DachengZhang commited on
Commit
e822f1d
1 Parent(s): 0641956

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +66 -60
README.md CHANGED
@@ -43,9 +43,9 @@ pipeline_tag: text-generation
43
  - [🥇 Company Introduction](#company-introduction)
44
  - [📜 Declarations & License](#declarations-license)
45
 
46
- # Model Introduction
47
 
48
- - Orion-14B-Chat is fine-tuned from Orion-14B-Base using a high-quality corpus of approximately 850,000 entries (only sft), and it also supports Chinese, English, Japanese, and Korean. It performs exceptionally well on the MT-Bench and AlignBench evaluation sets, significantly surpassing other models of the same parameter scale in multiple metrics.
49
 
50
  - The 850,000 fine-tuning corpus comprises two parts: approximately 220,000 manually curated high-quality datasets and 630,000 entries selected and semantically deduplicated from open-source data through model filtering. Among these, the Japanese and Korean data, totaling 70,000 entries, have only undergone basic cleaning and deduplication.
51
 
@@ -55,7 +55,9 @@ pipeline_tag: text-generation
55
  - The fine-tuned models demonstrate strong adaptability, excelling in human-annotated blind tests.
56
  - The long-chat version supports extremely long texts, extending up to 200K tokens.
57
  - The quantized versions reduce model size by 70%, improve inference speed by 30%, with performance loss less than 1%.
58
- ![](./assets/imgs/model_cap_en.png)
 
 
59
 
60
  - Orion-14B series models including:
61
  - **Orion-14B-Base:** A multilingual large language foundational model with 14 billion parameters, pretrained on a diverse dataset of 2.5 trillion tokens.
@@ -66,7 +68,7 @@ pipeline_tag: text-generation
66
  - **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%.
67
  - **Orion-14B-Chat-Int4:** A quantized chat model utilizing 4-bit integer weights.
68
 
69
- # Model Download
70
 
71
  Model release and download links are provided in the table below:
72
 
@@ -80,10 +82,10 @@ Model release and download links are provided in the table below:
80
  | 💼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) |
81
  | 📦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) |
82
 
83
- # Model Benchmarks
84
 
85
- ## 1. Base Model Benchmarks
86
- ### LLM evaluation results on examination and professional knowledge
87
  | Model | C-Eval | CMMLU | MMLU | AGIEval | Gaokao | BBH |
88
  |--------------------|----------|----------|----------|----------|----------|----------|
89
  | LLaMA2-13B | 41.4 | 38.4 | 55.0 | 30.9 | 18.2 | 45.6 |
@@ -93,7 +95,7 @@ Model release and download links are provided in the table below:
93
  | InternLM-20B | 58.8 | 59.0 | 62.1 | 44.6 | 45.5 | 52.5 |
94
  | **Orion-14B-Base** | **72.9** | **70.6** | **69.9** | **54.7** | 62.1 | **56.5** |
95
 
96
- ### LLM evaluation results on language understanding and common knowledge
97
  | Model |RACE-middle|RACE-high |HellaSwag | PIQA | Lambada | WSC |
98
  |--------------------|----------|----------|----------|----------|----------|----------|
99
  | LLaMA 2-13B | 63.0 | 58.9 | 77.5 | 79.8 | 76.5 | 66.3 |
@@ -103,7 +105,7 @@ Model release and download links are provided in the table below:
103
  | InternLM-20B | 86.4 | 83.3 | 78.1 | **80.3** | 71.8 | 68.3 |
104
  | **Orion-14B-Base** | **93.3** | **91.3** | 78.5 | 79.5 | **78.9** | **70.2** |
105
 
