Update README.md

README.md

@@ -77,6 +77,29 @@ from datasets import load_dataset
 dataset = load_dataset("BAAI/CCI3-Data")
 ```
 
+### Evaluation
+
+#### Setup
+Due to the mixed Chinese and English datasets, we chose Qwen2-0.5B model for datasets evaluation, each experiment with 100B tokens training. 
+
+We follow the same evaluation setup for all models using [FineWeb setup](https://github.com/huggingface/cosmopedia/tree/main/evaluation) with [lighteval](https://github.com/huggingface/lighteval) library.
+You can checkout the [evaluation script](./lighteval_tasks_v2.py) here.
+
+#### Results
+We conducted two types of experiments:
+1. Mixed Dataset Experiment: The ratio of English, code, and Chinese is 60% : 10% : 30%.
+2. Chinese Dataset Experiment: The Chinese ratio is 100%.
+
+For English datasets, we uniformly used [FineWeb-edu](https://huggingface.co/datasets/HuggingFaceFW/fineweb-edu/tree/main/sample/100BT); for code data, we used [StarCoder](https://huggingface.co/bigcode/starcoder); for Chinese datasets, we selected [wanjuan-v1](https://github.com/opendatalab/WanJuan1.0), [skypile](https://huggingface.co/datasets/Skywork/SkyPile-150B), and [cci3.0](https://huggingface.co/datasets/BAAI/CCI3-Data).
+
+For Mixed Dataset Experiment all evaluation metrics are averaged.
+![Mixed Dataset Experiment](./datasets_mix_metrics.png)
+
+For Chinese Dataset Experiment only chinese evaluation metrics are averaged.
+![Chinese Dataset Experiment](./chinese_dataset_metrics.png)
+
+
+
 ## User Agreement
 
 Users need to comply with the usage agreement of the CCI 3.0 dataset. You can view the agreement by clicking on the following link: （[View Usage Agreement](https://data.baai.ac.cn/resources/agreement/cci_usage_aggrement.pdf)）.