File size: 11,886 Bytes
4a00e43
 
 
 
 
 
 
cb42fb5
4a00e43
28e5488
 
 
 
 
 
 
 
 
 
 
 
 
8d08f8c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4a00e43
 
7c537d9
4a00e43
 
 
 
 
 
 
8e3115f
38cb3c2
8e3115f
38cb3c2
8e3115f
38cb3c2
8e3115f
38cb3c2
8e3115f
38cb3c2
7c537d9
4a00e43
8e3115f
38cb3c2
8e3115f
38cb3c2
8e3115f
38cb3c2
8e3115f
38cb3c2
7c537d9
4a00e43
8e3115f
38cb3c2
8e3115f
38cb3c2
8e3115f
38cb3c2
7c537d9
cb42fb5
4a00e43
 
 
 
7c537d9
cb42fb5
4a00e43
 
0e12c29
 
8d1d8ae
0e12c29
 
 
 
49bde7b
0e12c29
 
 
4a00e43
 
 
0e12c29
1c64301
0e12c29
4a00e43
 
 
38cb3c2
 
 
 
 
 
 
52f5445
38cb3c2
 
4a00e43
 
 
 
0e12c29
 
 
4a00e43
 
 
713e52d
 
 
 
 
5c7e293
8d08f8c
713e52d
 
 
cb42fb5
7c537d9
4a00e43
 
 
 
 
cb42fb5
84b38c9
4a00e43
bf2e6d8
 
 
a7a2b39
 
 
bf2e6d8
0e12c29
40eda1d
4a00e43
 
 
 
 
 
 
 
 
 
 
 
7bf2f17
4a00e43
 
 
 
7c537d9
302d0a4
 
bf2e6d8
 
302d0a4
 
 
 
bf2e6d8
0bfff66
bf2e6d8
 
 
302d0a4
bf2e6d8
38cb3c2
0e12c29
4a00e43
37cf514
0e12c29
aa490c1
49bde7b
aa490c1
0e12c29
3a48bd4
4a00e43
 
 
 
0e12c29
 
3a48bd4
4a00e43
 
 
 
0e12c29
 
3a48bd4
4a00e43
 
 
 
fbe5b7c
 
1e718ae
fbe5b7c
4a00e43
 
 
 
 
 
 
 
 
 
 
 
7c537d9
4a00e43
 
7c537d9
4a00e43
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
<!DOCTYPE html>
<html>
<head>
  <meta charset="utf-8">
  <meta name="description" content="Atla Selene Mini: A General Purpose Evaluation Model">
  <meta name="viewport" content="width=device-width, initial-scale=1">
  <title>Atla Selene Mini: A General Purpose Evaluation Model</title>

  <link href="https://fonts.googleapis.com/css?family=Google+Sans|Noto+Sans|Castoro" rel="stylesheet">
  <link rel="stylesheet" href="./static/css/bulma.min.css">
  <link rel="stylesheet" href="./static/css/bulma-carousel.min.css">
  <link rel="stylesheet" href="./static/css/bulma-slider.min.css">
  <link rel="stylesheet" href="./static/css/fontawesome.all.min.css">
  <link rel="stylesheet" href="https://cdn.jsdelivr.net/gh/jpswalsh/academicons@1/css/academicons.min.css">
  <link rel="stylesheet" href="./static/css/index.css">
  <link rel="icon" href="./static/images/favicon.svg">

  <script src="https://ajax.googleapis.com/ajax/libs/jquery/3.5.1/jquery.min.js"></script>
  <script defer src="./static/js/fontawesome.all.min.js"></script>
  <script src="./static/js/bulma-carousel.min.js"></script>
  <script src="./static/js/bulma-slider.min.js"></script>
  <script src="./static/js/index.js"></script>

  <style>
    @keyframes rainbow-shimmer {
      0% { background-position: 0% 50%; }
      50% { background-position: 100% 50%; }
      100% { background-position: 0% 50%; }
    }
    
    .rainbow-button {
      background: linear-gradient(45deg, #ff0000, #ff7f00, #ffff00, #00ff00, #0000ff, #8b00ff);
      background-size: 300% 300%;
      color: white !important;
      font-weight: bold;
      animation: rainbow-shimmer 5s ease infinite;
      transition: all 0.3s ease;
    }
    
    .rainbow-button:hover {
      transform: scale(1.05);
      box-shadow: 0 0 10px rgba(0,0,0,0.2);
    }
  </style>
</head>
<body>

