File size: 8,600 Bytes
190ef78 ece09fa 190ef78 ece09fa 99d4a87 ece09fa |
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 |
---
license: apache-2.0
datasets:
- openbmb/UltraInteract_sft
- openbmb/UltraInteract_pair
- openbmb/UltraFeedback
tags:
- reasoning
- preference_learning
- nca
pipeline_tag: text-generation
---
<div align="center">
<img src="https://huggingface.co/openbmb/Eurus-7b-sft/resolve/main/figures/Eurus-logo.png" width="200px">
**Eurus: A suite of open-source LLMs optimized for reasoning**
<p align="center">
<a href="#introduction"> Introduction</a> •
<a href="#evaluation">Evaluation</a>
</p>
</div>
# Links
- 📜 [Paper](https://arxiv.org/abs/2404.02078)
- 🤗 [Eurus Collection](https://huggingface.co/collections/openbmb/eurus-660bc40bec5376b3adc9d1c5)
- 🤗 UltraInteract
- [SFT](https://huggingface.co/datasets/openbmb/UltraInteract_sft)
- [Preference Learning](https://huggingface.co/datasets/openbmb/UltraInteract_pair)
- [GitHub Repo](https://github.com/OpenBMB/Eurus)
# Introduction
Eurux-8x22B-NCA is SFT and [NCA](https://arxiv.org/abs/2402.05369) fine-tuned from [Mixtral-8x22B](https://huggingface.co/mistral-community/Mixtral-8x22B-v0.1) on all multi-turn trajectory pairs in [UltraInteract](https://huggingface.co/openbmb/UltraInteract) and all pairs in [UltraFeedback](https://huggingface.co/openbmb/UltraFeedback).
It achieves superb reasoning performance as well as exellent chat & instruction-following capabilities.
## Evaluation
We conducted overall coding, math, reasoning, knowledge, instruction-following and chat benchmarking. Results are shown below:
<style type="text/css">
.tg {border-collapse:collapse;border-spacing:0;}
.tg td{border-color:black;border-style:solid;border-width:1px;font-family:Arial, sans-serif;font-size:14px;
overflow:hidden;padding:10px 5px;word-break:normal;}
.tg th{border-color:black;border-style:solid;border-width:1px;font-family:Arial, sans-serif;font-size:14px;
font-weight:normal;overflow:hidden;padding:10px 5px;word-break:normal;}
.tg .tg-c3ow{border-color:inherit;text-align:center;vertical-align:top}
.tg .tg-0pky{border-color:inherit;text-align:left;vertical-align:top}
</style>
<table class="tg">
<thead>
<tr>
<th class="tg-0pky" rowspan="2"><span style="font-weight:bold;color:#000">Model</span></th>
<th class="tg-c3ow" colspan="3"><span style="font-weight:bold;color:#000">Coding</span></th>
<th class="tg-c3ow" colspan="3"><span style="font-weight:bold;color:#000">Math</span></th>
<th class="tg-c3ow"><span style="font-weight:bold;color:#000">Reasoning</span></th>
<th class="tg-c3ow"><span style="font-weight:bold;color:#000">Knowledge</span></th>
<th class="tg-c3ow"><span style="font-weight:bold;color:#000">Ins-Following</span></th>
<th class="tg-c3ow"><span style="font-weight:bold;color:#000">Chat</span></th>
</tr>
<tr>
<th class="tg-c3ow"><span style="font-weight:bold;color:#000">HumanEval</span></th>
<th class="tg-c3ow"><span style="font-weight:bold;color:#000">MBPP</span></th>
<th class="tg-c3ow"><span style="font-weight:bold;color:#000">LeetCode</span></th>
<th class="tg-c3ow"><span style="font-weight:bold;color:#000">GSMPLUS</span></th>
<th class="tg-c3ow"><span style="font-weight:bold;color:#000">MATH</span></th>
<th class="tg-c3ow"><span style="font-weight:bold;color:#000">TheoremQA</span></th>
<th class="tg-c3ow"><span style="font-weight:bold;color:#000">BBH (CoT)</span></th>
<th class="tg-c3ow"><span style="font-weight:bold;color:#000">MMLU</span></th>
<th class="tg-c3ow"><span style="font-weight:bold;color:#000">IFEval</span></th>
<th class="tg-c3ow"><span style="font-weight:bold;color:#000">MT-Bench</span></th>
</tr>
</thead>
<tbody>
<tr>
<td class="tg-0pky"><span style="color:#000">GPT-3.5-Turbo</span></td>
<td class="tg-c3ow"><span style="color:#000">76.8</span> </td>
<td class="tg-c3ow"><span style="color:#000">82.5</span> </td>
<td class="tg-c3ow"><span style="color:#000">23.3</span> </td>
<td class="tg-c3ow"><span style="color:#000">61.2</span> </td>
<td class="tg-c3ow"><span style="color:#000">37.8</span> </td>
<td class="tg-c3ow"><span style="color:#000">35.6</span> </td>
<td class="tg-c3ow"><span style="color:#000">70.1</span> </td>
