Hugging Face's logo

Fate-Zero
/
Archer2.0-Code-1.5B-Preview

Safetensors
qwen2
Model card Files
xet
Community
1
Archer2.0-Code-1.5B-Preview
File size: 3,444 Bytes 
1c47874
 
 
 
 
78e4bad
 
 
 
 
 
 
 
 
 
 
 
 
 
 
954c68b
 
78e4bad
 
 
 
 
 
 
954c68b
 
 
 
 
25bec17
954c68b
 
 
 
1c47874
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
52f1293
c06fd27
 
 
 
 
1c47874
 
 
 
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
 
---
license: apache-2.0
---



<div align="center">

# ✨ Archer2.0

<div>
🏹️  Reinforcement Learning for Enhanced Reasoning in LLMs  🎯
</div>

</div>
<div>
<br>

<div align="center">

[![Notion](https://img.shields.io/badge/Notion-%23000000.svg?style=for-the-badge&logo=notion&logoColor=white)](https://rogue-canopy-54a.notion.site/ASPO-Asymmetric-Importance-Sampling-Policy-Optimization-2650e4c8c16a8034a5d3dfec358c9021)
[![Paper](https://img.shields.io/badge/Paper-%23000000?style=for-the-badge&logo=arxiv&logoColor=000&labelColor=white)](https://arxiv.org/abs/2510.06062)
[![Github](https://img.shields.io/badge/Code-000000?style=for-the-badge&logo=github&logoColor=000&logoColor=white)](https://github.com/wizard-III/Archer2.0)
[![Model](https://img.shields.io/badge/Model-fcd022?style=for-the-badge&logo=huggingface&logoColor=000&labelColor)](https://huggingface.co/Fate-Zero/Archer2.0-Code-1.5B-Preview)
[![Data](https://img.shields.io/badge/Data-fcd022?style=for-the-badge&logo=huggingface&logoColor=000&labelColor)](https://huggingface.co/datasets/Fate-Zero/Archer2.0-Code-1.5B)
[![知乎](https://img.shields.io/badge/知乎-0084FF?style=for-the-badge&logo=zhihu&logoColor=white)](https://zhuanlan.zhihu.com/p/1950989602023244983)

</div>

## Overview

**Archer2.0** marks a significant evolution from its predecessor through the introduction of **Asymmetric Importance Sampling Policy Optimization (ASPO)**, which is designed to overcome the fundamental limitations of **PPO-Clip**, effectively mitigating issues like **entropy collapse** and **repetitive outputs**, preventing **premature convergence**, and thereby enabling more advanced **reinforcement learning** capabilities.

<div align="center">
<img src="archer2.0.png" width="100%"/>
</div>

While our mathematical models are still in training and have not converged, we have evaluated Archer2.0 on the LiveCodeBench v5 and v6 code benchmarks. The results are detailed in the table below.

<table>
  <thead>
    <tr>
      <th rowspan="2">Method</th>
      <th colspan="2">LCB v5 (2024.08.01–2025.02.01)</th>
      <th colspan="2">LCB v6 (2025.02.01–2025.05.01)</th>
      <th rowspan="2">Avg.</th>
    </tr>
    <tr>
      <th>avg@8</th>
      <th>pass@8</th>
      <th>avg@16</th>
      <th>pass@16</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>DeepSeek-R1-1.5B</td>
      <td>16.7</td>
      <td>29.0</td>
      <td>17.2</td>
      <td>34.4</td>
      <td>17.0</td>
    </tr>
    <tr>
      <td>DAPO</td>
      <td>26.0</td>
      <td>40.5</td>
      <td>27.6</td>
      <td>43.5</td>
      <td>26.8</td>
    </tr>
    <tr>
      <td>DeepCoder-1.5B</td>
      <td>23.3</td>
      <td>39.1</td>
      <td>22.6</td>
      <td>42.0</td>
      <td>23.0</td>
    </tr>
    <tr>
      <td>Nemotron-1.5B</td>
      <td>26.1</td>
      <td>35.5</td>
      <td>29.5</td>
      <td>42.8</td>
      <td>27.8</td>
    </tr>
    <tr>
      <td><strong>Archer-Code-1.5B</strong></td>
      <td><strong>29.4</strong></td>
      <td><strong>43.7</strong></td>
      <td><strong>30.2</strong></td>
      <td><strong>45.8</strong></td>
      <td><strong>29.8</strong></td>
    </tr>
    <tr>
      <td><strong>Archer2.0-Code-1.5B-Preview</strong></td>
      <td><strong>31.5</strong></td>
      <td><strong>47.0</strong></td>
      <td><strong>30.5</strong></td>
      <td><strong>46.0</strong></td>
      <td><strong>31.0</strong></td>
    </tr>
  </tbody>
</table>