class="text-sm text-slate-600 dark:text-slate-300 mt-1">
+          Provide your model details. Submissions are queued (status: <span class="font-semibold">pending</span>) and evaluated automatically. Results will appear on the leaderboard when ready.
+        </p>
+
+        <form id="submitForm" class="mt-4 grid md:grid-cols-2 gap-4 md:gap-6">
+          <div>
+            <label class="label" for="modelProvider">Model Provider</label>
+            <input class="input" id="modelProvider" name="model_provider" required placeholder="e.g., TelcoAI Labs"/>
+          </div>
+          <div>
+            <label class="label" for="modelName">Model Name</label>
+            <input class="input" id="modelName" name="model_name" required placeholder="e.g., T-LLM-7B"/>
+          </div>
+          <div>
+            <label class="label" for="hfRepo">Hugging Face Repo</label>
+            <input class="input" id="hfRepo" name="hf_repo" required placeholder="e.g., telcoai/t-llm-7b"/>
+          </div>
+          <div>
+            <label class="label" for="contactEmail">Contact Email</label>
+            <input class="input" id="contactEmail" name="contact_email" required type="email" placeholder="[email protected]"/>
+          </div>
+          <div class="md:col-span-2">
+            <label class="label" for="notes">Notes (optional)</label>
+            <textarea class="input" id="notes" name="notes" rows="3" placeholder="Anything we should know about your model or expected behavior"></textarea>
+          </div>
+          <div class="md:col-span-2 flex items-center gap-2">
+            <input id="agree" type="checkbox" required class="h-4 w-4 accent-indigo-600"/>
+            <label for="agree" class="text-sm">I agree to have my model evaluated and results published.</label>
+          </div>
+          <div class="md:col-span-2 flex items-center gap-3">
+            <button class="btn btn-primary" type="submit">Submit to Queue</button>
+            <button class="btn btn-ghost" type="reset">Reset</button>
+            <span id="submitStatus" class="text-sm"></span>
+          </div>
+        </form>
+      </div>
+    </section>
+
+    <!-- About Tab -->
+    <section id="tab-about" class="hidden">
+      <div class="card p-5 md:p-6 lg:p-8 space-y-3">
+        <h2 class="text-lg md:text-xl font-bold" style="color: var(--text)">About this Leaderboard</h2>
+        <p class="text-sm text-slate-600 dark:text-slate-300">
+          This dashboard ranks LLMs on telecom-focused datasets. Each cell shows the score and the metric type used
+          (<span class="pill metric-badge">raw</span> or <span class="pill metric-badge-judge">llm-as-judge</span>). Energy & CO₂ appear on hover.
+        </p>
+      </div>
+    </section>
+  </main>
+
+  <div id="toast" class="fixed bottom-4 left-1/2 -translate-x-1/2 hidden">
+    <div class="rounded-xl bg-slate-900 text-white px-4 py-2 shadow-lg">✅ <span id="toastMsg">Done</span></div>
+  </div>
+
+  <script>
+    // ===== Config =====
+    const API_BASE = "";
+    const USE_MOCK_FALLBACK = false;
+    const MAX_COMPARE = 3;
+
+    // ===== State =====
+    const state = {
+      datasets: [],
+      models: [],
+      selectedDatasets: new Set(),
+      providerFilter: "",
+      search: "",
+      compare: new Set(),
+      dark: window.matchMedia && window.matchMedia('(prefers-color-scheme: dark)').matches,
+    };
+
+    // ===== Utils =====
+    const $ = (s)=>document.querySelector(s);
+    const $$ = (s)=>Array.from(document.querySelectorAll(s));
+    const showToast=(m)=>{const t=$("#toast");$("#toastMsg").textContent=m;t.classList.remove("hidden");setTimeout(()=>t.classList.add("hidden"),2200);};
