cArlIcon commited on
Commit
f77c5ff
1 Parent(s): d8029c8

update README

Browse files
Files changed (1) hide show
  1. README.md +26 -27
README.md CHANGED
@@ -9,24 +9,11 @@ license_link: LICENSE
9
  Yi
10
  </h1>
11
 
12
- <a href="https://github.com/01-ai/Yi/actions/workflows/ci.yml">
13
- <img src="https://github.com/01-ai/Yi/actions/workflows/ci.yml/badge.svg">
14
- </a>
15
- <a href="https://huggingface.co/01-ai">
16
- <img src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-01--ai-blue">
17
- </a>
18
- <a href="https://www.modelscope.cn/organization/01ai/">
19
- <img src="https://img.shields.io/badge/ModelScope-01--ai-blue">
20
- </a>
21
- <a href="https://github.com/01-ai/Yi/blob/main/LICENSE">
22
- <img src="https://img.shields.io/github/license/01-ai/yi">
23
- </a>
24
-
25
  </div>
26
 
27
  ## Introduction
28
 
29
- The **Yi** series models are large language models trained from scratch by developers at [01.AI](https://01.ai/). The first public release contains two base models with the parameter size of 6B and 34B. Besides, a specialized version with **200K** context window size is also provided.
30
 
31
  ## News
32
 
@@ -34,20 +21,30 @@ The **Yi** series models are large language models trained from scratch by devel
34
 
35
  ## Model Performance
36
 
37
- (FIXME)
 
 
 
 
 
 
 
 
 
 
 
 
38
 
39
- ## Usage
40
 
41
- ### 1. Download the model (optional)
42
 
43
- By default the model weights and tokenizer will be downloaded from [HuggingFace](https://huggingface.co/01-ai) automatically in the next step. You can also download them manually from the following places:
 
 
44
 
45
- - ModelScope (FIXME)
46
- - Mirror site (remember to extra the content with `tar`)
47
- - [Yi-6B.tar](https://01-ai.tos-cn-beijing.volces.com/yi/models/Yi-6B.tar)
48
- - [Yi-34B.tar](https://01-ai.tos-cn-beijing.volces.com/yi/models/Yi-34B.tar)
49
 
50
- ### 2. Run with docker
51
 
52
  The recommended approach to try out our models is through docker. We provide the following docker images.
53
 
@@ -56,7 +53,7 @@ The recommended approach to try out our models is through docker. We provide the
56
 
57
  Note that the `latest` tag always point to the latest code in the `main` branch. To test a stable version, please replace it with a specific [tag](https://github.com/01-ai/Yi/tags).
58
 
59
- #### 2.1 Try out the base model:
60
 
61
  ```bash
62
  docker run -it ghcr.io/01-ai/yi:latest python demo/text_generation.py
@@ -74,7 +71,7 @@ docker run -it \
74
 
75
  For more advanced usage, please refer the [doc](./demo/README.md).
76
 
77
- #### 2.2 Finetuning from the base model:
78
 
79
  ```bash
80
  docker run -it \
@@ -96,7 +93,7 @@ docker run -it \
96
 
97
  For more advanced usage like fine-tuning based on your custom data, please refer the [doc](./finetune/README.md).
98
 
