File size: 15,678 Bytes
b597d38
 
befff52
 
b721e32
b3aed60
b721e32
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1920b0e
 
 
70840e3
1920b0e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b597d38
c1bced4
1920b0e
 
c1bced4
 
 
 
 
 
 
568e186
c1bced4
 
f973676
c1bced4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3e9b5e2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c1bced4
 
 
 
125cb7c
c1bced4
 
 
 
 
3e9b5e2
c1bced4
 
 
 
 
3e9b5e2
c1bced4
babaf3f
c1bced4
 
 
 
 
 
 
 
 
3e9b5e2
c1bced4
 
 
 
 
 
 
568e186
 
125cb7c
 
3e9b5e2
 
125cb7c
568e186
c1bced4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
568e186
 
 
 
c1bced4
 
 
125cb7c
 
 
 
c1bced4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
---
license: mit
library_name: transformers
pipeline_tag: text-generation
datasets:
- yulan-team/YuLan-Mini-Datasets
- HuggingFaceFW/fineweb-edu
- bigcode/the-stack-v2
- mlfoundations/dclm-baseline-1.0
- math-ai/AutoMathText
- gair-prox/open-web-math-pro
- RUC-AIBOX/long_form_thought_data_5k
- internlm/Lean-Workbook
- internlm/Lean-Github
- deepseek-ai/DeepSeek-Prover-V1
- ScalableMath/Lean-STaR-base
- ScalableMath/Lean-STaR-plus
- ScalableMath/Lean-CoT-base
- ScalableMath/Lean-CoT-plus
- opencsg/chinese-fineweb-edu
- liwu/MNBVC
- vikp/textbook_quality_programming
- HuggingFaceTB/smollm-corpus
- OpenCoder-LLM/opc-annealing-corpus
- OpenCoder-LLM/opc-sft-stage1
- OpenCoder-LLM/opc-sft-stage2
- XinyaoHu/AMPS_mathematica
- deepmind/math_dataset
- mrfakename/basic-math-10m
- microsoft/orca-math-word-problems-200k
- AI-MO/NuminaMath-CoT
- HuggingFaceTB/cosmopedia
- MU-NLPC/Calc-ape210k
- manu/project_gutenberg
- storytracer/LoC-PD-Books
- allenai/dolma
language:
- en
- zh
tags:
- code
- math
arxiv: 2412.17743
model-index:
- name: YuLan-Mini
  results:
  - task:
      type: text-generation
    dataset:
      type: openai_humaneval
      name: HumanEval
    metrics:
    - name: pass@1
      type: pass@1
      value: 0.640
      verified: false
  - task:
      type: text-generation
    dataset:
      type: mbpp
      name: MBPP
    metrics:
    - name: pass@1
      type: pass@1
      value: 0.659
      verified: false
  - task:
      type: text-generation
    dataset:
      type: math-500
      name: MATH-500
    metrics:
    - name: maj@1
      type: maj@1
      value: 0.378
      verified: false
  - task:
      type: text-generation
    dataset:
      type: gsm8k
      name: GSM8K
    metrics:
    - name: maj@1
      type: maj@1
      value: 0.684
      verified: false
---

# Important Notice: This is a pre-trained **base model** without instruction-following capabilities. The **SFT version** will be released within a few weeks.


<div align=center>
<img src="assets/YuLan-logo.jpg" width="400px">
<h1>YuLan-Mini: An Open Data-efficient Language Model</h1>
<a href="https://github.com/RUC-GSAI/YuLan-Mini/blob/main/LICENSE"><img src="https://img.shields.io/badge/License-MIT-blue" alt="license"></a>
<a href="https://arxiv.org/abs/2412.17743" target="_blank"><img src=https://img.shields.io/badge/arXiv-b5212f.svg?logo=arxiv></a>
<a href="https://huggingface.co/collections/yulan-team/yulan-mini-676d214b24376739b00d95f3"><img alt="Static Badge" src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-blue?color=8A2BE2"></a>
<a href="https://github.com/RUC-GSAI/YuLan-Mini" target="_blank"><img src="https://img.shields.io/github/stars/RUC-GSAI/YuLan-Mini"></a>
</div>

YuLan-Mini is a lightweight language model with 2.4 billion parameters. It achieves performance comparable to industry-leading models trained on significantly more data, despite being pre-trained on only 1.08T tokens. The model excels particularly in the domains of **mathematics** and **code**. To facilitate reproducibility, we will open-source the relevant [pre-training resources](https://github.com/RUC-GSAI/YuLan-Mini/tree/main/pretrain).