+    const fmt=(n,d=2)=>(n==null||Number.isNaN(n))?"—":Number(n).toFixed(d);
+    const slug=(s)=>s.toLowerCase().replace(/[^a-z0-9]+/g,'-');
+
+    function savePrefs(){
+      localStorage.setItem('llm_lb_prefs', JSON.stringify({
+        selectedDatasets:[...state.selectedDatasets],
+        datasets:[...state.datasets],
+        providerFilter:state.providerFilter,
+        search:state.search, dark:state.dark,
+        compare:[...state.compare],
+      }));
+    }
+    function loadPrefs(){
+      try{
+        const p=JSON.parse(localStorage.getItem('llm_lb_prefs')||'{}');
+        if(p.selectedDatasets) state.selectedDatasets=new Set(p.selectedDatasets);
+        if(p.providerFilter) state.providerFilter=p.providerFilter;
+        if(p.search) state.search=p.search;
+        if(typeof p.dark==='boolean') state.dark=p.dark;
+        if(p.compare) state.compare=new Set(p.compare);
+      }catch{}
+    }
+
+    // ===== Mock =====
+    function mockDatasets(){ return [
+      "3GPP-TSG", "NetBench", "TeleQna", "TeleLogs", "TeleMath",
+    ]; }
+    function mockResults(){
+      const providers=["Qwen","Qwen","OpenAI","OpenAI","DeepSeek","ByteDance", "LLama"];
+      const models=["Qwen3-32B","QwQ-32B","GPT-OSS-120B","GPT-OSS-20B","R1-Distill-Llama-70B","Seed-OSS-36B", "Llama-8B"];
+      const repos=["qwen/qwen3-32b","qwen/qwq-32b","openai/gpt-oss-120b","openai/gpt-oss-20b","deepseek/r1-distill-llama-70b","bytedance/seed-oss-36b", "llama/llama-8B-instruct"];
+      const ds=mockDatasets(); const now=new Date().toISOString();
+      const mtypes=["raw","llm-as-judge"];
+      const rnd=(a)=>a[Math.floor(Math.random()*a.length)];
+      const rScore=()=>Math.round((45+Math.random()*50)*100)/100;
+      return {models: providers.map((p,i)=>({
+        provider:p, name:models[i], repo:repos[i], updated_at:now,
+        scores: ds.map(d=>({dataset_name:d, metric_type:rnd(mtypes), score:rScore(),
+          energy_consumed:+(0.1+Math.random()*2.4).toFixed(3),
+          co2_consumed:+(0.05+Math.random()*0.95).toFixed(3)}))
+      }))};
+    }
+
+    // ===== API =====
+    async function apiGet(path){
+      try{ const r=await fetch(`${API_BASE}${path}`); if(!r.ok) throw new Error(r.status); return await r.json(); }
+      catch(e){ if(USE_MOCK_FALLBACK){ if(path==='/api/datasets') return {datasets:mockDatasets()}; if(path==='/api/results') return mockResults(); } throw e; }
+    }
+    async function apiPost(path, body){
+      try{ const r=await fetch(`${API_BASE}${path}`,{method:'POST', headers:{'Content-Type':'application/json'}, body:JSON.stringify(body)}); if(!r.ok) throw new Error(r.status); return await r.json(); }
+      catch(e){ if(USE_MOCK_FALLBACK) return {status:'pending', id:`mock-${Date.now()}`}; throw e; }
+    }
+
+    // ===== Rendering =====
+    function visibleDatasets(){return state.selectedDatasets.size ? state.datasets.filter(d=>state.selectedDatasets.has(d)) : []; }
+    function modelRowMean(model, dsList){
+      const map=Object.fromEntries(model.scores.map(s=>[s.dataset_name,s]));
+      const vals=dsList.map(d=>map[d]?.score).filter(v=>typeof v==='number');
+      return vals.length? vals.reduce((a,b)=>a+b,0)/vals.length : null;
+    }
+
+    function renderDatasetFilters(){
+      const c=$("#datasetFilters"); c.innerHTML='';
+      state.datasets.forEach(d=>{
+        const checked=state.selectedDatasets.has(d);
+        c.insertAdjacentHTML('beforeend', `
+          <label class="flex flex-nowrap items-center gap-2 rounded-lg border" style="border-color: var(--border); padding:.5rem .6rem;">