106
- ### LLM evaluation results of OpenCompass testsets
107
  | Model | Average | Examination | Language | Knowledge | Understanding | Reasoning |
108
  |------------------|----------|----------|----------|----------|----------|----------|
109
  | LLaMA 2-13B | 47.3 | 45.2 | 47.0 | 58.3 | 50.9 | 43.6 |
@@ -113,7 +115,7 @@ Model release and download links are provided in the table below:
113
  | InternLM-20B | 59.4 | 62.5 | 55.0 | **60.1** | 67.3 | 54.9 |
114
  |**Orion-14B-Base**| **64.4** | **71.4** | **55.0** | 60.0 | **71.9** | **61.6** |
115
 
116
- ### Comparison of LLM performances on Japanese testsets
117
  | Model |**Average**| JCQA | JNLI | MARC | JSQD | JQK | XLS | XWN | MGSM |
118
  |--------------------|----------|----------|----------|----------|----------|----------|----------|----------|----------|
119
  | PLaMo-13B | 52.3 | 56.7 | 42.8 | 95.8 | 70.6 | 71.0 | 8.70 | 70.5 | 2.40 |
@@ -126,7 +128,7 @@ Model release and download links are provided in the table below:
126
  | Yi-34B | 67.1 | 83.8 | 61.2 | 95.2 | **86.1** | 78.5 | **27.2** | 69.2 | 35.2 |
127
  | **Orion-14B-Base** | **69.1** | **88.2** | **75.8** | 94.1 | 75.7 | **85.1** | 17.3 | **78.8** | **38.0** |
128
 
129
- ### Comparison of LLM performances on Korean testsets. n = 0 and n = 5 stand for n-shot prompts used in the evaluation
130
  |Model | **Average**<br>n=0&nbsp;&nbsp;n=5 | HellaSwag<br>n=0&nbsp;&nbsp;n=5 | COPA<br> n=0&nbsp;&nbsp;n=5 | BooIQ<br>n=0&nbsp;&nbsp;n=5 | SentiNeg<br>n=0&nbsp;&nbsp;n=5|
131
  |------------------|------------------------------|------------------------------|------------------------------|------------------------------|------------------------------|
132
  | KoGPT | 53.0 &nbsp;&nbsp; 70.1 | 55.9 &nbsp;&nbsp; 58.3 | 73.5 &nbsp;&nbsp; 72.9 | 45.1 &nbsp;&nbsp; 59.8 | 37.5 &nbsp;&nbsp; 89.4 |
@@ -137,7 +139,7 @@ Model release and download links are provided in the table below:
137
  | Yi-34B | 54.2 &nbsp;&nbsp; 72.1 | 44.6 &nbsp;&nbsp; 44.7 | 58.0 &nbsp;&nbsp; 60.6 | 65.9 &nbsp;&nbsp; 90.2 | 48.3 &nbsp;&nbsp; 92.9 |
138
  |**Orion-14B-Chat**|**74.5** &nbsp;&nbsp; **79.6**| 47.0 &nbsp;&nbsp; 49.6 | 77.7 &nbsp;&nbsp; 79.4 |**81.6** &nbsp;&nbsp; **90.7**|**92.4** &nbsp;&nbsp; **98.7**|
139
 
140
- ### Multilingual evaluation
141
  | Model | Train Lang | Japanese | Korean | Chinese | English |
142
  |--------------------|------------|----------|----------|----------|----------|
143
  | PLaMo-13B | En,Jp | 52.3 | * | * | * |
@@ -153,8 +155,8 @@ Model release and download links are provided in the table below:
153
  | **Orion-14B-Chat** | Multi | **69.1** | **79.5** | **67.9** | 67.3 |
154
 
155
 
156
- ## 2. Chat Model Benchmarks
157
- ### Chat model subjective evaluation of MTBench
158
  | Model | First-Turn | Second-Turn | **Average** |
159
  |----------------------|----------|----------|----------|
160
  | Baichuan2-13B-Chat | 7.05 | 6.47 | 6.76 |
@@ -164,7 +166,7 @@ Model release and download links are provided in the table below:
164
  | **Orion-14B-Chat** | **7.68** | **7.07** | **7.37** |
165
  \* use vllm for inference
166
 