---

## Model Downloads πŸ”—

|  Model  | Context Length | SFT |
|---------|----------------|-----|
| [YuLan-Mini](https://huggingface.co/yulan-team/YuLan-Mini) (Recommended) | 28K | ❎ |
| [YuLan-Mini-2.4B-4K](https://huggingface.co/yulan-team/YuLan-Mini-Intermediate-4K) | 4K | ❎ |
| YuLan-Mini-Instruct | Comming soon | βœ… |

---

## Features 🌟

<div align=center>
<img src="assets/main.png">
</div>

Our pre-training methodology improves training efficiency through three key innovations:

1. an elaborately designed **data pipeline** that combines data cleaning with data schedule strategies;
2. a systematic **optimization method** that can effectively mitigate training instability;
3. an effective **annealing approach** that integrate targeted data selection and long context training.


---
## Behchmarks 🌟

|      Models      | Model Size | # Train Tokens | Context Length | MATH 500 | GSM 8K | Human Eval | MBPP   | RACE Middle | RACE High | RULER  |
|:----------------|----------:|--------------:|--------------:|:--------|:------|:----------|:------|:-----------|:---------|:------|
|     MiniCPM      |    2.6B    |     1.06T      |       4K       |   15.00  |  53.83 |     50.00* |  47.31 |     56.61   |   44.27   |   N/A  |
|      Qwen-2      |    1.5B    |       7T       |      128K      |   22.60  | 46.90* |     34.80* | 46.90* |     55.77   |   43.69   |  60.16 |
|     Qwen2.5      |    0.5B    |      18T       |      128K      |   23.60  | 41.60* |     30.50* | 39.30* |     52.36   |   40.31   |  49.23 |
|     Qwen2.5      |    1.5B    |      18T       |      128K      |   **45.40**  | **68.50\*** |     37.20* | 60.20* |     **58.77**   |   44.33   |  <ins>68.26</ins> |
|     Gemma2       |    2.6B    |       2T       |       8K       |   18.30* | 30.30* |     19.50* | 42.10* |       -     |      -    |   N/A  |
|    StableLM2     |    1.7B    |       2T       |       4K       |     -    |  20.62 |      8.50* |  17.50 |     56.33   |   **45.06**   |   N/A  |
|    SmolLM2       |    1.7B    |      11T       |       8K       |   11.80  |    -   |     23.35  |  45.00 |     55.77   |   43.06   |   N/A  |
|    Llama3.2      |    3.2B    |       9T       |      128K      |    7.40  |    -   |     29.30  |  49.70 |     55.29   |   43.34   |  **77.06** |
|    YuLan-Mini    |    2.4B    |     1.04T      |       4K       |   32.60  |  66.65 |     <ins>61.60</ins>  |  **66.70** |     55.71   |   43.58   |   N/A  |
|    YuLan-Mini    |    2.4B    |     1.08T      |      28K       |  <ins>37.80</ins>  |  <ins>68.46</ins> |    **64.00**  |  <ins>65.90</ins>|     <ins>57.18</ins>   |   <ins>44.57</ins>   |  51.48 |


|      Models      | LAMBADA | MMLU  | CMMLU | CEval | HellaSwag | WinoGrande | StoryCloze | ARC-e | ARC-c |
|:----------------|:-------|:-----|:-----|:-----|:----------|:-----------|:-----------|:-----|:-----|
|   MiniCPM-2.6B   |  61.91  | 53.37 | 48.97 | 48.24 |   67.92    |     65.74   |     78.51   | 55.51 | 43.86 |
|   Qwen2-1.5B     |  64.68  | 55.90 | **70.76** | **71.94** |   66.11    |     66.14   |     77.60   | 62.21 | 42.92 |
|  Qwen2.5-0.5B    |  52.00  | 47.50 | 52.17 | 54.27 |   50.54    |     55.88   |     71.67   | 56.10 | 39.51 |
|  Qwen2.5-1.5B    |  62.12  | <ins>60.71</ins> | <ins>67.82</ins> | <ins>69.05</ins> |   67.18    |     64.48   |     76.80   | **71.51** | <ins>53.41</ins> |
|   Gemma2-2.6B    |    -    | 52.20*|   -   | 28.00*|   <ins>74.60*</ins>   |    **71.50\***   |       -     |   -   | **55.70\***|
| StableLM2-1.7B   |  66.15  | 40.37 | 29.29 | 26.99 |   69.79    |     64.64   |     <ins>78.56</ins>   | 54.00 | 40.78 |
|  SmolLM2-1.7B    |  <ins>67.42</ins>  | 51.91 | 33.46 | 35.10 |   72.96    |     67.40   |     **79.32**   | 44.82 | 35.49 |
|   Llama3.2-3B    |  **69.08**  | **63.40** | 44.44 | 44.49 |   **75.62**    |     <ins>67.48</ins>   |     76.80   | <ins>70.12</ins> | 48.81 |
|    YuLan-Mini    |  64.72  | 51.79 | 48.35 | 51.47 |   68.65    |     67.09   |     76.37   | 69.87 | 50.51 |
|    YuLan-Mini    |  65.67  | 49.10 | 45.45 | 48.23 |   67.22    |     67.24   |     75.89   | 67.47 | 49.32 |