+            <input type="checkbox" class="h-4 w-4 accent-indigo-600" data-dataset="${d}" ${checked?'checked':''}/>
+            <span class="text-sm" style="color: var(--text)">${d}</span>
+          </label>
+        `);
+      });
+      c.querySelectorAll('input[type="checkbox"]').forEach(cb=>{
+        cb.addEventListener('change', ()=>{
+          if(cb.checked) state.selectedDatasets.add(cb.dataset.dataset);
+          else state.selectedDatasets.delete(cb.dataset.dataset);
+          if([...c.querySelectorAll('input:checked')].length===0) state.selectedDatasets=new Set();
+          savePrefs(); renderTable(); updateChart();
+        });
+      });
+      $("#kpiDatasets").textContent = state.datasets.length;
+    }
+
+    function renderProviders(){
+      const provs=[...new Set(state.models.map(m=>m.provider))].sort();
+      const sel=$("#providerSelect");
+      sel.innerHTML='<option value="">All providers</option>'+provs.map(p=>`<option>${p}</option>`).join('');
+      if(state.providerFilter) sel.value=state.providerFilter;
+    }
+
+    function renderHeader(){
+      const head=$("#tableHeaderRow");
+      head.querySelectorAll('th[data-ds]').forEach(el=>el.remove());
+      visibleDatasets().forEach(d=>{
+        const th=document.createElement('th');
+        th.dataset.ds=d; th.className='p-3 text-left ds-col ds-head';
+        th.innerHTML = `<div class="font-semibold" style="color: var(--text)">${d}</div><div class="text-[11px] text-slate-500">score • metric</div>`;
+        head.appendChild(th);
+      });
+    }
+
+    function renderTable(){
+      renderHeader();
+      const ds=visibleDatasets();
+      const tb=$("#tableBody"); tb.innerHTML='';
+
+      let models=state.models.filter(m=>{
+        const text=(m.provider+" "+m.name+" "+m.repo).toLowerCase();
+        const okProvider=!state.providerFilter || m.provider===state.providerFilter;
+        const okSearch=!state.search || text.includes(state.search.toLowerCase());
+        return okProvider && okSearch;
+      }).map(m=>({...m, mean:modelRowMean(m, ds)}))
+        .sort((a,b)=>(b.mean??-1)-(a.mean??-1));
+
+      models.forEach((m, i)=>{
+        const sMap=Object.fromEntries(m.scores.map(s=>[s.dataset_name, s]));
+        const id=slug(m.provider+"-"+m.name);
+        const favOn=state.compare.has(id);
+
+        const row=document.createElement('tr');
+        row.className='border-b'; row.style.borderColor=getComputedStyle(document.documentElement).getPropertyValue('--border');
+        row.innerHTML = `
+          <td class="p-3 rank-col sticky-rank">${i+1}</td>
+          <td class="p-3 provider-col sticky-provider border-l">
+            <div class="font-medium">${m.provider}</div>
+          </td>
+          <td class="p-3 model-col sticky-model border-l">
+            <div class="flex items-center gap-2">
+              <button class="text-lg" data-fav="${id}" title="Add to compare">${favOn?'⭐':'☆'}</button>
+              <div class="font-semibold" style="color: var(--text)">${m.name}</div>
+            </div>
+            <div class="text-xs text-slate-500">${m.repo}</div>
+          </td>
+          <td class="p-3 mean-col sticky-mean border-l">
+            <div class="text-base font-semibold text-blue-500">${m.mean==null?'—':fmt(m.mean,2)}</div>
+            <!--div class="text-[11px] text-slate-500">mean across selected</div-->
+          </td>
+        `;
+        ds.forEach(d=>{
+          const s=sMap[d]; const mt=s?.metric_type;
+          // const badge= mt==='llm-as-judge' ? 'metric-badge-judge' : 'metric-badge';