167
- ### Chat model subjective evaluation of AlignBench
168
  | Model | Math. | Logi. | Basic. | Chi. | Comp. | Writ. | Role. | Prof. |**Avg.**|
169
  |--------------------|--------|--------|--------|--------|--------|--------|--------|--------|--------|
170
  | Baichuan2-13B-Chat | 3.76 | 4.07 | 6.22 | 6.05 | 7.11 | 6.97 | 6.75 | 6.43 | 5.25 |
@@ -175,8 +177,8 @@ Model release and download links are provided in the table below:
175
 
176
  \* use vllm for inference
177
 
178
- ## 3. LongChat Model Benchmarks
179
- ### LongChat evaluation of LongBench
180
  | Model | NarrativeQA|MultiFieldQA-en|MultiFieldQA-zh| DuReader | QMSum | VCSUM | TREC | TriviaQA | LSHT |RepoBench-P|
181
  |--------------------------|-----------|-----------|-----------|-----------|-----------|-----------|-----------|-----------|-----------|-----------|
182
  | 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 |
@@ -186,8 +188,8 @@ Model release and download links are provided in the table below:
186
  | Orion-14B-LongChat | 19.47 | 48.11 | 55.84 | **37.02** | **24.87** | 15.44 | **77.00** | 89.12 | **45.50** | 54.31 |
187
 
188
 
189
- ## 4. Chat RAG Model Benchmarks
190
- ### LLM evaluation results of self-built RAG testsets
191
  |Model|Effectiveness of Response(Keyword)|*Effectiveness of Response(subjective evaluation)|Quoting Ability|Fallback Ability|*AutoQA|*Data Extraction|
192
  |---------------------|------|------|------|------|------|------|
193
  | Baichuan2-13B-Chat | 85 | 76 | 1 | 0 | 69 | 51 |
@@ -197,8 +199,8 @@ Model release and download links are provided in the table below:
197
  | Orion-14B-Chat-RAG | 86 | 87 | 91 | 97 | 73 | 71 |
198
  \* means manual assessment
199
 
200
- ## 5. Chat Plugin Model Benchmarks
201
- ### LLM evaluation results of self-built plugin testsets
202
  |Model |Intent Recognition with Full Params |Intent Recognition with Missing Params |Non-Plugin Invocation Recognition |
203
  |-----------------------|--------|-----------|--------|
204
  | Baichuan2-13B-Chat | 25 | 0 | 0 |
@@ -206,20 +208,20 @@ Model release and download links are provided in the table below:
206
  | GPT-4 | **95** | 52.38 | 70 |
207
  | Orion-14B-Chat-Plugin | 92.5 | **60.32** | **90** |
208
 
209
- ## 6. Quantized Model Benchmarks
210
- ### Comparison of before and after quantization
211
  |Model |Size(GB)|Inference Speed(tokens/s)|C-Eval|CMMLU|MMLU|RACE|HellaSwag|
212
  |-------------------------|-------|-----|------|------|------|------|------|
213
  | OrionStar-14B-Base | 28.0 | 135 | 72.8 | 70.6 | 70.0 | 93.3 | 78.5 |
214
  | OrionStar-14B-Base-Int4 | 8.3 | 178 | 71.8 | 69.8 | 69.2 | 93.1 | 78.0 |
215
 
216
- # Model Inference
217
 
218
  Model weights, source code, and configuration needed for inference are published on Hugging Face, and the download link
219
  is available in the table at the beginning of this document. We demonstrate various inference methods here, and the
220
  program will automatically download the necessary resources from Hugging Face.
221
 
222
- ## Python Code
223
 
224
  ```python
225
  import torch
@@ -240,7 +242,7 @@ print(response)
240
  In the above Python code, the model is loaded with `device_map='auto'` to utilize all available GPUs. To specify the
241
  device, you can use something like `export CUDA_VISIBLE_DEVICES=0,1` (using GPUs 0 and 1).
242
 
243
- ## Command Line Tool
244
 
245
  ```shell
246
  CUDA_VISIBLE_DEVICES=0 python cli_demo.py
@@ -248,7 +250,7 @@ CUDA_VISIBLE_DEVICES=0 python cli_demo.py
248
 
