Commit
•
8d824d4
1
Parent(s):
5d06ec4
Update README.md
Browse files
README.md
CHANGED
@@ -1,11 +1,11 @@
|
|
1 |
---
|
2 |
license: other
|
3 |
license_name: orion
|
4 |
-
license_link: https://
|
5 |
widget:
|
6 |
-
- text: "
|
7 |
output:
|
8 |
-
text: "
|
9 |
pipeline_tag: text-generation
|
10 |
---
|
11 |
|
@@ -24,9 +24,9 @@ pipeline_tag: text-generation
|
|
24 |
|
25 |
<h4 align="center">
|
26 |
<p>
|
27 |
-
<b
|
28 |
-
<a href="
|
29 |
-
🤗 <a href="https://huggingface.co/OrionStarAI" target="_blank">HuggingFace
|
30 |
<p>
|
31 |
</h4>
|
32 |
|
@@ -34,83 +34,85 @@ pipeline_tag: text-generation
|
|
34 |
|
35 |
|
36 |
|
37 |
-
#
|
38 |
|
39 |
-
- [📖
|
40 |
-
- [🔗
|
41 |
-
- [🔖
|
42 |
-
- [📊
|
43 |
-
- [🥇
|
44 |
-
- [📜
|
45 |
|
46 |
-
#
|
47 |
|
48 |
-
- Orion-14B-
|
49 |
|
50 |
-
-
|
|
|
|
|
|
|
|
|
|
|
|
|
51 |
|
52 |
-
- Orion-14B
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
53 |
|
54 |
-
|
55 |
-
- **Orion-14B-Base:** 基于2.5万亿令牌多样化数据集训练处的140亿参数量级的多语言基座模型。
|
56 |
-
- **Orion-14B-Chat:** 基于高质量语料库微调的对话类模型,旨在为大模型社区提供更好的用户交互体验。
|
57 |
-
- **Orion-14B-LongChat:** 支持长度超过200K令牌上下文的交互,在长文本评估集上性能比肩专有模型。
|
58 |
-
- **Orion-14B-Chat-RAG:** 在一个定制的检索增强生成数据集上进行微调的聊天模型,在检索增强生成任务中取得了卓越的性能。
|
59 |
-
- **Orion-14B-Chat-Plugin:** 专门针对插件和函数调用任务定制的聊天模型,非常适用于使用代理的相关场景,其中大语言模型充当插件和函数调用系统。
|
60 |
-
- **Orion-14B-Base-Int4:** 一个使用4位整数进行量化的基座模型。它将模型大小显著减小了70%,同时提高了推理速度30%,仅引入了1%的最小性能损失。
|
61 |
-
- **Orion-14B-Chat-Int4:** 一个使用4位整数进行量化的对话模型。
|
62 |
|
63 |
-
|
64 |
|
65 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
66 |
|
67 |
-
|
68 |
-
|---------------------|-----------------------------------------------------------------------------------|------------------------------------------------------------------------------------------------|
|
69 |
-
| ⚾ 基座模型 | [Orion-14B-Base](https://huggingface.co/OrionStarAI/Orion-14B-Base) | [Orion-14B-Base](https://modelscope.cn/models/OrionStarAI/Orion-14B-Base/summary) |
|
70 |
-
| 😛 对话模型 | [Orion-14B-Chat](https://huggingface.co/OrionStarAI/Orion-14B-Chat) | [Orion-14B-Chat](https://modelscope.cn/models/OrionStarAI/Orion-14B-Chat/summary) |
|
71 |
-
| 📃 长上下文模型 | [Orion-14B-LongChat](https://huggingface.co/OrionStarAI/Orion-14B-LongChat) | [Orion-14B-LongChat](https://modelscope.cn/models/OrionStarAI/Orion-14B-LongChat/summary) |
|
72 |
-
| 🔎 检索增强模型 | [Orion-14B-Chat-RAG](https://huggingface.co/OrionStarAI/Orion-14B-Chat-RAG) | [Orion-14B-Chat-RAG](https://modelscope.cn/models/OrionStarAI/Orion-14B-Chat-RAG/summary) |
|
73 |
-
| 🔌 插件模型 | [Orion-14B-Chat-Plugin](https://huggingface.co/OrionStarAI/Orion-14B-Chat-Plugin) | [Orion-14B-Chat-Plugin](https://modelscope.cn/models/OrionStarAI/Orion-14B-Chat-Plugin/summary)|
|
74 |
-
| 💼 基座Int4量化模型 | [Orion-14B-Base-Int4](https://huggingface.co/OrionStarAI/Orion-14B-Base-Int4) | [Orion-14B-Base-Int4](https://modelscope.cn/models/OrionStarAI/Orion-14B-Base-Int4/summary) |
|
75 |
-
| 📦 对话Int4量化模型 | [Orion-14B-Chat-Int4](https://huggingface.co/OrionStarAI/Orion-14B-Chat-Int4) | [Orion-14B-Chat-Int4](https://modelscope.cn/models/OrionStarAI/Orion-14B-Chat-Int4/summary) |
|
76 |
|
77 |
-
|
78 |
-
|
79 |
-
|
80 |
-
| 模型名称 | C-Eval | CMMLU | MMLU | AGIEval | Gaokao | BBH |
|
81 |
|--------------------|----------|----------|----------|----------|----------|----------|
|
82 |
| LLaMA2-13B | 41.4 | 38.4 | 55.0 | 30.9 | 18.2 | 45.6 |
|
83 |
| Skywork-13B | 59.1 | 61.4 | 62.7 | 43.6 | 56.1 | 48.3 |
|
84 |
| Baichuan2-13B | 59.0 | 61.3 | 59.5 | 37.4 | 45.6 | 49.0 |
|
85 |
| QWEN-14B | 71.7 | 70.2 | 67.9 | 51.9 | **62.5** | 53.7 |
|
86 |
| InternLM-20B | 58.8 | 59.0 | 62.1 | 44.6 | 45.5 | 52.5 |
|
87 |
-
| **Orion-14B**
|
88 |
-
|
89 |
|
90 |
-
|
91 |
-
|
|
92 |
|--------------------|----------|----------|----------|----------|----------|----------|
|
93 |
| LLaMA 2-13B | 63.0 | 58.9 | 77.5 | 79.8 | 76.5 | 66.3 |
|
94 |