---

## Pre-Training Resources πŸ”§

To enhance research transparency and reproducibility, we are open-sourcing relevant [pre-training resources](https://github.com/RUC-GSAI/YuLan-Mini/blob/main/pretrain):

<details><summary>1. Pre-training and Evaluation Code</summary>

The pre-training and evaluation code will be released in a future update.
</details>



<details><summary>2. Intermediate Stage Checkpoints</summary>
The intermediate stage checkpoints are released in <a href="https://huggingface.co/collections/yulan-team/yulan-mini-676d214b24376739b00d95f3">YuLan-Mini</a>.

<table>
    <thead>
        <tr>
            <th>Stage</th>
            <th>Curriculum Phase</th>
            <th>4K Context</th>
            <th>28K Context</th>
            <th>Optimizer</th>
            <th>Inference Architecture</th>
            <th>LAMBADA <code>Acc</code></th>
            <th>GSM8K <code>Acc</code></th>
            <th>HumanEval <code>pass@1</code></th>
        </tr>
    </thead>
    <tbody>
        <tr>
            <td>Stable</td>
            <td>5</td>
            <td><a href="https://huggingface.co/yulan-team/YuLan-Mini-Phase5">YuLan-Mini-Phase5</a></td>
            <td></td>
            <td></td>
            <td><code>yulanmini</code></td>
            <td>53.85</td>
            <td>3.41</td>
            <td>12.26</td>
        </tr>
        <tr>
            <td>Stable</td>
            <td>10</td>
            <td><a href="https://huggingface.co/yulan-team/YuLan-Mini-Phase10">YuLan-Mini-Phase10</a></td>
            <td></td>
            <td></td>
            <td><code>yulanmini</code></td>
            <td>55.00</td>
            <td>9.57</td>
            <td>15.95</td>
        </tr>
        <tr>
            <td>Stable</td>
            <td>15</td>
            <td><a href="https://huggingface.co/yulan-team/YuLan-Mini-Phase15">YuLan-Mini-Phase15</a></td>
            <td></td>
            <td></td>
            <td><code>yulanmini</code></td>
            <td>55.81</td>
            <td>13.81</td>
            <td>16.99</td>
        </tr>
        <tr>
            <td>Stable</td>
            <td>20</td>
            <td><a href="https://huggingface.co/yulan-team/YuLan-Mini-Phase20">YuLan-Mini-Phase20</a></td>
            <td></td>
            <td>βœ…</td>
            <td><code>yulanmini</code></td>
            <td>55.81</td>
            <td>21.39</td>
            <td>20.79</td>
        </tr>
        <tr>
            <td>Stable</td>
            <td>25 (1T tokens)</td>
            <td><a href="https://huggingface.co/yulan-team/YuLan-Mini-Before-Annealing">YuLan-Mini-Before-Annealing</a></td>
            <td></td>
            <td>βœ…</td>
            <td><code>yulanmini</code></td>
            <td>55.67</td>
            <td>29.94</td>
            <td>34.06</td>
        </tr>
        <tr>
            <td></td>
            <td></td>
            <td></td>
            <td></td>
            <td></td>
            <td></td>
            <td></td>
            <td></td>
            <td></td>
        </tr>
        <tr>
            <td>Annealing</td>
            <td>26</td>
            <td>YuLan-Mini-4K</td>
            <td></td>
            <td></td>
            <td><code>llama</code>*</td>
            <td>64.72</td>
            <td>66.65</td>
            <td>61.60</td>
        </tr>
        <tr>
            <td>Annealing</td>
            <td>27</td>
            <td></td>
            <td><a href="https://huggingface.co/yulan-team/YuLan-Mini">YuLan-Mini</a></td>
            <td></td>
            <td><code>llama</code>*</td>
            <td>65.67</td>
            <td>68.46</td>
            <td>64.00</td>
        </tr>
    </tbody>
</table>

\*: For easier inference and deployment, we merged the re-parameterized added parameters and scaling factors into the final released models ([**YuLan-Mini**](https://huggingface.co/yulan-team/YuLan-Mini) and **YuLan-Mini-Intermediate-4K**), enabling it to run on the Llama architecture. However, these parameters are still retained in the intermediate checkpoints from the training process.

