Update README.md
Browse files
README.md
CHANGED
@@ -1,4 +1,16 @@
|
|
1 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
dataset_info:
|
3 |
- config_name: triplet
|
4 |
features:
|
@@ -68,3 +80,61 @@ configs:
|
|
68 |
- split: train
|
69 |
path: triplet-all/train-*
|
70 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
+
language:
|
3 |
+
- en
|
4 |
+
multilinguality:
|
5 |
+
- monolingual
|
6 |
+
size_categories:
|
7 |
+
- 1M<n<10M
|
8 |
+
task_categories:
|
9 |
+
- feature-extraction
|
10 |
+
- sentence-similarity
|
11 |
+
pretty_name: NLI for SimCSE
|
12 |
+
tags:
|
13 |
+
- sentence-transformers
|
14 |
dataset_info:
|
15 |
- config_name: triplet
|
16 |
features:
|
|
|
80 |
- split: train
|
81 |
path: triplet-all/train-*
|
82 |
---
|
83 |
+
|
84 |
+
# Dataset Card for NLI for SimCSE
|
85 |
+
|
86 |
+
This is a reformatting of the NLI for SimCSE Dataset used to train the [BGE-M3 model](https://huggingface.co/BAAI/bge-m3). See the full BGE-M3 dataset in [Shitao/bge-m3-data](https://huggingface.co/datasets/Shitao/bge-m3-data).
|
87 |
+
Despite being labeled as Natural Language Inference (NLI), this dataset can be used for training/finetuning an embedding model for semantic textual similarity.
|
88 |
+
|
89 |
+
## Dataset Subsets
|
90 |
+
|
91 |
+
### `triplet` subset
|
92 |
+
|
93 |
+
* Columns: "anchor", "positive", "negative"
|
94 |
+
* Column types: `str`, `str`, `str`
|
95 |
+
* Examples:
|
96 |
+
```python
|
97 |
+
{
|
98 |
+
'anchor': 'One of our number will carry out your instructions minutely.',
|
99 |
+
'positive': 'A member of my team will execute your orders with immense precision.',
|
100 |
+
'negative': 'We have no one free at the moment so you have to take action yourself.'
|
101 |
+
}
|
102 |
+
```
|
103 |
+
* Collection strategy: Reading the jsonl file in the `en_NLI_data` directory in [Shitao/bge-m3-data](https://huggingface.co/datasets/Shitao/bge-m3-data) and taking only the first negative.
|
104 |
+
* Deduplified: No
|
105 |
+
|
106 |
+
### `triplet-7` subset
|
107 |
+
|
108 |
+
* Columns: "anchor", "positive", "negative_1", "negative_2", "negative_3", "negative_4", "negative_5", "negative_6", "negative_7"
|
109 |
+
* Column types: `str`, `str`, `str`, `str`, `str`, `str`, `str`
|
110 |
+
* Examples:
|
111 |
+
```python
|
112 |
+
{
|
113 |
+
'anchor': 'One of our number will carry out your instructions minutely.',
|
114 |
+
'positive': 'A member of my team will execute your orders with immense precision.',
|
115 |
+
'negative_1': 'We have no one free at the moment so you have to take action yourself.',
|
116 |
+
'negative_2': 'A poodle is running through the grass.',
|
117 |
+
'negative_3': 'Investment and planning are growing industries in Jamaica.',
|
118 |
+
'negative_4': 'A bearded man is rocking out on an acoustic guitar',
|
119 |
+
'negative_5': 'The people are sunbathing on the beach.',
|
120 |
+
'negative_6': 'A construction worker installs a door.',
|
121 |
+
'negative_7': 'A crowd has gathered because of a dangerous situation.'
|
122 |
+
}
|
123 |
+
```
|
124 |
+
* Collection strategy: Reading the jsonl file in the `en_NLI_data` directory in [Shitao/bge-m3-data](https://huggingface.co/datasets/Shitao/bge-m3-data) and taking all samples that have 7 negatives (which is by far the majority).
|
125 |
+
* Deduplified: No
|
126 |
+
|
127 |
+
### `triplet-all` subset
|
128 |
+
|
129 |
+
* Columns: "anchor", "positive", "negative"
|
130 |
+
* Column types: `str`, `str`, `str`
|
131 |
+
* Examples:
|
132 |
+
```python
|
133 |
+
{
|
134 |
+
'anchor': 'One of our number will carry out your instructions minutely.',
|
135 |
+
'positive': 'A member of my team will execute your orders with immense precision.',
|
136 |
+
'negative': 'We have no one free at the moment so you have to take action yourself.'
|
137 |
+
}
|
138 |
+
```
|
139 |
+
* Collection strategy: Reading the jsonl file in the `en_NLI_data` directory in [Shitao/bge-m3-data](https://huggingface.co/datasets/Shitao/bge-m3-data) and taking only each negative, but making a separate sample with each of the negatives.
|
140 |
+
* Deduplified: No
|