| Skywork-13B | 87.6 | 84.1 | 73.7 | 78.3 | 71.8 | 66.3 |
|
95 |
| Baichuan 2-13B | 68.9 | 67.2 | 70.8 | 78.1 | 74.1 | 66.3 |
|
96 |
| QWEN-14B | 93.0 | 90.3 | **80.2** | 79.8 | 71.4 | 66.3 |
|
97 |
| InternLM-20B | 86.4 | 83.3 | 78.1 | **80.3** | 71.8 | 68.3 |
|
98 |
-
| **Orion-14B**
|
99 |
-
|
100 |
-
|
101 |
-
|
102 |
-
|
103 |
-
|
104 |
-
|
|
105 |
-
|
|
106 |
-
|
|
107 |
-
|
|
108 |
-
|
|
109 |
-
|
110 |
-
|
111 |
-
|
112 |
-
## 日语测试集评估结果
|
113 |
-
| 模型名称 |**Average**| JCQA | JNLI | MARC | JSQD | JQK | XLS | XWN | MGSM |
|
114 |
|--------------------|----------|----------|----------|----------|----------|----------|----------|----------|----------|
|
115 |
| PLaMo-13B | 52.3 | 56.7 | 42.8 | 95.8 | 70.6 | 71.0 | 8.70 | 70.5 | 2.40 |
|
116 |
| WebLab-10B | 50.7 | 66.6 | 53.7 | 82.1 | 62.9 | 56.2 | 10.0 | 72.0 | 2.40 |
|
@@ -120,22 +122,21 @@ pipeline_tag: text-generation
|
|
120 |
| Baichuan 2-13B | 57.1 | 73.7 | 31.3 | 91.6 | 80.5 | 63.3 | 18.6 | 72.2 | 25.2 |
|
121 |
| QWEN-14B | 65.8 | 85.9 | 60.7 | 97.0 | 83.3 | 71.8 | 18.8 | 70.6 | 38.0 |
|
122 |
| Yi-34B | 67.1 | 83.8 | 61.2 | 95.2 | **86.1** | 78.5 | **27.2** | 69.2 | 35.2 |
|
123 |
-
| **Orion-14B**
|
124 |
-
|
125 |
-
|
126 |
-
|
127 |
-
|
128 |
-
|
129 |
-
|
|
130 |
-
|
|
131 |
-
|
|
132 |
-
|
|
133 |
-
|
|
134 |
-
|
135 |
-
|
136 |
-
|
137 |
-
|
138 |
-
| 模型名称 | Train Lang | Japanese | Korean | Chinese | English |
|
139 |
|--------------------|------------|----------|----------|----------|----------|
|
140 |
| PLaMo-13B | En,Jp | 52.3 | * | * | * |
|
141 |
| Weblab-10B | En,Jp | 50.7 | * | * | * |
|
@@ -147,52 +148,76 @@ pipeline_tag: text-generation
|
|
147 |
| Qwen-14B | Multi | 65.8 | 73.7 | 64.5 | 65.4 |
|
148 |
| Llama2-13B | Multi | 46.3 | 63.7 | 41.4 | 55.3 |
|
149 |
| Yi-34B | Multi | 67.1 | 72.2 | 58.7 | **68.8** |
|
150 |
-
| **Orion-14B**
|
151 |
-
|
152 |
-
|
153 |
-
|
154 |
-
|
155 |
-
|
|
156 |
-
| Qwen-72B | 83.3 | 61.8 | 77.3 | 76.1 | 85.4 |
|
157 |
-
| Yi-34B | 81.8 | 82.6 | 76.3 | 73.1 | 82.0 |
|
158 |
-
| Orion-14B | 72.8 | 70.6 | 69.9 | 78.8 | 78.5 |
|
159 |
-
| Orion-14B(contaminated)| **92.7** | **82.9** | **85.4** | **78.5** | 85.8 |
|
160 |
-
|
161 |
-
|
162 |
-
## 对话模型标准评估
|
163 |
-
| 模型名称 | CMMLU | MMLU | BBH |HellaSwag | PIQA | WSC |
|
164 |
-
|----------------------|----------|----------|----------|----------|----------|----------|
|
165 |
-
| Baichuan2-13B-Chat | 58.4 | 57.0 | 49.9 | 66.9 | 77.6 | **71.2** |
|
166 |
-
| Qwen-14B-Chat | **70.0** | **66.4** | **58.0** | 65.2 | 74.0 | 66.3 |
|
167 |
-
| Llama2-13B-Chat | 38.7 | 54.6 | 40.2 | **78.2** | **78.8** | 68.3 |
|
168 |
-
| InternLM-20B-Chat | 52.2 | 52.5 | 35.3 | 69.2 | 76.7 | 61.5 |
|
169 |
-
| **Orion-14B-Chat** | 63.7 | 61.71 | 49.05 | 76.7 | 78.4 | 71.15 |
|
170 |
-
|
171 |
-
## 对话模型MTBench主观评估
|
172 |
-
| 模型名称 | 第一轮 | 第二轮 | **平均** |
|
173 |
|----------------------|----------|----------|----------|
|
174 |
| Baichuan2-13B-Chat | 7.05 | 6.47 | 6.76 |
|
175 |
| Qwen-14B-Chat | 7.30 | 6.62 | 6.96 |
|
176 |
| Llama2-13B-Chat | 7.10 | 6.20 | 6.65 |
|
177 |
| InternLM-20B-Chat | 7.03 | 5.93 | 6.48 |
|
178 |
| **Orion-14B-Chat** | **7.68** | **7.07** | **7.37** |
|
179 |
-
|
180 |
-
|
181 |
-
|
182 |
-
|
183 |
-
|
184 |
-
|
|
185 |
-
|
|
186 |
-
|
|
187 |
-
|
|
188 |
-
|
189 |
-
|
190 |
-
|
191 |
-
|
192 |
-
|
193 |
-
|
194 |
-
|
195 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
196 |
|
197 |
```python
|
198 |
import torch
|
@@ -204,106 +229,104 @@ model = AutoModelForCausalLM.from_pretrained("OrionStarAI/Orion-14B", device_map
|
|
204 |
torch_dtype=torch.bfloat16, trust_remote_code=True)
|
205 |
|
206 |
model.generation_config = GenerationConfig.from_pretrained("OrionStarAI/Orion-14B")
|
207 |
-
messages = [{"role": "user", "content": "
|
208 |
-
response = model.chat(tokenizer, messages, streaming=
|
209 |
print(response)