<section class="hero">
  <div class="hero-body">
    <div class="container is-max-desktop">
      <div class="columns is-centered">
        <div class="column has-text-centered">
          <h1 class="title is-1 publication-title">Atla Selene Mini:<br>A General Purpose Evaluation Model</h1>
          <div class="is-size-5 publication-authors">
            <span class="author-block">
              <a href="https://huggingface.co/inwaves" target="_blank">Andrei Alexandru</a><sup>1</sup>,</span>
            <span class="author-block">
              <a href="https://huggingface.co/NinaCalvi" target="_blank">Antonia Calvi</a><sup>1</sup>,</span>
            <span class="author-block">
              <a href="https://huggingface.co/HennersBro98" target="_blank">Henry Broomfield</a><sup>1</sup>,</span>
            <span class="author-block">
              <a href="https://huggingface.co/jacksongolden" target="_blank">Jackson Golden</a><sup>1</sup>,</span>
            <span class="author-block">
              <a href="https://huggingface.co/kaikaidai" target="_blank">Kyle Dai</a><sup>1</sup>,</span>
          </div>
          <div class="is-size-5 publication-authors">
            <span class="author-block">
              <a href="https://huggingface.co/mathias-atla" target="_blank">Mathias Leys</a><sup>1</sup>,</span>
            <span class="author-block">
              <a href="https://huggingface.co/MauriceBurg" target="_blank">Maurice Burger</a><sup>1</sup>,</span>
            <span class="author-block">
              <a href="https://huggingface.co/mbartolo" target="_blank">Max Bartolo</a><sup>2,3</sup>,</span>
            <span class="author-block">
              <a href="https://huggingface.co/RomanEngeler1805" target="_blank">Roman Engeler</a><sup>1</sup>,</span>
          </div>
          <div class="is-size-5 publication-authors">
            <span class="author-block">
              <a href="https://huggingface.co/spisupat" target="_blank">Sashank Pisupati</a><sup>1</sup>,</span>
            <span class="author-block">
              <a href="https://huggingface.co/tobydrane" target="_blank">Toby Drane</a><sup>1</sup>,</span>
            <span class="author-block">
              <a href="https://huggingface.co/youngsunpark" target="_blank">Young Sun Park</a><sup>1</sup></span>
          </div>

          <div class="is-size-5 publication-authors">
            <span class="author-block"><sup>1</sup>atla,</span>
            <span class="author-block"><sup>2</sup>University College London,</span>
            <span class="author-block"><sup>3</sup>Cohere</span>
          </div>

          <div class="column has-text-centered">
            <div class="publication-links">
              <!-- PDF Link -->
              <span class="link-block">
                <a href="https://arxiv.org/pdf/2501.17195v1" target="_blank"
                   class="external-link button is-normal is-rounded is-dark">
                  <span class="icon">
                      <i class="fas fa-file-pdf"></i>
                  </span>
                  <span>arXiv</span>
                </a>
              </span>
              <!-- HuggingFace Link -->
              <span class="link-block">
                <a href="https://hf.co/AtlaAI/Selene-1-Mini-Llama-3.1-8B" target="_blank"
                   class="external-link button is-normal is-rounded is-dark">
                  <span class="icon">
                      🤗
                  </span>
                  <span>HuggingFace</span>
                </a>
              </span>
              <!-- Github Link -->
              <span class="link-block">
                <a href="https://github.com/atla-ai/selene-mini" target="_blank"
                   class="external-link button is-normal is-rounded is-dark">
                  <span class="icon">
                      <i class="fab fa-github"></i>
                  </span>
                  <span>Cookbooks</span>
                </a>
              </span>
              <!-- Ollama Link -->
              <span class="link-block">
                <a href="https://ollama.com/atla/selene-mini" target="_blank"
                   class="external-link button is-normal is-rounded is-dark">
                  <span class="icon">
                      <i class="fas fa-code"></i>
                  </span>
                  <span>Ollama</span>
                </a>
              </span>
            </div>

            <!-- New API Sign-up Button -->
            <div class="publication-links" style="margin-top: 1rem;">
              <span class="link-block">
                <a href="https://www.atla-ai.com/sign-up-waitlist?utm_source=huggingface&utm_medium=community&utm_campaign=WL_HF_all_communitypost_sel1minilaunch" target="_blank"
                   class="external-link button is-normal is-rounded rainbow-button">
                  <span>Sign up for the API</span>
                </a>
              </span>
            </div>
          </div>
        </div>
      </div>
    </div>
  </div>
</section>

