janhq
/

Jan-v1-edge-gguf

@@ -21,11 +21,178 @@ library_name: transformers
 Jan-edge was developed through a two-phase post-training process. The first phase, **Supervised Fine-Tuning (SFT)**, transferred core capabilities from the `Jan-v1` teacher model to the smaller student. The second phase, **Reinforcement Learning with Verifiable Rewards (RLVR)** —the same method used in `Jan-v1` and `Lucy`—further optimized reasoning efficiency, tool use, and correctness. This staged approach delivers reliable results on complex, interactive workloads.
 ## Performance
-**Question Answering(SimpleQA)**
-**Chat & Instruction Following**
 ## Quick Start

 Jan-edge was developed through a two-phase post-training process. The first phase, **Supervised Fine-Tuning (SFT)**, transferred core capabilities from the `Jan-v1` teacher model to the smaller student. The second phase, **Reinforcement Learning with Verifiable Rewards (RLVR)** —the same method used in `Jan-v1` and `Lucy`—further optimized reasoning efficiency, tool use, and correctness. This staged approach delivers reliable results on complex, interactive workloads.
 ## Performance
+### Question Answering(SimpleQA)
+**Jan-edge** achieves **84.2% accuracy without a system prompt**, and **84.8% with a system prompt**. The small 0.6% gap highlights the model’s stability and reliability across prompting conditions — a critical property for real-world deployments where system prompts may vary.
+### Chat & Instruction Following
+<!DOCTYPE html>
+<html lang="en">
+<head>
+  <meta charset="UTF-8" />
+  <meta name="viewport" content="width=device-width, initial-scale=1.0"/>
+  <title>SimpleQA Benchmark Results</title>
+  <link rel="preconnect" href="https://fonts.googleapis.com"/>
+  <link rel="preconnect" href="https://fonts.gstatic.com" crossorigin/>
+  <link href="https://jan.ai/assets/images/general/logo-mark.svg" rel="stylesheet"/>
+  <link href="styles.css" rel="stylesheet"/>
+  <style>
+    :root{
+      --bg:#f8f9fa;
+      --fg:#111827;
+      --muted:#6b7280;
+      --card:#ffffff;
+      --rule:#e5e7eb;
+      --brand:#14b8a6;   /* Jan-v1 */
+      --edgeA:#0ea5e9;   /* Jan-edge no prompt */
+      --edgeB:#0284c7;   /* Jan-edge with system */
+      --lucy:#a78bfa;    /* Lucy */
+      --alt:#3b82f6;     /* Perplexity */
+      --oss:#86efac;     /* gpt-oss */
+      --gray:#d1d5db;    /* others */
+      --star:#fbbf24;
+    }
+    body{
+      font-family:'Inter',sans-serif;
+      background:var(--bg);
+      color:#333;
+      display:flex;
+      justify-content:center;
+      align-items:center;
+      padding:2rem;
+    }
+    .chart-container{
+      width:100%;
+      max-width:1100px;
+      background:var(--card);
+      border-radius:16px;
+      padding:2rem;
+      box-shadow:0 4px 12px rgba(0,0,0,.05);
+    }
+    .chart-header{
+      display:flex;
+      flex-direction:column;
+      align-items:center;
+      gap:.75rem;
+      margin-bottom:2rem;
+    }
+    .brand-logo{
+      width:40px;
+      height:40px;
+      object-fit:contain;
+    }
+    .titleblock{
+      display:flex;
+      flex-direction:column;
+      text-align:center;
+    }
+    .chart-title{margin:0;font-size:1.6rem;font-weight:700;color:var(--fg)}
+    .chart-subtitle{margin:.15rem 0 0;font-size:.95rem;color:var(--muted)}
+    .chart-area{
+      display:flex;justify-content:space-between;align-items:flex-end;
+      height:420px;padding:0 1rem;border-bottom:2px solid var(--rule);
+      position:relative;margin-bottom:120px;
+    }
+    .bar{
+      flex:1;margin:0 .4%;position:relative;display:flex;justify-content:center;align-items:flex-end;
+      border-radius:8px 8px 0 0;min-width:52px;
+    }