|
210 |
|
211 |
-
# 你好!我是一个人工智能助手,没有固定的名字。你可以随意称呼我。有什么我可以帮助你的吗?
|
212 |
```
|
213 |
|
214 |
-
|
215 |
-
|
216 |
|
217 |
-
##
|
218 |
|
219 |
```shell
|
220 |
CUDA_VISIBLE_DEVICES=0 python cli_demo.py
|
221 |
```
|
222 |
|
223 |
-
|
224 |
|
225 |
-
##
|
226 |
|
227 |
```shell
|
228 |
-
|
229 |
-
|
|
|
230 |
|
231 |
# chat model
|
232 |
-
CUDA_VISIBLE_DEVICES=0 python demo/text_generation.py --model OrionStarAI/Orion-14B-Chat --tokenizer OrionStarAI/Orion-14B-Chat --prompt
|
|
|
233 |
```
|
234 |
|
235 |
-
|
236 |
|
237 |
-
##
|
238 |
|
239 |
`````
|
240 |
-
|
241 |
-
Orion-14B
|
242 |
-
|
243 |
-
Orion-14B
|
244 |
-
|
245 |
-
|
246 |
-
|
247 |
-
|
248 |
-
|
249 |
-
|
250 |
-
|
251 |
-
|
252 |
-
|
253 |
-
|
254 |
-
|
255 |
-
|
256 |
-
|
257 |
-
|
258 |
-
|
259 |
-
从此以后,杰克和格里姆成为了好朋友,他们一起经历了许多惊险刺激的冒险,并用他们的勇气和智慧战胜了各种困难。
|
260 |
-
|
261 |
-
这个故事告诉我们,只要有勇气和决心,我们就能克服一切困难,实现自己的梦想。
|
262 |
`````
|
263 |
|
264 |
-
##
|
265 |
|
266 |
`````
|
267 |
-
|
268 |
Orion-14B:こんにちは、私の名前はChatMaxで、OrionStarによって開発されたAIアシスタントです。私はユーザーに正確で、客観的で、有用な回答を提供し、彼らが問題を解決し、情報を取得するのを助けることができます。
|
269 |
-
|
270 |
Orion-14B:안녕하세요, 제 이름은 ChatMax입니다. OrionStar에서 개발한 AI 어시스턴트입니다. 저는 사용자들에게 정확하고 객관적이며 유용한 답변을 제공하여 문제를 해결하고 정보를 얻는 데 도움을 줄 수 있습니다.
|
271 |
`````
|
272 |
|
273 |
-
#
|
274 |
-
|
275 |
-
猎户星空(OrionStar)是一家全球领先的服务机器人解决方案公司,成立于2016年9月。猎户星空致力于基于人工智能技术打造下一代革命性机器人,使人们能够摆脱重复的体力劳动,使人类的工作和生活更加智能和有趣,通过技术使社会和世界变得更加美好。
|
276 |
-
|
277 |
-
猎户星空拥有完全自主开发的全链条人工智能技术,如语音交互和视觉导航。它整合了产品开发能力和技术应用能力。基于Orion机械臂平台,它推出了ORION
|
278 |
-
STAR AI Robot Greeting、AI Robot Greeting Mini、Lucki、Coffee
|
279 |
-
Master等产品,并建立了Orion机器人的开放平台OrionOS。通过为 **真正有用的机器人而生** 的理念实践,它通过AI技术为更多人赋能。
|
280 |
-
|
281 |
-
凭借7年AI经验积累,猎户星空已推出的大模型深度应用“聚言”,并陆续面向行业客户提供定制化AI大模型咨询与服务解决方案,真正帮助客户实现企业经营效率领先同行目标。
|
282 |
-
|
283 |
-
**猎户星空具备全链条大模型应用能力的核心优势**,包括拥有从海量数据处理、大模型预训练、二次预训练、微调(Fine-tune)、Prompt
|
284 |
-
Engineering 、Agent开发的全链条能力和经验积累;拥有完整的端到端模型训练能力,包括系统化的数据处理流程和数百张GPU的并行模型训练能力,现已在大政务、云服务、出海电商、快消等多个行业场景落地。
|
285 |
|
286 |
-
|
|
|
|
|
|
|
287 |
|
288 |
-
|
|
|
|
|
|
|
|
|
289 |
|
290 |
-
|
291 |
|
292 |
-
|
293 |
-
Orion-14B 模型用于未经适当安全审查和备案的互联网服务。
|
294 |
|
295 |
-
|
296 |
-
|
297 |
-
|
|
|
|
|
|
|
|
|
|
|
298 |
|
299 |
-
##
|
300 |
|
301 |
-
|
302 |
-
|
|
|
303 |
|
304 |
-
#
|
305 |
|
306 |
-
|
307 |
|
308 |
![](./assets/imgs/wechat_group.jpg)
|
309 |
|
|
|
1 |
---
|
2 |
license: other
|
3 |
license_name: orion
|
4 |
+
license_link: https://huggingface.co/OrionStarAI/Orion-14B-Chat/blob/main/ModelsCommunityLicenseAgreement
|
5 |
widget:
|
6 |
+
- text: "Hi!"