+          const title= s ? `Energy: ${fmt(s.energy_consumed,3)} kWh\nCO₂: ${fmt(s.co2_consumed,3)} kg` : '';
+          const cell=document.createElement('td');
+          cell.className='p-3 ds-col'; cell.title=title;
+          cell.innerHTML = s ? (
+            mt==='llm-as-judge' ? `
+                <div class="flex item-center">
+                    <div class="text-base font-semibold" style="color: var(--text)">${fmt(s.score,2)}</div>
+                    <div class="pill metric-badge-judge ms-2">${mt}</div>
+                </div>
+            ` : 
+            `
+                <div class="flex item-center">
+                    <div class="text-base font-semibold" style="color: var(--text)">${fmt(s.score,2)}</div>
+                </div>
+            `
+          ) : '—';
+          row.appendChild(cell);
+        });
+        tb.appendChild(row);
+      });
+
+      $("#kpiModels").textContent=models.length;
+
+      $$('button[data-fav]').forEach(b=>{
+        b.onclick=()=>{
+          const id=b.dataset.fav;
+          if(state.compare.has(id)) state.compare.delete(id);
+          else{
+            if(state.compare.size>=MAX_COMPARE){ showToast(`You can compare up to ${MAX_COMPARE} models.`); return; }
+            state.compare.add(id);
+          }
+          savePrefs(); renderTable(); updateChart();
+        };
+      });
+    }
+
+    // ===== Chart =====
+    let chart;
+    function updateChart(){
+      const ds=visibleDatasets();
+      const labels=ds;
+      const datasets=[];
+      const chosen=state.models.filter(m=>state.compare.has(slug(m.provider+"-"+m.name)));
+      chosen.forEach(m=>{
+        const map=Object.fromEntries(m.scores.map(s=>[s.dataset_name, s.score]));
+        datasets.push({
+          label:`${m.provider} / ${m.name}`,
+          data: labels.map(d=>map[d] ?? null),
+          tension:.25, spanGaps:true, borderWidth:3, pointRadius:3, pointHoverRadius:5,
+        });
+      });
+
+      const ctx=document.getElementById('scoresChart').getContext('2d');
+      if(!chart){
+        chart=new Chart(ctx,{
+          type:'line',
+          data:{labels, datasets},
+          options:{
+            responsive:true,
+            maintainAspectRatio:false,  // uses #chartWrap height (clamp -> responsive)
+            plugins:{
+              legend:{ display:true, labels:{ boxWidth:18, usePointStyle:true }},
+              tooltip:{ mode:'index', intersect:false }
+            },
+            interaction:{ mode:'nearest', intersect:false },
+            scales:{
+              y:{ beginAtZero:true, max:100, title:{display:true, text:'Score'}, grid:{ drawBorder:false }},
+              x:{ grid:{ display:false } }
+            },
+            layout:{ padding:0 }
+          }
+        });
+      }else{
+        chart.data.labels=labels;
+        chart.data.datasets=datasets;
+        chart.update();
+      }
+    }
+
+    // ===== Tabs / Theme / Export =====
+    function bindTabs(){
+      $$(".tab").forEach(btn=>{
+        btn.addEventListener('click', ()=>{
+          const t=btn.dataset.tab;
+          $$(".tab").forEach(b=>b.classList.remove('tab-active')); btn.classList.add('tab-active');
+          ["leaderboard","submit","about"].forEach(x=>{
+            const el=document.getElementById(`tab-${x}`); (x===t)?el.classList.remove('hidden'):el.classList.add('hidden');
+          });
+        });
+      });
+    }
+    function applyTheme(){ 
+        document.documentElement.classList.toggle('dark', state.dark); 
+        $("#themeToggle").textContent= state.dark ? '☀️' : '🌙'; savePrefs(); 
+    }
+    
+    function exportCSV(){
+      const ds=visibleDatasets();