249
  This command-line tool is designed for chat scenarios, and thus, it does not support calling the base model.
250
 
251
- ## Direct Script Inference
252
 
253
  ```shell
254
 
@@ -260,32 +262,31 @@ CUDA_VISIBLE_DEVICES=0 python demo/text_generation.py --model OrionStarAI/Orion-
260
 
261
  ```
262
 
263
- # Example Output
264
 
265
- ## Casual Chat
266
 
267
  `````
268
  User: Hello
269
  Orion-14B: Hello! How can I help you today? If you have any questions or need assistance, feel free to ask.
270
  User: What functions do you have?
271
  Orion-14B: As an AI language model, I have a wide range of functions and capabilities, including but not limited to:
272
-
273
- Answering general knowledge questions: I can provide information on various topics such as history, science, geography, and more.
274
- Providing definitions and explanations: I can explain complex concepts and provide definitions for words and phrases.
275
- Assisting with tasks: I can help with simple tasks like setting reminders, converting units of measurement, and even ordering food from restaurants.
276
- Generating text: I can write short stories, essays, articles, and other types of content based on prompts given to me.
277
- Translating languages: I can translate text from one language to another.
278
- Chatting: I can engage in conversation with users and respond to their queries in a conversational manner.
279
- Summarizing long texts: I can summarize longer texts like articles, books, and research papers.
280
- Generating random numbers and names: I can generate random numbers within a specified range and also suggest names based on certain criteria.
281
- Playing games: I can play simple games like Hangman, Trivia, and more.
282
- Providing recommendations: I can recommend movies, music, books, and other forms of entertainment based on user preferences.
283
- 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.
284
  User: Tell me a joke.
285
  Orion-14B: Sure, here's a classic one-liner: Why don't scientists trust atoms? Because they make up everything.
286
  `````
287
 
288
- ## Japanese & Korean Chat
289
 
290
  `````
291
  User:自己を紹介してください
@@ -294,22 +295,10 @@ User:자기소개를 해주세요.
294
  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.
295
  `````
296
 
297
- # Company Introduction
298
-
299
- OrionStar is a leading global service robot solutions company, founded in September 2016. OrionStar is dedicated to
300
- using artificial intelligence technology to create the next generation of revolutionary robots, allowing people to break
301
- free from repetitive physical labor and making human work and life more intelligent and enjoyable. Through technology,
302
- OrionStar aims to make society and the world a better place.
303
-
304
- OrionStar possesses fully self-developed end-to-end artificial intelligence technologies, such as voice interaction and
305
- visual navigation. It integrates product development capabilities and technological application capabilities. Based on
306
- the Orion robotic arm platform, it has launched products such as OrionStar AI Robot Greeting, AI Robot Greeting Mini,
307
- Lucki, Coffee Master, and established the open platform OrionOS for Orion robots. Following the philosophy of "Born for
308
- Truly Useful Robots", OrionStar empowers more people through AI technology.
309
 
310
- # Declarations, License
311
 
312
- ## Declarations
313
 
314
  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.
315
  Additionally, we request users not to use the Orion-14B model for internet services without proper security review and filing.
@@ -320,15 +309,32 @@ problems arise due to the use of the Orion-14B open-source model, including but
320
  issues, public opinion risks, or any risks and issues arising from the model being misled, abused, disseminated, or
321
  improperly utilized, we will not assume any responsibility.
322
 
323
- ## License
324
 
325
  Community use of the Orion-14B series models
326
  - For code, please comply with [Apache License Version 2.0](./LICENSE)<br>
327
  - For model, please comply with [【Orion-14B Series】 Models Community License Agreement](./ModelsCommunityLicenseAgreement)
328
 
329
- # Contact Us
330
 
331
- Email: ai@orionstar.com
332
 
333
- ![](./assets/imgs/wechat_group.jpg)
 
