Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,55 @@
|
|
1 |
---
|
2 |
license: apache-2.0
|
|
|
|
|
|
|
|
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
license: apache-2.0
|
3 |
+
datasets:
|
4 |
+
- zaemyung/IteraTeR_plus
|
5 |
+
language:
|
6 |
+
- en
|
7 |
---
|
8 |
+
# DElIteraTeR-RoBERTa-Intent-Span-Detector
|
9 |
+
This model was obtained by fine-tuning [roberta-large](https://huggingface.co/roberta-large) on [IteraTeR+](https://huggingface.co/datasets/zaemyung/IteraTeR_plus) `multi_sent` dataset.
|
10 |
+
|
11 |
+
Paper: [Improving Iterative Text Revision by Learning Where to Edit from Other Revision Tasks](https://aclanthology.org/2022.emnlp-main.678/) <br>
|
12 |
+
Authors: Zae Myung Kim, Wanyu Du, Vipul Raheja, Dhruv Kumar, and Dongyeop Kang
|
13 |
+
|
14 |
+
## Usage
|
15 |
+
```python
|
16 |
+
import torch
|
17 |
+
from transformers import AutoTokenizer, AutoModelForTokenClassification
|
18 |
+
|
19 |
+
tokenizer = AutoTokenizer.from_pretrained("zaemyung/DElIteraTeR-RoBERTa-Intent-Span-Detector")
|
20 |
+
|
21 |
+
# update tokenizer with special tokens
|
22 |
+
INTENT_CLASSES = ['none', 'clarity', 'fluency', 'coherence', 'style', 'meaning-changed'] # `meaning-changed` is not used
|
23 |
+
INTENT_OPENED_TAGS = [f'<{intent_class}>' for intent_class in INTENT_CLASSES]
|
24 |
+
INTENT_CLOSED_TAGS = [f'</{intent_class}>' for intent_class in INTENT_CLASSES]
|
25 |
+
INTENT_TAGS = set(INTENT_OPENED_TAGS + INTENT_CLOSED_TAGS)
|
26 |
+
special_tokens_dict = {'additional_special_tokens': ['<bos>', '<eos>'] + list(INTENT_TAGS)}
|
27 |
+
tokenizer.add_special_tokens(special_tokens_dict)
|
28 |
+
|
29 |
+
model = AutoModelForTokenClassification.from_pretrained("zaemyung/DElIteraTeR-RoBERTa-Intent-Span-Detector")
|
30 |
+
|
31 |
+
id2label = {0: "none", 1: "clarity", 2: "fluency", 3: "coherence", 4: "style", 5: "meaning-changed"}
|
32 |
+
|
33 |
+
before_text = '<bos>I likes coffee?<eos>'
|
34 |
+
model_input = tokenizer(before_text, return_tensors='pt')
|
35 |
+
model_output = model(**model_input)
|
36 |
+
softmax_scores = torch.softmax(model_output.logits, dim=-1)
|
37 |
+
pred_ids = torch.argmax(softmax_scores, axis=-1)[0].tolist()
|
38 |
+
pred_intents = [id2label[_id] for _id in pred_ids]
|
39 |
+
|
40 |
+
tokens = tokenizer.convert_ids_to_tokens(model_input['input_ids'][0])
|
41 |
+
|
42 |
+
for token, pred_intent in zip(tokens, pred_intents):
|
43 |
+
print(f"{token}: {pred_intent}")
|
44 |
+
|
45 |
+
"""
|
46 |
+
<s>: none
|
47 |
+
<bos>: none
|
48 |
+
I: fluency
|
49 |
+
Ġlikes: fluency
|
50 |
+
Ġcoffee: none
|
51 |
+
?: none
|
52 |
+
<eos>: none
|
53 |
+
</s>: none
|
54 |
+
"""
|
55 |
+
```
|