|
7 |
output:
|
8 |
+
text: "Hello! How can I help you today?"
|
9 |
pipeline_tag: text-generation
|
10 |
---
|
11 |
|
|
|
24 |
|
25 |
<h4 align="center">
|
26 |
<p>
|
27 |
+
<b>🌐English</b> |
|
28 |
+
<a href="https://huggingface.co/OrionStarAI/Orion-14B-Chat/blob/main/README_cn.md">🇨🇳中文</a><br><br>
|
29 |
+
🤗 <a href="https://huggingface.co/OrionStarAI" target="_blank">HuggingFace Mainpage</a> | 🤖 <a href="https://modelscope.cn/organization/OrionStarAI" target="_blank">ModelScope Mainpage</a><br>🎬 <a href="https://huggingface.co/spaces/OrionStarAI/Orion-14B-App-Demo" target="_blank">HuggingFace Demo</a> | 🎫 <a href="https://modelscope.cn/studios/OrionStarAI/Orion-14B-App-Demo/summary" target="_blank">ModelScope Demo</a>
|
30 |
<p>
|
31 |
</h4>
|
32 |
|
|
|
34 |
|
35 |
|
36 |
|
37 |
+
# Table of Contents
|
38 |
|
39 |
+
- [📖 Model Introduction](#model-introduction)
|
40 |
+
- [🔗 Model Download](#model-download)
|
41 |
+
- [🔖 Model Benchmark](#model-benchmark)
|
42 |
+
- [📊 Model Inference](#model-inference)
|
43 |
+
- [🥇 Company Introduction](#company-introduction)
|
44 |
+
- [📜 Declarations & License](#declarations-license)
|
45 |
|
46 |
+
# Model Introduction
|
47 |
|
48 |
+
- Orion-14B series models are open-source multilingual large language models trained from scratch by OrionStarAI. The base model is trained on 2.5T multilingual corpus, including Chinese, English, Japanese, Korean, etc, and it exhibits superior performance in these languages.
|
49 |
|
50 |
+
- The Orion-14B series models exhibit the following features:
|
51 |
+
- Among models with 20B-parameter scale level, Orion-14B-Base model shows outstanding performance in comprehensive evaluations.
|
52 |
+
- Strong multilingual capabilities, significantly outperforming in Japanese and Korean testsets.
|
53 |
+
- The fine-tuned models demonstrate strong adaptability, excelling in human-annotated blind tests.
|
54 |
+
- The long-chat version supports extremely long texts, extending up to 200K tokens.
|
55 |
+
- The quantized versions reduce model size by 70%, improve inference speed by 30%, with performance loss less than 1%.
|
56 |
+
![](./assets/imgs/model_cap_en.png)
|
57 |
|
58 |
+
- Orion-14B series models including:
|
59 |
+
- **Orion-14B-Base:** A multilingual large language foundational model with 14 billion parameters, pretrained on a diverse dataset of 2.5 trillion tokens.
|
60 |
+
- **Orion-14B-Chat:** A chat-model fine-tuned on a high-quality corpus aims to provide an excellence interactive experience for users in the large model community.
|
61 |
+
- **Orion-14B-LongChat:** This model is optimized for long context lengths more than 200k tokens and demonstrates performance comparable to proprietary models on long context evaluation sets.
|
62 |
+
- **Orion-14B-Chat-RAG:** A chat-model fine-tuned on a custom retrieval augmented generation dataset, achieving superior performance in retrieval augmented generation tasks.
|
63 |
+
- **Orion-14B-Chat-Plugin:** A chat-model specifically tailored for plugin and function calling tasks, ideal for agent-related scenarios where the LLM acts as a plugin and function call system.
|
64 |
+
- **Orion-14B-Base-Int4:** A quantized base model utilizing 4-bit integer weights. It significantly reduces the model size by 70% and increases the inference speed by 30% while incurring a minimal performance loss of only 1%.
|
65 |
+
- **Orion-14B-Chat-Int4:** A quantized chat model utilizing 4-bit integer weights.