 
 
334
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
43
  - [🥇 Company Introduction](#company-introduction)
44
  - [📜 Declarations & License](#declarations-license)
45
 
46
+ # 1. Model Introduction
47
 
48
+ - Orion-14B-Chat is fine-tuned from Orion-14B-Base using a high-quality corpus of approximately 850,000 entries (only sft), and it also supports Chinese, English, Japanese, and Korean. It performs exceptionally well on the MT-Bench and AlignBench evaluation sets, significantly surpassing other models of the same parameter scale in multiple metrics. For details, please refer to [tech report](https://github.com/OrionStarAI/Orion/blob/master/doc/Orion14B_v3.pdf).
49
 
50
  - The 850,000 fine-tuning corpus comprises two parts: approximately 220,000 manually curated high-quality datasets and 630,000 entries selected and semantically deduplicated from open-source data through model filtering. Among these, the Japanese and Korean data, totaling 70,000 entries, have only undergone basic cleaning and deduplication.
51
 
 
55
  - The fine-tuned models demonstrate strong adaptability, excelling in human-annotated blind tests.
56
  - The long-chat version supports extremely long texts, extending up to 200K tokens.
57
  - The quantized versions reduce model size by 70%, improve inference speed by 30%, with performance loss less than 1%.
58
+ <div align="center">
59
+ <img src="./assets/imgs/model_cap_en.png" alt="model_cap_en" width="50%" />
60
+ </div>
61
 
62
  - Orion-14B series models including:
63
  - **Orion-14B-Base:** A multilingual large language foundational model with 14 billion parameters, pretrained on a diverse dataset of 2.5 trillion tokens.
 
68
  - **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%.
69
  - **Orion-14B-Chat-Int4:** A quantized chat model utilizing 4-bit integer weights.
70
 
71
+ # 2. Model Download
72
 
73
  Model release and download links are provided in the table below:
74
 
 
82
  | 💼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) |
83
  | 📦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) |
84
 
85
+ # 3. Model Benchmarks
86
 
87
+ ## 3.1. Base Model Orion-14B-Base Benchmarks
88
+ ### 3.1.1. LLM evaluation results on examination and professional knowledge
89
  | Model | C-Eval | CMMLU | MMLU | AGIEval | Gaokao | BBH |
90
  |--------------------|----------|----------|----------|----------|----------|----------|
91
  | LLaMA2-13B | 41.4 | 38.4 | 55.0 | 30.9 | 18.2 | 45.6 |
 
95
  | InternLM-20B | 58.8 | 59.0 | 62.1 | 44.6 | 45.5 | 52.5 |
96
  | **Orion-14B-Base** | **72.9** | **70.6** | **69.9** | **54.7** | 62.1 | **56.5** |
97
 
98
+ ### 3.1.2. LLM evaluation results on language understanding and common knowledge
99
  | Model |RACE-middle|RACE-high |HellaSwag | PIQA | Lambada | WSC |
100
  |--------------------|----------|----------|----------|----------|----------|----------|
101
  | LLaMA 2-13B | 63.0 | 58.9 | 77.5 | 79.8 | 76.5 | 66.3 |
 
105
  | InternLM-20B | 86.4 | 83.3 | 78.1 | **80.3** | 71.8 | 68.3 |
106
  | **Orion-14B-Base** | **93.3** | **91.3** | 78.5 | 79.5 | **78.9** | **70.2** |
107
 
108
+ ### 3.1.3. LLM evaluation results of OpenCompass testsets
109
  | Model | Average | Examination | Language | Knowledge | Understanding | Reasoning |
110
  |------------------|----------|----------|----------|----------|----------|----------|
111
  | LLaMA 2-13B | 47.3 | 45.2 | 47.0 | 58.3 | 50.9 | 43.6 |
 
115
  | InternLM-20B | 59.4 | 62.5 | 55.0 | **60.1** | 67.3 | 54.9 |
116
  |**Orion-14B-Base**| **64.4** | **71.4** | **55.0** | 60.0 | **71.9** | **61.6** |
117
 
