Create README.md
Browse files
README.md
ADDED
@@ -0,0 +1,78 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
language: zh
|
3 |
+
tags:
|
4 |
+
- roformer-v2
|
5 |
+
- pytorch
|
6 |
+
- tf2.0
|
7 |
+
---
|
8 |
+
## 介绍
|
9 |
+
### tf版本
|
10 |
+
https://github.com/ZhuiyiTechnology/roformer
|
11 |
+
|
12 |
+
### pytorch版本+tf2.0版本
|
13 |
+
https://github.com/JunnYu/RoFormer_pytorch
|
14 |
+
|
15 |
+
## pytorch & tf2.0使用
|
16 |
+
```python
|
17 |
+
import torch
|
18 |
+
import tensorflow as tf
|
19 |
+
from transformers import BertTokenizer
|
20 |
+
from roformer import RoFormerForMaskedLM, TFRoFormerForMaskedLM
|
21 |
+
|
22 |
+
text = "今天[MASK]很好,我[MASK]去公园玩。"
|
23 |
+
tokenizer = BertTokenizer.from_pretrained("junnyu/roformer_v2_chinese_char_base")
|
24 |
+
pt_model = RoFormerForMaskedLM.from_pretrained("junnyu/roformer_v2_chinese_char_base")
|
25 |
+
tf_model = TFRoFormerForMaskedLM.from_pretrained(
|
26 |
+
"junnyu/roformer_v2_chinese_char_base", from_pt=True
|
27 |
+
)
|
28 |
+
pt_inputs = tokenizer(text, return_tensors="pt")
|
29 |
+
tf_inputs = tokenizer(text, return_tensors="tf")
|
30 |
+
# pytorch
|
31 |
+
with torch.no_grad():
|
32 |
+
pt_outputs = pt_model(**pt_inputs).logits[0]
|
33 |
+
pt_outputs_sentence = "pytorch: "
|
34 |
+
for i, id in enumerate(tokenizer.encode(text)):
|
35 |
+
if id == tokenizer.mask_token_id:
|
36 |
+
tokens = tokenizer.convert_ids_to_tokens(pt_outputs[i].topk(k=5)[1])
|
37 |
+
pt_outputs_sentence += "[" + "||".join(tokens) + "]"
|
38 |
+
else:
|
39 |
+
pt_outputs_sentence += "".join(
|
40 |
+
tokenizer.convert_ids_to_tokens([id], skip_special_tokens=True)
|
41 |
+
)
|
42 |
+
print(pt_outputs_sentence)
|
43 |
+
# tf
|
44 |
+
tf_outputs = tf_model(**tf_inputs, training=False).logits[0]
|
45 |
+
tf_outputs_sentence = "tf: "
|
46 |
+
for i, id in enumerate(tokenizer.encode(text)):
|
47 |
+
if id == tokenizer.mask_token_id:
|
48 |
+
tokens = tokenizer.convert_ids_to_tokens(tf.math.top_k(tf_outputs[i], k=5)[1])
|
49 |
+
tf_outputs_sentence += "[" + "||".join(tokens) + "]"
|
50 |
+
else:
|
51 |
+
tf_outputs_sentence += "".join(
|
52 |
+
tokenizer.convert_ids_to_tokens([id], skip_special_tokens=True)
|
53 |
+
)
|
54 |
+
print(tf_outputs_sentence)
|
55 |
+
# small
|
56 |
+
# pytorch: 今天[的||,||是||很||也]很好,我[要||会||是||想||在]去公园玩。
|
57 |
+
# tf: 今天[的||,||是||很||也]很好,我[要||会||是||想||在]去公园玩。
|
58 |
+
# base
|
59 |
+
# pytorch: 今天[我||天||晴||园||玩]很好,我[想||要||会||就||带]去公园玩。
|
60 |
+
# tf: 今天[我||天||晴||园||玩]很好,我[想||要||会||就||带]去公园玩。
|
61 |
+
# large
|
62 |
+
# pytorch: 今天[天||气||我||空||阳]很好,我[又||想||会||就||爱]去公园玩。
|
63 |
+
# tf: 今天[天||气||我||空||阳]很好,我[又||想||会||就||爱]去公园玩。
|
64 |
+
```
|
65 |
+
|
66 |
+
```
|
67 |
+
## 引用
|
68 |
+
Bibtex:
|
69 |
+
```tex
|
70 |
+
@misc{su2021roformer,
|
71 |
+
title={RoFormer: Enhanced Transformer with Rotary Position Embedding},
|
72 |
+
author={Jianlin Su and Yu Lu and Shengfeng Pan and Bo Wen and Yunfeng Liu},
|
73 |
+
year={2021},
|
74 |
+
eprint={2104.09864},
|
75 |
+
archivePrefix={arXiv},
|
76 |
+
primaryClass={cs.CL}
|
77 |
+
}
|
78 |
+
```
|