+      const headers=['Rank','Provider','Model','Repo','Mean',...ds];
+      const rows=[];
+      let models=state.models.map(m=>({...m, mean:modelRowMean(m, ds)})).sort((a,b)=>(b.mean??-1)-(a.mean??-1));
+      models=models.filter(m=>{
+        const text=(m.provider+" "+m.name+" "+m.repo).toLowerCase();
+        const okProvider=!state.providerFilter || m.provider===state.providerFilter;
+        const okSearch=!state.search || text.includes(state.search.toLowerCase());
+        return okProvider && okSearch;
+      });
+      models.forEach((m,i)=>{
+        const map=Object.fromEntries(m.scores.map(s=>[s.dataset_name, s]));
+        const row=[i+1, m.provider, m.name, m.repo, fmt(m.mean,2)];
+        ds.forEach(d=>{ const s=map[d]; row.push(s? `${fmt(s.score,2)} (${s.metric_type})`:''); });
+        rows.push(row);
+      });
+      const csv=[headers, ...rows].map(r=> r.map(x=>'"'+String(x).replaceAll('"','""')+'"').join(',')).join('\n');
+      const blob=new Blob([csv],{type:'text/csv;charset=utf-8;'}); const url=URL.createObjectURL(blob);
+      const a=document.createElement('a'); a.href=url; a.download='telecom-llm-leaderboard.csv'; a.click(); URL.revokeObjectURL(url);
+    }
+
+    // ===== Controls & Submit =====
+    function bindControls(){
+      $("#selectAllBtn").onclick=()=>{ state.selectedDatasets=new Set(state.datasets); renderDatasetFilters(); renderTable(); updateChart(); savePrefs(); };
+      $("#clearAllBtn").onclick=()=>{ state.selectedDatasets=new Set(); renderDatasetFilters(); renderTable(); updateChart(); savePrefs(); };
+      $("#searchInput").addEventListener('input', (e)=>{ state.search=e.target.value; savePrefs(); renderTable(); });
+      $("#providerSelect").addEventListener('change', (e)=>{ state.providerFilter=e.target.value; savePrefs(); renderTable(); });
+      $("#refreshBtn").onclick=init;
+      $("#exportCsvBtn").onclick=exportCSV;
+      $("#themeToggle").onclick=()=>{ state.dark=!state.dark; applyTheme(); };
+
+      $("#submitForm").addEventListener('submit', async (e)=>{
+        e.preventDefault();
+        const payload={
+          model_provider: $("#modelProvider").value.trim(),
+          model_name: $("#modelName").value.trim(),
+          hf_repo: $("#hfRepo").value.trim(),
+          contact_email: $("#contactEmail").value.trim(),
+          notes: $("#notes").value.trim(),
+        };
+        $("#submitStatus").textContent='Submitting…';
+        try{
+          const res=await apiPost('/api/submit', payload);
+          $("#submitStatus").innerHTML=`Status: <span class="font-semibold">${res.status}</span> (id: ${res.id})`;
+          showToast('Submission received — queued as pending'); e.target.reset();
+        }catch(err){ $("#submitStatus").textContent='Submission failed. See console.'; }
+      });
+    }
+
+    // ===== Init =====
+    async function init(){
+      loadPrefs(); applyTheme(); bindTabs(); bindControls();
+      try{ state.datasets=(await apiGet('/api/datasets')).datasets; } catch{ state.datasets=mockDatasets(); }
+      try{ state.models=(await apiGet('/api/results')).models; }catch{ state.models=mockResults().models; }
+      const last=state.models.map(m=>new Date(m.updated_at)).sort((a,b)=>b-a)[0];
+      if(last) $("#lastUpdated").textContent='Last updated '+last.toLocaleString();
+      if (!state.selectedDatasets.size) {
+        state.selectedDatasets = new Set (state.datasets);
+      }
+      renderDatasetFilters(); renderProviders(); renderTable(); updateChart();
+    }
+    init();
+  </script>
+</body>
+</html>

requirements.txt

@@ -0,0 +1,13 @@
+# FastAPI + server