118
+ ### 3.1.4. Comparison of LLM performances on Japanese testsets
119
  | Model |**Average**| JCQA | JNLI | MARC | JSQD | JQK | XLS | XWN | MGSM |
120
  |--------------------|----------|----------|----------|----------|----------|----------|----------|----------|----------|
121
  | PLaMo-13B | 52.3 | 56.7 | 42.8 | 95.8 | 70.6 | 71.0 | 8.70 | 70.5 | 2.40 |
 
128
  | Yi-34B | 67.1 | 83.8 | 61.2 | 95.2 | **86.1** | 78.5 | **27.2** | 69.2 | 35.2 |
129
  | **Orion-14B-Base** | **69.1** | **88.2** | **75.8** | 94.1 | 75.7 | **85.1** | 17.3 | **78.8** | **38.0** |
130
 
131
+ ### 3.1.5. Comparison of LLM performances on Korean testsets. n = 0 and n = 5 stand for n-shot prompts used in the evaluation
132
  |Model | **Average**<br>n=0&nbsp;&nbsp;n=5 | HellaSwag<br>n=0&nbsp;&nbsp;n=5 | COPA<br> n=0&nbsp;&nbsp;n=5 | BooIQ<br>n=0&nbsp;&nbsp;n=5 | SentiNeg<br>n=0&nbsp;&nbsp;n=5|
133
  |------------------|------------------------------|------------------------------|------------------------------|------------------------------|------------------------------|
134
  | KoGPT | 53.0 &nbsp;&nbsp; 70.1 | 55.9 &nbsp;&nbsp; 58.3 | 73.5 &nbsp;&nbsp; 72.9 | 45.1 &nbsp;&nbsp; 59.8 | 37.5 &nbsp;&nbsp; 89.4 |
 
139
  | Yi-34B | 54.2 &nbsp;&nbsp; 72.1 | 44.6 &nbsp;&nbsp; 44.7 | 58.0 &nbsp;&nbsp; 60.6 | 65.9 &nbsp;&nbsp; 90.2 | 48.3 &nbsp;&nbsp; 92.9 |
140
  |**Orion-14B-Chat**|**74.5** &nbsp;&nbsp; **79.6**| 47.0 &nbsp;&nbsp; 49.6 | 77.7 &nbsp;&nbsp; 79.4 |**81.6** &nbsp;&nbsp; **90.7**|**92.4** &nbsp;&nbsp; **98.7**|
141
 
142
+ ### 3.1.6. Multilingual evaluation
143
  | Model | Train Lang | Japanese | Korean | Chinese | English |
144
  |--------------------|------------|----------|----------|----------|----------|
145
  | PLaMo-13B | En,Jp | 52.3 | * | * | * |
 
155
  | **Orion-14B-Chat** | Multi | **69.1** | **79.5** | **67.9** | 67.3 |
156
 
157
 
158
+ ## 3.2. Chat Model Orion-14B-Chat Benchmarks
159
+ ### 3.2.1. Chat model subjective evaluation of MTBench
160
  | Model | First-Turn | Second-Turn | **Average** |
161
  |----------------------|----------|----------|----------|
162
  | Baichuan2-13B-Chat | 7.05 | 6.47 | 6.76 |
 
166
  | **Orion-14B-Chat** | **7.68** | **7.07** | **7.37** |
167
  \* use vllm for inference
168
 
169
+ ### 3.2.2. Chat model subjective evaluation of AlignBench
170
  | Model | Math. | Logi. | Basic. | Chi. | Comp. | Writ. | Role. | Prof. |**Avg.**|
171
  |--------------------|--------|--------|--------|--------|--------|--------|--------|--------|--------|
172
  | Baichuan2-13B-Chat | 3.76 | 4.07 | 6.22 | 6.05 | 7.11 | 6.97 | 6.75 | 6.43 | 5.25 |
 