<td class="tg-c3ow"><span style="color:#000">70.0</span> </td>
<td class="tg-c3ow"><span style="color:#000">56.6</span> </td>
<td class="tg-c3ow"><span style="color:#000">7.94</span> </td>
</tr>
<tr>
<td class="tg-0pky"><span style="color:#000">GPT-4</span></td>
<td class="tg-c3ow"><span style="color:#000">85.4</span> </td>
<td class="tg-c3ow"><span style="color:#000">83.5</span> </td>
<td class="tg-c3ow"><span style="color:#000">41.8</span> </td>
<td class="tg-c3ow"><span style="color:#000">85.6</span> </td>
<td class="tg-c3ow"><span style="color:#000">69.7</span> </td>
<td class="tg-c3ow"><span style="color:#000">52.4</span> </td>
<td class="tg-c3ow"><span style="color:#000">86.7</span> </td>
<td class="tg-c3ow"><span style="color:#000">86.4</span> </td>
<td class="tg-c3ow"><span style="color:#000">79.7</span> </td>
<td class="tg-c3ow"><span style="color:#000">8.96</span> </td>
</tr>
<tr>
<td class="tg-0pky"><span style="color:#000">Eurus-70b-NCA</span></td>
<td class="tg-c3ow"><span style="color:#000">79.3</span> </td>
<td class="tg-c3ow"><span style="color:#000">71.9</span> </td>
<td class="tg-c3ow"><span style="color:#000">33.3</span> </td>
<td class="tg-c3ow"><span style="color:#000">62.8</span> </td>
<td class="tg-c3ow"><span style="color:#000">41.7</span> </td>
<td class="tg-c3ow"><span style="color:#000">32.6</span> </td>
<td class="tg-c3ow"><span style="color:#000">80.0</span> </td>
<td class="tg-c3ow"><span style="color:#000">59.4</span> </td>
<td class="tg-c3ow"><span style="color:#000">49.2</span> </td>
<td class="tg-c3ow"><span style="color:#000">7.54</span> </td>
</tr>
<tr>
<td class="tg-0pky"><span style="color:#000">Eurux-8x22b-KTO</span></td>
<td class="tg-c3ow"><span style="color:#000">71.3</span> </td>
<td class="tg-c3ow"><span style="color:#000">68.9</span> </td>
<td class="tg-c3ow"><span style="color:#000">29.4</span> </td>
<td class="tg-c3ow"><span style="color:#000">68.3</span> </td>
<td class="tg-c3ow"><span style="color:#000">48.4</span> </td>
<td class="tg-c3ow"><span style="color:#000">35.3</span> </td>
<td class="tg-c3ow"><span style="color:#000">83.6</span> </td>
<td class="tg-c3ow"><span style="color:#000">75.9</span> </td>
<td class="tg-c3ow"><span style="color:#000">67.1</span> </td>
<td class="tg-c3ow"><span style="color:#000">8.58</span> </td>
</tr>
<tr>
<td class="tg-0pky"><span style="color:#000">Eurux-8x22b-NCA</span></td>
<td class="tg-c3ow"><span style="color:#000">75.0</span> </td>
<td class="tg-c3ow"><span style="color:#000">69.7</span> </td>
<td class="tg-c3ow"><span style="color:#000">35.0</span> </td>
<td class="tg-c3ow"><span style="color:#000">68.1</span> </td>
<td class="tg-c3ow"><span style="color:#000">49.0</span> </td>
<td class="tg-c3ow"><span style="color:#000">35.5</span> </td>
<td class="tg-c3ow"><span style="color:#000">83.5</span> </td>
<td class="tg-c3ow"><span style="color:#000">75.6</span> </td>
<td class="tg-c3ow"><span style="color:#000">67.1</span> </td>
<td class="tg-c3ow"><span style="color:#000">8.46</span> </td>
</tr>
</tbody>
</table>
## Usage
We apply tailored prompts for coding and math, consistent with UltraInteract data formats:
**Coding**
```
[INST] Write Python code to solve the task:
{Instruction} [/INST]
```
**Math-CoT**
```
[INST] Solve the following math problem step-by-step.
Simplify your answer as much as possible. Present your final answer as \\boxed{Your Answer}.
{Instruction} [/INST]
```
**Math-PoT**
```
[INST] Tool available:
[1] Python interpreter
When you send a message containing Python code to python, it will be executed in a stateful Jupyter notebook environment.
Solve the following math problem step-by-step.
Simplify your answer as much as possible.
{Instruction} [/INST]
```
## Citation
```
@misc{yuan2024advancing,
title={Advancing LLM Reasoning Generalists with Preference Trees},
author={Lifan Yuan and Ganqu Cui and Hanbin Wang and Ning Ding and Xingyao Wang and Jia Deng and Boji Shan and Huimin Chen and Ruobing Xie and Yankai Lin and Zhenghao Liu and Bowen Zhou and Hao Peng and Zhiyuan Liu and Maosong Sun},
year={2024},
eprint={2404.02078},
archivePrefix={arXiv},
primaryClass={cs.AI}
}
``` |