</details>

<details><summary>3. Optimizer States Before Annealing</summary>

<a href="https://huggingface.co/yulan-team/YuLan-Mini-Before-Annealing">YuLan-Mini-Before-Annealing</a>
</details>


<details><summary>4. The Used Open-Source Datasets </summary>

<a href="https://github.com/RUC-GSAI/YuLan-Mini/blob/main/pretrain/datasets">Used-Datasets-List</a>

</details>

<details><summary>5. Data Distribution for every phase</summary>

<a href="https://github.com/RUC-GSAI/YuLan-Mini/blob/main/pretrain/datasets/final.pdf">
  <div align=center>
    <img src="assets/data_distribution_for_every_phase.png">
  </div>
</a>

</details>

<details><summary>6. Synthetic Data</summary>

Data cleaning and synthesis pipeline:
<div align=center>
<img src="https://github.com/RUC-GSAI/YuLan-Mini/blob/main/assets/data-pipeline.png">
</div>

The synthetic data we are using is released in <a href="https://huggingface.co/collections/yulan-team/yulan-mini-676d214b24376739b00d95f3">YuLan-Mini-Datasets</a>

</details>


### What you can do with these pre-training resources

1. **Pre-train** your own LLM. You can use [our data](https://huggingface.co/yulan-team/YuLan-Mini-Datasets) and curriculum to train a model that's just as powerful as YuLan-Mini.
2. Perform your own **learning rate annealing**. During the annealing phase, YuLan-Mini's learning ability is at its peak. You can resume training from [the checkpoint before annealing](https://huggingface.co/yulan-team/YuLan-Mini-Before-Annealing) and use your own dataset for learning rate annealing.
3. **Fine-tune** the Instruct version of the LLM. You can use the [YuLan-Mini](https://huggingface.co/yulan-team/YuLan-Mini) base model to train your own Instruct version.
4. **Training dynamics** research. You can use YuLan-Mini's [intermediate checkpoints](https://huggingface.co/collections/yulan-team/yulan-mini-676d214b24376739b00d95f3) to explore internal changes during the pre-training process.
5. **Synthesize** your own data. You can use YuLan-Mini's [data pipeline](https://github.com/RUC-GSAI/YuLan-Mini) to clean and generate your own dataset.

---

## Quick Start πŸ’»

Below is a simple example for inference using Huggingface:

**Huggingface Inference Example**
```python
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM

# Load model and tokenizer
tokenizer = AutoTokenizer.from_pretrained("yulan-team/YuLan-Mini")
model = AutoModelForCausalLM.from_pretrained("yulan-team/YuLan-Mini", torch_dtype=torch.bfloat16)

# Input text
input_text = "Renmin University of China is"
inputs = tokenizer(input_text, return_tensors="pt")

# Completion
output = model.generate(inputs["input_ids"], max_new_tokens=100)
print(tokenizer.decode(output[0], skip_special_tokens=True))
```

**vLLM Serve Example**
```bash
vllm serve yulan-team/YuLan-Mini --dtype bfloat16
```

**SGLang Serve Example**
```bash
python -m sglang.launch_server --model-path yulan-team/YuLan-Mini --port 30000 --host 0.0.0.0
```

---

## The Team

YuLan-Mini is developed and maintained by [AI Box, Renmin University of China](http://aibox.ruc.edu.cn/).

## License

- The code in this repository is released under the [MIT License](./LICENSE).
- Policies regarding the use of model weights, intermediate optimizer states, and training data will be announced in future updates.
- Limitations: Despite our efforts to mitigate safety concerns and encourage the generation of ethical and lawful text, the probabilistic nature of language models may still lead to unexpected outputs. For instance, responses might contain bias, discrimination, or other harmful content. Please refrain from disseminating such content. We are not liable for any consequences arising from the spread of harmful information.

## Citation

If you find YuLan-Mini helpful for your research or development, please cite [our technical report](https://arxiv.org/abs/2412.17743):

```
@misc{hu2024yulanmini,
      title={YuLan-Mini: An Open Data-efficient Language Model}, 
      author={Yiwen Hu and Huatong Song and Jia Deng and Jiapeng Wang and Jie Chen and Kun Zhou and Yutao Zhu and Jinhao Jiang and Zican Dong and Wayne Xin Zhao and Ji-Rong Wen},
      year={2024},
      eprint={2412.17743},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2412.17743}, 
}
```