177
 
178
  \* use vllm for inference
179
 
180
+ ## 3.3. LongChat Model Orion-14B-LongChat Benchmarks
181
+ ### 3.3.1. LongChat evaluation of LongBench
182
  | Model | NarrativeQA|MultiFieldQA-en|MultiFieldQA-zh| DuReader | QMSum | VCSUM | TREC | TriviaQA | LSHT |RepoBench-P|
183
  |--------------------------|-----------|-----------|-----------|-----------|-----------|-----------|-----------|-----------|-----------|-----------|
184
  | 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 |
 
188
  | Orion-14B-LongChat | 19.47 | 48.11 | 55.84 | **37.02** | **24.87** | 15.44 | **77.00** | 89.12 | **45.50** | 54.31 |
189
 
190
 
191
+ ## 3.4. Chat RAG Model Benchmarks
192
+ ### 3.4.1. LLM evaluation results of self-built RAG testsets
193
  |Model|Effectiveness of Response(Keyword)|*Effectiveness of Response(subjective evaluation)|Quoting Ability|Fallback Ability|*AutoQA|*Data Extraction|
194
  |---------------------|------|------|------|------|------|------|
195
  | Baichuan2-13B-Chat | 85 | 76 | 1 | 0 | 69 | 51 |
 
199
  | Orion-14B-Chat-RAG | 86 | 87 | 91 | 97 | 73 | 71 |
200
  \* means manual assessment
201
 
202
+ ## 3.5. Chat Plugin Model Orion-14B-Chat-Plugin Benchmarks
203
+ ### 3.5.1. LLM evaluation results of self-built plugin testsets
204
  |Model |Intent Recognition with Full Params |Intent Recognition with Missing Params |Non-Plugin Invocation Recognition |
205
  |-----------------------|--------|-----------|--------|
206
  | Baichuan2-13B-Chat | 25 | 0 | 0 |
 
208
  | GPT-4 | **95** | 52.38 | 70 |
209
  | Orion-14B-Chat-Plugin | 92.5 | **60.32** | **90** |
210
 
211
+ ## 3.6. Quantized Model Orion-14B-Base-Int4 Benchmarks
212
+ ### 3.6.1. Comparison of before and after quantization
213
  |Model |Size(GB)|Inference Speed(tokens/s)|C-Eval|CMMLU|MMLU|RACE|HellaSwag|
214
  |-------------------------|-------|-----|------|------|------|------|------|
215
  | OrionStar-14B-Base | 28.0 | 135 | 72.8 | 70.6 | 70.0 | 93.3 | 78.5 |
216
  | OrionStar-14B-Base-Int4 | 8.3 | 178 | 71.8 | 69.8 | 69.2 | 93.1 | 78.0 |
217
 
218
+ # 4. Model Inference
219
 
220
  Model weights, source code, and configuration needed for inference are published on Hugging Face, and the download link
221
  is available in the table at the beginning of this document. We demonstrate various inference methods here, and the
222
  program will automatically download the necessary resources from Hugging Face.
223
 
224
+ ## 4.1. Python Code
225
 
226
  ```python
227
  import torch
 
242
  In the above Python code, the model is loaded with `device_map='auto'` to utilize all available GPUs. To specify the
243
  device, you can use something like `export CUDA_VISIBLE_DEVICES=0,1` (using GPUs 0 and 1).
244
 
245
+ ## 4.2. Command Line Tool
246
 
247
  ```shell
248
  CUDA_VISIBLE_DEVICES=0 python cli_demo.py
 