99
- #### 2.3 Quantization
100
 
101
  ```bash
102
  docker run -it \
@@ -128,4 +125,6 @@ Although we use data compliance checking algorithms during the training process
128
 
129
  ## License
130
 
131
- The source code in this repo is licensed under the [Apache 2.0 license](./LICENSE). The Yi series model must be adhere to the [Model License Agreement](./MODEL_LICENSE_AGREEMENT.txt).
 
 
 
9
  Yi
10
  </h1>
11
 
 
 
 
 
 
 
 
 
 
 
 
 
 
12
  </div>
13
 
14
  ## Introduction
15
 
16
+ The **Yi** series models are large language models trained from scratch by developers at [01.AI](https://01.ai/). The first public release contains two base models with the parameter size of 6B and 34B.
17
 
18
  ## News
19
 
 
21
 
22
  ## Model Performance
23
 
24
+ | Model | MMLU | CMMLU | C-Eval | GAOKAO | BBH | Commonsense Reasoning | Reading Comprehension | Math & Code |
25
+ | :------------ | :------: | :------: | :------: | :------: | :------: | :-------------------: | :-------------------: | :---------: |
26
+ | | 5-shot | 5-shot | 5-shot | 0-shot | 3-shot@1 | - | - | - |
27
+ | LLaMA2-34B | 62.6 | - | - | - | 44.1 | 69.9 | 68.0 | 26.0 |
28
+ | LLaMA2-70B | 68.9 | 53.3 | - | 49.8 | 51.2 | 71.9 | 69.4 | 36.8 |
29
+ | Baichuan2-13B | 59.2 | 62.0 | 58.1 | 54.3 | 48.8 | 64.3 | 62.4 | 23.0 |
30
+ | Qwen-14B | 66.3 | 71.0 | 72.1 | 62.5 | 53.4 | 73.3 | 72.5 | 39.8 |
31
+ | Skywork-13B | 62.1 | 61.8 | 60.6 | 68.1 | 41.7 | 72.4 | 61.4 | 24.9 |
32
+ | InternLM-20B | 62.1 | 59.0 | 58.8 | 45.5 | 52.5 | 78.3 | - | 26.0 |
33
+ | Aquila-34B | 67.8 | 71.4 | 63.1 | - | - | - | - | - |
34
+ | Falcon-180B | 70.4 | 58.0 | 57.8 | 59.0 | 54.0 | 77.3 | 68.8 | 34.0 |
35
+ | Yi-6B | 63.2 | 75.5 | 72.0 | 72.2 | 42.8 | 72.3 | 68.7 | 19.8 |
36
+ | **Yi-34B** | **76.3** | **83.7** | **81.4** | **82.8** | **54.3** | **80.1** | **76.4** | **37.1** |
37
 
 
38
 
39
+ While benchmarking open-source models, we have observed a disparity between the results generated by our pipeline and those reported in public sources (e.g. OpenCampus). Upon conducting a more in-depth investigation of this difference, we have discovered that various models may employ different prompts, post-processing strategies, and sampling techniques, potentially resulting in significant variations in the outcomes. Our prompt and post-processing strategy remains consistent with the original benchmark, and greedy decoding is employed during evaluation without any post-processing for the generated content. For scores that did not report by original author (including score reported with different setting), we try to get results with our pipeline.
40
 
41
+ To extensively evaluate model's capability, we adopted the methodology outlined in Llama2. Specifically, we included PIQA, SIQA, HellaSwag, WinoGrande, ARC, OBQA, and CSQA to assess common sense reasoning. SquAD, QuAC, and BoolQ were incorporated to evaluate reading comprehension. CSQA was exclusively tested using a 7-shot setup, while all other tests were conducted in a 0-shot configuration. Additionally, we introduced GSM8K (8-shot@1), MATH (4-shot@1), HumanEval (0-shot@1), and MBPP (3-shot@1) under the category "Math & Code". Due to technical constraints, we did not test Falcon-180 on QuAC and OBQA; the score is derived by averaging the scores on the remaining tasks. Since the scores for these two tasks are generally lower than the average, we believe that Falcon-180B's performance was not underestimated.
42
+
43
+ ## Usage
44
 
45
+ Feel free to [create an issue](https://github.com/01-ai/Yi/issues/new) if you encounter any problem when using the Yi series models.
 
 
 
46
 
47
+ ### 1. Run with docker
48
 
49
  The recommended approach to try out our models is through docker. We provide the following docker images.
50
 
 
53
 
54
  Note that the `latest` tag always point to the latest code in the `main` branch. To test a stable version, please replace it with a specific [tag](https://github.com/01-ai/Yi/tags).
55
 
56
+ #### 1.1 Try out the base model:
57
 
58
  ```bash
59
  docker run -it ghcr.io/01-ai/yi:latest python demo/text_generation.py
 
71
 
72
  For more advanced usage, please refer the [doc](./demo/README.md).
73
 
74
+ #### 1.2 Finetuning from the base model:
75
 
76
  ```bash
77
  docker run -it \
 
93
 
94
  For more advanced usage like fine-tuning based on your custom data, please refer the [doc](./finetune/README.md).
95
 
96
+ #### 1.3 Quantization
97
 
98
  ```bash
99
  docker run -it \
 
125
 
126
  ## License
127
 
128
+ The source code in this repo is licensed under the [Apache 2.0 license](https://github.com/01-ai/Yi/blob/main/LICENSE).
129
+ The Yi series model must be adhere to the [Model License Agreement](https://github.com/01-ai/Yi/blob/main/MODEL_LICENSE_AGREEMENT.txt).
130
+ For any questions related to licensing and copyright, please contact us ([[email protected]](mailto:[email protected])).