+fastapi>=0.110
+uvicorn[standard]>=0.29
+gunicorn>=21.2
+
+# Data validation
+pydantic>=2.5
+
+# Hugging Face queue integration used by server.py
+huggingface_hub>=0.20
+
+# load .env if you add it later; otherwise safe to remove
+python-dotenv>=1.0

server.py

@@ -0,0 +1,299 @@
+# server.py
+# FastAPI backend for the Telecom LLM Leaderboard
+# ------------------------------------------------
+# Developer:
+#   Name: Mohamed SANA
+#   Email: [email protected]
+#
+# Quickstart:
+#   pip install fastapi uvicorn "pydantic>=2" huggingface_hub python-dotenv
+#   uvicorn server:app --reload --port 8000
+#
+# Configure (optional) environment variables for real HF queue uploads:
+#   HF_TOKEN=<your token>
+#   QUEUE_REPO=<org-or-user/submission-queue-dataset>
+#   RESULTS_REPO=<org-or-user/results-dataset>  # (not used in this fake generator)
+#   HF_HOME=<custom cache dir>  # optional
+#   USE_LOCAL_DATA=<set true to load data from local file>
+#   LOCAL_DATA_FILE=<the filename to load>
+#
+# The frontend expects:
+#   GET  /api/datasets
+#   GET  /api/results
+#   POST /api/submit
+
+from fastapi import FastAPI, Body, HTTPException
+from fastapi.middleware.cors import CORSMiddleware
+from fastapi.responses import FileResponse
+from pydantic import BaseModel, EmailStr, Field
+from datetime import datetime
+from typing import List, Literal, Optional, Dict, Any
+import os, json, random
+
+# Optional: Hugging Face Hub for queue upload
+HF_AVAILABLE = False
+try:
+    from huggingface_hub import HfApi
+    HF_AVAILABLE = True
+except Exception:
+    HF_AVAILABLE = False
+
+# ---------------- Config ----------------
+HF_TOKEN = os.getenv("HF_TOKEN")
+QUEUE_REPO = os.getenv("QUEUE_REPO", "msana/test-submission")       # e.g., "your-org/telecom-eval-queue"
+RESULTS_REPO = os.getenv("RESULTS_REPO", "msana/test-results")   # future: read results remotely if desired
+CACHE_PATH = os.getenv("HF_HOME", ".") # cache dir
+USE_LOCAL_DATA = os.getenv("USE_LOCAL_DATA", True) # set False to load data from results repo
+LOCAL_DATA_FILE = os.getenv("LOCAL_DATA_FILE", "benchmark.json") # the local data to loadt
+DATASETS = os.getenv("DATASETS", "3GPP-TSG;NetBench;TeleQna;TeleLogs;TeleMath")
+
+EVAL_REQUESTS_PATH = os.path.join(CACHE_PATH, "submission-queue")
+EVAL_RESULTS_PATH = os.path.join(CACHE_PATH, "results-queue")
+
+# ---------------- App ----------------
+app = FastAPI(title="Telecom LLM Leaderboard API", version="1.0.0")
+app.add_middleware(
+    CORSMiddleware, allow_origins=["*"], allow_credentials=True,
+    allow_methods=["*"], allow_headers=["*"]
+)
+
+# ---------------- Data Models ----------------
+MetricType = Literal["raw", "llm-as-judge"]
+
+class Score(BaseModel):
+    dataset_name: str
+    metric_type: MetricType
+    score: float
+    energy_consumed: float
+    co2_consumed: float
+
+class ModelResult(BaseModel):
+    provider: str
+    name: str
+    repo: str
+    updated_at: str
+    scores: List[Score]
+
+class SubmitPayload(BaseModel):
+    model_provider: str = Field(..., examples=["TelcoAI"])
+    model_name: str = Field(..., examples=["T-LLM-7B"])
+    hf_repo: str = Field(..., examples=["telcoai/t-llm-7b"])
+    contact_email: EmailStr
+    notes: Optional[str] = ""
+
+
+
+if isinstance(DATASETS, str):
+    if ';' in DATASETS:
+        DATASETS = DATASETS.split(";")
+    elif '|' in DATASETS:
+        DATASETS = DATASETS.split("|")
+    elif ',' in DATASETS:
+        DATASETS = DATASETS.split(",")
+    else:
+        DATASETS = DATASETS.split(" ")
+
+    DATASETS = list(map(lambda s: s.strip(), DATASETS))
+
+
+
+# ---------- fake data for test purpose
+FAKE_MODELS = [
