Create README.md
Browse files
README.md
CHANGED
@@ -1,27 +1,25 @@
|
|
1 |
---
|
2 |
-
|
3 |
-
|
4 |
-
|
5 |
-
|
6 |
-
|
7 |
-
|
8 |
-
|
9 |
-
|
10 |
-
- name: task
|
11 |
-
dtype: string
|
12 |
-
splits:
|
13 |
-
- name: train
|
14 |
-
num_bytes: 1027206414.0
|
15 |
-
num_examples: 2095288
|
16 |
-
- name: validation
|
17 |
-
num_bytes: 47771886.0
|
18 |
-
num_examples: 63296
|
19 |
-
- name: test
|
20 |
-
num_bytes: 48695495.0
|
21 |
-
num_examples: 65919
|
22 |
-
download_size: 562394652
|
23 |
-
dataset_size: 1123673795.0
|
24 |
---
|
25 |
-
# Dataset Card for "zero-shot-label-nli"
|
26 |
|
27 |
-
[
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
+
license: apache-2.0
|
3 |
+
task_categories:
|
4 |
+
- zero-shot-classification
|
5 |
+
- text-classification
|
6 |
+
task_ids:
|
7 |
+
- natural-language-inference
|
8 |
+
language:
|
9 |
+
- en
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
10 |
---
|
|
|
11 |
|
12 |
+
[tasksource](https://github.com/sileod/tasksource) classification tasks recasted as natural language inference.
|
13 |
+
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
|
14 |
+
).
|
15 |
+
|
16 |
+
Inputs that are text pairs are separated by a newline (\n).
|
17 |
+
```python
|
18 |
+
from transformers import pipeline
|
19 |
+
classifier = pipeline(model="sileod/deberta-v3-base-tasksource-nli")
|
20 |
+
classifier(
|
21 |
+
"I have a problem with my iphone that needs to be resolved asap!!",
|
22 |
+
candidate_labels=["urgent", "not urgent", "phone", "tablet", "computer"],
|
23 |
+
)
|
24 |
+
```
|
25 |
+
`sileod/deberta-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).
|