feat: support Model Yuan2.0, a new generation Fundamental Large Language Model developed by IEIT System

[cauwulixuan]

Why are these changes needed?

Yuan2.0 is a new generation Fundamental Large Language Model developed by IEIT System. We have published all three models, Yuan 2.0-102B, Yuan 2.0-51B, and Yuan 2.0-2B. And we provide relevant scripts for pretraining, fine-tuning, and inference services for other developers. Yuan2.0 is based on Yuan1.0, utilizing a wider range of high-quality pre training data and instruction fine-tuning datasets to enhance the model's understanding of semantics, mathematics, reasoning, code, knowledge, and other aspects.

1. As we can see, Yuan2.0 provided an chat template
2. The model is Yuan2.0.
3. when loading tokenizer, several special should be set tokens explicitly.
4. set model registry info, based on Yuan2.0 README

Here gives some demos:

user: 您好。
assistant: 您好，请问您需要哪方面的帮助？

{'conv_template': 'yuan2', 'prompt': '您好。<sep>', 'outputs': '您好，请问您需要哪方面的帮助？', 'speed (token/s)': 5.69}

user: 给我写首诗，主题是明月。
assistant: 明月生在水中央，
轻云浮动望不穷。<n>夜色清光洒满空，<n>辉辉点点映江枫。<n>徘徊独步望明月，<n>空明月色照唐三。<n>愿您片刻得平安，<n>愿您早日展雄风。<n>祝您身体健康，<n>愿您前程似锦腾。

{'conv_template': 'yuan2', 'prompt': '您好。<n>您好，请问您需要哪方面的帮助？<n>给我写首诗，主题是明月。<sep>', 'outputs': '明月生在水中央，\n轻云浮动望不穷。<n>夜色清光洒满空，<n>辉辉点点映江枫。<n>徘徊独步望明月，<n>空明月色照唐三。<n>愿您片刻得平安，<n>愿您早日展雄风。<n>祝您身体健康，<n>愿您前程似锦腾。', 'speed (token/s)': 32.89}

user: 写一段knn算法。
assistant: 以下是使用Python编写的简单高斯混合模型（KNN）算法的代码，用于对输入数据进行K近邻分类。
```python
import numpy as np
def KNN(X, y, k):
    # 获取数据集大小和样本数量
    n = len(X)
    # 创建一个向量X，包含所有输入特征
    X = np.array([x for x in X])
    # 创建一个向量y，包含所有输入标签
    y = np.array([y for y in y])
    # 创建一个向量X，包含所有输入特征的索引
    X_indices = np.where(np.array(range(n)).reshape(-1, 1) == y)[0]
    # 创建一个空的向量Y，用于存储每个样本的预测标签
    Y = np.zeros(n)
    # 对每个样本进行K近邻分类
    for i in range(n):
        # 获取当前样本的标签
        current_label = y[i]
        # 获取当前样本的k近邻样本的标签
        nearest_label = X_indices[np.argmin(np.linalg.norm(X[i] - X)**2, axis=1)]
        # 根据K值和当前样本的标签进行比较，选择最近的标签作为预测标签
        if current_label == nearest_label:
            Y[i] = current_label
        else:
            Y[i] = nearest_label
    # 对预测标签进行排序，并返回结果
    return sorted(Y, key=lambda x: Y[X_indices.index(X[np.argmin(np.abs(X[np.argmin(K]-X[np.argmin(K)]))])])

{'conv_template': 'yuan2', 'prompt': '您好。您好，请问您需要哪方面的帮助？给我写首诗，主题是明月。明月生在水中央，\n轻云浮动望不穷。夜色清光洒满空，辉辉点点映江枫。徘徊独步望明月，空明月色照唐三。愿您片刻得平安，愿您早日展雄风。祝您身体健康，愿您前程似锦腾。写一段knn算法。', 'outputs': '以下是使用Python编写的简单高斯混合模型（KNN）算法的代码，用于对输入数据进行K近邻分类。\npython\nimport numpy as np\ndef KNN(X, y, k):\n # 获取数据集大小和样本数量\n n = len(X)\n # 创建一个向量X，包含所有输入特征\n X = np.array([x for x in X])\n # 创建一个向量y，包含所有输入标签\n y = np.array([y for y in y])\n # 创建一个向量X，包含所有输入特征的索引\n X_indices = np.where(np.array(range(n)).reshape(-1, 1) == y)[0]\n # 创建一个空的向量Y，用于存储每个样本的预测标签\n Y = np.zeros(n)\n # 对每个样本进行K近邻分类\n for i in range(n):\n # 获取当前样本的标签\n current_label = y[i]\n # 获取当前样本的k近邻样本的标签\n nearest_label = X_indices[np.argmin(np.linalg.norm(X[i] - X)**2, axis=1)]\n # 根据K值和当前样本的标签进行比较，选择最近的标签作为预测标签\n if current_label == nearest_label:\n Y[i] = current_label\n else:\n Y[i] = nearest_label\n # 对预测标签进行排序，并返回结果\n return sorted(Y, key=lambda x: Y[X_indices.index(X[np.argmin(np.abs(X[np.argmin(K]-X[np.argmin(K)]))])])\n', 'speed (token/s)': 29.11}

user:

## Checks

- [ ] I've run `format.sh` to lint the changes in this PR.
- [ ] I've included any doc changes needed.
- [ ] I've made sure the relevant tests are passing (if applicable).