|
66 |
|
67 |
+
# Model Download
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
68 |
|
69 |
+
Model release and download links are provided in the table below:
|
70 |
|
71 |
+
| Model Name | HuggingFace Download Links | ModelScope Download Links |
|
72 |
+
|-------------------------|-----------------------------------------------------------------------------------|-------------------------------------------------------------------------------------------------|
|
73 |
+
| ⚾Orion-14B-Base | [Orion-14B-Base](https://huggingface.co/OrionStarAI/Orion-14B-Base) | [Orion-14B-Base](https://modelscope.cn/models/OrionStarAI/Orion-14B-Base/summary) |
|
74 |
+
| 😛Orion-14B-Chat | [Orion-14B-Chat](https://huggingface.co/OrionStarAI/Orion-14B-Chat) | [Orion-14B-Chat](https://modelscope.cn/models/OrionStarAI/Orion-14B-Chat/summary) |
|
75 |
+
| 📃Orion-14B-LongChat | [Orion-14B-LongChat](https://huggingface.co/OrionStarAI/Orion-14B-LongChat) | [Orion-14B-LongChat](https://modelscope.cn/models/OrionStarAI/Orion-14B-LongChat/summary) |
|
76 |
+
| 🔎Orion-14B-Chat-RAG | [Orion-14B-Chat-RAG](https://huggingface.co/OrionStarAI/Orion-14B-Chat-RAG) | [Orion-14B-Chat-RAG](https://modelscope.cn/models/OrionStarAI/Orion-14B-Chat-RAG/summary) |
|
77 |
+
| 🔌Orion-14B-Chat-Plugin | [Orion-14B-Chat-Plugin](https://huggingface.co/OrionStarAI/Orion-14B-Chat-Plugin) | [Orion-14B-Chat-Plugin](https://modelscope.cn/models/OrionStarAI/Orion-14B-Chat-Plugin/summary) |
|
78 |
+
| 💼Orion-14B-Base-Int4 | [Orion-14B-Base-Int4](https://huggingface.co/OrionStarAI/Orion-14B-Base-Int4) | [Orion-14B-Base-Int4](https://modelscope.cn/models/OrionStarAI/Orion-14B-Base-Int4/summary) |
|
79 |
+
| 📦Orion-14B-Chat-Int4 | [Orion-14B-Chat-Int4](https://huggingface.co/OrionStarAI/Orion-14B-Chat-Int4) | [Orion-14B-Chat-Int4](https://modelscope.cn/models/OrionStarAI/Orion-14B-Chat-Int4/summary) |
|
80 |
|
81 |
+
# Model Benchmarks
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
82 |
|
83 |
+
## 1. Base Model Benchmarks
|
84 |
+
### LLM evaluation results on examination and professional knowledge
|
85 |
+
| Model | C-Eval | CMMLU | MMLU | AGIEval | Gaokao | BBH |
|
|
|
86 |
|--------------------|----------|----------|----------|----------|----------|----------|
|
87 |
| LLaMA2-13B | 41.4 | 38.4 | 55.0 | 30.9 | 18.2 | 45.6 |
|
88 |
| Skywork-13B | 59.1 | 61.4 | 62.7 | 43.6 | 56.1 | 48.3 |
|
89 |
| Baichuan2-13B | 59.0 | 61.3 | 59.5 | 37.4 | 45.6 | 49.0 |
|
90 |
| QWEN-14B | 71.7 | 70.2 | 67.9 | 51.9 | **62.5** | 53.7 |
|
91 |
| InternLM-20B | 58.8 | 59.0 | 62.1 | 44.6 | 45.5 | 52.5 |
|
92 |
+
| **Orion-14B-Base** | **72.9** | **70.6** | **69.9** | **54.7** | 62.1 | **56.5** |
|
|
|
93 |
|
94 |
+
### LLM evaluation results on language understanding and common knowledge
|
95 |
+
| Model |RACE-middle|RACE-high |HellaSwag | PIQA | Lambada | WSC |
|
96 |
|--------------------|----------|----------|----------|----------|----------|----------|
|
97 |
| LLaMA 2-13B | 63.0 | 58.9 | 77.5 | 79.8 | 76.5 | 66.3 |
|
98 |
| Skywork-13B | 87.6 | 84.1 | 73.7 | 78.3 | 71.8 | 66.3 |
|
99 |
| Baichuan 2-13B | 68.9 | 67.2 | 70.8 | 78.1 | 74.1 | 66.3 |
|
100 |
| QWEN-14B | 93.0 | 90.3 | **80.2** | 79.8 | 71.4 | 66.3 |
|
101 |
| InternLM-20B | 86.4 | 83.3 | 78.1 | **80.3** | 71.8 | 68.3 |
|
102 |
+
| **Orion-14B-Base** | **93.3** | **91.3** | 78.5 | 79.5 | **78.9** | **70.2** |
|
103 |
+
|
104 |
+
### LLM evaluation results of OpenCompass testsets
|
105 |
+
| Model | Average | Examination | Language | Knowledge | Understanding | Reasoning |
|
106 |
+
|------------------|----------|----------|----------|----------|----------|----------|
|
107 |
+
| LLaMA 2-13B | 47.3 | 45.2 | 47.0 | 58.3 | 50.9 | 43.6 |
|
108 |
+
| Skywork-13B | 53.6 | 61.1 | 51.3 | 52.7 | 64.5 | 45.2 |
|
109 |
+
| Baichuan 2-13B | 49.4 | 51.8 | 47.5 | 48.9 | 58.1 | 44.2 |
|
110 |
+
| QWEN-14B | 62.4 | 71.3 | 52.67 | 56.1 | 68.8 | 60.1 |
|
111 |
+
| InternLM-20B | 59.4 | 62.5 | 55.0 | **60.1** | 67.3 | 54.9 |
|
112 |