250
 
251
  This command-line tool is designed for chat scenarios, and thus, it does not support calling the base model.
252
 
253
+ ## 4.3. Direct Script Inference
254
 
255
  ```shell
256
 
 
262
 
263
  ```
264
 
265
+ ## 4.4 Example Output
266
 
267
+ ### 4.4.1. Casual Chat
268
 
269
  `````
270
  User: Hello
271
  Orion-14B: Hello! How can I help you today? If you have any questions or need assistance, feel free to ask.
272
  User: What functions do you have?
273
  Orion-14B: As an AI language model, I have a wide range of functions and capabilities, including but not limited to:
274
+ Answering general knowledge questions: I can provide information on various topics such as history, science, geography, and more.
275
+ Providing definitions and explanations: I can explain complex concepts and provide definitions for words and phrases.
276
+ Assisting with tasks: I can help with simple tasks like setting reminders, converting units of measurement, and even ordering food from restaurants.
277
+ Generating text: I can write short stories, essays, articles, and other types of content based on prompts given to me.
278
+ Translating languages: I can translate text from one language to another.
279
+ Chatting: I can engage in conversation with users and respond to their queries in a conversational manner.
280
+ Summarizing long texts: I can summarize longer texts like articles, books, and research papers.
281
+ Generating random numbers and names: I can generate random numbers within a specified range and also suggest names based on certain criteria.
282
+ Playing games: I can play simple games like Hangman, Trivia, and more.
283
+ Providing recommendations: I can recommend movies, music, books, and other forms of entertainment based on user preferences.
284
+ 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.
 
285
  User: Tell me a joke.
286
  Orion-14B: Sure, here's a classic one-liner: Why don't scientists trust atoms? Because they make up everything.
287
  `````
288
 
289
+ ### 4.4.2. Japanese & Korean Chat
290
 
291
  `````
292
  User:自己を紹介してください
 
295
  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.
296
  `````
297
 
 
 
 
 
 
 
 
 
 
 
 
 
298
 
299
+ # 5. Declarations, License
300
 
301
+ ## 5.1. Declarations
302
 
303
  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.
304
  Additionally, we request users not to use the Orion-14B model for internet services without proper security review and filing.
 
309
  issues, public opinion risks, or any risks and issues arising from the model being misled, abused, disseminated, or
310
  improperly utilized, we will not assume any responsibility.
311
 
312
+ ## 5.2. License
313
 
314
  Community use of the Orion-14B series models
315
  - For code, please comply with [Apache License Version 2.0](./LICENSE)<br>
316
  - For model, please comply with [【Orion-14B Series】 Models Community License Agreement](./ModelsCommunityLicenseAgreement)
317
 
 
318
 
319
+ # 6. Company Introduction
320
 
321
+ OrionStar is a leading global service robot solutions company, founded in September 2016. OrionStar is dedicated to
322
+ using artificial intelligence technology to create the next generation of revolutionary robots, allowing people to break
323
+ free from repetitive physical labor and making human work and life more intelligent and enjoyable. Through technology,
324
+ OrionStar aims to make society and the world a better place.
325
 
326
+ OrionStar possesses fully self-developed end-to-end artificial intelligence technologies, such as voice interaction and
327
+ visual navigation. It integrates product development capabilities and technological application capabilities. Based on
328
+ the Orion robotic arm platform, it has launched products such as OrionStar AI Robot Greeting, AI Robot Greeting Mini,
329
+ Lucki, Coffee Master, and established the open platform OrionOS for Orion robots. Following the philosophy of "Born for
330
+ Truly Useful Robots", OrionStar empowers more people through AI technology.
331
+
332
+ **The core strengths of OrionStar lies in possessing end-to-end AI application capabilities,** including big data preprocessing, large model pretraining, fine-tuning, prompt engineering, agent, etc. With comprehensive end-to-end model training capabilities, including systematic data processing workflows and the parallel model training capability of hundreds of GPUs, it has been successfully applied in various industry scenarios such as government affairs, cloud services, international e-commerce, and fast-moving consumer goods.
333
+
334
+ Companies with demands for deploying large-scale model applications are welcome to contact us.
335
+ **Enquiry Hotline: 400-898-7779**<br>
336
+ **E-mail: ai@orionstar.com**
337
+
338
+ <div align="center">
339
+ <img src="./assets/imgs/wechat_group.jpg" alt="wechat" width="40%" />
340
+ </div>