+    ("Qwen", "Qwen3-32B", "qwen/qwen3-32b"),
+    ("Qwen", "QwQ-32B", "qwen/qwq-32b"),
+    ("OpenAI", "GPT-OSS-120B", "openai/gpt-oss-120b"),
+    ("OpenAI", "GPT-OSS-20B", "openai/gpt-oss-20b"),
+    ("DeepSeek", "R1-Distill-Llama-70B", "deepseek/r1-distill-llama-70b"),
+    ("ByteDance", "Seed-OSS-36B", "bytedance/seed-oss-36b"),
+    ("LLama", "Llama-8B", "llama/llama-8B-instruct"),
+]
+
+def _random_score() -> float:
+    return round(random.uniform(45.0, 62.0), 2)
+
+def generate_fake_model(provider: str, name: str, repo: str) -> ModelResult:
+    metric_types = ["raw", "llm-as-judge"]
+    now = datetime.utcnow().isoformat() + "Z"
+    scores = []
+    for d in DATASETS:
+        mt = random.choice(metric_types)
+        scores.append(Score(
+            dataset_name=d,
+            metric_type=mt, score=_random_score(),
+            energy_consumed=round(random.uniform(0.1, 2.5), 3),
+            co2_consumed=round(random.uniform(0.05, 1.0), 3),
+        ))
+    return ModelResult(provider=provider, name=name, repo=repo, updated_at=now, scores=scores)
+
+def generate_fake_results() -> Dict[str, Any]:
+    models = [generate_fake_model(p, n, r) for (p, n, r) in FAKE_MODELS]
+    return {"models": [m.model_dump() for m in models]}
+
+
+
+def load_benchmark() -> Dict[str, Any]:
+    with open(LOCAL_DATA_FILE, 'r') as f:
+        return json.load(f)
+
+# ---------------- HF Upload Helper ----------------
+def push_submission_to_queue(eval_entry: Dict[str, Any]) -> str:
+    """
+    Writes a JSON file locally under HF_HOME/submission-queue/<provider>/<model>_submission_<version>.json
+    and uploads it to the dataset repo specified by QUEUE_REPO (if configured).
+    Returns a submission id string.
+    """
+    model_provider = eval_entry.get("model_provider", "unknown-provider")
+    model_name = eval_entry.get("model_name", "unknown-model")
+    version = eval_entry.get("version", "v1")
+
+    OUT_DIR = os.path.join(EVAL_REQUESTS_PATH, model_provider)
+    os.makedirs(OUT_DIR, exist_ok=True)
+    out_path = os.path.join(OUT_DIR, f"{model_name}_submission_{version}.json")
+
+    with open(out_path, "w", encoding="utf-8") as f:
+        f.write(json.dumps(eval_entry, ensure_ascii=False, indent=2))
+
+    upload_ok = False
+    if HF_AVAILABLE and HF_TOKEN and QUEUE_REPO:
+        api = HfApi(token=HF_TOKEN)
+        # path in repo: <provider>/<file>
+        print("-------------", out_path.split("submission-queue/"), out_path)
+        if '\\' in out_path:
+            out_path = out_path.replace('\\', '/')
+        path_in_repo = out_path.split("submission-queue/")[1]
+
+        api.upload_file(
+            path_or_fileobj=out_path,
+            path_in_repo=path_in_repo,
+            repo_id=QUEUE_REPO,
+            repo_type="dataset",
+            commit_message=f"Add {model_name} to eval queue",
+        )
+        upload_ok = True
+
+    return f"{model_provider}-{model_name}-{version}{'' if upload_ok else '-local'}"
+
+# ---------------- Routes ----------------
+
+@app.get("/", include_in_schema=False)
+async def index():
+    return FileResponse("index.html")
+
+
+@app.get("/api/health")
+def health():
+    return {"ok": True, "huggingface_available": HF_AVAILABLE, "queue_repo": QUEUE_REPO, "result_repo": RESULTS_REPO}
+
+@app.get("/api/datasets")
+def get_datasets():
+    return {"datasets": DATASETS}
+
+@app.get("/api/models")
+def get_submitted_models():
+    models = []
+
+    if HF_AVAILABLE and HF_TOKEN and QUEUE_REPO:
+        api = HfApi(token=HF_TOKEN)
+        # path in repo: <provider>/<file>
+        api.snapshot_download(
