Datasets:
Update README.md
Browse files
README.md
CHANGED
@@ -26,7 +26,34 @@ dataset_info:
|
|
26 |
num_examples: 9400
|
27 |
download_size: 104413879
|
28 |
dataset_size: 188946124.0
|
|
|
|
|
|
|
|
|
|
|
|
|
29 |
---
|
30 |
-
# Dataset Card for "multilingual-zero-shot-label-nli"
|
31 |
|
32 |
-
[
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
26 |
num_examples: 9400
|
27 |
download_size: 104413879
|
28 |
dataset_size: 188946124.0
|
29 |
+
license: other
|
30 |
+
task_categories:
|
31 |
+
- zero-shot-classification
|
32 |
+
- text-classification
|
33 |
+
task_ids:
|
34 |
+
- natural-language-inference
|
35 |
---
|
|
|
36 |
|
37 |
+
[mtasksource](https://github.com/sileod/tasksource) classification tasks recasted as natural language inference.
|
38 |
+
This dataset is intended to improve label understanding in [zero-shot classification HF pipelines](https://huggingface.co/docs/transformers/main/main_classes/pipelines#transformers.ZeroShotClassificationPipeline
|
39 |
+
).
|
40 |
+
|
41 |
+
Inputs that are text pairs are separated by a newline (\n).
|
42 |
+
```python
|
43 |
+
from transformers import pipeline
|
44 |
+
classifier = pipeline(model="sileod/mdeberta-v3-base-tasksource-nli")
|
45 |
+
classifier(
|
46 |
+
"I have a problem with my iphone that needs to be resolved asap!!",
|
47 |
+
candidate_labels=["urgent", "not urgent", "phone", "tablet", "computer"],
|
48 |
+
)
|
49 |
+
```
|
50 |
+
[mdeberta-v3-base-tasksource-nli](https://huggingface.co/sileod/mdeberta-v3-base-tasksource-nli) will include `label-nli` in its training mix (a relatively small portion, to keep the model general, but note that nli models work for label-like zero shot classification without specific supervision (https://aclanthology.org/D19-1404.pdf).
|
51 |
+
|
52 |
+
|
53 |
+
```
|
54 |
+
@article{sileo2023tasksource,
|
55 |
+
title={tasksource: A Dataset Harmonization Framework for Streamlined NLP Multi-Task Learning and Evaluation},
|
56 |
+
author={Sileo, Damien},
|
57 |
+
year={2023}
|
58 |
+
}
|
59 |
+
```
|