Commit
•
ef7c556
1
Parent(s):
49c9969
Update README.md
Browse files
README.md
CHANGED
|
@@ -15,7 +15,8 @@ language:
|
|
| 15 |
- tr
|
| 16 |
- ur
|
| 17 |
- vu
|
| 18 |
-
- zh
|
|
|
|
| 19 |
tags:
|
| 20 |
- zero-shot-classification
|
| 21 |
- text-classification
|
|
@@ -42,6 +43,7 @@ As of December 2021, mDeBERTa-base is the best performing multilingual base-size
|
|
| 42 |
```python
|
| 43 |
from transformers import AutoTokenizer, AutoModelForSequenceClassification
|
| 44 |
import torch
|
|
|
|
| 45 |
|
| 46 |
model_name = "MoritzLaurer/mDeBERTa-v3-base-mnli-xnli"
|
| 47 |
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
|
@@ -53,10 +55,8 @@ hypothesis = "Emmanuel Macron is the President of France"
|
|
| 53 |
input = tokenizer(premise, hypothesis, truncation=True, return_tensors="pt")
|
| 54 |
output = model(input["input_ids"].to(device)) # device = "cuda:0" or "cpu"
|
| 55 |
prediction = torch.softmax(output["logits"][0], -1).tolist()
|
| 56 |
-
|
| 57 |
label_names = ["entailment", "neutral", "contradiction"]
|
| 58 |
prediction = {name: round(float(pred) * 100, 1) for pred, name in zip(prediction, label_names)}
|
| 59 |
-
|
| 60 |
print(prediction)
|
| 61 |
```
|
| 62 |
|
|
@@ -87,7 +87,7 @@ average | ar | bg | de | el | en | es | fr | hi | ru | sw | th | tr | ur | vu |
|
|
| 87 |
## Limitations and bias
|
| 88 |
Please consult the original DeBERTa-V3 paper and literature on different NLI datasets for potential biases.
|
| 89 |
|
| 90 |
-
##
|
| 91 |
If you use this model, please cite: Laurer, Moritz, Wouter van Atteveldt, Andreu Salleras Casas, and Kasper Welbers. 2022. ‘Less Annotating, More Classifying – Addressing the Data Scarcity Issue of Supervised Machine Learning with Deep Transfer Learning and BERT - NLI’. Preprint, June. Open Science Framework. https://osf.io/74b8k.
|
| 92 |
|
| 93 |
## Ideas for cooperation or questions?
|
|
|
|
| 15 |
- tr
|
| 16 |
- ur
|
| 17 |
- vu
|
| 18 |
+
- zh
|
| 19 |
+
license: mit
|
| 20 |
tags:
|
| 21 |
- zero-shot-classification
|
| 22 |
- text-classification
|
|
|
|
| 43 |
```python
|
| 44 |
from transformers import AutoTokenizer, AutoModelForSequenceClassification
|
| 45 |
import torch
|
| 46 |
+
device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")
|
| 47 |
|
| 48 |
model_name = "MoritzLaurer/mDeBERTa-v3-base-mnli-xnli"
|
| 49 |
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
|
|
|
| 55 |
input = tokenizer(premise, hypothesis, truncation=True, return_tensors="pt")
|
| 56 |
output = model(input["input_ids"].to(device)) # device = "cuda:0" or "cpu"
|
| 57 |
prediction = torch.softmax(output["logits"][0], -1).tolist()
|
|
|
|
| 58 |
label_names = ["entailment", "neutral", "contradiction"]
|
| 59 |
prediction = {name: round(float(pred) * 100, 1) for pred, name in zip(prediction, label_names)}
|
|
|
|
| 60 |
print(prediction)
|
| 61 |
```
|
| 62 |
|
|
|
|
| 87 |
## Limitations and bias
|
| 88 |
Please consult the original DeBERTa-V3 paper and literature on different NLI datasets for potential biases.
|
| 89 |
|
| 90 |
+
## Citation
|
| 91 |
If you use this model, please cite: Laurer, Moritz, Wouter van Atteveldt, Andreu Salleras Casas, and Kasper Welbers. 2022. ‘Less Annotating, More Classifying – Addressing the Data Scarcity Issue of Supervised Machine Learning with Deep Transfer Learning and BERT - NLI’. Preprint, June. Open Science Framework. https://osf.io/74b8k.
|
| 92 |
|
| 93 |
## Ideas for cooperation or questions?
|