+            repo_id=QUEUE_REPO, local_dir=EVAL_REQUESTS_PATH, repo_type="dataset", tqdm_class=None, etag_timeout=30
+        )
+
+        models_filepaths = []
+
+        for root, _, files in os.walk(EVAL_REQUESTS_PATH):
+            # We should only have json files in model results
+            if len(files) == 0 or any([not f.endswith(".json") for f in files]):
+                continue
+
+            for file in files:
+                models_filepaths.append(os.path.join(root, file))
+
+        for filepath in models_filepaths:
+            try:
+                with open(filepath, 'r', encoding="utf-8") as f:
+                    model: dict = json.load(f)
+
+                    # --- security issue
+                    model.pop("contact_email")
+
+                    models.append(model)
+            except:
+                continue
+    
+    return {"models": models}
+
+
+@app.get("/api/results")
+def get_results():
+    results = []
+
+    if USE_LOCAL_DATA:
+        return {"models": load_benchmark()}
+
+    if HF_AVAILABLE and HF_TOKEN and QUEUE_REPO:
+        api = HfApi(token=HF_TOKEN)
+        # path in repo: <provider>/<file>
+        api.snapshot_download(
+            repo_id=RESULTS_REPO, local_dir=EVAL_RESULTS_PATH, repo_type="dataset", tqdm_class=None, etag_timeout=30
+        )
+
+        model_result_filepaths = []
+
+        for root, _, files in os.walk(EVAL_RESULTS_PATH):
+            # We should only have json files in model results
+            if len(files) == 0 or any([not f.endswith(".json") for f in files]):
+                continue
+
+            for file in files:
+                model_result_filepaths.append(os.path.join(root, file))
+
+        for filepath in model_result_filepaths:
+            try:
+                with open(filepath, 'r', encoding="utf-8") as f:
+                    results.append(json.load(f))
+            except:
+                continue
+        
+            # print("----------------", results)
+
+    return {"models": results}
+
+
+    # For simplicity, this returns generated fake data.
+    # If you want to read from a RESULTS_REPO, implement a reader here that loads
+    # per-model folders and aggregates JSON files into the required shape.
+    # return generate_fake_results()
+
+@app.post("/api/submit")
+def submit_model(payload: SubmitPayload = Body(...)):
+    ts = datetime.utcnow().strftime("%Y%m%dT%H%M%SZ")
+    eval_entry = {
+        "model_provider": payload.model_provider,
+        "model_name": payload.model_name,
+        "hf_repo": payload.hf_repo,
+        "contact_email": payload.contact_email,
+        "notes": payload.notes or "",
+        "status": "pending",
+        "version": ts,
+        "submitted_at": datetime.utcnow().isoformat() + "Z",
+    }
+
+    # Is the model info correctly filled?
+    try:
+        _ = HfApi.model_info(repo_id=payload.model_name)
+    except Exception as e:
+        # If queue upload fails, still persist locally and report an error
+        raise HTTPException(status_code=500, detail=f"Could not get your model information. Please fill it up properly.")
+    
+    try:
+        submission_id = push_submission_to_queue(eval_entry)
+        return {"status": "pending", "id": submission_id}
+    except Exception as e:
+        # If queue upload fails, still persist locally and report an error
+        raise HTTPException(status_code=500, detail=f"Failed to queue submission: {e}")
+
+# ---------------- Optional: run via `python server.py` ----------------
+if __name__ == "__main__":
+    import uvicorn
+    uvicorn.run("server:app", host="0.0.0.0", port=7600, reload=True)