+    .bar .value{position:absolute;top:-26px;font-weight:600;font-size:.86rem;color:var(--fg)}
+    .bar .value.highlight{font-weight:700;display:flex;flex-direction:column;align-items:center;top:-36px}
+    .bar .value .star{font-size:.9rem;margin-bottom:2px;color:var(--star)}
+    .bar .label{
+      position:absolute;bottom:-54px;left:50%;transform:translateX(-50%) rotate(-45deg);
+      white-space:nowrap;font-size:.76rem;color:#4b5563;font-weight:500;
+    }
+    .bar .label.emph{font-weight:700}
+    .note{font-size:.85rem;color:var(--muted);margin-top:.5rem;font-style:italic;text-align:center}
+  </style>
+</head>
+<body>
+  <div class="chart-container">
+    <div class="chart-header">
+      <img src="./logo.png" alt="Jan Logo" class="brand-logo"/>
+      <div class="titleblock">
+        <h2 class="chart-title">SimpleQA Benchmark Results</h2>
+        <p class="chart-subtitle">Comprehensive accuracy on question answering • Serper MCP evaluation</p>
+      </div>
+    </div>
+    <div class="chart-area">
+      <!-- Jan-v1 -->
+      <div class="bar" style="height:91.1%; background:var(--brand);">
+        <span class="value highlight" style="top: -44px;"><span class="star">★</span>91.1%</span>
+        <span class="label emph" style="color:var(--brand)">Jan-v1</span>
+      </div>
+      <!-- Perplexity Pro -->
+      <div class="bar" style="height:90.6%; background:var(--alt);">
+        <span class="value" style="top: -23px;">90.6%</span>
+        <span class="label">Perplexity Pro</span>
+      </div>
+      <!-- Qwen3-4B-2507 -->
+      <div class="bar" style="height:86.5%; background:var(--gray);">
+        <span class="value">86.5%</span>
+        <span class="label">Qwen3-4B-2507</span>
+      </div>
+      <!-- gpt-oss-20B -->
+      <div class="bar" style="height:86.3%; background:var(--oss);">
+        <span class="value">86.3%</span>
+        <span class="label">gpt-oss-20B</span>
+      </div>
+      <!-- Jan-edge (system prompt) -->
+      <div class="bar" style="height:84.8%; background:var(--edgeB);">
+        <span class="value">84.8%</span>
+        <span class="label emph" style="color:var(--edgeB)">Jan-edge (system)</span>
+      </div>
+      <!-- Jan-edge (no prompt) -->
+      <div class="bar" style="height:84.2%; background:var(--edgeA);">
+        <span class="value">84.2%</span>
+        <span class="label emph" style="color:var(--edgeA)">Jan-edge (no prompt)</span>
+      </div>
+      <!-- Jan-nano-128k -->
+      <div class="bar" style="height:83.2%; background:var(--gray);">
+        <span class="value">83.2%</span>
+        <span class="label">Jan-nano-128k</span>
+      </div>
+      <!-- Jan-nano -->
+      <div class="bar" style="height:80.7%; background:var(--gray);">
+        <span class="value">80.7%</span>
+        <span class="label">Jan-nano</span>
+      </div>
+      <!-- Jan-nano (YaRN) -->
+      <div class="bar" style="height:79.7%; background:var(--gray);">
+        <span class="value">79.7%</span>
+        <span class="label">Jan-nano (YaRN)</span>
+      </div>
+      <!-- Lucy (no prompt) -->
+      <div class="bar" style="height:78.3%; background:var(--lucy);">
+        <span class="value">78.3%</span>
+        <span class="label emph" style="color:var(--lucy)">Lucy</span>
+      </div>
+      <!-- DeepSeek-V3 -->
+      <div class="bar" style="height:78.2%; background:var(--gray);">
+        <span class="value">78.2%</span>
+        <span class="label">DeepSeek-V3</span>
+      </div>
+    </div>
+    <p class="note">
+      *Jan-edge benchmarked at 84.2% (no system prompt) and 84.8% (with system prompt), indicating stable performance with minimal reliance on prompt engineering.*
+    </p>
+  </div>
+</body>
+</html>
 ## Quick Start