<section class="section" style="padding-top: 0;">
  <div class="container is-max-desktop">
    <!-- Logo -->
    <div class="columns is-centered has-text-centered">
      <div class="column is-2">
        <div style="max-width: 200px; margin: 0 auto;">
          <img src="figs/atla-logo.png" alt="Atla Logo" style="width: 100%; height: auto;">
        </div>
      </div>
    </div>

    <!-- Abstract -->
    <div class="columns is-centered has-text-centered">
      <div class="column is-four-fifths">
        <h2 class="title is-3">Abstract</h2>
        <div class="content has-text-justified">
          <p>
            We introduce Atla Selene Mini, a state-of-the-art small language model-as-a-judge (SLMJ). Selene Mini is a general-purpose evaluator that outperforms the best SLMJs and GPT-4o-mini on overall performance across 11 out-of-distribution benchmarks, spanning absolute scoring, classification, and pairwise preference tasks. It is the highest-scoring 8B generative model on RewardBench, surpassing strong baselines like GPT-4o and specialized judges.
          </p>
          <p>
            To achieve this, we develop a principled data curation strategy that augments public datasets with synthetically generated critiques and ensures high quality through filtering and dataset ablations. We train our model on a combined direct preference optimization (DPO) and supervised fine-tuning (SFT) loss, and produce a highly promptable evaluator that excels in real-world scenarios.
          </p>
          <p>
            Selene Mini shows dramatically improved zero-shot agreement with human expert evaluations on financial and medical industry datasets. It is also robust to variations in prompt format. Preliminary results indicate that Selene Mini is the top-ranking evaluator in a live, community-driven <a href="https://huggingface.co/blog/arena-atla" target="_blank">Judge Arena</a>. We release the model weights on <a href="https://hf.co/AtlaAI/Selene-1-Mini-Llama-3.1-8B" target="_blank">HuggingFace</a> and <a href="https://ollama.com/atla/selene-mini" target="_blank">Ollama</a> to encourage widespread community adoption.
          </p>
        </div>
      </div>
    </div>

    <!-- Demo Video -->
    <div class="columns is-centered">
      <div class="column is-four-fifths">
        <div class="content has-text-centered">
          <video controls width="800" autoplay loop muted>
            <source src="figs/demo.mp4" type="video/mp4">
            Your browser does not support the video element.
          </video>
          <p class="subtitle">
            Demo of Atla Selene Mini on our&thinsp;<a href="https://huggingface.co/spaces/AtlaAI/selene" target="_blank">playground</a>
          </p>
        </div>
      </div>
    </div>

    <!-- Key Results -->
    <div class="columns is-centered">
      <div class="column is-four-fifths">
        <h2 class="title is-3 has-text-centered">Key Results</h2>
        <div class="content has-text-justified">
          <div class="columns is-centered has-text-centered">
            Read the full technical report&nbsp;<a href="https://arxiv.org/pdf/2501.17195v1" target="_blank">here</a>
          </div>
          <figure class="image">
            <img src="figs/Fig1.png" alt="Performance comparison">
            <figcaption>
              <b>Figure 1:</b> Atla Selene Mini outperforms current state-of-the-art SLMJs: a) Overall task-average performance, comparing Atla Selene Mini (black) with the best and most widely used SLMJs. b) Breakdown of performance by task type and benchmark.
            </figcaption>
          </figure>

          <figure class="image">
            <img src="figs/Fig2.png" alt="Data curation strategy">
            <figcaption>
              <b>Figure 2:</b> Data curation strategy: The process of transforming a candidate dataset (left) into the final training mix (right). Yellow boxes indicate filtering steps, purple represents synthetic generation of chosen and rejected pairs for preference optimization.
            </figcaption>
          </figure>

          <figure class="image">
            <img src="figs/Fig3.png" alt="Real-world evaluation">
            <figcaption>
              <b>Figure 3:</b> Real-world evaluation: a) Performance on domain-specific industry benchmarks b) Performance on RewardBench with different prompt formats c) Performance measured by ELO scores in Judge Arena.
            </figcaption>
          </figure>

          <div class="columns is-centered has-text-centered">
            Our larger model from the Selene family will be released soon. Sign up to our &thinsp;<a href="https://www.atla-ai.com/sign-up-waitlist" target="_blank">waitlist</a>&thinsp; to get first access.
          </div>
        </div>
      </div>
    </div>
  </div>
</section>

<footer class="footer">
  <div class="container">
    <div class="content has-text-centered">
      <p>
        © 2025 Atla AI
      </p>
    </div>
  </div>
</footer>

</body>
</html>