+
|**Orion-14B-Base**| **64.4** | **71.4** | **55.0** | 60.0 | **71.9** | **61.6** |
|
113 |
+
|
114 |
+
### Comparison of LLM performances on Japanese testsets
|
115 |
+
| Model |**Average**| JCQA | JNLI | MARC | JSQD | JQK | XLS | XWN | MGSM |
|
|
|
|
|
116 |
|--------------------|----------|----------|----------|----------|----------|----------|----------|----------|----------|
|
117 |
| PLaMo-13B | 52.3 | 56.7 | 42.8 | 95.8 | 70.6 | 71.0 | 8.70 | 70.5 | 2.40 |
|
118 |
| WebLab-10B | 50.7 | 66.6 | 53.7 | 82.1 | 62.9 | 56.2 | 10.0 | 72.0 | 2.40 |
|
|
|
122 |
| Baichuan 2-13B | 57.1 | 73.7 | 31.3 | 91.6 | 80.5 | 63.3 | 18.6 | 72.2 | 25.2 |
|
123 |
| QWEN-14B | 65.8 | 85.9 | 60.7 | 97.0 | 83.3 | 71.8 | 18.8 | 70.6 | 38.0 |
|
124 |
| Yi-34B | 67.1 | 83.8 | 61.2 | 95.2 | **86.1** | 78.5 | **27.2** | 69.2 | 35.2 |
|
125 |
+
| **Orion-14B-Base** | **69.1** | **88.2** | **75.8** | 94.1 | 75.7 | **85.1** | 17.3 | **78.8** | **38.0** |
|
126 |
+
|
127 |
+
### Comparison of LLM performances on Korean testsets. n = 0 and n = 5 stand for n-shot prompts used in the evaluation
|
128 |
+
|Model | **Average**<br>n=0 n=5 | HellaSwag<br>n=0 n=5 | COPA<br> n=0 n=5 | BooIQ<br>n=0 n=5 | SentiNeg<br>n=0 n=5|
|
129 |
+
|------------------|------------------------------|------------------------------|------------------------------|------------------------------|------------------------------|
|
130 |
+
| KoGPT | 53.0 70.1 | 55.9 58.3 | 73.5 72.9 | 45.1 59.8 | 37.5 89.4 |
|
131 |
+
| Polyglot-ko-13B | 69.6 73.7 |**59.5** **63.1**|**79.4** **81.1**| 48.2 60.4 | 91.2 90.2 |
|
132 |
+
| LLaMA 2-13B | 46.7 63.7 | 41.3 44.0 | 59.3 63.8 | 34.9 73.8 | 51.5 73.4 |
|
133 |
+
| Baichuan 2-13B | 52.1 58.7 | 39.2 39.6 | 60.6 60.6 | 58.4 61.5 | 50.3 72.9 |
|
134 |
+
| QWEN-14B | 53.8 73.7 | 45.3 46.8 | 64.9 68.9 | 33.4 83.5 | 71.5 95.7 |
|
135 |
+
| Yi-34B | 54.2 72.1 | 44.6 44.7 | 58.0 60.6 | 65.9 90.2 | 48.3 92.9 |
|
136 |
+
|**Orion-14B-Chat**|**74.5** **79.6**| 47.0 49.6 | 77.7 79.4 |**81.6** **90.7**|**92.4** **98.7**|
|
137 |
+
|
138 |
+
### Multilingual evaluation
|
139 |
+
| Model | Train Lang | Japanese | Korean | Chinese | English |
|
|
|
140 |
|--------------------|------------|----------|----------|----------|----------|
|
141 |
| PLaMo-13B | En,Jp | 52.3 | * | * | * |
|
142 |
| Weblab-10B | En,Jp | 50.7 | * | * | * |
|
|
|
148 |
| Qwen-14B | Multi | 65.8 | 73.7 | 64.5 | 65.4 |
|
149 |
| Llama2-13B | Multi | 46.3 | 63.7 | 41.4 | 55.3 |
|
150 |
| Yi-34B | Multi | 67.1 | 72.2 | 58.7 | **68.8** |
|
151 |
+
| **Orion-14B-Chat** | Multi | **69.1** | **79.5** | **67.9** | 67.3 |
|
152 |
+
|
153 |
+
|
154 |
+
## 2. Chat Model Benchmarks
|
155 |
+
### Chat model subjective evaluation of MTBench
|
156 |
+
| Model | First-Turn | Second-Turn | **Average** |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
157 |
|----------------------|----------|----------|----------|
|
158 |
| Baichuan2-13B-Chat | 7.05 | 6.47 | 6.76 |
|
159 |
| Qwen-14B-Chat | 7.30 | 6.62 | 6.96 |
|
160 |
| Llama2-13B-Chat | 7.10 | 6.20 | 6.65 |
|
161 |
| InternLM-20B-Chat | 7.03 | 5.93 | 6.48 |
|
162 |
| **Orion-14B-Chat** | **7.68** | **7.07** | **7.37** |
|
163 |
+
\* use vllm for inference
|
164 |
+
|
165 |
+
### Chat model subjective evaluation of AlignBench
|
166 |
+
| Model | Math. | Logi. | Basic. | Chi. | Comp. | Writ. | Role. | Prof. |**Avg.**|
|
167 |
+
|--------------------|--------|--------|--------|--------|--------|--------|--------|--------|--------|
|
168 |
+
| Baichuan2-13B-Chat | 3.76 | 4.07 | 6.22 | 6.05 | 7.11 | 6.97 | 6.75 | 6.43 | 5.25 |
|
169 |
+
| Qwen-14B-Chat |**4.91**|**4.71**|**6.90**| 6.36 | 6.74 | 6.64 | 6.59 | 6.56 |**5.72**|
|
170 |
+
| Llama2-13B-Chat | 3.05 | 3.79 | 5.43 | 4.40 | 6.76 | 6.63 | 6.99 | 5.65 | 4.70 |
|
171 |
+
| InternLM-20B-Chat | 3.39 | 3.92 | 5.96 | 5.50 |**7.18**| 6.19 | 6.49 | 6.22 | 4.96 |
|
172 |
+
| **Orion-14B-Chat** | 4.00 | 4.24 | 6.18 |**6.57**| 7.16 |**7.36**|**7.16**|**6.99**| 5.51 |
|
173 |
+
|
174 |
+
