Update README.md
Browse files
README.md
CHANGED
@@ -13,6 +13,61 @@ https://github.com/ZhuiyiTechnology/roformer-v2
|
|
13 |
### pytorch版本+tf2.0版本
|
14 |
https://github.com/JunnYu/RoFormer_pytorch
|
15 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
16 |
### 安装
|
17 |
- pip install roformer==0.4.3
|
18 |
|
|
|
13 |
### pytorch版本+tf2.0版本
|
14 |
https://github.com/JunnYu/RoFormer_pytorch
|
15 |
|
16 |
+
## 评测对比
|
17 |
+
### CLUE-dev榜单分类任务结果,base+large版本。
|
18 |
+
|
19 |
+
| | iflytek | tnews | afqmc | cmnli | ocnli | wsc | csl |
|
20 |
+
| :-----: | :-----: | :---: | :---: | :---: | :---: | :---: | :---: |
|
21 |
+
| BERT | 60.06 | 56.80 | 72.41 | 79.56 | 73.93 | 78.62 | 83.93 |
|
22 |
+
| RoBERTa | 60.64 | 58.06 | 74.05 | 81.24 | 76.00 | 87.50 | 84.50 |
|
23 |
+
| RoFormer | 60.91 | 57.54 | 73.52 | 80.92 | 76.07 | 86.84 | 84.63 |
|
24 |
+
| RoFormerV2<sup>*</sup> | 60.87 | 56.54 | 72.75 | 80.34 | 75.36 | 80.92 | 84.67 |
|
25 |
+
| GAU-α | 61.41 | 57.76 | 74.17 | 81.82 | 75.86 | 79.93 | 85.67 |
|
26 |
+
| RoFormer-pytorch(本仓库代码) | 60.60 | 57.51 | 74.44 | 80.79 | 75.67 | 86.84 | 84.77 |
|
27 |
+
| RoFormerV2-pytorch(本仓库代码) | **62.87** | 59.03 | **76.20** | 80.85 | 79.73 | 87.82 | **91.87** |
|
28 |
+
| GAU-α-pytorch(Adafactor) | 61.18 | 57.52 | 73.42 | 80.91 | 75.69 | 80.59 | 85.5 |
|
29 |
+
| GAU-α-pytorch(AdamW wd0.01 warmup0.1) | 60.68 | 57.95 | 73.08 | 81.02 | 75.36 | 81.25 | 83.93 |
|
30 |
+
| RoFormerV2-large-pytorch(本仓库代码) | 61.75 | **59.21** | 76.14 | 82.35 | **81.73** | **91.45** | 91.5 |
|
31 |
+
| Chinesebert-large-pytorch | 61.25 | 58.67 | 74.70 | **82.65** | 79.63 | 87.83 | 84.97 |
|
32 |
+
|
33 |
+
|
34 |
+
### CLUE-1.0-test榜单分类任务结果,base+large版本。
|
35 |
+
|
36 |
+
| | iflytek | tnews | afqmc | cmnli | ocnli | wsc | csl |
|
37 |
+
| :-----: | :-----: | :---: | :---: | :---: | :---: | :---: | :---: |
|
38 |
+
| RoFormer-pytorch(本仓库代码) | 59.54 | 57.34 | 74.46 | 80.23 | 73.67 | 80.69 | 84.57 |
|
39 |
+
| RoFormerV2-pytorch(本仓库代码) | **63.15** | 58.24 | 75.42 | 80.59 | 74.17 | 83.79 | 83.73 |
|
40 |
+
| GAU-α-pytorch(Adafactor) | 61.38 | 57.08 | 74.05 | 80.37 | 73.53 | 74.83 | **85.6** |
|
41 |
+
| GAU-α-pytorch(AdamW wd0.01 warmup0.1) | 60.54 | 57.67 | 72.44 | 80.32 | 72.97 | 76.55 | 84.13 |
|
42 |
+
| RoFormerV2-large-pytorch(本仓库代码) | 61.85 | **59.13** | **76.38** | 80.97 | 76.23 | **85.86** | 84.33 |
|
43 |
+
| Chinesebert-large-pytorch | 61.54 | 58.57 | 74.8 | **81.94** | **76.93** | 79.66 | 85.1 |
|
44 |
+
|
45 |
+
### 注:
|
46 |
+
- 其中RoFormerV2<sup>*</sup>表示的是未进行多任务学习的RoFormerV2模型,该模型苏神并未开源,感谢苏神的提醒。
|
47 |
+
- 其中不带有pytorch后缀结果都是从[GAU-alpha](https://github.com/ZhuiyiTechnology/GAU-alpha)仓库复制过来的。
|
48 |
+
- 其中带有pytorch后缀的结果都是自己训练得出的。
|
49 |
+
- 苏神代码中拿了cls标签后直接进行了分类,而本仓库使用了如下的分类头,多了2个dropout,1个dense,1个relu激活。
|
50 |
+
|
51 |
+
```python
|
52 |
+
class RoFormerClassificationHead(nn.Module):
|
53 |
+
def __init__(self, config):
|
54 |
+
super().__init__()
|
55 |
+
self.dense = nn.Linear(config.hidden_size, config.hidden_size)
|
56 |
+
self.dropout = nn.Dropout(config.hidden_dropout_prob)
|
57 |
+
self.out_proj = nn.Linear(config.hidden_size, config.num_labels)
|
58 |
+
|
59 |
+
self.config = config
|
60 |
+
|
61 |
+
def forward(self, features, **kwargs):
|
62 |
+
x = features[:, 0, :] # take <s> token (equiv. to [CLS])
|
63 |
+
x = self.dropout(x)
|
64 |
+
x = self.dense(x)
|
65 |
+
x = ACT2FN[self.config.hidden_act](x) # 这里是relu
|
66 |
+
x = self.dropout(x)
|
67 |
+
x = self.out_proj(x)
|
68 |
+
return x
|
69 |
+
```
|
70 |
+
|
71 |
### 安装
|
72 |
- pip install roformer==0.4.3
|
73 |
|