\* use vllm for inference
|
175 |
+
|
176 |
+
## 3. LongChat Model Benchmarks
|
177 |
+
### LongChat evaluation of LongBench
|
178 |
+
| Model | NarrativeQA|MultiFieldQA-en|MultiFieldQA-zh| DuReader | QMSum | VCSUM | TREC | TriviaQA | LSHT |RepoBench-P|
|
179 |
+
|--------------------------|-----------|-----------|-----------|-----------|-----------|-----------|-----------|-----------|-----------|-----------|
|
180 |
+
| GPT-3.5-Turbo-16k | **23.60** | **52.30** | **61.20** | 28.70 | 23.40 | **16.00** | 68.00 | **91.40** | 29.20 | 53.60 |
|
181 |
+
| LongChat-v1.5-7B-32k | 16.90 | 41.40 | 29.10 | 19.50 | 22.70 | 9.90 | 63.50 | 82.30 | 23.20 | 55.30 |
|
182 |
+
| Vicuna-v1.5-7B-16k | 19.40 | 38.50 | 43.00 | 19.30 | 22.80 | 15.10 | 71.50 | 86.20 | 28.80 | 43.50 |
|
183 |
+
| Yi-6B-200K | 14.11 | 36.74 | 22.68 | 14.01 | 20.44 | 8.08 | 72.00 | 86.61 | 38.00 | **63.29** |
|
184 |
+
| Orion-14B-LongChat | 19.47 | 48.11 | 55.84 | **37.02** | **24.87** | 15.44 | **77.00** | 89.12 | **45.50** | 54.31 |
|
185 |
+
|
186 |
+
|
187 |
+
## 4. Chat RAG Model Benchmarks
|
188 |
+
### LLM evaluation results of self-built RAG testsets
|
189 |
+
|Model|Effectiveness of Response(Keyword)|*Effectiveness of Response(subjective evaluation)|Quoting Ability|Fallback Ability|*AutoQA|*Data Extraction|
|
190 |
+
|---------------------|------|------|------|------|------|------|
|
191 |
+
| Baichuan2-13B-Chat | 85 | 76 | 1 | 0 | 69 | 51 |
|
192 |
+
| Qwen-14B-Chat | 79 | 77 | 75 | 47 | 68 | 72 |
|
193 |
+
| Qwen-72B-Chat(Int4) | 87 | 89 | 90 | 32 | 67 | 76 |
|
194 |
+
| GPT-4 | 91 | 94 | 96 | 95 | 75 | 86 |
|
195 |
+
| Orion-14B-Chat-RAG | 86 | 87 | 91 | 97 | 73 | 71 |
|
196 |
+
\* means manual assessment
|
197 |
+
|
198 |
+
## 5. Chat Plugin Model Benchmarks
|
199 |
+
### LLM evaluation results of self-built plugin testsets
|
200 |
+
|Model |Intent Recognition with Full Params |Intent Recognition with Missing Params |Non-Plugin Invocation Recognition |
|
201 |
+
|-----------------------|--------|-----------|--------|
|
202 |
+
| Baichuan2-13B-Chat | 25 | 0 | 0 |
|
203 |
+
| Qwen-14B-Chat | 55 | 0 | 50 |
|
204 |
+
| GPT-4 | **95** | 52.38 | 70 |
|
205 |
+
| Orion-14B-Chat-Plugin | 92.5 | **60.32** | **90** |
|
206 |
+
|
207 |
+
## 6. Quantized Model Benchmarks
|
208 |
+
### Comparison of before and after quantization
|
209 |
+
|Model |Size(GB)|Inference Speed(tokens/s)|C-Eval|CMMLU|MMLU|RACE|HellaSwag|
|
210 |
+
|-------------------------|-------|-----|------|------|------|------|------|
|
211 |
+
| OrionStar-14B-Base | 28.0 | 135 | 72.8 | 70.6 | 70.0 | 93.3 | 78.5 |
|
212 |
+
| OrionStar-14B-Base-Int4 | 8.3 | 178 | 71.8 | 69.8 | 69.2 | 93.1 | 78.0 |
|
213 |
+
|
214 |
+
# Model Inference
|
215 |
+
|
216 |
+
Model weights, source code, and configuration needed for inference are published on Hugging Face, and the download link
|
217 |
+
is available in the table at the beginning of this document. We demonstrate various inference methods here, and the
|
218 |
+
program will automatically download the necessary resources from Hugging Face.
|
219 |
+
|
220 |
+
## Python Code
|
221 |
|
222 |
```python
|
223 |
import torch
|
|
|
229 |
torch_dtype=torch.bfloat16, trust_remote_code=True)
|
230 |
|
231 |
model.generation_config = GenerationConfig.from_pretrained("OrionStarAI/Orion-14B")
|
232 |
+
messages = [{"role": "user", "content": "Hello, what is your name? "}]
|
233 |
+
response = model.chat(tokenizer, messages, streaming=False)
|
234 |
print(response)
|
235 |
|
|
|
236 |
```
|
237 |
|
238 |
+
In the above Python code, the model is loaded with `device_map='auto'` to utilize all available GPUs. To specify the
|
239 |
+
device, you can use something like `export CUDA_VISIBLE_DEVICES=0,1` (using GPUs 0 and 1).
|
240 |
|
241 |
+
## Command Line Tool
|
242 |
|
243 |
```shell
|
244 |
CUDA_VISIBLE_DEVICES=0 python cli_demo.py
|
245 |
```
|
246 |
|
247 |
+
This command-line tool is designed for chat scenarios, and thus, it does not support calling the base model.
|
248 |
|
249 |
+
## Direct Script Inference
|
250 |
|
251 |
```shell
|
252 |
+
|
253 |
+
# base model
|
254 |
+
CUDA_VISIBLE_DEVICES=0 python demo/text_generation_base.py --model OrionStarAI/Orion-14B --tokenizer OrionStarAI/Orion-14B --prompt hello
|
255 |
|
256 |
# chat model
|
257 |
+
CUDA_VISIBLE_DEVICES=0 python demo/text_generation.py --model OrionStarAI/Orion-14B-Chat --tokenizer OrionStarAI/Orion-14B-Chat --prompt hi
|
258 |
+
|
259 |
```
|
260 |
|
261 |
+
# Example Output
|
262 |
|
263 |
+
## Casual Chat
|
264 |
|
265 |
`````
|
266 |
+
User: Hello
|
267 |
+
Orion-14B: Hello! How can I help you today? If you have any questions or need assistance, feel free to ask.
|
268 |
+
User: What functions do you have?
|
269 |
+
Orion-14B: As an AI language model, I have a wide range of functions and capabilities, including but not limited to:
|
270 |
+
|
271 |
+
Answering general knowledge questions: I can provide information on various topics such as history, science, geography, and more.
|
272 |
+
Providing definitions and explanations: I can explain complex concepts and provide definitions for words and phrases.
|
273 |
+
Assisting with tasks: I can help with simple tasks like setting reminders, converting units of measurement, and even ordering food from restaurants.
|
274 |
+
Generating text: I can write short stories, essays, articles, and other types of content based on prompts given to me.
|
275 |
+
Translating languages: I can translate text from one language to another.
|
276 |
+
Chatting: I can engage in conversation with users and respond to their queries in a conversational manner.
|
277 |
+
Summarizing long texts: I can summarize longer texts like articles, books, and research papers.
|
278 |
+
Generating random numbers and names: I can generate random numbers within a specified range and also suggest names based on certain criteria.
|
279 |
+
Playing games: I can play simple games like Hangman, Trivia, and more.
|
280 |
+
Providing recommendations: I can recommend movies, music, books, and other forms of entertainment based on user preferences.
|
281 |
+
These are just some of the many functions that I am capable of performing. If you have a specific task in mind, feel free to ask me about it.
|
282 |
+
User: Tell me a joke.
|
283 |
+
Orion-14B: Sure, here's a classic one-liner: Why don't scientists trust atoms? Because they make up everything.
|
|
|
|
|
|
|
|
|
284 |
`````
|
285 |
|
286 |
+
## Japanese & Korean Chat
|
287 |
|
288 |
`````
|
289 |
+
User:自己を紹介してください
|
290 |
Orion-14B:こんにちは、私の名前はChatMaxで、OrionStarによって開発されたAIアシスタントです。私はユーザーに正確で、客観的で、有用な回答を提供し、彼らが問題を解決し、情報を取得するのを助けることができます。
|
291 |
+
User:자기소개를 해주세요.
|
292 |
Orion-14B:안녕하세요, 제 이름은 ChatMax입니다. OrionStar에서 개발한 AI 어시스턴트입니다. 저는 사용자들에게 정확하고 객관적이며 유용한 답변을 제공하여 문제를 해결하고 정보를 얻는 데 도움을 줄 수 있습니다.
|
293 |
`````
|
294 |
|
295 |
+
# Company Introduction
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
296 |
|
297 |
+
OrionStar is a leading global service robot solutions company, founded in September 2016. OrionStar is dedicated to
|
298 |
+
using artificial intelligence technology to create the next generation of revolutionary robots, allowing people to break
|
299 |
+
free from repetitive physical labor and making human work and life more intelligent and enjoyable. Through technology,
|
300 |
+
OrionStar aims to make society and the world a better place.
|
301 |
|
302 |
+
OrionStar possesses fully self-developed end-to-end artificial intelligence technologies, such as voice interaction and
|
303 |
+
visual navigation. It integrates product development capabilities and technological application capabilities. Based on
|
304 |
+
the Orion robotic arm platform, it has launched products such as OrionStar AI Robot Greeting, AI Robot Greeting Mini,
|
305 |
+
Lucki, Coffee Master, and established the open platform OrionOS for Orion robots. Following the philosophy of "Born for
|
306 |
+
Truly Useful Robots", OrionStar empowers more people through AI technology.
|
307 |
|
308 |
+
# Declarations, License
|
309 |
|
310 |
+
## Declarations
|
|
|
311 |
|
312 |
+
We strongly urge all users not to use the Orion-14B model for any activities that may harm national or social security or violate the law.
|
313 |
+
Additionally, we request users not to use the Orion-14B model for internet services without proper security review and filing.
|
314 |
+
We hope all users abide by this principle to ensure that technological development takes place in a regulated and legal environment.
|
315 |
+
We have done our best to ensure the compliance of the data used in the model training process. However, despite our
|
316 |
+
significant efforts, unforeseen issues may still arise due to the complexity of the model and data. Therefore, if any
|
317 |
+
problems arise due to the use of the Orion-14B open-source model, including but not limited to data security
|
318 |
+
issues, public opinion risks, or any risks and issues arising from the model being misled, abused, disseminated, or
|
319 |
+
improperly utilized, we will not assume any responsibility.
|
320 |
|
321 |
+
## License
|
322 |
|
323 |
+
Community use of the Orion-14B series models
|
324 |
+
- For code, please comply with [Apache License Version 2.0](./LICENSE)<br>
|
325 |
+
- For model, please comply with [【Orion-14B Series】 Models Community License Agreement](./ModelsCommunityLicenseAgreement)
|
326 |
|
327 |
+
# Contact Us
|
328 |
|
329 |
+
Email: ai@orionstar.com
|
330 |
|
331 |
![](./assets/imgs/wechat_group.jpg)
|
332 |
|